# F -E A C

#### Document technical information

Format pdf
Size 1.5 MB
First found Jun 9, 2017

#### Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

#### Transcript

Forty-Eighth
Asilomar Conference on
Signals, Systems, and Computers
November 2–5, 2014
FINAL PROGRAM & ABSTRACTS
Asilomar Hotel
Conference Grounds
FORTY-EIGHTH
ASILOMAR CONFERENCE ON
SIGNALS, SYSTEMS & COMPUTERS
IEEE Signal Processing Society
CONFERENCE COMMITTEE
General Chair
Prof. Roger Woods
School of Electronics, Electrical
Engineering and Computer Science
Queen’s University Belfast
Belfast, BT3 9DT, UK
Publicity Chair
Linda S. DeBrunner
Department of Electrical &
Computer Engineering
Florida State University
Tallahassee, FL 32310-6046
E-mail:
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
Technical Program Chair
Prof. Geert Leus
Delft University of Technology
Fac. of Electrical Engineering, Mathematics
and Computer Science
Mekelweg 4
2628CD Delft, The Netherlands
Finance Chair
Ric Romero
Department of Electrical &
Computer Engineering
Monterey, CA 93943-5121
E-mail: [email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
Conference Coordinator
Monique P. Fargues
Department of Electrical &
Computer Engineering
Monterey, CA 93943
E-mail: [email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
Electronic Media Chair
Prof. Marios S. Pattichis
Department of Electrical & Computer
Engineering
MSC01 1100
1 University of New Mexico
ECE Bldg., Room 125
Albuquerque, NM 87131-0001
Publication Chair
Michael Matthews
ATK Space Systems
10 Ragsdale Drive, Suite 201
Monterey, CA 93940
E-mail:
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
Student Paper Contest Chair
Prof. Joseph R. Cavallaro
Rice University
Dept. of Electrical and Computer
Engineering
6100 Main Street, MS 380
Houston, TX 77005
1
Welcome from the General Chair
Prof. Roger Woods, Queen’s University Belfast, UK
Welcome to the 48th Asilomar Conference on Signals, Systems, and Computers! I have
had a long involvement with the Conference since my first publication in 1997 when I was
immediately struck by the unique nature of the Asilomar conference environment. The
picturesque sand dunes and warm sunshine provide a wonderful backdrop to a conference that
allows easy access to, and interaction with key researchers. Understandably, over the years, I
have needed little persuasion to attend. There will never be a better opportunity to capture the
attention of a key researcher in your area of expertise than at Asilomar!
The technical program was crafted expertly by the Technical Program Chair, Geert Leus, and his
team of Technical Area Chairs: Shengli Zhou, Zhengdao Wang, Bhaskar Rao, Michael Rabbat,
Zhi Tian, Visa Koivunen, Selin Aviyente, Jorn Janneck, Mohsin Jamali, and Matt McKay. I
would like to thank Geert and his team for assembling a high quality program with 437 accepted
papers and 163 invited papers. The student paper contest this year has been chaired by Joe
Cavallaro and he has selected a total of 11 submissions. The student finalists will present poster
presentations to the judges on Sunday afternoon and, of course, everyone is welcome to attend.
The awards for the top three papers will be made at the plenary session. A key innovation this
year has been to inculcate two major themes, brain machine interface and neural networks, and
processing of high dimensional large scale data.
This year’s plenary talk will be given by Professor Georgios B. Giannakis, from the University
of Minnesota. I am pleased to have such a high profile speaker with a strong background in
signal processing across a wide range of applications. Georgios will describe signal processing
techniques to handle massive datasets which are noisy, incomplete, vulnerable to cyber-attacks
and have outliers. The growth of Big Data represents a major ongoing challenge for humanity.
The derivation of suitable data processing techniques is a vital activity and I am especially
looking forward to seeing what can be accomplished in this area. Georgios has had a long
engagement with the conference having acted as part of the technical committee as early as 1993
and presented his first paper at Asilomar in 1988.
I am privileged to have served as this year’s General Chair. I hope that you enjoy the 2014
Conference programme whilst taking some time out to encounter the very special environment
and atmosphere that Asilomar has to offer.
Prof. Roger Woods
Queen’s University Belfast, UK, June 2014
2
Conference Steering Committee
DR. RALPH D. HIPPENSTIEL
San Diego, CA 92126
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. MONIQUE P. FARGUES
President & Chair
Electrical & Computer Eng. Dept.
Code EC/Fa
Monterey, CA 93943-5121
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. W. KENNETH JENKINS
Electrical Eng. Dept.
The Pennsylvania State University
209C Electrical Engineering West
University Park, PA 16802-2705
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. SHERIF MICHAEL
Secretary
Electrical & Computer Eng. Dept.
Code EC/Mi
Monterey, CA 93943-5121
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. FRANK KRAGH
Electrical & Computer Eng. Dept.
Code EC/Kr
Monterey, CA 93943-5121
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. RIC ROMERO
Treasurer
Electrical & Computer Eng. Dept.
Code EC/Rr
Monterey, CA 93943-5121
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
DR. MICHAEL B. MATTHEWS
Publications Chair
ATK Space Systems
10 Ragsdale Drive, Suite 201
Monterey, CA 93940
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. SCOTT ACTON
Electrical & Computer Eng. Dept.
University of Virginia
P.O. Box 400743
Charlottesville, VA 22904-4743
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
DR. MARIOS PATTICHIS
Electrical & Computer Eng. Dept.
MSC01 1100
1 University of New Mexico
ECE Bldg., Room: 229A
Albuquerque, NM 87131-000
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. MAITE BRANDT-PEARCE
Electrical & Computer Eng. Dept.
University of Virginia
P.O. Box 400743
Charlottesville, VA 22904
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. JAMES A. RITCEY
Electrical Eng. Dept.
Box 352500
University of Washington
Seattle, Washington 98195
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. LINDA DEBRUNNER
Publicity Chair
Electrical & Computer Eng. Dept.
Florida State University
2525 Pottsdamer Street, Room A-341-A
Tallahassee, FL 32310-6046
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
DR. MICHAEL SCHULTE
AMD
11400 Cherisse Dr.
Austin, TX 78739
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. EARL E. SWARTZLANDER, JR.
Electrical & Computer Eng. Dept.
University of Texas at Austin
Austin, TX 78712
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. VICTOR DEBRUNNER
Electrical & Computer Eng. Dept.
Florida State University
2525 Pottsdamer Street, Room A-341-A
Tallahassee, FL 32310-6046
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. KEITH A. TEAGUE
School Electrical & Computer Engineering / 202ES
Oklahoma State University
Stillwater, OK 74078
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. MILOS ERCEGOVAC
Computer Science Dept.
University of California at Los Angeles
Los Angeles, CA 90095
DR. MILOŠ DOROSLOVAČKI
General Program Chair (ex officio) Electrical and Computer Engineering Dept.
George Washington University
Washington, DC
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. BENJAMIN FRIEDLANDER
Computer Eng. Dept.
University of California
1156 High Street, MS:SOE2
Santa Cruz, CA 95064
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. ROBERT HEATH
General Program Chair (ex officio) Electrical & Computer Eng. Dept.
The University of Texas at Austin
Austin, TX 78712
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
PROF. FREDRIC J. HARRIS
Electrical Eng. Dept.
San Diego State University
San Diego, CA 92182
[email protected]/* <![CDATA[ */!function(t,e,r,n,c,a,p){try{t=document.currentScript||function(){for(t=document.getElementsByTagName('script'),e=t.length;e--;)if(t[e].getAttribute('data-cfhash'))return t[e]}();if(t&&(c=t.previousSibling)){p=t.parentNode;if(a=c.getAttribute('data-cfemail')){for(e='',r='0x'+a.substr(0,2)|0,n=2;a.length-n;n+=2)e+='%'+('0'+('0x'+a.substr(n,2)^r).toString(16)).slice(-2);p.replaceChild(document.createTextNode(decodeURIComponent(e)),c)}p.removeChild(t)}}catch(u){}}()/* ]]> */
3
Year 2012
Year 2013
2014 Asilomar Technical Program Committee
Chairman
Prof. Geert Leus
Delft University of Technology
2014 Asilomar
Technical Program Committee Members
A: Communications Systems
Prof. Shengli Zhou
University of Connecticut
F: Biomedical Signal and Image Processing
Prof. Selin Aviyente
Michigan State University
Prof. Zhengdao Wang
Iowa State University
G: Architecture and Implementation
Prof. Jörn W. Janneck
Lund University
B: MIMO Communications and Signal
Processing
University of California San Diego
H: Speech
Image and Video Processing
Prof. Mohsin M. Jamali
University of Toledo
C: Networks
Prof. Michael Rabbat
McGill University
Vice Chair
Prof. Matthew McKay
Hong Kong University of Science and Technology
D: Signal Processing and Adaptive Systems
Prof. Zhi (Gerry) Tian
Michigan Technological University
Student Paper Contest Chair
Prof. Joseph R. Cavallaro
Rice University
E: Array Signal Processing
Prof. Visa Koivunen
Aalto University
4
2014 Asilomar Conference Session Schedule
Sunday Afternoon, November 2, 2014
3:00–7:00 pm
4:00–6:30 pm
7:00–9:00 pm
Registration — Merrill Hall
Student Paper Contest — Heather
Welcoming Dessert Reception — Merrill Hall
Monday Morning, November 3, 204
7:30–9:00 am
Breakfast – Crocker Dining Hall
8:00 am–6:00 pmRegistration
8:15–9:45 am
MA1a — Conference Welcome and Plenary Session — Chapel
9:45–10:15 am
Coffee Social
10:15–11:55 am MORNING SESSIONS
MA1b
Learning and Optimization for Big Data
MA2b
EEG Based Brain Computer Interface
MA3b
Underwater Wireless Networks
MA4b
Physical Layer Security I
MA5b
Image and Video Processing
MA6b
Sparse Estimation and Learning in Multi-Channel and Array Systems
MA7b
Architectures for Detection and Decoding
MA8b1 Synchronization and Channel Estimation (Poster)
MA8b2 Relaying (Poster)
MA8b3 Active Sensing and Target Recognition (Poster)
MA8b4 Physiological Signal Processing (Poster)
12:00–1:00 pm
Lunch – Crocker Dining Hall
Monday Afternoon, November 3, 2014
1:30–5:10 pm
AFTERNOON SESSIONS
MP1a
Big Data Analytics
MP1b
Tensor-Based Signal Processing
MP2a
Neural Engineering and Signal Processing
MP2b
Brain Connectomics
MP3a
Compressed Sensing I
MP3b
Compressed Sensing II
MP4a
Underwater Acoustic Communications and Networking
MP4b
Massive MIMO I
MP5a
Smart Grid: Learning and Optimization
MP5b
Image and Video Quality
MP6a
Array Calibration
MP6b
Wireless Localization
MP7a
Resource-aware and Domain-specific Computing
MP7b
Detection and Estimation for Networked Data
MP8a1 Network Resource Allocation and Localization (Poster)
MP8a2 Bioinformatics and Medical Imaging (Poster)
MP8a3 Source Separation and Array Processing (Poster)
MP8a4 Digital Communications (Poster)
MP8a5 Image and Speech Processing (Poster)
Monday Evening, November 3, 2014
6:00–9:30 pm
Conference Cocktail/Social — Merrill Hall
The Cocktail/Social takes the place of Monday’s dinner. No charge for conference attendees and a
guest.
5
2014 Asilomar Conference Session Schedule
(continued)
Tuesday Morning, November 4, 2014
7:30–9:00 am
Breakfast — Crocker Dining Hall
8:00 am–5:00 pmRegistration
8:15–11:55 am MORNING SESSIONS
TA1a
High Dimensional and Large Volume Data
TA1b
Big Data Signal Processing
TA2a
Neural Spike Train Analysis
TA2b
Dynamic Brain Functional Connectivity
TA3a
Distributed Optimization over Networks
TA3b
TA4a
Enhanced MIMO for LTE-A and 5G Systems
TA4b
TA5a
TA5b
Historic Photographic Paper Identification via Textural Similarity Assessment
TA6a
TA6b
Statistical Inference in Smart Grids
TA7a
Computer Arithmetic I
TA7b
MIMO Sensing
TA8a1 Channel Estimation and MIMO Feedback (Poster)
TA8a2 Image Processing I (Poster)
TA8a3 Signal Processing for Communications (Poster)
TA8b1 Multiuser and Cellular Systems (Poster)
TA8b2 Computer Arithmetic II (Poster)
TA8b3 Array Processing Methods (Poster)
TA8b4 Compressed Sensing III (Poster)
12:00–1:00 pm
Lunch – Crocker Dining Hall
Tuesday Afternoon, November 4, 2014
1:30–5:35 pm
AFTERNOON SESSIONS
TP1a
Covariance Mining
TP1b
Large-Scale Learning and Optimization
TP2a
Bioinformatics and DNA Computing
TP2b
Echo Cancellation
TP3a
Machine Learning
TP3b
Sparse Signal Recovery
TP4a
Optical Communications
TP4b
Energy Harvesting Wireless Communications
TP5a
Speech Enhancement
TP5b
TP6a
TP6b
Many-Core Platforms
TP7a
Design Methodologies for Signal Processing
TP7b
Optical Wireless Communications
TP8a1
TP8a2
Signal Processing Methods (Poster)
TP8a3
Image Processing II (Poster)
TP8a4
Sensor and Wireless Networks (Poster)
TP8b1
Topics in Communication Systems (Poster)
TP8b2
Relays, Cognitive, Cooperative, and Heterogeneous Networks (Poster)
TP8b3
Signal Processing Architectures (Poster)
TP8b4
Signal Processing Theory and Applications (Poster)
Tuesday Evening
Open Evening — Enjoy the Monterey Peninsula
6
2014 Asilomar Conference Session Schedule
(continued)
Wednesday Morning, November 5, 2014
7:30–9:00 am
8:00 am–12:00 pm
Breakfast — Crocker Dining Hall
Registration — Copyright forms must be turned in before the registration closes at 12:00
noon.
8:15–11:55 am
MORNING SESSIONS
WA1a
MIMO Design for mmWave Systems
WA1b
Massive MIMO II
WA2a
5G and Energy Efficient Cellular Networks
WA2b
Mobile Health
WA3a
Sparse Learning and Estimation
WA3b
WA4a
Physical Layer Security II
WA4b
Coding and Decoding
WA5a
Information Processing for Social and Sensor Networks
WA5b
Document Processing and Synchronization
WA6a
WA6b
Distributed Detection and Optimization
WA7a
Implementation of Wireless Systems
WA7b
Video Coding Architecture and Design
12:00–1:00 pm
Lunch — Meal tickets may be purchased at registration desk. This meal is not included in
the registration.
Student Paper Contest
Heather - Sunday, November 2, 2014, 4:00 - 6:30 pm
Track A
Track B
Track C
Track D
Track E
Track F
Track G
Track H
“Everlasting Secrecy in Disadvantaged Wireless Environments against Sophisticated Eavesdroppers”
Azadeh Sheikholeslami, Dennis Goeckel, Hossein Pishro-nik, UMASS-Amherst, United States
“On Physical Layer Secrecy of Collaborative Compressive Detection”
Bhavya Kailkhura, Thakshila Wimalajeewa, Pramod Varshney, Syracuse University, United States
“Max-Min Fairness in Compact MU-MIMO Systems: Can the Matching Network Play a Role?”
Yahia Hassan, Armin Wittneben, ETH Zurich, Switzerland
“On the Convergence Rate of Swap-collide Algorithm for Simple Task Assignment”
Sam Safavi, Usman A. Khan, Tufts University, United States
“Secrecy Outage Analysis of Cognitive Wireless Sensor Networks”
Satyanarayana Vuppala, Jacobs University Bremen, Germany; Weigang Liu, Tharmalingam
Ratnarajah, University of Edinburgh, United Kingdom; Giuseppe Abreu, Jacobs University Bremen,
Germany
“Subspace Learning from Extremely Compressed Measurements”
Martin Azizyan, Akshay Krishnamurthy, Aarti Singh, Carnegie Mellon University, United States
“Abstract Algebraic-Geometric Subspace Clustering”
Manolis Tsakiris, Rene Vidal, Johns Hopkins University, United States
“Calibrating Nested Sensor Arrays with Model Errors”
Keyong Han, Peng Yang, Arye Nehorai, Washington University in St. Louis, United States
“Whitening 1/f-type Noise in Electroencephalogram Signals for Steady-State Visual Evoked Potential
Brain-Computer Interfaces”
Alan Paris, Azadeh Vosoughi, George Atia, University of Central Florida, United States
“Hybrid Floating-Point Modules with Low Area Overhead on a Fine-Grained Processing Core”
Jon Pimentel, Bevan Baas, University of California, Davis, United States
“Crowdsourced Study of Subjective Image Quality”
Deepti Ghadiyaram, Alan Bovik, University of Texas at Austin, United States
7
2014 Asilomar Conference Session Schedule
Coffee breaks will be at 9:55 am and 3:10 pm. (except Monday morning when refreshments will
be served outside Chapel from 9:45–10:15 am)
Monday, November 3, 2014
CONFERENCE OPENING AND PLENARY
SESSION 8:15 – 9:45 am, LOCATED IN CHAPEL
1. Welcome from the General Chairperson:
Prof. Roger Woods
Queen’s University of Belfast
2. Session MA1a Distinguished Lecture for the 2014 Asilomar Conference
Learning Tools for Big Data Analytics
Prof. Georgios B. Giannakis
University of Minnesota, USA
Abstract
We live in an era of data deluge. Pervasive sensors collect massive amounts of information on every bit
of our lives, churning out enormous streams of raw data in various formats. Mining information from
unprecedented volumes of data promises to limit the spread of epidemics and diseases, identify trends
in financial markets, learn the dynamics of emergent social-computational systems, and also protect
critical infrastructure including the smart grid and the Internet’s backbone network. While Big Data can be
definitely perceived as a big blessing, big challenges also arise with large-scale datasets. The sheer volume
of data makes it often impossible to run analytics using a central processor and storage, and distributed
processing with parallelized multi-processors is preferred while the data themselves are stored in the cloud.
As many sources continuously generate data in real time, analytics must often be performed “on-the-fly”
and without an opportunity to revisit past entries. Due to their disparate origins, massive datasets are noisy,
incomplete, prone to outliers, and vulnerable to cyber-attacks. These effects are amplified if the acquisition
and transportation cost per datum is driven to a minimum. Overall, Big Data present challenges in which
resources such as time, space, and energy, are intertwined in complex ways with data resources. Given these
challenges, ample signal processing opportunities arise. This keynote lecture outlines ongoing research in
novel models applicable to a wide range of Big Data analytics problems, as well as algorithms to handle
the practical challenges, while revealing fundamental limits and insights on the mathematical trade-offs
involved.
Biography
Georgios B. Giannakis received his Diploma in Electrical Engineering from the National Technical
University of Athens, Greece, 1981. From 1982 to 1986 he was with the University of Southern California,
where he received his MSc. in Electrical Engineering (1983), MSc. in Mathematics (1986), and Ph.D. in
Electrical Engineering (1986). He became a Fellow of the IEEE in 1997. Since 1999, he has been a Professor
with the University of Minnesota where he now holds an ADC Chair in Wireless Telecommunications in
the ECE Department, and serves as director of the Digital Technology Center. His general interests span the
areas of communications, networking and statistical signal processing – subjects on which he has published
more than 370 journal papers, 630 conference papers, 20 book chapters, two edited books and two research
monographs (h-index 108). Current research focuses on sparsity and big data analytics, wireless cognitive
radios, mobile ad hoc networks, renewable energy, power grid, gene-regulatory, and social networks. He
is the (co-) inventor of 22 patents issued, and the (co-) recipient of 8 best paper awards from the IEEE
Signal Processing (SP) and Communications Societies, including the G. Marconi Prize Paper Award in
Wireless Communications. He also received Technical Achievement Awards from the SP Society (2000),
from EURASIP (2005), a Young Faculty Teaching Award, and the G. W. Taylor Award for Distinguished
Research from the University of Minnesota. He is a Fellow of EURASIP, and has served the IEEE in a
number of posts, including that of a Distinguished Lecturer for the IEEE-SP Society.
8
Program of 2014
Asilomar Conference
on
Signals, Systems, and Computers
Technical Program Chairman
Prof. Geert Leus
Delft University of Technology
9
Track D – Signal Processing and Adaptive Systems
Session: MAb1 – Learning and Optimization for Big Data
Co-Chairs: Konstantinos Slavakis, University of Minnesota and Nicholas D. Sidiropoulos,
University of Minnesota
MA1b-1
FLEXA: A Fast Parallel Algorithm for Big-Data Optimization
10:15 AM
Francisco Facchinei, Simone Sagratella, University of Rome, Italy; Gesualdo Scutari, University of Buffalo, the
State University of New York, United States
We propose a decomposition framework for the parallel optimization of the sum of a differentiable function and a (block)
separable nonsmooth, convex one. The latter term is usually employed to enforce structure in the solution, typically sparsity.
Our framework is very flexible and includes both fully parallel Jacobi schemes and Gauss-Seidel (i.e., sequential) ones, as well
as virtually all possibilities “in between” with only a subset of variables updated at each iteration. Our theoretical convergence
results improve on existing ones, and numerical results on LASSO and logistic regression problems show that the new method
consistently outperforms existing algorithms.
MA1b-2
Fast and Robust Bootstrap in Analysing Large Multivariate Datasets
10:40 AM
Shahab Basiri, Esa Ollila, Visa Koivunen, Aalto University, Finland
In this paper we investigate applications of fast and robust boostrapping (Salibian-Barrera and Zamar, 2002) in analysing large
volume and high-dimensional data sets. Smaller datasets are resampled from the original data set with replacement. The full
data set may be comprised of subsets stored in many locations because of its large volume. Conventional bootstrap estimators
are particularly sensitive to outliers. Hence, statistically robust approximate bootstrap estimates and confidence intervals are
computed by solving fixed-point estimating equations for multivariate data. The considered estimators have bounded loss
functions and are quantitatively robust having a high breakdown point. Confidence intervals may be used for identifying
sparseness present in high-dimensional signals. Different resampling strategies are considered obtaining the bootstrap estimates.
Different strategies for combining the estimates from resample realizations are considered. Statistical properties of the estimators
are established and their computational complexity is studied as well.
MA1b-3
11:05 AM
Clustering High-Dimensional Dynamical Systems on Low-Rank Matrix Manifolds
Konstantinos Slavakis, X. Wang, G. Lerman, University of Minnesota, United States
Based on the low-rank representation ability of autoregressive moving average (ARMA) models, this paper introduces a novel
algorithm for clustering ARMA modeled high-dimensional dynamical systems into submanifolds placed on low-rank matrix
manifolds. Sparse coding and the tangent spaces of the underlying manifold are utilized to reveal the low-dimensional structure
of the observed data. Such structure is efficiently employed by spectral clustering to segment data into clusters which are even
allowed to intersect. Extensive validation on real data demonstrates the superior performance of the proposed method over stateof-the-art techniques on important action identification applications.
MA1b-4
11:30 AM
Adaptive Estimation from Big Data via Censored Stochastic Approximation
Dimitrios Berberidis, University of Minnesota, Twin Cities, United States; Gang Wang, Beijing Institute of
Technology, China; Georgios Giannakis, Vassilis Kekatos, University of Minnesota, Twin Cities, United States
The era of ‘’Big Data’’ is undoubtedly upon us with 2.5 quintillion bytes of data generated every day. Nonetheless, a significant
percentage of the data accrued can be ‘’thrown away’’ or ‘’reduced’’ while maintaining a certain quality of statistical inference.
By capitalizing on data redundancy, interval censoring is leveraged here to cope with the scarcity of resources needed for
exchanging, storing, and processing Big Data. Using censored data, a novel online maximum likelihood algorithm is developed
that is shown to be convergent in the mean and mean-square error sense. Simulated tests corroborate its efficacy relative to
competing alternatives.
10
Track F – Biomedical Signal and Image Processing
Session: MAb2 – EEG Based Brain Computer Interface
Chair: Murat Akcakaya, Northeastern University
MA2b-1
Decoding the Focus of Auditory Attention from Single-Trial EEG Signals
10:15 AM
Lenny Varghese, Inyong Choi, Siddharth Rajaram, Courtney Pacheco, Barbara Shinn-Cunningham, Boston
University, United States
Auditory stimuli produce attention-modulated responses detectable from electroencephalography (EEG) signals. The focus of
attention to individual elements within a sound mixture can be determined from these signals if sound “streams” are temporally
decorrelated from one another. We discuss the physiological origins of these brain signals, ongoing work in towards decoding
these signals on a “single trial” basis, and how such results compare to decoding EEG signals using visual stimuli.
MA2b-2
10:40 AM
Auditory Considerations for a Motor Imagery Brain-Computer Interface for Speech
Synthesizer Control
Jonathan Brumberg, Jeremy Burnison, University of Kansas, United States
We report on a sensorimotor rhythm (SMR) brain-computer interface (BCI) for controling a speech synthesizer with
instantaneous auditory output. Subjects first listen to three acoustically presented vowel stimuli while imagining three different
limb movements. Subjects then repeat the sounds using the BCI- synthesizer under motor imagery control. Using an adaptive
filter technique, the decoder predicts and synthesizes vowel features from the SMR for auditory output. Here we focus on the
implications of auditory perception on control of the BCI, specifically, whether users rely on BCI output that is perceptually
equivalent to the stimulus rather than an exact replication.
MA2b-3
Single-Trial Identification of Failed Memory Retrieval
11:05 AM
Eunho Noh, University of California, San Diego, United States; Matthew Mollison, Tim Curran, University of
Colorado Boulder, United States; Virginia de Sa, University of California, San Diego, United States
We show that it is possible to distinguish unsuccessfully retrieved from successfully retrieved studied items based on single-trial
scalp EEG activity recorded during the test phase of 3 separate recognition memory experiments. The likelihood of remembering
a study item for trials with the 10% highest and lowest classifier outputs were 0.80 and 0.45 respectively. This suggests that the
classifier outputs are reflecting the level of retrieval during the test phase. These findings combined with previous single-trial
results predicting successful memory encoding from EEG recorded during the study phase will provide a basis for a passive BCI
system for improving memory.
MA2b-4
Utilization of Temporal Trial Dependency in ERP based BCIs
11:30 AM
Umut Orhan, CorTech, LLC, United States; Delia Fernandez-Canellas, Universitat Politècnica de Catalunya, Spain;
Murat Akcakaya, Dana H. Brooks, Deniz Erdogmus, Northeastern University, United States
In most event related potential (ERP) based brain computer interfaces (BCI) that utilizes electroencephalography (EEG), features
corresponding to the stimuli are extracted after the application of a window covering the expected duration of the response.
Especially for the paradigms with shorter inter stimulus interval compared to the expected duration of the response, such an
approach not only causes dependencies of the consecutive trials to be lost but also might decrease the efficiency of the utilization
of the temporal information in the signal. Alternative to the classical approach, we propose a graphical model that considers the
dependency between consecutive trials to make more informed decisions based on the characteristics of the signal. Additionally,
we propose modeling and identification of the system from visual stimuli to EEG to potentially increase the efficiency of
utilization of the temporal information.
11
Track C – Networks
Session: MAb3 – Underwater Wireless Networks
Chair: Milica Stojanovic, Northeastern University
MA3b-1
On the Feasibility of Fully Wireless Remote Control for ROVs
10:15 AM
Federico Favaro, Filippo Campagnaro, Paolo Casari, Michele Zorzi, University of Padova, Italy
In this paper, we explore the possibility of controlling a Remotely Operated Vehicle (ROV) via a fully wireless control channel.
As a first step, we review the expected bit rate offered by optical, acoustic as well as radio-frequency underwater communication
technologies, as a function of the distance between the transmitter and the receiver. We then discuss the Quality-of-Service (QoS)
requirements of services offered by a typical ROV and discuss which can be supported by a given technology at a given distance.
Finally we simulate the performance of the system during missions of interest, and conclude by discussing the effectiveness of
wireless control methods for ROVs.
MA3b-2
10:40 AM
Modeling Realistic Underwater Acoustic Networks using Experimental Data
Mandar Chitre, Gabriel Chua, National University of Singapore, Singapore
Since underwater network experiments are logistically challenging and expensive to conduct, many researchers evaluate the
performance of their protocols through simulation. The validity of simulation results strongly depends on the accuracy of the
channel model used. We use data from underwater network experiments to model the spatiotemporal variability of network
performance. This approach allows researchers to test protocols in realistic simulation environments driven by representative
experimental datasets, long after the experiments are conducted.
MA3b-3
11:05 AM
Scalable Collision-Tolerant Localization in Underwater Acoustic Sensor Networks
Hamid Ramezani, Geert Leus, Technical University of Delft, Netherlands; Milica Stojanovic, Northeastern
University, United States
In this paper, we consider the joint problem of packet scheduling and localization in a multi-hop underwater acoustic sensor
network where the anchors and sensor nodes are distributed in an operating area at random. Briefly, the anchors broadcast their
packets with a known probability density function, e.g., Poisson distribution. Based on its communication range, each sensor
node collects the successfully received packets, and uses them for self-localization. Taking into account the interference power,
collision probability, and noise power, the proposed scheme adjusts the anchors’ packet transmission rate in such a way that
each sensor node in the network can localize itself with a predefined probability. Here, we will focus on the simplicity of the
implementation, localization time, and probability of successful self-localization.
MA3b-4
New Frontiers in Underwater Acoustic Communications
11:30 AM
Andrew Singer, Thomas Riedl, University of Illinois at Urbana Champaign, United States
This talk will discuss one of the most challenging digital communications channels on the planet, the underwater acoustic
channel. Through aggressive use of signal processing and forward error correction, a number of highly successful methods have
been developed in our research group for achieving unprecedented data rates through the underwater acoustic medium. When
an underwater acoustic modem is installed on a mobile platform such as an underwater vehicle, a buoy, or a surface vessel,
Doppler effects distort the acoustic signal significantly. The acoustic path between a surface vessel and an underwater vehicle, for
example, can experience Mach numbers of one percent and more which can be catastrophic if not compensated dynamically. We
derive a sample-by-sample, recursive resampling technique, in which time-varying Doppler is explicitly modeled, tracked and
compensated. Integrated into an iterative turbo equalization-based receiver, this novel Doppler compensation technique achieves
unprecedented communication performance in field tests and simulations. Some of our field data stems from the MACE10
experiment conducted in the shallow waters 100 km south of Martha’s Vineyard, MA. Under challenging conditions (harsh
multi-path, ranges up to 7.2 km, SNRs down to 2 dB and relative speeds up to 3 knots) we obtained a data rate of 40 kbits/s
using 10 kHz of bandwidth. Additional experimental results will be discussed from at-sea tests as well as tests in our acoustic
communications tanks on campus. This talk will also cover the essential elements of joint equalization and decoding, or so-called
turbo equalization and the essential role that it plays in making these systems robust and effective.
12
Track A – Communications Systems
Session: MAb4 – Physical Layer Security I
Chair: Pramod Varshney, Syracuse University
MA4b-1
On Physical Layer Secrecy of Collaborative Compressive Detection
10:15 AM
Bhavya Kailkhura, Thakshila Wimalajeewa, Pramod Varshney, Syracuse University, United States
This paper considers the problem of collaborative compressive detection under a physical layer secrecy constraint. More
specifically, we consider the problem where the network operates in the presence of an eavesdropper who wants to discover the
state of the nature being monitored by the system. It is shown that, the security performance of the system can be improved by
using cooperating trustworthy nodes that assist the Fusion Center (FC) by providing falsified data to the eavesdroppers. We also
consider the problem of determining optimal system parameters which maximize the detection performance at the FC, while
ensuring perfect secrecy at the eavesdropper.
MA4b-2
Converse Results for Secrecy Generation over Channels
10:40 AM
Himanshu Tyagi, University of California, San Diego, United States; Shun Watanabe, University of Tokushima,
Japan
We revisit the problem of secret key agreement in channel models, where in addition to a noisy, albeit secure channel, the
terminals have access to a noiseless public communication channel. We show a strong converse for the secret key capacity in
the point-to-point model and give upper bounds for the general case. Underlying our proofs is a recently discovered single-shot
converse for secret key rates in multiterminal source models.
MA4b-3
Robust Transmission over Wiretap Channels with Secret Keys
11:05 AM
Rafael F. Schaefer, H. Vincent Poor, Princeton University, United States
The compound wiretap channel models the problem of secure communication under channel uncertainty in the presence of an
eavesdropper who must be kept ignorant of transmitted messages. In this paper, the compound wiretap channel with secret keys
is studied, where transmitter and legitimate receiver share an additional secret key of a fixed rate. This paper studies how the
channel uncertainty and the secret key influence the corresponding secrecy capacity.
MA4b-4
Secret Key-Private Key Generation for Multiple Terminals
11:30 AM
Huishuai Zhang, Syracuse University, United States; Lifeng Lai, Worcester Polytechnic Institute, United States;
Yingbin Liang, Huishuai Zhang, Syracuse University, United States
The problem of simultaneously generating a secret key (SK) and private key (PK) pair among multiple terminals is studied, in
which each terminal observes a component of correlated sources. All terminals are required to generate a common secret key
concealed from an eavesdropper that has access to public discussion, while two designated terminals are required to generate an
extra private key concealed from both the eavesdropper and the remaining terminals. Bounds on the SK-PK capacity region are
derived for the general case. For pairwise independent network (PIN model), the SK-PK capacity region is established.
Track H – Speech, Image and Video Processing
Session: MAb5 – Image and Video Processing
Chair: Marios S. Pattichis, University of New Mexico
MA5b-1
Robust Image Recognition by Multi-Kernel Dictionary Learning
10:15 AM
Rituparna Sarkar, Sedat Ozer, Scott Acton, Kevin Skadron, University of Virginia, United States
Many recent studies discussed the problem of selecting and combining the salient features from a pool of feature-types and
showed that such techniques yield higher accuracy on average than only selecting features from a single feature-type in image
retrieval and classification applications. In this paper, we approach this problem as selection of the salient feature-types from a
13
pool of feature-types rather than selecting the individual features. Our approach utilizes multiple kernels within the dictionarylearning framework where a combination of dictionary atoms represents individual categories and category specific feature
combination parameters and weights are determined by the multiple kernel technique.
MA5b-2
Robust Dual-Band MWIR/LWIR Infrared Target Tracking
10:40 AM
Chuong Nguyen, Joseph Havlicek, University of Oklahoma, United States; Guoliang Fan, Oklahoma State
University, United States; John Caulfield, Cyan Systems, United States; Marios Pattichis, University of New
Mexico, United States
We introduce an SIR particle filter for tracking civilian targets including vehicles and pedestrians in dual-band midwave/
longwave infrared imagery as well as a novel dual-band track consistency check for triggering appearance model updates.
Because of the paucity of available dual-band data, we constructed a custom sensor to acquire the test sequences. The proposed
algorithm is robust against magnification changes, aspect changes, and clutter and successfully tracked all 17 cases tested,
including two partial occlusions. Future work is needed to comprehensively evaluate performance of the algorithm against stateof-the-art video trackers, especially considering the relatively small number of previous dual-band tracking results that have
appeared.
MA5b-3
Crowdsourced Study of Subjective Image Quality
11:05 AM
Deepti Ghadiyaram, Alan Bovik, University of Texas at Austin, United States
We designed and created a new image quality database that models diverse realistic image distortions and artifacts that affect
images that are captured using modern mobile devices. We also designed and implemented a new online crowdsourcing system,
which we are using to conduct a very large-scale, on-going, multi-month image quality assessment (IQA) subjective study,
wherein a wide range of diverse observers record their judgments of image quality. Our database currently consists of over
220,000 opinion scores on 1,163 characteristically distorted images evaluated by over 5000 human observers.
MA5b-4
Detecting Coronal Holes for Solar Activity Modeling
11:30 AM
Marios Pattichis, University of New Mexico, United States; Rachel Hock, AFRL/RVBXS Space Vehicles
Directorate, United States; Venkatesh Jatla, University of New Mexico, United States; Carl Henney, Charles Arge,
AFRL/RVBXS Space Vehicles Directorate, United States
The paper focuses on the development of coronal hole detection methods for use in physical models of solar activity. The
problem is motivated from the need to provide physical models with accurate detection of coronal holes. For each method, we
use a new optimization approach for determining the best parameter values. Optimization is based on a new matching metric that
compares clusters of automatically detected coronal holes against a manually annotated database. Validation of the approach is
performed on the manually annotated database using leave-one-out.
Track E – Array Signal Processing
Session: MAb6 – Sparse Estimation and Learning in Multi-Channel and
Array Systems
Co-Chairs: Palghat P. Vaidyanathan, California Institute of Technology and Piya Pal,
University of Maryland
MA6b-1
10:15 AM
Characterization of Orthogonal Subspaces for Alias-Free Reconstruction of Damped
Complex Exponential Modes in Sparse Arrays
Pooria Pakrooh, Ali Pezeshki, Louis L. Scharf, Colorado State University, United States
In this work, we consider the problem of parameter estimation for p damped complex exponentials, from the observation of
non-uniform samples of their weighted and damped sum. This problem arises in many areas such as modal analysis, speech
processing, system identification and direction of arrival estimation. For the case of DOA estimation, it is shown that for specific
choices of sparse sensor geometries such as coprime and nested arrays the DOA problem is identifiable using MUSIC. We
are interested in the estimation of the mode parameters through characterization of the orthogonal subspace of the generalized
Vandermonde matrix associated with the signal component of the sensor measurements. This characterization becomes useful
when we are interested in maximum likelihood or least squares estimation of the modes from noisy measurements. Here, we can
14
use iterative quadratic maximum likelihood or modified least squares to come up with a 2p-parameter characterization of the
orthogonal subspace. After estimating the parameters representing the orthogonal subspace, we find the roots of two polynomials
associated with these coefficients and match up the roots. We show that for certain sparse geometries, such as the coprime array,
matching up the roots removes aliasing and yields the actual modes in the noise-free case. Also, we derive the stochastic CramerRao Bound (CRB) in the estimation of the mode parameters. In the special case of DOA estimation, we look for the best array
geometries that minimize the CRB in a certain SNR region. We study the sensitivity of our parameterization of the orthogonal
subspace to sensor location errors. This can be considered another factor in determining the best geometries for sparse arrays.
Naturally, all of our developments also apply to estimation of complex exponential modes from time series data.
MA6b-2
10:40 AM
Exploiting Sparsity during the detection of High-Order QAM Signals in Large Dimension
MIMO Systems
Oleg Tanchuk, Bhaskar Rao, University of California, San Diego, United States
This paper proposes a receiver for multiple-input multiple-output (MIMO) systems for various constellation sizes and channel
knowledge at the receiver. The detector is composed of multiple stages. During the first stage, linear MMSE filter is employed
and nearest neighbor quantization is performed resulting in a sub-optimal estimate. During the next stage, the residual in the
measurement vector is calculated and the detector focuses on the error vector which has additional structure. The all-zero vector
and lowest energy vectors have the largest priors. As a result the error vector is often sparse (has few non-zero components),
allowing sparse signal recovery techniques such as Sparse Bayesian Learning to be employed during the detection step. Large
number of antennas allows Gaussian approximations to take effect, simplifying and minimizing some of the dependencies
between error and noise vectors.
MA6b-3
Structured Sparse Representation with Low-Rank Interference
11:05 AM
Minh Dao, Yuanming Suo, Sang (Peter) Chin, Trac Tran, Johns Hopkins University, United States
This paper provides an efficient framework for multiple-measurement representation where the underlying signals exhibit
sparsity properties over some proper dictionaries but the measurements are largely corrupted by interference sources. Under
assumption that the interference component forms a low-rank structure, the proposed model extracts the interference by
minimizing its nuclear norm while simultaneously promoting structured-sparsity representation of multiple correlated signals. An
efficient algorithm based on alternating direction method of multipliers is also proposed. Extensive experiments are conducted on
various applications: hyperspectral chemical plume classification, robust speech recognition in noisy environments, and synthetic
aperture radar image recovery to verify the method’s effectiveness.
MA6b-4
Grid-Less Algorithms for Identifying More Spectral Lines Than Sensors.
11:30 AM
Piya Pal, University of Maryland, College Park, United States; P. P. Vaidyanathan, California Institute of
Technology, United States
We consider the problem of estimating the sparse spectrum of a signal from only a few samples measured at a sub-Nyquist
rate. We specifically consider the situ- ation where the signal exhibits a line spectrum. Considering a Wide Sense Stationary
signal model, the authors had previously proposed novel sampling techniques that could identify O(M^2) spectral lines using
only O(M) samples. Recently, the authors have proposed a new approach exploiting the low rank structure of the covariance
matrix, to identify these O(M^2) spectral lines. This new method does not suffer from basis mismatch, nor does it need to know
the number of spectral lines apriori. In this paper, we study the performance of the proposed low rank minimization approach,
especially when correlation estimates deviate from their ideal values and/or when they are corrupted with additive noise. We also
demonstrate how other grid-less algorithms, such as “atomic norm minimization” or “TV-norm minimization” can be adopted to
identify O(M^2) spectral lines using suitable sub- Nyquist sampling schemes, and compare their performance both analytically
and empirically. The results in this paper extend a series of recent work on convex optimization based approaches for sparse
line spectrum estimation, by improving the guarantees on the number of resolvable lines through the use of suitable sampling
schemes.
15
Track G – Architecture and Implementation
Session: MAb7 – Architectures for Detection and Decoding
Chair: Joseph R. Cavallaro, Rice University
MA7b-1
10:15 AM
A Reduced-Complexity Iterative Decoding Scheme for Quasi-Cyclic Low-Density ParityCheck Codes
Shu Lin, Keke Liu, Juane Li, University of California, Davis, United States
Quasi cyclic low-density parity-check (QC-LDPC) codes are the most preferred type of LDPC codes for error control in
communication and data storage systems due to their encoding and decoding implementation advantages over the other types
of LDPC codes. This paper presents a reduced-complexity iteratively revolving decoding scheme for QC-LDPC codes which is
devised based on the quasi-cyclic structure of the parity-check matrices of these codes. A high-rate and long QC-LDPC code is
used to demonstrate the effectiveness of the proposed decoding scheme.
MA7b-2
Efficient Adaptive List Successive Cancellation Decoder for Polar Codes
10:40 AM
Chuan Zhang, National Mobile Communications Research Laboratory, China; Zhongfeng Wang, Broadcom
Corporation, United States; Xiaohu You, National Mobile Communications Research Laboratory, China
Because of their ability to provably achieve symmetric capacities of binary-input discrete memory-less channels (B-DMCs), polar
codes have been receiving significant attentions from researchers. In this paper, an adaptive list successive cancellation (SC)
decoder is proposed for polar codes. In the proposed list decoder, the maximum list size L_max is fixed, but the actual utilized
list size L can be adaptively reduced once the setup threshold metric is met. Simulation results show that the proposed adaptive
list decoder achieves similar performance as that of the classic list decoder, whereas requires less computation complexity. The
hardware architecture for the proposed decoder is also proposed. Compared to the existing adaptive scheme for polar decoders,
the proposed approach turns out to be more implementation friendly.
MA7b-3
11:05 AM
Decoder Diversity Architectures for Finite Alphabet Iterative Decoders for LDPC Codes
Bane Vasic, University of Arizona, United States; David Declercq, Universite de Cergy-Pontoise, France; Shiva
Planjery, Codelucida, United States
We present a finite alphabet iterative decoders (FAIDs), a new type of decoders for low-density parity check (LDPC) codes,
which outperform much more complex belief-propagation-based counterparts in the error floor region. The FAID variable node
update is a simple Boolean map, and we show that by varying this map one can achieve a class of decoders capable of correcting
wide range of distinct error patterns uncorrectable by a single FIAD. We call this concept decoding diversity, and present a
low-complexity architecture and error performance analysis of the FAID diversity decoder for column-weight three LDPC codes
using only hard-decisions from the channel.
MA7b-4
11:30 AM
Asynchronous Design for Precision-Scaleable Energy-Efficient LDPC Decoder
Jingwei Xu, Tiben Che, Ehsan Rohani, Gwan Choi, Texas A&M university, United States
This paper presents a low-density parity-check (LDPC) decoder design that uses scalable-precision calculation (SPC) and
asynchronous circuit techniques to reduce power consumption. The decoder configures the computation precision to minimize
circuit-level switching necessary for given target bit-error rate. The asynchronous circuit approach guarantees the completion of
each compute-and-forward phase at necessary voltage levels. The voltage level is scheduled to ensure completion of minimum
necessary decoding iterations. The proposed scheme is studied for the specific application of WiMAX to reduce the power
consumption at a desired quality of service (QoS). The proposed design is implemented and evaluated on Nangate 45nm library.
16
Track A – Communications Systems
Session: MAb8 – Synchronization and Channel Estimation 10:15 AM–11:55 AM
Chair: Shengli Zhou, University of Connecticut
MA8b1-1
Frequency Tracking with Intermittent Wrapped Phase Measurement Using the RaoBlackwellized Particle Filter
Maryam Eslami Rasekh, Upamanyu Madhow, University of California, Santa Barbara, United States; Raghuraman
Mudumbai, University of Iowa, United States
We consider the problem of frequency and phase tracking with intermittent measurements. Since accurate one-shot frequency
estimation requires long measurement epochs, to minimize overhead, we consider using phase-only measurements. Based on
the concept of a Rao-Blackwellized particle filter, a particle filter is utilized to deduce the unwrapped phase from the wrapped
phase measurements, while an extended Kalman filter (EKF) estimates the frequency offset and provides phase predictions that
are used to update particle weights. By jittering the timing of measurement epochs, frequency aliasing due to phase wrapping is
disambiguated. The efficacy of the proposed method is demonstrated on measured data.
MA8b1-2
Improving IEEE 1588v2 Time Synchronization Performance with Phase Locked Loop
Rico Jahja, Suk-seung Hwang, Goo-Rak Kwon, Jae-young Pyun, Seokjoo Shin, Chosun University, Indonesia
IEEE 1588 is one of the packet-based clock synchronization protocols. Different clock quality in each device will cause
inaccurate clock synchronization. In order to mitigate such kind of error, Phase Locked Loop (PLL) could be the solution of it.
In this paper, we propose a method that is consisted of the combination of IEEE 1588 and PLL to mitigate both queuing delay
variation from the network congestion as well as clock error because of the clock drift. The experiments show that our method
achieved sub-microsecond clock accuracy in faster period than the existing method.
MA8b1-3
Superimposed Pilots based Secure Communications for Multiple Antenna System
Yejian Chen, Bell Laboratories, Alcatel-Lucent, Germany
In this paper, we investigate secure communications by introducing superimposed pilots for multiple antenna system. The
superimposed pilots enable the trellis-based joint channel tracking and data detection for the user of interest. Further, by adjusting
the power ratio between the data symbol and superimposed pilot symbol, the secure capacity region can be established. The user
of interest can appropriately select the Forward Error Correction (FEC) code rate, to prevent any possible eavesdropping. In this
paper, we present the achievable secure capacity region for multiple antenna system, and verify it via Monte Carlo simulation as
well.
MA8b1-4
An Improved ESPRIT-Based Blind CFO Estimation Algorithm In OFDM Systems
Yen-Chang Pan, See-May Phoong, National Taiwan University, Taiwan; Yuan-Pei Lin, National Chiao Tung
University, Taiwan
Carrier frequency offset (CFO) is an important issue in orthogonal frequency division multiplexing (OFDM) systems. It destroys
the orthogonality between subcarriers and causes degradation in system performance. This paper presents an improved algorithm
for the existing ESPRIT-based CFO estimation method. The proposed method estimates the CFO by taking the determinant of
a matrix within the ESPRIT algorithm. It is a simple modification to the original ESPRIT-based method and its computational
overhead is low. The performance analysis shows that the proposed method can achieve a lower mean square error.
MA8b1-5
Blind, Low Complexity Estimation of Time and Frequency Offsets in OFDM Systems
Rohan Ramlall, University of California, Irvine, United States
A novel blind time and frequency offset estimator for orthogonal frequency division multiplexing (OFDM) systems is presented.
The estimator exploits a basic assumption of OFDM: the channel order is less than or equal to the length of the cyclic prefix.
Under this assumption, it is shown that the last sample of the received cyclic prefix is not corrupted by intersymbol interference
(ISI) and this sample can be used to estimate the time of arrival (TOA) in multipath channels. It is demonstrated that the proposed
estimator identifies the correct TOA with a higher probability than existing estimators for medium to high signal-to-noise ratios.
17
MA8b1-6
Efficient NLOS Optical Wireless Channel Estimation based on Sparse Pulse
Xiaoke Zhang, Chen Gong, Zhengyuan Xu, University of Science and Technology of China, China
We study the channel estimation problem for the non-line of sight (NLOS) optical wireless communications with coherent phase
optical sources, for example Laser signals. Such channel estimation problem is fundamentally different from the conventional
channel estimation problem for wireless electro-magnetic communication, since only the squared norm of the channel finitelength filter output is observed from the received signal. Thus the channel estimation methods for wireless electro-magnetic
communication cannot be applied. We propose a sparse pilot pulse-based channel estimation method, and the corresponding
preliminary numerical results. Further optimization of the proposed channel estimation remains for our work in the next step.
MA8b1-7
Channel Estimation and Precoder Design for Millimeter-Wave Communications: The
Sparse Way
Philip Schniter, Ohio State University, United States; Akbar Sayeed, Wisconsin, United States
We propose spectrally and computationally efficient methods for space-time channel estimation and precoding applicable to
millimeter-wave communication systems, which operate at high frequencies (30-300 GHz) over large bandwidths (>1 GHz).
Our methods exploit the fact that such channels are much sparser (in both angle and delay domains) than their microwave
counterparts, allowing accurate channel estimation from relatively few measurements. Furthermore, they leverage the MIMO
virtual-channel model, fast algorithms to compute its coefficients, and aperature-domain windowing methods to ensure its
sparsity.
Track B – MIMO Communications and Signal Processing
Session: MAb8 – Relaying
10:15 AM–11:55 AM
Chair: Guiseppe Caire, TU Berlin
MA8b2-1
Performance Analysis of Fixed Gain MIMO AF Relaying with Co-Channel Interferences
Min Lin, Min Li, PLA University of Science and Technology, China; Wei-Ping Zhu, Concordia University, Canada;
Kang An, PLA University of Science and Technology, China
This paper investigates the outage performance of a two-hop multiple-input and multiple-output (MIMO) amplify-and-forward
(AF) relay network. Specifically, by applying the MRT and MRC for the transmitter and receiver of each hop, respectively,
we first obtain the output signal-to-interference-plus-noise ratio with multiple co-channel interferences (CCIs) and noise at the
relay. Then, we present the closed-form outage probability (OP) expression of the consider AF relay network. Finally, computer
simulations are provided to demonstrate the validity of the derived theoretical formulas, and indicate the effects of antenna
combinations, CCI and power allocation on the outage performance of the consider two-hop AF relaying.
MA8b2-2
On Carrier-Cooperation in Parallel Gaussian MIMO Relay Channels with Partial Decodeand-Forward
Christoph Hellings, Wolfgang Utschick, Technische Universität München, Germany
It is known that parallel relay channels are not separable, i.e., the capacity with joint processing of the subchannels can be
higher than the sum of the individual capacities. The same holds for the data rates achievable using partial decode-and-forward
with Gaussian input signals. However, in this paper, we show that for parallel Gaussian MIMO relay channels, it is sufficient
to allow the relay to remap information from one subchannel to another between the decoding and the re-encoding. A carriercooperative transmission in the sense of spreading transmit symbols over several subchannels does not bring advantages in terms
of achievable rate.
MA8b2-3
Enhanced Relay Cooperation via Rate Splitting
Ivana Maric, Dennis Hui, Ericsson, United States
In wireless networks, mixed cooperative strategies in which relays in favorable positions decode-and-forward and the rest
quantize via short message noisy network coding (SNNC) have been shown to outperform existing cooperative strategies (e.g.,
decode-and-forward or compress-and-forward). We propose a novel relaying scheme that improves the performance of such
18
mixed cooperative strategies. In the proposed scheme, superposition coding is incorporated into SNNC encoding to enable partial
interference cancellation at DF relays. The achievable rate with proposed scheme is derived for the discrete two-relay network
and evaluated in the Gaussian case where gains over the rate achievable without rate splitting are demonstrated.
MA8b2-4
Alternate versus Simultaneous Relaying in MIMO Cellular Relay Networks: A Degrees of
Freedom Study
Aya Salah, Amr El-Keyi, Nile University, Egypt; Mohammed Nafie, Cairo University, Egypt
A two-hop cellular relay network consisting of two source-destination pairs equipped with M antennas is considered where each
source is assisted by two decode-and-forward relays operating in half-duplex mode and the relays are equipped with N antennas.
The DoF of the system is investigated for both simultaneous and alternate relaying configurations. For each relay configuration,
an outer bound on the degrees of freedom (DoF) is developed. A new achievable scheme is proposed that meets the upper bound
on the maximum DoF for all values of M and N except for M<N<5M/2.
MA8b2-5
Low-Complexity Two-Way AF MIMO Relay Strategy for Wireless Relay Networks
Kanghee Lee, Republic of Korea Air Force, Republic of Korea; Visvakumar Aravinthan, Sunghoon Moon, Wichita
State University, United States; Jongbum Ryou, Sungo Kim, Changki Moon, Inha Hyun, Republic of Korea Air
Force, Republic of Korea
In this paper, a two-way amplify-and-forward (AF) wireless relay network consisting of two sources with multiple M antennas
and one relay with multiple N antennas is investigated with a self-interference cancelation process. The relay amplifying matrix
that has the low computational complexity is optimized under the relay transmit power constraint based on zero-forcing (ZF)
criterion with a singular value decomposition (SVD) method. The sum rate behavior is analytically and numerically studied.
Numerical results show that the eigen beamforming is more competitive than the equal-gain beamforming for the two-way
relaying systems.
MA8b2-6
Blind Self-Interference Cancellation for Full-Duplex Relays
Gustavo Gonzalez, Fernando Gregorio, Juan Cousseau, CONICET - Universidad Nacional del Sur, Argentina
Full-duplex relays improve spectral efficiency but transmission and reception in the same frequency band produce coupling
between the transmitter and receiver side signals. This self-interference requires to be removed to allow a proper operation.
The cyclic prefix employed in modern modulations produces a periodic autocorrelation that is perturbed by the relay coupling.
We propose a criterion, and an associated blind adaptive algorithm, to estimate the self-interference that employs the known
autocorrelation pattern of the source signal as a reference. We present an study of the mean squared error of the criterion and
analyze the stationary points of the algorithm.
Track E – Array Signal Processing
Session: MAb8 – Active Sensing and Target Recognition
10:15 AM–11:55 AM
Chair: Mark R. Bell, Purdue University
MA8b3-1
Proximal Constrained Waveform Design Algorithms for Cognitive Radar STAP
Pawan Setlur, Wright State Research Institute, United States; Muralidhar Rangaswamy, Air Force Research
Laboratory, United States
Waveform design is an important component of the fully adaptive radar construct. In this paper we consider waveform design
for radar space time adaptive processing (STAP), accounting for the waveform dependence of the clutter correlation matrix. Due
to this dependence, in general, the joint problem of receiver weight vector optimization and radar waveform design becomes an
intractable optimization problem. We derive proximal constrained algorithms which, at each step, optimizes the STAP weight
vector and waveform independently. Unlike traditional STAP techniques, these algorithms are numerically stable, and can be
used in the practical training data starved STAP scenarios. Our simulations reveal a non-increasing error variance at the output of
the filter.
19
MA8b3-2
The Generalized Sinusoidal Frequency Modulated Waveform for High Duty Cycle Active
Sonar
David Hague, John Buck, University of Massachusetts Dartmouth, United States
High Duty Cycle Active Sonar (HDCAS) continuously transmits waveforms and processes target echoes to revisit the target
scene more often than conventional pulsed active sonar. The main challenge in HDCAS waveform design is to maintain large
instantaneous bandwidth while revisiting the target scene often. This research proposes a new diverse pulse train for HDCAS
that is composed of Generalized Sinusoidal Frequency Modulated (GSFM) waveforms. Since each GSFM waveform in the pulse
train is nearly orthogonal to the others, each pulse can be processed separately providing the high target revisit rates necessary for
HDCAS while also attaining large instantaneous bandwidth.
MA8b3-3
Concurrent Exploration of Orthogonal Waveform and Co-Prime Array for Quick and
High Resolution Scanning
Shuo Yang, Xin Wang, Xuehong Lin, Stony Brook University, United States
Co-prime array has been introduced lately to achieve O(MN) degree of freedom with M+N sensors. In existing studies on the
application of co-prime array, each or both sub-arrays of a co-prime array work in coherent mode where all the antennas of an
array transmit the same signal. In this work, we would like to investigate the potential benefit of using orthogonal waveform
transmissions like that done MIMO radar case in co-prime array. In particular, we apply the co-prime array for signal scanning,
and our analysis indicates that the use of the two techniques can significantly speed up the scanning process.
MA8b3-4
On Bayesian Transmit Signal Design using Information Theory
Mir H. Mahmood, NextNav LLC, United States; Mark R. Bell, Purdue University, United States
A signal design problem that maximizes the mutual information between the received signal and target impulse response
under Bayes decision theoretic framework is investigated. Owing to its analytical intractability, a new approximate problem
is formulated, and its solution presented in the form of a fixed point signal design equation along with an efficient numerical
algorithm to design MI-maximizing signals. The energy allocation of the designed signals is compared to SNR-maximizing
signals. Simulation results of the detection performance show that while SNR-maximizing signals provide overall better detection
performance, MI-maximizing signals provide better conditional detection performance for less dominant target classes.
MA8b3-5
Improved Distributed Automatic Target Recognition Performance by Exploiting Dominant
Scatterer Spatial Diversity
John Wilcher, William Melvin, Georgia Tech Research Institute, United States; Aaron Lanterman, Georgia Institute
of Technology, United States
Radar automatic target recognition (ATR) is examined from the viewpoint of improving classification performance through
spatial diversity. Recent radar target classification performance improvements have been confirmed using multiple target
reflectivity returns. This paper extends the use of multiple target looks by examining the impact of dominant scatterer placement
and physical characteristics on target classification rates. Multiple, spatially diverse high range resolution (HRR) profiles are
exploited to show progressive improvement in classification rates as such additional target perspectives are included in the
classification algorithm. Percentage of correct classification (PCC) improvements exceeding 10% is demonstrated.
MA8b3-6
Semi-Supervised Classification of Terrain Features in Polarimetric SAR Images using H/A/
alpha and the General Four-Component Scattering Power Decompositions
Stephen Dauphin, Sandia National Laboratories, United States; Margaret Cheney, Colorado State University, United
States; Derek West, Robert Riley, Sandia National Laboratories, United States
In an effort to enhance image segmentation and classification of terrain features in a fully polarimetric SAR image, this paper
explores the utility of the ordered coupling of SLIC superpixel segmentation of the General Four-Component Scattering Power
decomposition multi-channel product with a subsequent H/A/alpha decomposition on resulting superpixels. The resulting
superpixels can be classified into discrete terrain feature categories by comparing them to hand-picked reference samples using a
semi-supervised routine.
20
MA8b3-7
A Super-Resolving Near-Field Holographic Method for Underwater EM Signature
Modeling
Hatim Alqadah, Naval Research Laboratory, United States; Nicolas Valdivia, US Naval Research Laboratory,
United States
This work is concerned with fast and stable back-projection of near-field electromagnetic array measurements in underwater
environments. Such procedures are used for near-real time prediction of underwater structure radiation patterns. The conductivity
of the water introduces severe instability due to the presence of evanescent waves in the data. We demonstrate a stable superresolution technique based on a sparse decomposition of the measurements into a finite sum of elementary electric and/or
magnetic dipole sources. The resulting sparse regularization objective function is minimized via fast iterative thresholding
methods. Back -projection results using experimental underwater AC and DC range data are presented.
MA8b3-8
Limitations and Capabilities of the Fractional Spectrogram Analysis Tool for SAR-Based
Detection of Multiple Vibrating Targets
Adebello Jelili, Balu Santhanam, Majeed Hayat, University of New Mexico, United States
SAR-based detection and estimation of vibration signatures of ground objects using the micro-Doppler effect has gained attention
in recent work [4,6,9,10]. Ground target vibrations that introduce phase modulation in the SAR return signals are examined
using standard pre-processing, followed by analysis via the recently introduced fractional spectrograms [8]. The capabilities and
limitations of this new time-frequency analysis tool, for multiple vibrating target detection and classification, are investigated.
Track F – Biomedical Signal and Image Processing
Session: MAb8 – Physiological Signal Processing
10:15 AM–11:55 AM
Chair: Alessio Medda, Georgia Tech
MA8b4-1
Sample-Based Cross-Frequency Coupling Analysis with CFAR Detection
Charles Creusere, Nathan McRae, Mark Norman, Philip Davis, New Mexico State University, United States
In the proposed paper, we introduce a new approach for cross-frequency coupling analysis as applied to electroencephalograph
(EEG) signals. Our approach consists of a low-complexity signal analysis block which is well-suited to implementation as an
integrated circuit followed by constant false alarm rate (CFAR) detection. In the preliminary results presented here, we see that
the proposed framework provides good detection while effectively rejecting false alarms. In the final paper, we will include full
receiver operating curves (ROCs) as well for more precise validation.
MA8b4-2
Classification of Human Viewers using SVM
Philip Davis, Charles Creusere, Jim Kroger, New Mexico State University, United States
Subject identification and authentication using electroencephalograph (EEG) signals has been gaining interest in the biometric
field due to the decrease in price in EEG systems and the extremely positive results that researchers have seen. Here, we evaluate
biometric identification using linear support vector machine (SVM) classification on subjects watching short video clips. In
particular, cepstral coefficient feature vectors are formed for each of the 128-channels of our EEG system. We explore the effects
on classification of using individual versus grouped channels, different video types, and number of channels used. Furthermore,
we also consider which regions of the head give the best classification results.
MA8b4-3
Activity Recognition using Statistical Gait Parameters from a Single Accelerometer
Andrew Vaughan, Alessio Medda, Brian Liu, Shean Phelps, Georgia Tech Research Institute, United States
Wearable sensor systems represent an increasingly viable approach to short and long term motion classification and gait
estimation primarily due to decreased size, cost and broad applicability. A simple model of an acceleration response, acquired
during different levels and types of activities, was obtained using statistical mixture models. The proposed algorithm uses this
information to categorize activity in order to achieve the goal of estimating human performance through motion classification and
gait estimation. Mixture parameters obtained from the signal of a single chest-worn accelerometer are used in a semi-supervised
learning approach allowing for the robust classification of gait across subjects.
21
MA8b4-4
Intra-Patient and Inter-Patient Seizure Prediction from Spatial-Temporal EEG Features
Shuoxin Ma, Daniel Bliss, Arizona State University, United States
In this paper, an algorithm for intra-patient and inter-patient seizure prediction from invasive electroencephalography (EEG) is
proposed. Multi-channel EEG are pre-processed and built into spatial-temporal covariance matrices, from which multivariate
features are extracted. A support vector machine~(SVM) is trained with the features of classified data to predict the un-classified
data. The cross-validation shows that the proposed algorithm achieves an outstanding performance, with the area under receiver
operating characteristic~(ROC) curve of 0.977 for intra-patient and 0.822 for inter-patient prediction. The significance test further
proves that the result is reliable, with p-values of 0.00 and 0.10 for intra-patient and inter-patient prediction, respectively.
MA8b4-5
Effective Connectivity in fMRI from Mutual Prediction Approach
Marisel Villafañe-Delgado, Selin Aviyente, Michigan State University, United States
Estimation of effective connectivity in neurophysiological signals has gained great popularity in recent years. Widely used,
Granger causality depends on the temporal precedence of the signal’s amplitude and assumes the signals are linear and stationary.
In this work an alternative model-free method for estimation of effective connectivity based on the signal’s instantaneous phases
is assessed. Mutual prediction approach is implemented by estimating the instantaneous phases from the Reduced Interference
Rihaczek time-frequency distribution. Results on both simulated and fMRI data indicate that the proposed method is comparable
to Granger’s causality.
MA8b4-6
Whitening 1/f-type Noise in Electroencephalogram Signals for Steady-State Visual Evoked
Potential Brain-Computer Interfaces
Alan Paris, Azadeh Vosoughi, George Atia, University of Central Florida, United States
A method is proposed to whiten 1-f-type background noise in electroencephalogram data by a non-linear spectral transformation
from the frequency domain to a newly-defined alpha-pitch domain. Based on the alpha-pitch spectra of steady-state visual evoked
potentials, an algorithm called octave-averaged spectral rectification is applied which simultaneously attenuates 1-f noise while
enhancing resonance peaks. This has important potential benefits for gamma-band brain-computer interfaces.
MA8b4-7
Adaptive Learning of Behavioral Tasks for Patients with Parkinson’s Disease Using Signals
from Deep Brain Stimulation
Nazanin Zaker, University of Denver, United States; Arindam Dutta, Alexander Maurer, Arizona State University,
United States; Jun Zhang, University of Denver, United States; Sara Hanrahan, Adam Hebb, Colorado Neurological
Institute, United States; Narayan Kovvali, Antonia Papandreou-Suppappola, Arizona State University, United States
We propose adaptive learning methods for identifying different behavioral tasks of patients with Parkinson’s disease (PD). The
methods use local field potential signals that were collected during Deep Brain Stimulation implantation surgeries. Using timefrequency signal processing methods, features are first extracted and then clustered using two different methods. The first method
uses a hybrid model that combines support vector machines and hidden Markov models. The second method does not require any
a priori information and uses Dirichlet process Gaussian mixture models. We demonstrate the performance of both methods and
discuss the advantages of each method under different conditions.
Track D – Signal Processing and Adaptive Systems
Session: MPa1 – Big Data Analytics
Chair: Ali Tajer, Rensselaer Polytechnic Institute
MP1a-1
Universal Sequential Outlier Hypothesis Testing
1:30 PM
Yun Li, Sirin Nitinawarat, Venugopal Veeravalli, University of Illinois at Urbana-Champaign, United States
Multiple observation sequences are collected, a few of which are outliers. Observations in an outlier sequence are generated by a
different mechanism from that generating the typical sequences. The goal is to design a universal test to best discern the outlier
sequences with the fewest observations on average. This problem was treated by us previously in the fixed sample size setting
with an arbitrary number of outliers [1], and in the sequential setting with at most one outlier [2]. In this work, our previous
findings are generalized to the sequential setting with multiple outliers.
22
MP1a-2
Parsimonious Models for Random Variables and Stochastic Processes
1:55 PM
Weiyu Xu, University of Iowa, United States
In this paper, we will discuss various parsimonious models for random variables and random processes. High dimensional data
generated from this model can be described by low-dimensional parameters. We will further discuss how to leverage these
parsimonious models in signal processing applications, including network tomography and sparse principal component analysis.
MP1a-3
2:20 PM
Fundamental Limits on Information-Friction Energy of Big-Data Computing
Majid Mahzoon, Pulkit Grover, Carnegie Mellon University, India
Big-Data processing consumes a non-negligible fraction of world energy. This work provides fundamental limits on info-friction
energy of example Big Data computations.
MP1a-4
Quickest Search Over Correlated Sequences
2:45 PM
Ali Tajer, Wayne State University, United States
This paper considers the problem of searching over a collection of sequences that are generated by one of the two possible
distributions $F_0$ and $F_1$ and designs the quickest sequential detection procedure for identifying one sequence that is
generated according to $F_1$. Generation of the sequences obeys a known dependency kernel such that the prior probability
that a sequence is generated by $F_1$ is governed by the distributions of the rest of sequences. The optimal quickest sequential
detection procedure, that is the procedure that strikes a balance between detection quality and decision delay as two opposing
performance measures, is characterized.
Track D – Signal Processing and Adaptive Systems
Session: MPb1 – Tensor-Based Signal Processing
Chair: Eric Moreau, University of Toulon
MP1b-1
3:30 PM
Memory-Efficient Parallel Computation of Tensor and Matrix Products for Big Tensor
Decomposition
Niranjay Ravindran, Nicholas Sidiropoulos, Shaden Smith, George Karypis, University of Minnesota, United States
Low-rank tensor decomposition has many applications in signal processing and machine learning, and is becoming increasingly
important for analyzing big data. An significant challenge is the computation of the intermediate products which can be much
larger than the final result of the computation, or even the original tensor. We propose a scheme that effectively allows in-place
updates of intermediate matrices. We also propose a memory-efficient tensor compression scheme. The resulting algorithms can
be parallelized, and do not require sparsity -- although it can be exploited if present.
MP1b-2
3:55 PM
Recent Advances on Tensor Models and their Relevance for Multidimensional Data
Processing
Salah Bourennane, Julien Marot, Ecole Centrale Marseille - Institut Fresnel, France
This paper is mainly focused on reducing various types of noise in hyperspectral images. Several denoising methods based on
tensor models, such as Tucker and PARAFAC, are proposed for different types of random noise in hyperspectral images. They
can not only obtain great denoising performances but also preserve the rare signals in hyperspectral images.
MP1b-3
4:20 PM
Tensor-Based Channel Estimation for Non-Regenerative Two-Way Relaying Networks
with Multiple Relays
Jianshu Zhang, Kristina Naskovska, Martin Haardt, Ilmenau University of Technology, Germany
In this paper we investigate a two-way relaying network with multiple amplify-and-forward relays where both the user terminals
(UTs) and the relays can have multiple antennas. To improve the system performance, the UTs need channel knowledge
of all relevant channels. Therefore, we propose a tensor-based channel estimation method based on the block component
23
decomposition (BCD) as well as the CANDECOMP/PARAFAC (CP) decomposition. The proposed method is analytic, i.e.,
iterations are not required. We also derive the design criteria for the corresponding relay amplification matrices and the training
sequences. Simulation results demonstrate the performance of the proposed channel estimation method.
MP1b-4
Fast Non-Unitary Simultaneous Diagonalization of Third-Order Tensors
4:45 PM
Victor Maurandi, Eric Moreau, University of Toulon, France
We consider the problem of non-orthogonal joint diagonalization of a set of real-valued third-order tensors. This appears in many
signal processing problems and it is instrumental in source separation. We propose a new Jacobi-like algorithm based on a special
parameterization of the so-called diagonalizing matrix. One important point is that each Jacobi estimation parameters is done
entirely analytic using an appropriate criterion and based on an alternate estimation. Numerical simulations illustrate the overall
very good performances of the proposed algorithm.
Track F – Biomedical Signal and Image Processing
Session: MPa2 – Neural Engineering and Signal Processing
Chair: Ervin Sejdic, University of Pittsburgh
MP2a-1
1:30 PM
Electroencephalography-based Alzheimer’s Disease Diagnosis: Where we are at Now and
Tiago Falk, Institut National de la Recherche Scientifique, Canada
This paper will present an overview of existing electroencephalography (EEG) based Alzheimer’s disease (AD) diagnosis
tools, along with their limitations and advantages. Examples will be given from both an ERP (event related potential) and a
quantitative EEG perspective. The paper will conclude with a roadmap of possible future multimodal tools that can be developed
with emerging technologies, such as functional Transcranial Doppler sonography (fTCD-EEG) and functional near-infrared
spectroscopy (fNIRS-EEG).
MP2a-2
EEG Event Detection Using Big Data
1:55 PM
Iyad Obeid, Amir Harati, Joseph Picone, Temple University, United States
Although signal processing has attempted for reading EEGs, these efforts have mostly yielded poor results. These algorithms,
based largely on heuristic methods or trained on modest data sets, have lacked the statistical power to adequately generalize
their performance over the great variability seen in the clinic. Here, we present our efforts to overcome these limitations by
using machine learning algorithms that are trained on a Big Data corpus comprising over 22,000 clinical recordings made in the
Neurology department at Temple University Hospital over a ten-year span.
MP2a-3
2:20 PM
A Source Localization Approach to Creating a Neural Interface with the Peripheral
Nervous System
Jose Zariffa, Toronto Rehabilitation Institute - University Health Network, Canada
Selectively monitoring the messages encoded in the electrical activity of peripheral nerves would enable us to improve the control
of neuroprosthetic devices, which interface with the nervous system to help restore function after neurological injuries. By using
multi-contact nerve cuff electrodes, which measure the electric potentials at several locations on the surface of the nerve, we can
approach this task as an inverse problem of source localization. We will review our group’s foundational work on this problem in
a rat sciatic nerve model, and present recent developments.
24
MP2a-4
2:45 PM
A Picture is Worth a Thousand Words: Some Examples of the Utility of Biomedical Image
Processing in Brain Research
Negar Memarian, University of California, Los Angeles, United States
Investigating the neural bases of various behaviors, emotions, neurological diseases, or neuropsychological disorders is a widely
studied and yet very mysterious realm of research. Technical advances in structural and functional neuroimaging have spurred the
development of sophisticated analysis techniques. In this talk, I explain the rationale for use of quantitative approaches in brain
image analysis and draw examples from my interdisciplinary research collaboration with neuroscientists to illustrate the promise
of signal/image processing and pattern recognition for achieving a more objective and accurate knowledge of the human brain.
Track F – Biomedical Signal and Image Processing
Session: MPb2 – Brain Connectomics
Chair: Dimitri Van De Ville, EPFL
MP2b-1
3:30 PM
Brain-Network Continua Revealed with Multivariate Performance Metrics.
Stephen Strother, Baycrest and University of Toronto, Canada
The brain is thought to contain a hierarchy of “networks” that dynamically reconfigure themselves to meet task demands.
Evidence for such hierarchies in BOLD fMRI data sets has emerged from our attempts to build a quantitative framework for
comparing the performance of multivariate pattern analysis models (e.g., linear discriminants (LD), support vector machines
(SVM), logistic regression (LR)). I will describe the pseudo-Receiver Operating Characteristic performance plot defined
by subsampled measures of prediction (P) plotted against spatial reproducibility (R) that is central to such performance
measurements. Using these (P, R) plots as a function of model regularization we have shown that the choice and tuning of
regularisation is generally more important than the particular linear model chosen. Moreover, we have also demonstrated that LD
models regularized with dimensionality reduction using principal components (PC) are particularly well suited to reveal taskdependent hierarchies in the brain’s covarying network structure. During task performance an LD model regularized with PCs
added in order of largest to smaller variance traces out a characteristic (P, R) curve shape as a function of the number of PCs (q).
Such (P, R) curves typically start with Pmin equivalent to random guessing for q=1, and rise to a task-dependent Pmax for q ϵ
[20,100]. Values of R typically start near Rmax for q=1, decline and then reach a local maximum before Pmax, and then decrease
rapidly with increasing q. I will compare and contrast the (P, R) curves of several related tasks with different levels of difficulty.
Finally I will present evidence that the resulting normalized, spatial salience profiles as a function of increasing P and q reflect
the regional hierarchies of the underlying brain-network continua adapted to meet particular task demands.
MP2b-2
Learning with Multi-Site fMRI Graph Data
3:55 PM
Richiardi, Stanford University, United States
Neuroimaging data collection is very costly, and acquisition is commonly distributed across multiple sites. However, factors such
as different noise characteristics or inhomogeneities make it difficult to successfully combine multi-site functional imaging data.
Correlation estimators necessary for computing functional connectivity graphs can also be sensitive to extreme signal values.
Here, we propose to find a stable subspace by using a discriminative projection that does not only minimise site differences, but
also preserves discriminative class information. We compare our method with the “regressing-out” approach on synthetic and real
data.
MP2b-3
Using Computer Vision to Understand Biological Vision
4:20 PM
Dmitri Chklovskii, Simons Center for Data Analysis, United States
How do the networks of neurons in our brain give rise to our behavior, feelings, and thoughts? In attacking this question we
follow the strategy used in reverse engineering of computer chips. We image brain structure at nanometer resolution and, by
using computer vision algorithms, reconstruct neuronal wiring diagrams, or connectomes. Then, by using recent algorithms
from signal processing and online learning, we infer the function of these connectomes. By applying this strategy to the simpler
brain of a fruit fly we have made significant progress towards understanding its visual system, a small but efficient and powerful
computing device.
25
MP2b-4
4:45 PM
Dynamic Functional Connectivity: Probing Spontaneous Network Reorganization
Dimitri Van De Ville, Nora Leonardi, École Polytechnique Fédérale de Lausanne / University of Geneva,
Switzerland
Functional connectivity (FC) during resting state has already revealed significant insights into functional integration of the
brain during task, but also at rest. However, increasing evidence points towards continuously fluctuating FC across the duration
of a scan. Using unsupervised learning techniques, groups of co-evolving connections, or reproducible patterns of dynamic
FC (dFC), have been revealed in fluctuating FC. In particular, based on results from k-means clustering and sliding-window
correlations, it has recently been hypothesized that dFC may cycle through several discrete FC states. Alternatively, it has been
proposed to represent dFC as a linear combination of multiple FC patterns using principal component analysis. As it is unclear
whether sparse or non-sparse combinations of FC patterns are most appropriate, the goal of our study was to evaluate the impact
of sparsity by performing an empirical evaluation of simulated, task-based and resting-state dFC. Therefore, we applied sparse
matrix factorization methods for which we varied the sparsity constraints and then studied accuracy, reproducibility and stability
of ensuing representations of dFC. During subject-driven tasks, dFC was well described by alternating FC states in accordance
with the nature of the data. The estimated FC patterns showed a rich structure with combinations of known functional networks
enabling accurate identification of the three different tasks. During rest, dFC was better described by multiple FC patterns that
overlap. The executive control networks, which are critical for working memory, appeared grouped alternately with externallyor internally-oriented networks. These results suggest that sparse combinations of FC patterns can provide an economical and
meaningful solution to disentangle resting-state dFC.
Track D – Signal Processing and Adaptive Systems
Session: MPa3 – Compressed Sensing I
Chair: Aleksandar Dogandzic, Iowa State University
MP3a-1
Robust Line Spectral Estimation
1:30 PM
Gongguo Tang, Colorado School of Mines, United States; Parikshit Shah, Badri Bhaskar, University of WisconsinMadison, United States; Benjamin Recht, University of California, Berkeley, United States
Line spectral estimation is a classical signal processing problem that finds numerous applications in array signal processing and
speech analysis. We propose a robust approach for line spectral estimation based on atomic norm minimization that is able to
recover the spectrum exactly even when the observations are corrupted by arbitrary but sparse outliers. The resulting optimization
problem is reformulated as a semidefinite program. Our work extends previous work on robust uncertainty principles by allowing
the frequencies to assume values in a continuum rather than a discrete set.
MP3a-2
On the Applicability of Matrix Completion on MIMO Radars
1:55 PM
Shunqiao Sun, Athina Petropulu, Rutgers University, United States
It was recently shown that networked MIMO radars with sparse sensing and matrix completion (MC) can significantly reduce
the volume of data required for accurate target detection and estimation. This paper studies the applicability of MC theory on the
data matrices that arise in colocated MIMO radars using uniform linear arrays. It is shown that the coherence is directly related
to transmit waveforms, and that when the waveforms are orthogonal the optimum choice is for them to be white noise-type
functions.
MP3a-3
Subspace Learning from Extremely Compressed Measurements
2:20 PM
Martin Azizyan, Akshay Krishnamurthy, Aarti Singh, Carnegie Mellon University, United States
We consider learning the principal subspace of a large set of vectors from an extremely small number of compressive
measurements of each vector. Our theoretical results show that even a constant number of measurements per column suffices to
approximate the principal subspace to arbitrary precision, provided that the number of vectors is large. This result is achieved
by a simple algorithm that computes the eigenvectors of an estimate the covariance matrix. The main insight is to exploit an
averaging effect that arises from applying a different random projection to each vector. We provide a number of simulations
confirming our theoretical results.
26
MP3a-4
Analysis of Misfocus Effects in Compressive Optical Imaging
2:45 PM
Wenbing Dang, Ali Pezeshki, Randy Bartels, Colorado State University, United States
In this paper, we investigate the effect of misfocus on the performance of compressive optical imaging. Misfocus exists either
because the object is out of focus, relative to the focal plane of the optical system, or because the object is extended along the
optical axis. We derive theoretical bounds on the error in reconstructing the sparse field, as a function of demagnification and
misfocus. Our numerical results show that compressed sensing is robust to misfocus at modest demagnification factors. However,
at high demagnifications, which is typical in optical microscopy, compressive imaging performance degrades considerably.
Track D – Signal Processing and Adaptive Systems
Session: MPb3 – Compressed Sensing II
Chair: George Atia, University of Central Florida
MP3b-1
Filter Design for a Compressive Sensing Delay Estimation Framework
3:30 PM
Misagh Khayambashi, Lee Swindlehurst, University of California, Irvine, United States
Compressive sensing aims to find efficient signal acquisition and recovery techniques with the aid of prior knowledge of the
signal. While traditionally applied to sparse vectors, CS has been extended to analog signals with more general structures. The
use of CS in delay-Doppler estimation has recently received attention. We adopt a CS frameworks for delay-Doppler estimation
and optimize the deployed filter. The optimization criterion is the Bayesian Cr\’amer Rao Bound of delay estimation. An iterative
algorithm is proposed to solve the optimization problem and the results are compared with the prototype filter design.
MP3b-2
3:55 PM
Davood Mardani, George Atia, University of Central Florida, United States
The problem of detecting the existence of a sparse signal in noise is considered. While it is well-established that significant
compressive gains are achievable on account of sparsity, the question is whether further gains could be achieved when exact
signal reconstruction is not necessary. In contrast to prior work, which considered non-adaptive strategies, a sequential
adaptive feedback approach with a stopping rule is proposed for compressive signal detection. Two sources of gain are studied,
namely,compression gains due to adaptation, and computational gains via effective algorithms that fuse newly acquired
measurements and previous information.
MP3b-3
A Recursive Way for Sparse Reconstruction of Parametric Spaces
4:20 PM
Oguzhan Teke, Bilkent University, Turkey; Ali Cafer Gurbuz, TOBB University of Economics and Technology,
Turkey; Orhan Arikan, Bilkent University, Turkey
A novel recursive framework for sparse reconstruction of continuous parameter spaces is proposed by adaptive partitioning and
discretization of the parameter space together with expectation maximization type iterations. Any sparse solver or reconstruction
technique can be used within the proposed recursive framework . Initial results show that proposed technique improves the
parameter estimation performance of classical sparse solvers while achieving Cram\’er-Rao lower bounds on the tested problems.
Final paper will present more detail on both the development of the algorithm and simulation results.
MP3b-4
Subspace Methods for Recovery of Low Rank & Joint Sparse Matrices
4:45 PM
Sampurna Biswas, Mathews Jacob, Soura Dasgupta, University of Iowa, United States
We consider the recovery of a low rank and jointly sparse matrix from under-sampled measurements of its columns. This
is highly relevant in recovery of dynamic MRI data with high spatio-temporal resolution, where each column of the matrix
correspond to a frame in the image time series; which is low-rank since the frames are highly correlated. Classical multiple
measurement vector setup measures all the snapshots using the same matrix. We measure each snapshot using a different
measurement matrix. We show that this reduces the total number of measurements, especially when the rank (X) is much smaller
than than its sparsity.
27
Track A – Communications Systems
Session: MPa4 – Underwater Acoustic Communications and Networking
Chair: Zhaohui Wang, Michigan Technological University
MP4a-1
1:30 PM
Experimental Study of Secret Key Generation in Underwater Acoustic Channels
Yi Huang, University of Connecticut, United States; Lifeng Lai, Worcester Polytechnic Institute, United States;
Shengli Zhou, Zhijie Shi, University of Connecticut, United States
The predefined secret keys are often used to encrypt the data information in underwater acoustic (UWA) communications.
However, it’s not safe once the keys are stolen by eavesdroppers. We propose a new method that can generate secret keys
dynamically based on the frequency measurements of the channel in orthogonal frequency-division multiplexing (OFDM)
systems, where randomness and correlation of mutual channels are exploited to update the keys at regular intervals. Specifically,
the amplitude of each tone is quantized to 1-bit information, resulting a correlated binary sequence. Then low-density paritycheck (LDPC) codes are used to extract keys according to the Slepian-Wolf coding theorem. We will use experiment results to
verify the randomness and correlation between channels, and show that the secret keys can be generated in a low speed due to the
time-varying characteristics of UWA channels.
MP4a-2
1:55 PM
Random Linear Packet Coding for Fading Channels: Joint Power and Rate Control
Rameez Ahmed, Milica Stojanovic, Northeastern University, United States
Random linear packet coding is considered for channels that experience fading and have long propagation delay, such as the
underwater acoustic channels. Previously, we employed power control (adjusting the transmission power according to the
channel gain) and rate control (adjusting the number of coded packets according to the channel gain) to counteract the effects of
fading. For such policies, it can be shown that there exists an optimal number of coded packets (when employing power control)
or optimal transmission power (when employing rate control) for which the energy required per bit of information transmitted
is minimized. In this paper, we present a strategy to jointly optimize the transmission power and the number of coded packets,
such that the average energy per bit is minimized. We demonstrate the algorithm performance in simulation, as well as with
experimental data from two at-sea experiments.
MP4a-3
2:20 PM
Underwater Acoustic Communications in Great Lakes and in Oceans: What is the
Difference?
Wensheng Sun, Mohsen Jamalabdollahi, Zhaohui Wang, Seyed Zekavat, Michigan Technological University,
United States
Underwater acoustic (UWA) communications and networking have drawn considerable attentions in last decades, while a
majority of the research efforts have been paid to oceanic environment. Given that the Great Lakes occupy a shoreline of
an approximately identical length as that of the US east coast, investigation on the UWA communication performance in
Great Lakes is of great importance. Based on both numerical modeling and data sets collected in field experiments, this work
characterizes the acoustic channel in Lake Superior in terms of the signal transmission loss and the channel multipath structure.
By comparing the channel characteristics with that in oceans, an appropriate transceriver design for UWA communications in the
Great Lakes is studied.
MP4a-4
2:45 PM
Information-Guided Pilot Insertion for Capacity Improvement in OFDM Underwater
Acoustic Communications
Xilin Cheng, Colorado State University, United States; Miaowen Wen, Xiang Cheng, Peking University, China;
Liuqing Yang, Colorado State University, United States
Knowledge of channel state information at the receiver is essential for the recovery of the information-conveying symbols. In
OFDM underwater acoustic communication systems, acquisition of the channel state information is usually achieved with the
aid of pilots. Unlike traditional OFDM underwater acoustic communication systems, where the pilot positions are fixed, we
allow a random pilot insertion for the enhancement of the system spectral efficiency. In other words, the pilot positions will carry
28
Track B – MIMO Communications and Signal Processing
Session: MPb4 – Massive MIMO I
MP4b-1
3:30 PM
Jsdm and Multi-Cell Networks: Handling Inter-Cell Interference Through Long-Term
Antenna Statistics
Ansuman Adhikary, University of Southern California, United States; Giuseppe Caire, Technical University Berlin,
Germany
Joint Space Division and Multiplexing (JSDM) is a novel paradigm for clearly splitting the functions of multiuser MIMO
precoding into the concatenation of blocks: a pre-beamforming projection that depends only on the channel long0term
statistics (channel covariance matrices), and a precoder that achieve multiplexing gain over the effective channel including
pre-beamforming. In our previous work, we showed that this structure is capacity achieving when the users can be grouped
into sets with same covariance matrix, such that the dominant channel eigenspaces form a tall unitary basis. In addition, we
showed that this case is always achieved in the massive MIMO limit (large number of antennas at the base station), and under
some assumptions on the users and scattering environemnt distribution. Furthermore, JSDM is naturally suited for a hybrid
implementation of massive MIMO, where the pre-beamforming matrix can be implemented in the analog RF domain. In this
paper, we extend our study of JSDM to a multi-cell system, where the pre-beamforming projections are jointly selected across
cells and together with user scheduling, such that inter-cell interference can be handled by using only the channel covariances
(i.e., the so-called long-term’’ antenna statistics), which are much easier to collect than the classical instantaneous channel state
information.
MP4b-2
3:55 PM
Enabling Massive MIMO Systems in the FDD Mode thanks to D2D Communications
Haifan Yin, Laura Cottatellucci, David Gesbert, Eurecom, France
Massive MIMO systems offer an attractive solution in enhancing the data rate performance of wireless networks through the
use of large arrays at the base station side. Such arrays offer high beamforming and interference mitigation capabilities without
requiring complex inter-cell coordination approaches. By and large, such concepts have been confined to TDD systems due to
the prohibitive cost of (i) estimating channels and (ii) conveying the information back to infrastructure, in non reciprocal setups
like FDD. This paper offers a re-visit of this issue in the context of cellular networks enabled with some form of proximity D2D
communication. We show how such D2D communications, when combined with recently unveiled statistical properties of large
array channels, can help make massive MIMO amenable to an FDD implementation
MP4b-3
Massive MIMO As a Cyber-Weapon
4:20 PM
Massive MIMO is a revolutionary technology in wireless access, where large numbers of coherently operating base station
antennas serve many terminals in the same time-frequency resource via spatial multiplexing. Massive MIMO relies on channel
reciprocity and TDD operation, where the base station learns the downlink channel from uplink training. In this paper we
consider a scenario where massive MIMO technology falls in the hands of an adversary, who uses the large array of antennas
to jam a conventional (primary) communication link that operates in TDD. The jammer adaptively learns the slot timing of the
primary system, and estimates all channels within each coherence interval. It is shown that a low-output-power jammer can cause
significant harm to the primary link, without prior knowledge of neither the propagation channels nor the slot timing.
MP4b-4
Large Antenna Array and Propagation Environment Interaction
4:45 PM
Xiang Gao, Meifang Zhu, Fredrik Rusek, Fredrik Tufvesson, Ove Edfors, Lund University, Sweden
In traditional MIMO, propagation conditions are often considered stationary over the entire antenna array. In massive MIMO
systems, where arrays can span over large physical dimensions, the situation is quite different. For instance, significant variations
in signal strength, due to shadowing, can be experienced across a large array. These effects vary with propagation environment,
in which the array is placed, and influence achievable sum-rates. We characterize these variations for several measured
propagation scenarios in the 2.6 GHz frequency range and illustrate how power variations and correlation properties change
along the array.
29
Track C – Networks
Session: MPa5 – Smart Grid: Learning and Optimization
Chair: Gonzalo Mateos, University of Minnesota
MP5a-1
Dynamic Attacks on Power Systems Economic Dispatch
1:30 PM
Jinsub Kim, Lang Tong, Robert Thomas, Cornell University, United States
A dynamic data attack on a power system aimed at making the real time economic dispatch infeasible is considered. As a manin-the-middle attack, the attack modifies part of meter measurements such that the control center is misled with an incorrect
system state estimate, which affects the computation of real time economic dispatch. Two attack mechanisms are considered. The
first is an opportunistic approach where the attacker waits for a chance of a successful attack and launches an attack in a single
state estimation period. The second is a dynamic attack strategy where the attacker gradually drifts the system state toward the
infeasible region for real time economic dispatch. The efficacy of the proposed attacks is demonstrated by numerical experiments
with the IEEE 14-bus network and the IEEE 118-bus network.
MP5a-2
1:55 PM
Line Outage Detection in Power Transmission Networks Via Message Passing Algorithms
Jianshu Chen, University of California, Los Angeles, United States; Yue Zhao, Andrea Goldsmith, Stanford
University, United States; H. Vincent Poor, Princeton University, United States
Detecting multiple simultaneous line outages in power transmission networks is known to be a challenging problem due to the
number of hypotheses that grows exponentially with the network size. This paper proposes a low complexity message passing
algorithm that exploits the sparse structure of the network topology in power systems. First, a factor graph is established
that jointly characterizes the power system and the monitoring sensor network. Exploiting the power flow equations in
power systems, novel techniques that extend the conventional message passing algorithms are developed. Simulation results
demonstrate that the developed algorithm can effectively detect multiple simultaneous line outages.
MP5a-3
Online Learning Approaches for Dynamic Optimal Power Flow
2:20 PM
Seung-Jun Kim, Georgios Giannakis, University of Minnesota, United States
Optimal power flow is a critical control task for reliable and efficient operation of power systems. Significant challenges are
anticipated in future power systems, as a substantial level of renewable generation and vehicle electrification is accommodated,
imposing volatile and uncertain dynamics to the grid. In this work, online learning approaches, which do not require elaborate
models and yet provide provable performance guarantees, are adopted to tackle dynamic optimal power flow. The efficacy of the
proposed techniques is verified through numerical tests.
MP5a-4
Decentralized Primary Frequency Control in Power Networks
2:45 PM
Changhong Zhao, Steven Low, California Institute of Technology, United States
We augment existing generator-side primary frequency control with load-side control that are local, ubiquitous, and continuous.
The mechanisms on both the generator and the load sides are decentralized in that their control decisions are functions of locally
measurable frequency deviations. These local algorithms interact over the network through nonlinear power flows. We design
the local frequency feedback control so that any equilibrium point of the closed-loop system is the solution to an optimization
problem that minimizes the total generation cost and user disutility subject to power balance across entire network. With
Lyapunov method we derive a sufficient condition ensuring an equilibrium point of the closed-loop system is asymptotically
stable. Simulation demonstrates improvement in both the transient and steady-state performance over the traditional control only
on the generators, even when the total control capacity remains the same.
30
Track H – Speech, Image and Video Processing
Session: MPb5 – Image and Video Quality
Chair: Pamela C. Cosman, University of California, San Diego
MP5b-1
Real-Time 3D Rotation Smoothing for Video Stabilization
3:30 PM
Chao Jia, Zeina Sinno, Brian Evans, University of Texas at Austin, United States
We propose two real-time motion smoothing algorithms for video stabilization using a 3D rotational camera motion model. Both
algorithms aim at smoothing 3D rotation matrix sequences in a causal way. The first algorithm smooths the rotation sequences
in a way similar to 1st-order IIR filtering. The second uses sequential probabilistic estimation under a constant angular velocity
model. We exploit the manifold structure of the rotation matrices so that the proposed algorithms directly smooth the rotation
sequences on the manifold. In addition, we introduce a projection step in order to guarantee that no black borders intrude into the
stabilized video frames.
MP5b-2
3:55 PM
Joint Source-Channel Rate-Distortion Optimization with Motion Information Sharing for
H.264/AVC Video-Plus-Depth Coding
Yueh-Lun Chang, University of California, San Diego, United States; Yuan Zhang, Communication University of
China, China; Pamela Cosman, University of California, San Diego, United States
Video-plus-depth coding has been suggested as an efficient tool to support three-dimensional television. In this paper, we propose
a motion information sharing encoding scheme with an end-to-end rate-distortion model for H.264/AVC coding of videoplus-depth sequences. Experimental results with the proposed encoding scheme show PSNR gains of up to 1 dB for the depth
sequence under a packet loss environment.
MP5b-3
Image Assisted Upsampling of Depth Map via Nonlocal Similarity
4:20 PM
Wentian Zhou, Xin Li, Daryl Reynolds, West Virginia University, United States
The depth resolution of TOF cameras is poor and the resulting depth maps are noisy. Therefore, it is highly desirable to develop
powerful image processing tools to enhance the resolution and suppress the noise of depth maps. In this paper, we propose a new
image-assisted upsampling method for depth map. A spatially adaptive iterative singular-value thresholding (SAIST) with imageguided patch clustering strategy is developed and compared with previous image-guided depth map upsampling techniques.
Overall, the proposed scheme is capable of better preserving salient global structure information. Extensive experimental results
are reported to justify the superiority of the proposed method.
MP5b-4
Video De-Interlacing Using Asymmetric Nonlocal-Means Filtering
4:45 PM
Roozbeh Dehghannasiri, Texas A&M University, United States
This paper presents a de-interlacing method based on the nonlocal-means (NL-means) approach. In the NL-means, every
interpolated pixel is set to a weighted combination of its neighboring pixels. We use an asymmetric NL-means scheme where
the pixel in the neighborhood and the pixel being interpolated can be at the center or off-center of their patches. To calculate
NL-means weights, we initially find estimates of patches using a spatio-temporal edge-based method which employs a risk
function for interpolation. This function reflects the performance of the interpolation direction on the original adjacent pixels.
Experimental results verify the superior performance of our method.
31
Track E – Array Signal Processing
Session: MPa6 – Array Calibration
Chair: Visa Koivunen, Aalto University
MP6a-1
Bilinear Compressed Sensing for Array Self-Calibration
1:30 PM
Benjamin Friedlander, University of California, Santa Cruz, United States; Thomas Strohmer, University of
California, Davis, United States
We consider array self-calibration by joint estimation of directions-of-arrival and gain/phase parameters. Previous approaches
were mainly based on eignestructure methods (such as MUSIC) which have difficulties with correlated sources and small number
of snapshots. A different solution is presented here based on a sparse signal model and the use of mixed-norm minimization
of a cost function. Two approaches are presented: one based on alternating minimization and the other on tensor completion.
The proposed method does not have the disadvantages of eigenstructure based methods. It offers performance similar to that of
maximum likelihood estimation but has a lower computational cost.
MP6a-2
Calibrating Nested Sensor Arrays with Model Errors
1:55 PM
Keyong Han, Peng Yang, Arye Nehorai, Washington University in St. Louis, United States
We consider the problem of direction of arrival (DOA) estimation based on a nonuniform linear nested array. In practice, the
actual sensor gain and phase are often perturbed from their nominal values. In this paper, we investigate the calibration problem
for perturbed nested arrays, proposing corresponding robust algorithms to estimate both the model errors and the DOAs. The
partial Toeplitz structure of the covariance matrix is employed to estimate the gain errors, and the sparse total least squares is
used to deal with the phase error issue. Numerical examples are provided to verify the effectiveness of the proposed strategies.
MP6a-3
2:20 PM
A New Method for DOA Estimation in the Presence of Unknown Mutual Coupling of an
Antenna Array
Eric Wei-Jhong Ding, Borching Su, National Taiwan University, Taiwan
In this paper, we studied the problem of DOA estimation in the presence of mutual coupling. The proposed method for
reconstructing the steering vector can estimate the DOAs in the situation of a large number of source signals with unknown
mutual coupling. The main advantage of the proposed method is that the capacity of the number of source signals is higher than
other methods in the case of high accuracy system model. Moreover, the proposed method can work well by applying the spatial
smoothing technique even if source signals are coherent. Simulation results confirm the advantage of the proposed method.
MP6a-4
2:45 PM
An Angular Sampling Theorem for the Usable Frequency Range of Antenna Array
Calibration Measurements
Chung-Cheng Ho, Scott Douglas, Southern Methodist University, United States
For high precision direction-finding using antenna arrays, a highly-accurate calibration data set is required. This paper focuses
on the sampling requirements of such data sets to ensure an accurate direction-of-arrival (DOA) estimate, in which interpolation
across angle and frequency is assumed. Limited discussions of this issue can be found in the literature, and a quantified procedure
to avoid aliasing has not been presented. We establish a link between calibration angle spacing and the maximum operational
frequency of the array to avoid aliasing when computing DOAs from this calibration data. Numerical evaluations show the
validity of our geometric aliasing model.
32
Track E – Array Signal Processing
Session: MPb6 – Wireless Localization
Chair: Petar M. Djuric, Stony Brook University
MP6b-1
3:30 PM
Direct Localization of Emitters Using Widely Spaced Sensors in Multipath Environments
Nil Garcia, New Jersey Institute of Technology, United States; Marco Lops, Universita degli Studi di Cassino, Italy;
Martial Coulon, University of Toulouse, France; Alexander Haimovich, New Jersey Institute of Technology, United
States; Jason Dabin, Space and Naval Warfare Systems Command - Systems Center Pacific, United States
The problem of localizing multiple emitters using widely spaced sensors is commonly solved by estimating parameters such as
time-of-arrivals. Techniques in which sensors estimate intermediate parameters independently are suboptimal relative to direct
localization, where emitter locations are estimated jointly. We propose a direct localization method for frequency-selective
channels for known transmitted signals. The proposed method exploits the sparsity of the emitters as well as differences in the
properties of line of sight versus multipath components of signals received at the sensors. It is shown that the proposed method
has superior accuracy and is robust to sensors with blocked line-of-sight.
MP6b-2
Millimeter-Wave Personal Radars for 3D Environment Mapping
3:55 PM
Anna Guerra, Francesco Guidi, Davide Dardari, University of Bologna, Italy
The future availability of millimeter-wave technology in next generation smartphones will allow the exploitation of compact
massive antenna arrays thus enabling new interesting applications such as those related to the personal radar concept. In this
paper we analyze the possibility of accurate environment mapping through millimeter-wave personal radars. The impact on map
reconstruction accuracy of signal bandwidth, number of antennas, and non-idealities is investigated.
MP6b-3
Simultaneous Tracking and RSS Model Calibration by Robust Filtering
4:20 PM
Juan Manuel Castro-Arvizu, Universitat Politècnica de Catalunya, Spain; Jordi Vilà-Valls, Pau Closas, Centre
Tecnològic de Telecomunicacions de Catalunya, Spain; Juan Fernández-Rubio, Universitat Politècnica de
Catalunya, Spain
Received Signal Strength (RSS) localization is widely used due to its simplicity and availability in most mobile devices. The
RSS channel model is defined by the propagation losses and the shadow fading. These parameters might vary over time because
of changes in the environment. In this paper, the problem of tracking a mobile node by RSS measurements is addressed, while
simultaneously estimating a two-slope RSS model. The methodology considers a Kalman filter with Interacting Multiple Model
architecture, coupled to an on-line estimation of the observation’s variance. The performance of the method is shown through
numerical simulations in realistic scenarios.
MP6b-4
Proximity Detection with RFID in the Internet of Things
4:45 PM
Miodrag Bolic, Majed Rostamian, University of Ottawa, United States; Petar Djuric, Stony Brook University,
United States
In the “Internet of Things” (IoT), the things will be able to sense, communicate, and interact. They will also exchange data,
information and knowledge, and locate themselves and other things that surround them. In order to be able to interact, the things
need to recognize that they are in proximity of other things. It is anticipated that the most widespread components of the IoT
will be passive radio frequency identification (RFID) tags because they are inexpensive and provide automatic identification.
However, passive RFID tags are not seen as capable of performing complex operations, such as proximity detection and
localization, which will be required in future networks. In this paper, we describe existing problems with current RFID systems
and survey potential solutions for proximity detection. Then we present a new RFID device called “Sense-a-Tag” (ST) that can
passively detect and decode backscatter signals from tags in its proximity. We show that when STs are added to a standard RFID
system, the problems of proximity detection and localization in the IoT with RFID tags can readily be resolved. We demonstrate
the feasibility of an ST-based RFID system by experiments of proximity detection.
33
Track G – Architecture and Implementation
Session: MPa7 – Resource-aware and Domain-specific Computing
Chair: Frank Hannig, Friedrich-Alexander University Erlangen-Nurnberg
MP7a-1
1:30 PM
Partial Expansion of Dataflow Graphs for Resource-Aware Scheduling of Multicore Signal
Processing Systems
George Zaki, IGI Technologies, United States; William Plishker, Shuvra Bhattacharyya, University of Maryland,
College Park, United States; Frank Fruth, Texas Instruments, United States
The complex design spaces associated with state-of-the-art, multicore signal processing systems pose significant challenges in
realizing designs with high productivity and quality. The Partial Expansion Graph (PEG) implementation model was developed
to help address these challenges by enabling more efficient exploration of the scheduling design space for multicore digital signal
processors. The PEG allows designers and design tools to systematically adjust and adapt the amount of parallelism exposed from
applications depending on the targeted platform. In this paper, we develop new algorithms for scheduling and mapping systems
implemented using PEGs.
MP7a-2
1:55 PM
Performance Analysis of Weakly-Consistent Scenario-Aware Dataflow Graphs
Marc Geilen, TU Eindhoven, Netherlands; Joachim Falk, University of Erlangen-Nuremberg, Germany; Christian
Haubelt, Universität Rostock, Germany; Twan Basten, TU Eindhoven, Netherlands; Bart Theelen, TNO-ESI,
Netherlands; Sander Stuijk, TU Eindhoven, Netherlands
The timed dataflow model of computation is a useful performance analysis tool for Electronic System Level Design automation
and embedded software synthesis. It is used to model systems, including platform mapping and resource scheduling, of
components communicating and synchronizing in regular patterns. Its determinism gives it strong analysability properties and
makes it less subject to state-space explosion problems. Because of its monotonic temporal behaviour it can provide hard realtime guarantees on throughput and latency. It is expressive enough to cover a fairly large class of applications and platforms.
The trend however, in both embedded applications and their platforms is to become more dynamic, reaching the limits of what
the model can express and analyse with tight performance guarantees. Scenario-aware dataflow (SADF) is an extension that
allows more dynamism to be expressed, introducing a controlled amount of non-determinism into the model to represent different
scenarios of behaviour. The combination of a relatively infrequent switching between scenarios and still deterministic dataflow
behaviour within scenarios stretches the expressiveness of the model while keeping sufficient analysability. In this report we
investigate so-called weakly consistent graphs in which the scenario changes are not tightly coupled with periods of repetitive
behaviour of the static dataflow behaviour in scenarios as in previous methods. We define their semantics in terms of (max,+)
algebra and we introduce a method to analyse throughput using a generalisation of (max,+)-automata.
MP7a-3
2:20 PM
Application-driven Reconfiguration of Shared Resources for Timing Predictability of
MPSoC Platforms
Deepak Gangadharan, Ericles Sousa, Vahid Lari, Frank Hannig, Juergen Teich, University of Erlangen-Nuremberg,
Germany
The growing demand of compute-intensive applications has resulted in the widespread acceptance of heterogeneous MPSoC
platforms. The primary reason for this trend is due to the better performance and power efficiency exhibited by the heterogeneous
architectures consisting of standard processor cores and hardware accelerators. However, multiple processors accessing
shared resources such as cache, memory, and bus leads to significant contention on them, thereby decreasing the performance.
Moreover, the effect of shared resource contention worsens in the presence of multiple application scenarios having different
execution requirements. To mitigate this problem, we first propose an application-driven runtime reconfiguration approach for
shared resources to adapt to the varying execution requirements. We also present a performance analysis framework to take
the reconfiguration into consideration while providing timing guarantees for the applications. The results from the analytical
framework are then validated using experimental evaluation for different algorithms from the video processing domain.
34
MP7a-4
2:45 PM
Accelerating the Dynamic Time Warping Distance Measure using Logarithmic Arithmetic
Joseph Tarango, University of California, Riverside / Intel, United States; Eamonn Keogh, Philip Brisk, University
of California, Riverside, United States
We are designing an application-specific processor that has been customized for the dynamic time warping (DTW) application,
a widely used similarity measure in time-series data mining. Most DTW reference software is floating-point, and previous
hardware implementations either use floating-point or fixed-point binary arithmetic. In this paper, we will evaluate the possibility
to implement DTW using a logarithmic number system (LNS). We will rewrite portions of the algorithm to make them more
amenable to LNS arithmetic, and we will design customized arithmetic operations and instruction set extensions that employ
LNS. We believe that this will enable a cheaper and more efficient implementation of the algorithm in comparison with known
state-of-the-art techniques.
Track C – Networks
Session: MPb7 – Detection and Estimation for Networked Data
Chair: Yue Lu, Harvard University
MP7b-1
Detecting Convoys in Networks of Short-Range Sensors
3:30 PM
Sean Lawlor, Michael Rabbat, McGill University, Canada
We consider the problem of detecting groups of vehicles that are traveling together as a convoy. License plate recognition
sensors are an emerging technology which can be used to solve this problem. The sensors are deployed throughout road networks
across the world and meta-data about the vehicles passing in front of each sensor is collected. These provide discrete, irregularly
sampled, time series information about where vehicles are traveling. This paper proposes a method to solve the problem of
detecting convoys utilizing irregularly sampled time series information about objects moving between sensors.
MP7b-2
3:55 PM
Distributed SPRT for Gaussian Binary Hypothesis Testing: Performance Analysis and
Anit Sahu, Soummya Kar, Carnegie Mellon University, United States
This paper studies the problem of sequential Gaussian binary hypothesis testing in a distributed multi-agent network. A
distributed sequential probability ratio test (SPRT) type algorithm of the consensus-plus-innovations form is proposed, in which
the agents update their decision statistics by simultaneously processing local observations (innovations) sensed sequentially over
time and information obtained from neighboring agents (consensus). Local detection thresholds achieving pre-specified type I
and type II error probabilities are derived and large deviation exponents for the stopping time distributions are obtained.
MP7b-3
Denoising of Network Graphs using Topology Diffusion
4:20 PM
PDE based diffusion has been utilized for image denoising for more than two decades. It is known that the process of diffusion
preserves the edges and object boundaries making it a suitable preprocessing for edge detection. Synergetic to these efforts, in
this work, we apply diffusion to network graphs leading to an efficient algorithm for removing anomalous structures from the
network topology. The driving force for the diffusion process in our work is the clustering propensity that exists in real social
networks. The proposed diffusion enhances the boundaries of communities which makes it a suitable step prior to community
detection.
MP7b-4
4:45 PM
Optimal Hypothesis Testing with Combinatorial Structure: Applications in Graph
Detection
Yue M. Lu, Harvard University, United States
We consider hypothesis testing with combinatorial structure: the alternate and null hypotheses differ in an unknown subset of the
variables. For example, given Gaussian measurements on the nodes of a graph, we test whether one out of a collection of possible
subgraphs has elevated mean; or, in a time series of such measurements, whether there is a walk on graph with elevated mean.
We provide in many cases sharp bounds for generalized error exponents describing the error decay rate. We illustrate a phase
transition in the performance and show that if the SNR scales appropriately, genie-aided performance is achievable.
35
Track C – Networks
Session: MPa8 – Network Resource Allocation and Localization 1:30 PM–3:10
PM
Chair: Michael Rabbat, McGill University
MP8a1-1
Optimal Scheduling Policies and the Performance of the CDF Scheduling
PhuongBang Nguyen, Bhaskar Rao, University of California, San Diego, United States
We seek to characterize the performance properties of scheduling policies for wireless systems that are based on Cumulative
Density Functions (CDF) of the channels. We first derive optimal scheduling policies for a two-user system under max-sum rate
and max-min rate performance criteria subject to the same temporal resource constraints as those under CDF-based policies. The
behaviors of the CDF schemes are then compared against those of the optimal policies. We illustrate the differences in scheduling
decision boundaries as well as the sub-optimality in rate performance of the CDF-based policies.
MP8a1-2
Joint Interference and User Association Optimization in Cellular Wireless Networks
Changkyu Kim, Russell Ford, Sundeep Rangan, New York University, Polytechnic School of Engineering, United
States
In cellular wireless networks, user association refers to the problem of assigning mobile users to base station cells. This paper
considers a general class of utility maximization problems for joint optimization of mobile user associations and bandwidth
and power allocations. The formulation can incorporate a large class of network topologies, interference models, SNR-to-rate
mappings and network constraints. While the problem is nonconvex, our main contribution shows that the optimization admits
a separable dual decomposition. This property enables fast computation of upper bounds on the utility as well as an efficient,
distributed implementation for approximate local optimization via augmented Lagrangian techniques.
MP8a1-3
Throughput Maximization in Wireless Powered Communication Networks with Energy
Saving
Rui Wang, Donald Brown, Worcester Polytechnic Institute, United States
This paper considers optimal time and energy allocation to maximize the sum throughput of a wireless powered communication
networks for the case when the nodes can save energy for later time blocks in a time division multiple access scenario with
where a wireless access point transmits to a group of users which harvest the energy and then use this energy to transmit back to
the access point. To maximize the sum-throughput over L blocks, the initial optimization problem is separated into two subproblems, which can be solved efficiently.
MP8a1-4
Optimal Flow Bifurcation in Networks with Dual Base Station Connectivity and Non-ideal
Backhaul
Amitav Mukherjee, Hitachi America, Ltd., United States
In the dual connectivity architecture in LTE-Advanced systems, a user with multiple transceivers can be simultaneously
connected to two eNBs that operate independently at the radio layer. One of them is designated as a master eNB (MeNB) which
can choose to forward a portion of the UE traffic to the secondary eNB (SeNB) over a non-ideal backhaul link, and serve the
remaining data by itself. We examine how to optimally bifurcate the traffic flow when the MeNB has only statistical information
of the SeNB channel state information. A fluid model approximation is used to derive individual and global myopic policies.
MP8a1-5
Joint Sequential Target State Estimation and Clock Synchronization in Wireless Sensor
Networks
Jichuan Li, Arye Nehorai, Washington University in St. Louis, United States
Clock Synchronization is crucial to a wireless sensor network but often difficult to maintain. In this paper, we propose a joint
estimation method to estimate both target states and clock synchronization status based on sensor observations from a wireless
sensor network. We build a multi-sensor state-space model to connect clock synchronization status with sequential target state
transition. We divide the joint estimation problem into state and parameter estimation problems that alternate to update their
36
own estimates via the expectation-maximization algorithm. We use the Monte Carlo method in the state estimation to handle
nonlinear functions and non-Gaussian noise, and stochastic approximation in the EM algorithm to compensate for Monte Carlo
approximation error. Numerical examples demonstrate that the proposed joint estimation method yields lower estimation error
than a target estimation method.
MP8a1-6
High-Accuracy Vehicle Position Estimation Using a Cooperative Algorithm with Anchors
and Probe Vehicles
Ramez L. Gerges, First Responder System Testbed (FiRST), United States; John J. Shynk, University of California,
Santa Barbara, United States
This paper describes a method for achieving lane differentiation without requiring expensive GNSS receivers in every vehicle.
Lane identification on a highway can be used to optimize transportation network safety, by improving a vehicle’s ability to
avoid collision and to safely complete lane-change maneuvers. Using a cooperative technique, a vehicle communicates with road
infrastructure and nearby connected vehicles to achieve sub-meter lane identification. We assume that a small subset of vehicles
have high-accuracy GPS localization, which we refer to as probe vehicles. Other vehicles use location information from the probe
vehicles and fixed infrastructure landmarks to improve their position estimates.
MP8a1-7
Statistical Scheduling of Economic Dispatch and Energy Reserves of Hybrid Power
Systems with High Renewable Energy Penetration
Yi Gu, Huaiguang Jiang, University of Denver, United States; Yingchen Zhang, National Renewable Energy
Laboratory, United States; David Wenzhong Gao, University of Denver, United States
A statistical scheduling approach to economic dispatch and energy reserves is proposed in this paper. The proposed approach
focuses on minimizing the overall power production cost with considerations of renewable energy uncertainty, transmission
cost, and power system security. For our comprehensive numerical study, the proposed approach will be simulated and studied
in the IEEE 24-bus reliability test system (IEEE-RTS), which is the commonly used for evaluating power system stability and
reliability.
Track F – Biomedical Signal and Image Processing
Session: MPa8 – Bioinformatics and Medical Imaging
1:30 PM–3:10 PM
Chair: George Atia, University of Central Florida
MP8a2-1
Comparison and Integration of Genomic Profiles Predict Brain Cancer Survival and Drug
Targets
Katherine Aiello, Orly Alter, University of Utah, United States
Recently, we demonstrated the effectiveness of the generalized singular value decomposition (GSVD) in comparing patientmatched genomic profiles and identifying a pattern of DNA copy-number aberrations that correlates with glioblastoma survival.
Here, we show cross-platform validation of this signature, bringing it a step closer to the clinic. Surprisingly, we also find that the
prognostic value of the signature is maintained for lower-grade astrocytomas (LGAs). GSVD of LGA patient-matched profiles
uncovers molecular similarities and differences between glioblastoma and LGA that suggest drug targets.
MP8a2-2
Tensor GSVD for Comparison of Two Large-Scale Multidimensional Datasets
Theodore Schomay, Preethi Sankaranarayanan, Katherine Aiello, Orly Alter, University of Utah, United States
There exists a fundamental need for frameworks that can simultaneously compare and contrast large-scale data tensors, e.g.,
biomedical datasets recording multiple aspects of a disease across a set of patients. We describe a novel exact and unique
simultaneous decomposition for two tensors that generalizes the GSVD to a tensor GSVD (tGSVD). In an application to
biomedical datasets, we show that tGSVD comparisons of two ovarian cancer patient- and platform-matched genomic datasets
from The Cancer Genome Atlas predict survival and drug targets.
37
MP8a2-3
An Efficient ADMM-based Sparse Reconstruction Strategy for Multi-Level Sampled MRI
Joshua Trzasko, Eric Borisch, Paul Weavers, Armando Manduca, Phillip Young, Stephen Riederer, Mayo Clinic,
United States
Sparsity-driven image reconstruction is a promising paradigm for improving the spatial, temporal, and contrast resolution
in magnetic resonance imaging (MRI). However, high computational ex-pense continues to inhibit the translation of these
techniques into routine clinical practice. In many MRI acquisition protocols (e.g., time-resolved CAPR), the sampling operator
can be factored into a uniform and non-uniform component. In this work, we present a novel alternating direction methodof-multipliers (ADMM) strategy for sparse Cartesian SENSE-type MRI reconstruction that explicitly targets such multi-level
sampling protocols, and demonstrate that this framework ena-bles 3D contrast-enhanced MR angiogram (CE-MRA) sparse
reconstruction in only seconds.
MP8a2-4
Multiscale Functional Networks in Human Resting State Functional MRI
Jacob Billings, Emory University, United States; Alessio Medda, Georgia Tech Research Institute, United States;
Shella Keilholz, Georgia Institute of Technology / Emory University, United States
Recent advent of fast imaging techniques for MRI application allow whole brain coverage with sub-second resolution, opening
the door for new data-driven computational techniques that can harvest the information contained in the data. To this end, we
examine the use of wavelet based spectral decomposition and hierarchical clustering for resting state functional MRI analysis.
Wavelet packets naturally enable short time spectral decomposition with minimal temporal window lengths across multiple
frequency ranges, while hierarchical clustering is used for organizing broadband and filtered fMRI data into functional network.
This method was applied to human group data from 5 volunteers from the 1000 Functional Connectomes database. Preliminary
results show that this technique can produce spatially distinct networks on multiple scales
MP8a2-5
Piecewise Linear Slope Estimation
Atul Ingle, William Sethares, Tomy Varghese, James Bucklew, University of Wisconsin-Madison, United States
This paper presents a method for directly estimating slope values in a noisy piecewise linear function. By imposing a Markov
structure on the sequence of slopes, piecewise linear fitting is posed as maximum a posteriori estimation problem. A dynamic
program efficiently solves this by traversing a linearly growing trellis. A pseudo-EM algorithm to estimate model parameters
from data is proposed and its convergence behavior is analyzed. Ultrasound shear wave imaging is presented as the primary
application. The algorithm is general enough for applicability in other fields, as suggested by another application to estimation of
shifts in financial interest rate data.
MP8a2-6
Fast Magnetic Resonance Parametric Imaging via Model-Based Low-Rank Matrix
Factorization
Parisa Amiri Eliasi, New York University, Polytechnic School of Engineering, United States; Li Feng, Ricardo
Otazo, New York University, School of Medicine, United States; Sundeep Rangan, New York University,
Polytechnic School of Engineering, United States
Magnetic Resonance Parametric Imaging is a recently-proposed method that permits quantitative determination of MR
parameters such as the T1 and T2 relaxation times. In contrast to conventional MRI, one or more encoding parameters in the
RF excitation are randomly varied over the scan and tissue parameters are inferred from the temporal response to the excitation.
This work presents a novel low-rank model-based matrix-factorization method for joint reconstruction and parameter estimation
suitable for highly accelerated (i.e. highly undersampled) scans. The method is demonstrated on T2 cardiac breath-hold imaging
with varying spin echo times.
38
MP8a2-7
A Signal Model for Forensic DNA Mixtures
Ullrich Mönich, Massachusetts Institute of Technology, United States; Catherine Grgicak, Boston University,
United States; Viveck Cadambe, Yonglin Wu, Massachusetts Institute of Technology, United States; Genevieve
Wellner, Boston University, United States; Ken Duffy, National University of Ireland Maynooth, Ireland; Muriel
Médard, Massachusetts Institute of Technology, United States
For forensic purposes, short tandem repeat allele signals are used as DNA fingerprints. The interpretation of signals measured
from samples has traditionally been conducted by thresholding. More quantitative approaches have recently been developed, but
not for the purposes of identifying an appropriate signal model. By analyzing data from 643 single person samples, we develop
such a signal model. Our analysis suggests that additive noise is best modeled via log-normal distributions and that variability
in peak heights is well described as Gaussian. This is a crucial step towards the development of principled techniques for mixed
sample signal deconvolution.
Track E – Array Signal Processing
Session: MPa8 – Source Separation and Array Processing
1:30 PM–3:10 PM
Chair: Douglas Cochran, Arizona State University
MP8a3-1
Forward - Backward Greedy Algorithms for Signal Demixing
Nikhil Rao, Parikshit Shah, Stephen Wright, University of Wisconsin, United States
In many signal processing applications, one wishes to separate mixtures of signals. Common among these are the separation of
sparse and low rank components in image and video processing, sparse and group sparse models in multitask learning and spikes
and sinusoids in source separation problems. An underlying theme for performing such recovery is the notion of simplicity”
of the components with respect to a given basis or frame. For specific problems of interest, many methods exist to perform
recovery, but an approach that generalizes to arbitrary notions of simplicity has not been forthcoming. In this paper, we propose
a framework for signal demixing when the components are defined to be simple in a fairly arbitrary manner. Despite being very
general, our method remains computationally simple and can be used in a variety of practical applications.
MP8a3-2
An Extended Family of Bounded Component Analysis Algorithms
Huseyin Atahan Inan, Alper Tunga Erdogan, Koc University, Turkey
Bounded Component Analysis (BCA) is a recent concept proposed as an alternative method for Blind Source Separation
problem. BCA enables the separation of dependent as well as independent sources from their mixtures under the practical
assumption on source boundedness. This article extends the optimization setting of a recent BCA approach which can be used to
produce a variety of BCA algorithms. The article also provides examples of objective functions and the corresponding iterative
algorithms. The numerical examples illustrate the advantages of proposed BCA examples regarding the correlated source
separation capability over the state of the art ICA based approaches.
MP8a3-3
Source Separation in Noisy and Reverberant Environment using Miniature Microphone
Array
Shuo Li, Milutin Stanacevic, Stony Brook University, United States
In the unique framework that combines spatial sensing and independent component analysis, we recover the impinging acoustic
sources and estimate their direction of arrival on the miniature four microphone array. We examine and quantify the performance
of the proposed algorithm under different acquisition noise and reverberant conditions. With artificially generated microphone
signals using room acoustic model, the algorithm demonstrates over 10 dB separation in moderate reverberant environments.
With the recordings from a miniature microphone array in a typical conference-room environment, the algorithm demonstrates
over 10 dB separation performance if the angular distance between two speech sources is over 30 degrees.
39
MP8a3-4
Competitive Algorithm Blending for Enhanced Source Separation
Keith Gilbert, Karen Payton, University of Massachusetts Dartmouth, United States
This paper studies the performance of competitive blending of outputs of multiple source separation algorithms running in
parallel. The proposed blending system can operate in three modes; two that can incorporate prior information to control a source
separtion network or influence blending, and a third that operates in a fully blind mode. Although derivation of the blending
methods is given within the context of the instantaneous case, extension to the convolutive case is given. Results of example
applications show that the blending methods can produce an better overall set of source estimates than any constituent algorithm.
MP8a3-5
Design of Coprime DFT Arrays and Filter Banks
Chun-Lin Liu, P. P. Vaidyanathan, California Institute of Technology, United States
Coprime arrays in array processing offer an enhanced degree of freedom of O(MN) given M+N sensors, where M and N
are coprime and are associated with the interelement spacing and the number of sensors. Direction-of-arrival estimation and
beamforming are popular applications of coprime arrays. The performance in some of these applications is based on coprime
DFT filter banks (coprime DFTFBs), which cascade an M-channel DFTFB and an N-channel DFTFB to achieve equivalently
MN-channel filter banks. However, the practical design of coprime DFTFBs, which is important in applications of coprime
arrays, has not been well-studied. In this paper, the practical design for coprime DFTFBs is related to interpolated FIR (IFIR)
filter designs. Based on M, N, filter orders, and peak ripples, a systematic design procedure is proposed. Our design leads to an
additional parameter λ that controls tradeoffs between the bandwidths of passbands and stopbands. Different approximations for
transition bands are considered to estimate λ. A design example for different λ is also presented.
MP8a3-6
The Differential Geometry of Asymptotically Efficient Subspace Estimation
Thomas Palka, Raytheon, United States; Richard Vaccaro, University of Rhode Island, United States
Subspace estimation is often a prelude to parameter estimation. The underlying parameterization constrains the set of subspaces
of interest and the singular value decomposition, which is the maximum likelihood (ML) estimator when rank is the only
limitation, is not the ML subspace estimator for the parameter constrained problem. Using the Stiefel manifold formulation of
the standard problem we establish intrinsic Cramer-Rao bounds for the constrained subspace estimation problem. In addition we
establish an asymptotic ML formulation for the constrained problem which has a closed-form solution for the important special
case of harmonic signals on uniformly spaced sensor arrays.
MP8a3-7
Effects of Network Topology on the Conditional Distributions of Surrogated Generalized
Coherence Estimates
Lauren Crider, Douglas Cochran, Arizona State University, United States
The generalized coherence estimate has an established history as a test statistic in multiple-channel signal detection. Recent
work has considered the use of maximum-entropy matrix completion when elements are missing from the gram matrix from
which the generalized coherence estimate is computed. This is desirable in sensor network settings, for example, where direct
communication is not available between every pair of nodes in the network. This paper examines the role of network topology in
determining the conditional distributions of the statistic obtained by the matrix completion process under both signal-present and
signal-absent hypotheses.
MP8a3-8
Maximum Energy Sequential Matrix Diagonalisation for Parahermitian Matrices
Jamie Corr, Keith Thompson, Stephan Weiss, University of Strathclyde, United Kingdom; John McWhirter, Cardiff
University, United Kingdom; Ian Proudler, Loughbourgh University, United Kingdom
Sequential matrix diagonalisation (SMD) refer to a family of algorithms to iteratively approximate a polynomial matrix
eigenvalue decomposition. Key is to transfer as much energy as possible from off-diagonal elements to the diagonal per iteration,
which has led to fast converging SMD versions involving judicious shifts within the polynomial matrix. Through an exhaustive
search, this paper determines the optimum shift in terms of energy transfer. Although costly to implement, this scheme yields
an important benchmark to which limited search strategies can be compared. In simulations, multiple-shift SMD algorithms can
perform within 10% of the optimum energy transfer per iteration step.
40
Track G – Architecture and Implementation
Session: MPa8 – Digital Communications
1:30 PM–3:10 PM
Chair: James Glenn-Anderson, Supercomputer Systems Inc.
MP8a4-1
High-throughput DOCSIS Upstream QC-LDPC Decoder
Bei Yin, Michael Wu, Rice University, United States; Christopher Dick, Xilinx Incorporated, United States; Joseph
R. Cavallaro, Rice University, United States
In this paper, we propose a new scheduling scheme that can significantly improve the decoding throughput of the layered
min-sum LDPC (Low Density Parity Check) decoder for the high-throughput data over cable service interface specification
(DOCSIS) standard. The new scheduling scheme reduces the data dependency in the LDPC decoder by rescheduling the two-step
min calculation. The scheme also avoids the data conflict in the decoder. We then implement the decoder using fully parallel
architecture on Xilinx FPGA with high level synthesis tool. The resulting design can achieve 1.25\,Gbps throughput with
204.8\,MHz clock frequency and only 25\% device usage.
MP8a4-2
On the Performance of LDPC and Turbo Decoder Architectures with Unreliable Memories
Joao Andrade, Instituto de Telecomunicações, Universidade de Coimbra, Portugal; Aida Vosoughi, Guohui Wang,
Rice University, United States; Georgios Karakonstantis, Andreas Burg, Telecommunication Circuits Lab, EPFL,
Switzerland; Gabriel Falcao, Vitor Silva, Instituto de Telecomunicações, Universidade de Coimbra, Portugal; Joseph
R. Cavallaro, Rice University, United States
In this paper, we investigate and compare the impact of faulty bit-cells on the performance of a Turbo and LDPC channel decoder
based on realistic memory failure models. Our study determines the upper-bounds on the number of bit-cell faults in the different
memory modules and iterations that can be tolerated by the inherent error resilience of such codes. We also show how selective
protection of some bits in the various memory modules at specific decoding iterations can improve the error resilience under high
fault rates.
MP8a4-3
Successive Cancellation List Polar Decoder using Log-likelihood Ratios
Bo Yuan, Keshab K. Parhi, University of Minnesota, Twin Cities, United States
Successive cancellation list (SCL) decoding algorithm is a powerful method that can help polar codes achieve excellent errorcorrecting performance. However, the current SCL algorithm and decoders are based on likelihood or log-likelihood forms,
which render high hardware complexity. In this paper, we propose a log-likelihood-ratio (LLR)-based SCL (LLR-SCL) decoding
algorithm, which only needs half the computation and storage complexity than the conventional one. Then, based on the proposed
algorithm, we develop low-complexity VLSI architectures for LLR-SCL decoders. Analysis results show that the proposed LLRSCL decoder achieves 50% reduction in hardware and 98% improvement in hardware efficiency, respectively.
MP8a4-4
60 GHz Synthetic Aperture Radar for Short-Range Imaging: Theory and Experiments
Babak Mamandipoor, University of California, Santa Barbara, United States; Greg Malysa, Amin Arbabian,
Stanford University, United States; Upamanyu Madhow, University of California, Santa Barbara, United States;
Karam Noujeim, Anritsu Co., United States
We report on preliminary experimental results, and associated theory, for a 60 GHz synthetic aperture radar (SAR) testbed for
short-range (sub-meter) imaging. Our testbed consists of a monostatic radar with synchronized transmitter and receiver, with
lateral motion (over 10-30 cm) providing the SAR geometry, and range resolution provided by stepped frequency continuous
wave (SFCW) signals covering a band exceeding 6 GHz. Centimeter (cm) level resolution is achieved in both cross-range and
slant-range. Computation of the Cramer-Rao Lower Bound (CRLB) and normalized cross-correlation between responses to closeby scatterers show how the crossrange resolution is indeed limited by the SAR geometry.
41
MP8a4-5
A Systematic Procedure for Deriving Block-Parallel, Power Efficient, Digital Filter
Architectures for High- Speed Data Conversion
Paraskevas Argyropoulos, Hanoch Lev-Ari, Northeastern University, United States
A systematic and highly intuitive procedure for deriving the digital filter realization of an arbitrary block-parallel and/or
pipelined discrete-time transfer function is proposed. The method is based on simple block diagram and multi-rate properties
and is intended for low-power processing by reducing (a) the required system clock frequency and (b) the number of required
multipliers by facilitating arithmetic resource sharing. Several detailed design examples are provided. The technique and
examples presented are intended to serve as a high-speed filter realization reference for Digital and DSP ASIC developers.
MP8a4-6
Distributed Synchronization of a Testbed Network with USRP N200 Radio Boards
Gilberto Berardinelli, Jakob L. Buthler, Fernando M. L. Tavares, Oscar Tonelli, Dereje A. Wassie, Farhood
Hakhamaneshi, Troels B. Sørensen, Preben Mogensen, Aalborg University, Denmark
In this paper, we evaluate the feasibility of a distributed synchronization technique in a testbed network based on Universal
Software Radio Peripheral N200 (USRP N200) hardware boards. We consider a setup with 8 neighbor nodes, and different
deployment conditions (open space with line-of-sight and nodes located in different office rooms). Measurements collected on a
testbed server demonstrate the capability of our technique of aligning the timing of the nodes with a precision of few µs.
MP8a4-7
Design Study of a Short-Range Airborne UAV Radar for Human Monitoring
Sevgi Zubeyde Gurbuz, TOBB University of Economics and Technology, Turkey; Muhsin Alperen Bolucek,
Tunahan Kirilmaz, TUALCOM Communication and RF Technologies, Turkey; Unver Kaynak, TOBB University of
Economics and Technology, Turkey
UAVs have many attributes making them advantageous in many surveillance applications, such the control of borders against
illegal trafficking. An important task in border control is to monitor the activities of any people in the region. In particular,
discrimination between friendly and non-cooperative targets is of high importance. This work studies the top-level design and
tradeoffs involved in the development of a short-range UAV for human monitoring, such as type of UAV, power and weight
requirements, frequency and type of radar, and signal processing algorithms. Additionally, the compatibility of the TAN100
UAV for a human monitoring mission is evaluated.
MP8a4-8
Max-Min Fairness in Compact MU-MIMO Systems: Can the Matching Network Play a
Role?
Yahia Hassan, Armin Wittneben, ETH Zurich, Switzerland
We consider a MU-MIMO system with compact antenna array at the receiver that uses successive interference cancellation. Such
a system is characterized by spatial correlation, antenna coupling and noise correlation. For the two cases of time sharing and no
time sharing we design the matching network to alter the capacity region in order to maximize the minimum user rate. Our results
show substantial performance enhancement in comparison to the conventional matching network that ignores coupling and
the one maximizing the sum rate. We view our results as a step in connecting the two worlds of information theory and circuit
theory.
Track H – Speech, Image and Video Processing
Session: MPa8 – Image and Speech Processing
1:30 PM–3:10 PM
Chair: Linda S. DeBrunner, Florida State University
MP8a5-1
Acoustic Echo and Noise Cancellation using Kalman Filter in a Modiﬁed GSC Framework
Subhash Tanan, Karan Nathwani, Ayush Jain, Rajesh M Hegde, Indian Institute of Technology Kanpur, India; Ruchi
Rani, Abhijit Tripathy, Samsung R&D Institute India Delhi, India
In this paper a novel method for acoustic echo and noise cancellation in a generalized sidelobe canceler framework (GSC)
is described. The primary contribution of this work is the development of multichannel adaptive Kalman filter (MCAKF) in
a modified GSC framework. Additionally both the near end speech signal and noise is assumed to be unknown in this work.
42
In contrast to conventional GSC, the MCAKF developed in this context is used to estimate the noise and residual echo. The
performance of proposed method in terms of both perceptual evaluation and distant speech recognition on the ARCTIC database
is motivating.
MP8a5-2
Paper Texture Classification via Multi-Scale Restricted Boltzman Machines
Arash Sangari, William Sethares, University of Wisconsin-Madison, United States
The performance of two Restricted Boltzman Machine (RBM) algorithms are compared in the paper texture classification
application when utilizing a multi-scale Local Binary Pattern sampling method. In the first approach, a separate RBM is trained
for each texture-type to estimate the probability distribution. In the second approach, a Deep Belief Net, which consists of a
cascade of RBM layers, is used to extract texture features which are then input into a logistic regression layer. The classification
performance of the two methods are compared in detail.
MP8a5-3
Regularized Logistic Regression Based classification for Infrared Images
Golrokh Mirzaei, Mohsin M. Jamali, University of Toledo, United States; Jeremy Ross, Peter Gorsevski, Verner
Bingman, Bowling Green State University, United States
There is an increase in bird and bat mortality near wind farms. It is desirable to document and quantify behavior and activity of
birds/bats near wind farms. Infrared Imaging is a useful monitoring method. However, IR images do not provide information as
whether it is a bird, bat or insect. A logistic regression classifier is used to provide category information of targets. There are no
priori known database of images available for birds, and bats. We have extracted and used the features of targets for classification
purposes. A database of labeled category based on the features has been created.
MP8a5-4
Localizing Near and Far Field Acoustic Sources with Distributed Microphone Arrays
Martin Weiss Hansen, Jesper Rindom Jensen, Mads Græsbøll Christensen, Aalborg University, Denmark
In this paper, we consider acoustic source localization using distributed microphone arrays. TDOAs are estimated using a
recently proposed method based on joint DOA and range estimation. The TDOAs are used to estimate the location of an acoustic
source using a recently proposed method, based on a 4D parameter space. The performance of the proposed method for acoustic
source localization is compared to GCC-PHAT, and a method based on joint DOA and pitch estimation. Results show a decrease
in the error of the estimated position when joint DOA and range estimation is used for TDOA estimation.
MP8a5-5
Graph Wavelet Transform: Application to Image Segmentation
Alp Ozdemir, Selin Aviyente, Michigan State University, United States
Recently, there has been a lot of work on extending traditional signal processing methods to irregular domains such as graphs.
Graph wavelet transform offers a multiscale analysis of graphs similar to traditional wavelets. Similar to wavelets which are
effective at detecting transients in a signal, graph wavelets can be used to detect discontinuities of functions defined on graphs.
In this paper, we use this realization to propose a graph wavelet based approach to image segmentation. The images are first
transformed to the graph domain and the graph wavelet transform is used to detect the discontinuities in the pixel domain.
MP8a5-6
Histogram Transform Model Using MFCC Features for Text-Independent Speaker
Identification
Hong Yu, Zhanyu Ma, Beijing University of Posts and Telecommunications, China; Minyue Li, Jun Guo, Google,
Inc., Sweden
A novel text-independent speaker identification method is proposed in this paper. This method uses the MFCCs and the dynamic
information among adjacent frames as feature set to capture the speaker’s characteristics. The PDF of these super MFCCs
features is estimated by the recently proposed histogram transform (HT) method, which generated more train data in a randomly
transform way to realize the histogram PDF estimation and receded the discontinuity problem of the common multivariate
histograms computing. Compared to the conventional PDF estimation method, such as Gaussian mixture model, the HT model
shows promising improvement in a SI task.
43
Track E – Array Signal Processing
Session: TAa1 – High Dimensional and Large Volume Data Chair: Sergiy Vorobyov, Aalto University
TA1a-1
8:15 AM
Tensor Restricted Isomety Property for Multilinear Sparse System of Genomic Interactions
Alexandra Fry, Carmeliza Navasca, University of Alabama at Birmingham, United States
We model multilinear genomic interactions through a sparse tensor equation. We show that a tensor restricted isometry property
is necessary to find the sparse unique solution in the multilinear system. This solution will drastically reduce the number of
experiments needed to assess which genes and combinations of genes are necessary for viral replication. In addition, we also look
at a novel numerical method for approximating the sparse unique solution.
TA1a-2
Analysis of a Separable STAP Algorithm for Very Large Arrays
8:40 AM
Jie Chen, Feng Jiang, A. Lee Swindlehurst, University of California, Irvine, United States
Studies of massive MIMO in wireless communications have recently attracted significant attention. The benefits of very large
arrays can also be exploited in space-time adaptive processing (STAP). In this paper, we analyze the performance of a reduceddimension separable STAP algorithm. We study the performance of it for clairvoyant interference covariance matrices with
orthogonality assumptions on the steering vectors, and show that in the asymptotic sense this simplified scheme performs as well
as the fully adaptive STAP method. Appealing to random matrix theory, we finally do an analysis for the case that the covariance
matrix is estimated using secondary data.
TA1a-3
9:05 AM
Spatial-Temporal Characterization of Synchrophasor Measurement Systems - A Big Data
Approach for Smart Grid System Situational Awareness
Huaiguang Jiang, University of Denver, United States; Lei Huang, Electric Power Research Institute, China
Southern Power Grid, China; Jun Zhang, University of Denver, United States; Yingchen Zhang, National Renewable
Energy Laboratory, United States; David Wenzhong Gao, University of Denver, United States
An approach for fully characterizing a synchrophasor measurement system is proposed in this paper, which aims to provide
substantial data volume reduction while keep comprehensive information from synchrophasor measurements in time and spatial
domains. Specifically, the optimal synchrophasor sensor placement problem is modeled and solved. A fault detection and
identification approach approach is designed for monitoring the power systems. Several IEEE standard systems are employed to
validate and evaluate the proposed approach.
TA1a-4
9:30 AM
Performance Analysis of the Tucker HOSVD for Extracting Low-Rank Structure from
Multiple Signal-Plus-Noise Matrices
Himanshu Nayar, Rajesh Nadakuditi, University of Michigan, Ann Arbor, United States
The Tucker HOSVD (also known as Tucker3ALS , 3-Mode PCA and 3-Mode SVD) is a popular algorithm for uncovering
structure from tensor datacubes. This algorithm has been successfully used in many signal processing, machine learning and data
mining applications. In this work, we use recent results from random matrix theory to analyze the performance of the HOSVD
algorithm relative to an SVD based alternative . The analysis brings into sharp focus when and the extent to which the HOSVD
improves estimation performance. We illustrate the predicted performance improvement using numerical simulations and on a
background subtraction application from computer vision.
44
Track D – Signal Processing and Adaptive Systems
Session: TAb1 – Big Data Signal Processing
Chair: Georgios B. Giannakis, University of Minnesota
TA1b-1
10:15 AM
A Comparison of Clustering and Missing Data Methods for Health Sciences
Ran Zhao, Claremont Graduate University, United States; Deanna Needell, Claremont McKenna College, United
States; Christopher Johansen, Jerry Grenard, Claremont Graduate University, United States
In this paper, we compare and analyze clustering methods with missing data in health behavior research. In particular, we
propose and analyze the use of compressive sensing’s matrix completion along with spectral clustering to cluster health related
data. The empirical tests and real data results show that these methods can outperform standard methods like LPA and FIML, in
terms of lower misclassification rates in clustering and better matrix completion performance in missing data problems.
TA1b-2
10:40 AM
Discovery of Principles of Nature from Matrix and Tensor Modeling of Large-Scale
Molecular Biological Data
Orly Alter, University of Utah, United States
We will describe the use of matrix and tensor decompositions in comparing and integrating different types of large-scale
molecular biological data, from different studies of cell division and cancer and from different organisms, to computationally
predict previously unknown physical, cellular and evolutionary mechanisms that govern the activity of DNA and RNA. We will
present novel generalizations of the singular value decomposition as well as experimental verification and validation of some of
the computational predictions.
TA1b-3
Big Data Clustering Using Random Sampling and Consensus
11:05 AM
Panagiotis Traganitis, Konstantinos Slavakis, Georgios Giannakis, University of Minnesota, United States
The advent of Big Data has introduced new challenges and opportunities in the fields of signal processing and machine
learning. Clustering huge numbers of high-dimensional data is an undoubtedly complex and time consuming task even for
high performance computers. The present paper puts forth a novel randomized dimensionality reduction algorithm for k-means
clustering of high-dimensional data based on a random sampling and consensus approach. Simulations assess performance of the
proposed method, and compare it with state-of-the-art k-means techniques that rely on random projections.
TA1b-4
Classification of Streaming Big Data with Misses
11:30 AM
Fatemeh Sheikholeslami, Morteza Mardani, Georgios Giannakis, University of Minnesota, United States
Big Data’ classification is hindered by the large volume of data, missing observations, and the need for real-time processing.
This paper aims at learning a support-vector-machine (SVM) classifier from such data on the fly.’ Leveraging the decomposable
structure of the decision variable, a novel approach is put forth to cope with misses by treating them as unknown scalars. This
amounts to first nullifying the missing features, followed by running SVM with a slack variable per datum to rectify the possible
data missclassification. Preliminary tests corroborate the effectiveness of the novel approach.
Track F – Biomedical Signal and Image Processing
Session: TAa2 – Neural Spike Train Analysis
Chair: Rebecca Willett, University of Wisconsin-Madison
TA2a-1
8:15 AM
Neural Spike Train Denoising by Point Process Re-weighted Iterative Smoothing
Demba Ba, Massachusetts Institute of Technology, United States; Behtash Babadi, University of Maryland, College
Park, United States; Emery Brown, Massachusetts Institute of Technology / Harvard University, United States
We propose a novel iteratively re-weighted least-squares (IRLS) algorithm, termed Point process Re-weighted Iterative
SMoothing (PRISM), to solve the total variation denoising problem in one dimension using observations from a point process.
PRISM can be implemented using well-established point process analogues of the Kalman smoother. We use a connection
45
between the Expectation-Maximization algorithm and IRLS to prove the convergence of PRISM. We apply PRISM to neural
spiking data acquired from an epileptic patient during general anesthesia induced using the drug propofol, and demonstrate that
the algorithm can capture robustly the onset of loss of consciousness at the millisecond time scale.
TA2a-2
8:40 AM
Neurally Inspired Objective Function for Subspace Tracking and Online Feature Learning
Dmitri Chklovskii, Simons Center for Data Analysis, United States
In most approaches to subspace modeling, such as PCA or SVD, and feature learning, such as Sparse Dictionary Learning,
existing algorithms are derived by minimizing a (regularized) signal representation error. Although such approaches can be
straightforwardly implemented in the online setting, the objective function is typically non-convex and the neural network
implementation of such online algorithm often requires biologically implausible non-local learning rules. Here, we propose to
minimize the difference between the input and the modeled instance covariance matrix. We show that minimization of such a
cost function online naturally leads to a neural network architecture with local learning rules. Our derivation yields both Hebbian
and anti-Hebbian learning rules and predicts dependence of learning rate on activity history in agreement with biological
observations.
TA2a-3
Tracking Influence in Dynamic Neural Networks
9:05 AM
Rebecca Willett, University of Wisconsin-Madison, United States; Eric Hall, Duke University, United States
Cascading chains of spikes are a salient feature of many biological neural networks. This talk addresses the challenge of tracking
how the spikes within a neural network stimulate or influence future spike activity. We adopt an online learning framework wellsuited to streaming data, using a multivariate Hawkes model to encapsulate autoregressive features of observed spikes within the
neural network. Recent work on online learning in dynamic environments is leveraged not only to exploit the dynamics within
the neural network, but also to track that network structure as it evolves. Regret bounds and experimental results demonstrate that
the proposed method (with no prior knowledge of the network) performs nearly as well as would be possible with full knowledge
of the network.
TA2a-4
9:30 AM
A Design and Implementation Framework for Unsupervised High-resolution Recursive
Filters in Neuromotor Prosthesis Applications
Islam Badreldin, Karim Oweiss, University of Florida, United States
In cortically-controlled neuromotor prostheses, the initialization of decoders for motor impaired patients requires concurrently
measured neural and motor imagery/observation data. Additionally, the decoder implementation poses a scalability challenge
with an increasing number of decoded neurons. We propose a new decoder design framework in which the decoder initialization
is unsupervised, the decoder is implemented using recursive filters that can operate at high-resolution sampling of the neural data
minimizing the delay introduced in the system, and the decoder gives a smooth control signal expressed by the span of neural
data projections onto a low-dimensional space with desirable features for the control task.
Track F – Biomedical Signal and Image Processing
Session: TAb2 – Dynamic Brain Functional Connectivity
TA2b-1
10:15 AM
Functional Connectivity Differences in Brain Networks: Contributions of Shared and
Unshared Variance
Michael Cole, Rutgers University, United States; Grega Repovs?, University of Ljubljana, United States; Alan
Anticevic, Yale University, United States
Investigations of functional connectivity – the statistical dependence among brain activity time series – have provided important
insights into the brain systems underlying behavior and cognition. An increasing number of such studies gain these insights by
focusing on differences in functional connectivity – between groups, individuals, time periods, or task conditions. One important
theoretical issue unaddressed by current approaches is the distinction between shared and unshared variance. This distinction
is apparent when considering that adding random fluctuations to one of two tested time series (altering unshared variance)
consistently reduces functional connectivity estimates (e.g., correlations, mutual information, coherence), even though this
does not systematically affect the amplitude or number of corresponding fluctuations (shared variance) between those two time
series. This is inconsistent with the common notion of functional connectivity as the amount of inter-region communication.
46
We identified methods for isolating differences in shared variance using both simulations and functional magnetic resonance
imaging (fMRI). These new approaches may be important for proper characterization of brain network dynamics across groups,
individuals, time, and conditions, with many or all neuroscientific recording methods (e.g., fMRI, electroencephalography, local
field potentials, multi-unit recording).
TA2b-2
10:40 AM
Beyond Brain Maps: Functional Connectivity versus Task-Based Activations in Mental
State Prediction
Irina Rish, IBM T. J. Watson Research Center, United States
Functional connectivity gives rise to a wide range of network-based features that can be much more informative about a
particular mental state than the traditional task-based responses (voxel activations). For example, in our fMRI-based analysis
of schizophrenia, we observe disruptions of functional networks that cannot be explained by local (area-based) changes in
activations, are global in nature as they affect long-distance correlations, and can be also leveraged to achieve high classification
accuracy 93% when discriminating between schizophrenic vs control subjects. We also discuss several other examples, such as
cocaine addiction and pain, and functional network disruptions associated with such disorders.
TA2b-3
Approaches for Capturing Dynamic Connectivity States in fMRI data
11:05 AM
Vince Calhoun, University of New Mexico, United States
Most fMRI connectivity analysis is focused on generating maps of average connectivity over a 5-10 minute or longer scan
session. In this talk I will discuss recent efforts to capture transient, recurring connectivity states and a series of tools for making
inferences about individuals in this case. Such approaches enable us to evaluate how long certain individuals or groups spend in
each state or in one versus multiple states, and are an extremely promising way to study brain connectivity which is very likely
both highly dynamic within an individual over time and highly variable across individuals.
TA2b-4
Characterizing whole Brain Modulatory Interactions in Resting-State
11:30 AM
Bharat Biswal, New Jersey Institute of Technology, United States
Functional connectivity between two brain regions have been shown to be modulated by other brain regions using resting-state
functional magnetic resonance imaging (fMRI), which may provide novel insight on dynamics of brain connectivity. However,
the spatial distributions and specificity of the modulatory interactions have not been thoroughly examined. The current study
investigated modulatory interactions across 169 regions of interest (ROIs) across whole brain using physiophysiological
interaction (PPI) analysis on a resting-state fMRI dataset of 125 subjects. Among all the modulatory interactions cross whole
brain, there were considerably greater number of negative mdoulatory interactions than positive effects, e.g. in more cases the
increase of activity in one region was associated with decreased functional connectivity between two other regions. Lower
(medial) visual regions, the cingulate/medial frontal regions and the basal ganglia showed large numbers of positive modulatory
interaction effects, while higher (lateral) visual regions, the anterior cingulate and precuneus of the default mode network, and the
bilateral precental gyrus regions showed large numbers of negative effects. Lastly, the ROIs were divided into five modules based
on graph theory-based analysis, and we observed disproportionally higher number of significant positive modulatory interactions
between three regions within one module, suggesting increased within module processing efficiency through positive modulatory
interactions. In contrast, there were disproportionally higher number of significant negative modulatory interactions between
regions from two modules, suggesting a tendency of between modules segregation through negative modulatory interactions.
These results have implications on large scale brain network dynamics.
47
Track C – Networks
Session: TAa3 – Distributed Optimization over Networks
Chair: Philippe Ciblat, TELECOM ParisTech
TA3a-1
8:15 AM
The ADMM Algorithm for Distributed Averaging: Convergence Rates and Optimal
Parameter Selection
Euhanna Ghadimi, Andr´e Teixeira, Royal Institute of Technology-KTH, Sweden; Michael Rabbat, McGill
University, Canada; Mikael Johansson, Royal Institute of Technology-KTH, Sweden
We derive the optimal step-size and over- relaxation parameter that minimizes the convergence time of two ADMM-based
algorithms for distributed averaging. Our study shows that the convergence times for given step-size and over-relaxation
parameters depend on the spectral properties of the normalized Laplacian of the underlying communication graph. Motivated by
this, we optimize the edge-weights of the communication graph to improve the convergence speed even further. The performance
of the ADMM algorithms with our parameter selection are compared with alternatives from the literature in extensive numerical
simulations on random graphs.
TA3a-2
8:40 AM
Roula Nassif, Cédric Richard, André Ferrari, Université de Nice Sophia-Antipolis, France; Ali H. Sayed, University
of California, Los Angeles, France
Diffusion adaptation enables networked agents to interact cooperatively to estimate a common parameter of interest, or a
collection of related parameters that we refer to as tasks. In multitask learning, different clusters within the network are interested
in estimating their own parameter models, and interactions among adjacent clusters are motivated by their desire to enforce
certain measures of similarly between their tasks. In prior work on multitask diffusion adaptation, we focused on the behavior
of synchronous networks where all agents act synchronously in updating their estimates. In this work, we examine adaptation
and learning under asynchronous conditions where networks are subject to various sources of uncertainties such as changing
topologies, random link failures, and agents turning on and off randomly. We pursue a detailed mean-square analysis and
examine how asynchronous events interfere with the learning performance.
TA3a-3
9:05 AM
On the Convergence of an Alternating Direction Penalty Method for Nonconvex Problems
Sindri Magnússon, P. Chathuranga Weeraddana, KTH Royal Institute of Technology, Sweden; Michael Rabbat,
McGill University, Canada; Carlo Fischione, KTH Royal Institute of Technology, Sweden
Distributed and scalable methods for solving large scale structured convex optimization problems have recently been heavily
investigated. However, the requirement of scalability is also crucial for more challenging nonconvex problems, e.g. the optimal
power flow problem in smart grids and general consensus problems in multi-agent networks. Problem structures commonly
exploited to develop distributed/scalable algorithms for convex problems can often be found in large scale nonconvex problems
as well. This paper investigate approaches for solving separable optimization problems with linear coupling constraints, where
the objective functions are nonconvex.
TA3a-4
Decentralized Regression with Asynchronous Sub-Nyquist Sampling
9:30 AM
Hoi To Wai, Anna Scaglione, University of California, Davis, United States
When capturing a sensor field, local analog to digital converters are not coherent in time and are also characterized by a sampling
frequency offset. Some of these delays may be due to replay attacks in the system. The paper explores how to solve decentralized
optimization problems can lead to a consensus in the estimate of a latent parameter vector signal that is Band Limited, when the
sensor observations are plagued by heterogenous sampling frequency and time offsets.
48
Track B – MIMO Communications and Signal Processing
Session: TAb3 – Latest Coding Advances
Chair: Hamid Jafarkhani, University of California, Irvine
TA3b-1
Joint Space-Time Code Designs for Multiple Access Channels
10:15 AM
Tianyi Xu, InterDigital Communications, Inc., United States; Xiang-Gen Xia, University of Delaware, United States
A joint code design for multiple access channels (MAC) was proposed by G$\ddot{a}$rtner and B$\ddot{o}$lcskei, based on
the characterization of the error events. In this paper, we propose two new code designs. The first one has a better performance
and a higher symbol rate for two-user MAC with two transmit antennas at both users, while the same diversity order is achieved
as the design by G$\ddot{a}$rtner and B$\ddot{o}$lcskei when both users are in errors. The other one is a systematic design for
multiple users and multiple transmit antennas, which achieves the maximum diversity across all the users when errors occur in a
group of users.
TA3b-2
Quantized Distributed Reception Techniques for MIMO Wireless Systems
10:40 AM
Junil Choi, David Love, Purdue University, United States
Distributed reception techniques leverage nodes spread over a geographic area that can pass only a few bits of information to a
centralized processor. Using this quantized information, the centralized processor aims to recover the transmitted signal. In this
paper, we propose distributed reception techniques for multiple antenna spatial multiplexing transmission. We show how the
distributed array can efficiently recover the transmitted vector.
TA3b-3
11:05 AM
Generalized Spatial Modulation for Large-Scale MIMO Systems: Analysis and Detection
Theagarajan Lakshmi Narasimhan, Patchava Raviteja, Ananthanarayanan Chockalingam, Indian Institute of Science,
India
Generalized Spatial modulation (GSM) uses $n_t$ antenna elements but fewer radio frequency chains ($n_{rf}$) at the
transmitter. Spatial modulation and spatial multiplexing are special cases of GSM with $n_{rf}=1$ and $n_{rf}=n_t$,
respectively. In GSM, apart from conveying information bits through $n_{rf}$ modulation symbols, information bits are also
conveyed through the indices of the active $n_{rf}$ transmit antennas. In this paper, we derive analytical bounds on the codeword and bit error probabilities of maximum likelihood detection in GSM. The bounds are shown to be tight at medium to
high signal-to-noise ratios (SNR). We also present a low-complexity detection algorithm based on reactive tabu search (RTS)
for GSM in large-scale MIMO systems. Simulation results show that the proposed algorithm performs well and scales well in
complexity.
TA3b-4
11:30 AM
Bandwidth Analysis of Low-Complexity Decoupling Networks for Multiple Coupled
Antennas
Ding Nie, Bertrand Hochwald, University of Notre Dame, United States
In MIMO communication systems where coupled antennas are driven by independent amplifiers, a decoupling network is often
used to eliminate power reflection from the antennas. The realization of the decoupling network is not unique, and different
realizations have major differences in complexity and bandwidth. Recent designs for low-complexity networks show that
a minimum of $N^2+N$ impedances are generally needed to achieve decoupling for $N$ antennas. This paper provides a
perturbation analysis of the bandwidth, and shows that a decoupling network with $N^2+2N$ impedances can often achieve a
much wider bandwidth than $N^2+N$ impedances. We illustrate using a three-antenna example.
49
Track B – MIMO Communications and Signal Processing
Session: TAa4 – Enhanced MIMO for LTE-A and 5G Systems
Chair: Fred Vook, Nokia Siemens Networks
TA4a-1
8:15 AM
3D Channel Models for Elevation Beamforming and FD-MIMO in LTE-A and 5G
Jianzhong (Charlie) Zhang, Yang Li, Young-Han Nam, Samsung, United States
Elevation beamforming and Full Dimension MIMO (FD-MIMO) is an active area of research and standardization in 3GPP
LTE-A, and initial investigation has shown that promising (3-5x) cell capacity and cell edge gain can be achieved with these
new techniques. In an FD-MIMO system, a 2-dimensional (2D) active array is deployed to support both elevation beamforming
and conventional azimuth beamforming, which resulting in much higher MU-MIMO capability compared to conventional
MIMO systems. To evaluate effectiveness of elevation beamforming and FD-MIMO, a 3-dimention (3D) channel model has
been recently developed in 3GPP. This paper summarizes the latest 3D channel model development and highlights the main
differences compared to a legacy 2D channel model. Meanwhile, we also report several key insights gained from this recent study
in 3GPP on 3D channel model, including practical scenarios, system design considerations, as well as initial calibration results
and its implications.
TA4a-2
Advanced Antenna Solutions for 5G Wireless Access
8:40 AM
Erik Dahlman, Stefan Parkvall, David Astely, Hugo Tullberg, Ericsson, Sweden
The use of multiple transmit antennas at the base station and device side play an important role already in current 4G/LTE
systems, enhancing system performance and extending the data rates that can be provided to the end user. For the future (5G)
wireless-access solution advanced antenna solutions are expected to play an even more pronounced role. 5G wireless access
needs to provide substantially higher data rates, up to the multi-Gbps range in specific scenarios and with hundreds of Mbps
to be generally available in urban/suburban environments, as well handle traffic volumes hundreds of times higher than today.
Advanced multi-antenna transmission will be key to fulfill both the requirements. At the same time, 5G wireless is expected to
extend to frequency-range-of-operation beyond 10 GHz and into the mmw range. The corresponding smaller wave length will
be an enabler of more advanced antenna configurations with a much larger number of controllable antenna elements compared
to the antenna configurations of today. In this paper we discuss advanced antenna solutions for 5G wireless access, what are the
opportunities, alternatives, and possible obstacles.
TA4a-3
Multi-Layer Precoding for Full-Dimensional MIMO Systems
9:05 AM
Ahmed Alkhateeb, University of Texas at Austin, United States; Geert Leus, Delft University of Technology,
Netherlands; Robert W. Heath Jr., University of Texas at Austin, United States
Full-dimensional multiple-input multiple-output (FD-MIMO) systems boost spectral efficiency by offering orders of magnitude
increase in multiplexing gains. The overhead associated with estimating and acquiring channel state information for the large
number of antennas, however, is a performance limiting factor. In this paper, we consider FD-MIMO downlink channel, and
propose low-complexity channel estimation and 3D beamforming algorithms that exploit the spatial correlation of the large
MIMO channel. The performance of the proposed techniques is analyzed and compared with other FD-MIMO precoding
algorithms.
TA4a-4
Massive MIMO for mmWave systems
9:30 AM
Frederick Vook, Timothy Thomas, Nokia Solutions and Networks, United States
Multiple-Input, Multiple-Output (MIMO) technology has been successfully deployed on a wide scale in current “4G” cellular
systems and is expected to play a key role in “5G” systems, which will likely be rolled out in the year 2020 and beyond. Given
the current state of world-wide spectrum allocations for broadband communications, it is anticipated that a class of 5G systems
will be deployed in the cm-wave (3-30 GHz) and mm-wave (30-300 GHz) bands. As a result, the unique characteristics and
challenges of these bands have led to significant research efforts into the design and performance tradeoffs associated with
deploying MIMO technology in these bands. In this paper, we present a framework for beamforming and MIMO technology for
mm-wave systems were we focus on techniques for obtaining channel state information, various implementation and architecture
issues, and overall system performance. We focus on the application of large scale arrays with “Massive MIMO” to the
enhanced-local-area deployment scenario.
50
Track A – Communications Systems
Session: TAb4 – Cognitive Radio I
Chair: Paul de Kerret, Eurecom
TA4b-1
10:15 AM
Statistically Coordinated Precoding for the MISO Cognitive Radio Channel
Paul de Kerret, Miltiades Filippou, David Gesbert, Eurecom, France
We study a cognitive radio setting where the two transmitters aim at coordinating to maximize the rate of the secondary user
subject to a primary rate constraint. Considering a realistic channel state information scenario where each transmitter has solely
access to the instantaneous knowledge of its direct channel, we let the transmitters exploit their statistical knowledge of the multiuser channel to coordinate. This setting gives rise to a Team Decisional problem. We develop a novel coordination scheme where
the transmitters coordinate without any exchange of information or any iteration.
TA4b-2
10:40 AM
Simultaneous Detection and Estimation based Spectrum Sharing in Cognitive Radio
Networks
Jyoti Mansukhani, Priyadip Ray, Indian Institute of Technology Kharagpur, India; Pramod Varshney, Syracuse
University, United States
A new spectrum sharing strategy based on coupled detection and estimation is proposed for cognitive radio networks. The
strategy consists of the following steps: First the secondary user (SU) listens to the spectrum allocated to the primary user (PU)
to detect the state of the PU and transmits if the PU is inactive. However, if the PU is active, the SU estimates the PU location
and then makes a decision about the reliability of the estimate. The SU transmits via beamforming, with a null in the estimated
direction of the PU, only when the estimate is classified as reliable.
TA4b-3
11:05 AM
Interference-Temperature Limit for Cognitive Radio Networks with MIMO Primary Users
Cristian Lameiro, University of Cantabria, Spain; Wolfgang Utschick, Technische Universität München, Germany;
Ignacio Santamaria, University of Cantabria, Spain
In this paper, we derive the interference-temperature (IT) limit for a multi-antenna primary user (PU) that has a rate constraint.
While in the case of a single-stream PU there is a one-to-one mapping between IT and achievable rate, this property does not
hold anymore when a multiple-input multiple-output (MIMO) system is considered. For a MIMO user, we derive a closed-form
expression for the maximum IT that can be tolerated by identifying the worst-case interference covariance matrix, which results
in a multilevel waterfilling problem.
TA4b-4
Competitive Dynamic Pricing under Demand Uncertainty
11:30 AM
Yixuan Zhai, Qing Zhao, University of California, Davis, United States
We consider a multi-seller dynamic pricing problem with unknown demand models. In this problem, each seller offers prices
sequentially to a stream of potential customers and observes either success or failure in each sale attempt. The underlying demand
model is unknown and can take a finite number of possible forms. The problem is formulated as a repeated game with incomplete
information. Based on likelihood ratio test, we develop a dynamic pricing strategy that leads to an efficient sequential Nash
Equilibrium and offers a finite regret with respect to the ideal case of a known demand model.
51
Track H – Speech, Image and Video Processing
Session: TAa5 – Recent Advances in Speech Coding
Chair: Tokunbo Ogunfunmi, Santa Clara University
TA5a-1
8:15 AM
Large Margin Nearest Neighborhood Metric Learning for I-Vector Based Speaker
Verification
Waquar Ahmad, Harish Karnick, Rajesh M Hegde, Indian Institute of Technology Kanpur, India
A new large margin nearest neighborhood metric learning (LMNN) method for i-vector based speaker verification is proposed
in this paper. In general, a verification decision is taken by computing the cosine distance between the i-vectors of the test
utterance and the claimed identity. LMNN metric is learned from the examples and can be viewed as a linear transformation of
the input i-vector space of the training and test utterance. In this work, the metric is learned with the objective of reducing the
distance between the i-vectors of same class of speaker, while impostors are separated by a large margin. The metric learned in
this manner leads to a better speaker verification performance. Speaker verification experiments are then conducted on the NIST
2008 and YOHO speaker verification databases. Experimental results indicate a reasonable improvement in performance, when
compared to i-vector based speaker verification methods which use conventional cosine scoring.
TA5a-2
8:40 AM
Performance Enhanced Scalable Wideband Speech Coding for IP Networks
Tokunbo Ogunfunmi, Koji Seto, Santa Clara University, United States
Abstract—The scalable wideband speech coding scheme based on the internet low bitrate codec (iLBC) was previously presented
and achieved speech quality equivalent to ITU-T G.729.1 at high bit rates for wideband signals. The codec adopted a split-band
structure which employed the narrowband codec based on the iLBC coding scheme to encode both the lower- and higher-band
signals. This structure provided the interoperability with the narrowband codec in a core layer and achieved high speech quality
at high bit rates; however, the performance was limited at low bit rates. In this paper, we explore various approaches to improve
performance under both clean and lossy channel conditions. The bit rate scalable structure is used to achieve flexibility in terms
of bit rates for voice over IP (VoIP) applications as previously employed, whereas the use of different coding schemes for the
core layer and the enhancement layers is explored without regard to the interoperability. The proposed codec is expected to be
more robust to packet loss than state-of-the-art wideband codecs and still achieves similar voice quality to those codecs under
clean channel condition.
TA5a-3
9:05 AM
Adaptive Control of Applying Band-Width for Post Filter of Speech Coder Depending on
Pitch Frequency
Hironobu Chiba, Univ. of Tsukuba, Japan; Yutaka Kamamoto, Takehiro Moriya, Noboru Harada, Nippon Telegraph
and Telephone Corp., Japan; Shigeki Miyabe, Takeshi Yamada, Shoji Makino, Univ. of Tsukuba, Japan
Most speech codecs utilize a post-filter that emphasizes pitch structures to enhance perceptual quality at the decoder. Particularly,
the bass post-filter used in ITU-T G.718 performs an adaptive pitch enhancement for a lower fixed frequency band. This paper
describes a new post-filtering method to improve the bass post-filter by means of a frame-by frame adaptive control of frequency
band and gain depending on the pitch frequency of decoded signal. We have confirmed the improvement of the speech quality
with the developed method through the objective and subjective evaluations.
TA5a-4
9:30 AM
Classification of Sonorant Consonants Utilizing Empirical Mode Decomposition
Ashkan Ashrafi, San Diego State University, United States; Stanley Wenndt, Air Force Research Laboratory, United
States
In this paper, a method to classify nasal utterance among sonorant consonants utilizing empirical mode decomposition (EMD)
is introduced. In this method, each audio signal is divided into overlapping 20 millisecond frames. Then each frame’s signal
is decomposed by using the EMD. Four different features are extracted from each frame to create a vector. These vectors
are employed to train a support vector machine (SVM) with radial basis functions. A different set of audio signals are used
to validate the SVM model. The results show an overall correct identification rate of 91.19\% for nasals and 89.74\% for
semivowels
52
Track H – Speech, Image and Video Processing
Session: TAb5 – Historic Photographic Paper Identification via Textural
Similarity Assessment
Co-Chairs: Andrew G. Klein, Worcester Polytechnic Institute and Patrice Abry, Ecole
Superieure de Lyon (CNRS)
TA5b-1
Automated Surface Texture Classification of Photographic Print Media
10:15 AM
Paul Messier, Paul Messier LLC, United States; Richard Johnson, Cornell University, United States
This paper introduces a project to automatically characterize the surface texture of inkjet and black and white photographic
printing media. Texture is a key feature in the manufacture and use of photographic paper. Raking light photomicrographs reveal
texture through a stark rendering of highlights and shadows. Two raking light datasets were created for inkjet and black and white
print media. Using different approaches, four university teams were successful in detecting affinities and outliers built into these
datasets, demonstrating the feasibility of automatic texture classification. These methods have applications for cultural heritage
scholarship and for industrial quality control and design.
TA5b-2
Eigentextures: An SVD Approach to Automated Paper Classification
10:40 AM
William Sethares, Atul Ingle, Tomas Krc, University of Wisconsin, United States; Sally Wood, Santa Clara
University, United States
This paper investigates a method of texture classification using the eigentexture approach. This method assembles a collection
of small patches from each class of photographic paper; the patches are gathered into a large matrix and simplified so as to
retain only the most relevant eigendirections using an SVD. During the classification stage, a number of similarly-sized patches
are drawn from an unknown photographic paper. Each of these patches is compared to the classes, and a similarity measure is
created by a kind of voting procedure. Whichever class receives the most votes is the most likely class. This paper provides two
kinds of analyses. First is a rate-of-convergence result that can be used to estimate parameters within the model and that can be
used to estimate the quality of the classification. Second is an analysis of a randomized-resampling implementation that reduces
the computational burden as the algorithm is scaled for larger data sets. The advantages and disadvantages of this procedure are
investigated in the Historic Photo Paper Classification dataset and in a collection of inkjet papers organized to compare different
classification methods.
TA5b-3
Texture Classification via Area-Scale Analysis of Raking Light Images
11:05 AM
Andrew G. Klein, Western Washington University, United States; Anh Do, Christopher Brown, Worcester
Polytechnic Institute, United States; Philip Klausmeyer, WAM, United States
An image processing algorithm for photographic paper texture classification is developed based on area-scale analysis. This
analysis has been applied in surface metrology, and relies on the fact that the measured area of a surface depends on the scale
of observation. By comparing relative areas at various scales, the technique can compute topological similarity of two surfaces.
Results show the algorithm is successful in detecting affinities among similarity groupings within the HPPC dataset.
TA5b-4
11:30 AM
Hyperbolic Wavelet Transform for Historic Photographic Paper Classification Challenge
Stephane Roux, Patrice Abry, ENS Lyon, France; Herwig Wendt, ENSHEEIT-IRIT, France; Stephane Jaffard, Paris
Est University, France
Photographic paper texture characterization constitutes a challenging image processing task and an important stake both for
manufacturers and art museums. The present contribution shows how the Hyperbolic Wavelet Transform, thanks to its joint
multiscale and anisotropic natures, permits to achieve an accurate photographic paper texture analysis. A cepstral-type distance,
constructed on the coefficients of the Hyperbolic Wavelet Transform, is then used to measure similarity between pairs of paper
textures. Postprocessing of the the similarity matrix enables to achieve a relevant non supervised classification of photographic
papers. This methodology is applied to a test dataset made available in the framework of the Historic Photographic Paper
Classification Challenge, led by the Museum of Modern Art (NYC).
53
Track E – Array Signal Processing
Session: TAa6 – Compressive Methods in Radar
Chair: Athina Petropulu, Rutgers University
TA6a-1
Sparse Arrays, MIMO, and Compressive Sensing for GMTI Radar
8:15 AM
Haley Kim, Alexander Haimovich, New Jersey Institute of Technology, United States
Despite advances in radar, important gaps remain in the ability of ground moving target indicator (GMTI) radars to detect
slow moving targets embedded in ground clutter. This work proposes a GMTI radar combining synergistic elements from four
themes: space-time adaptive processing (STAP), random arrays, multiple-input multiple-output (MIMO) radar, and compressive
sensing. STAP supports joint space-time processing for detecting moving targets in ground clutter. Large, random arrays are
undersampled adaptive arrays that support improved angle-Doppler resolution and lower minimum detectable velocity (MDV), at
the cost of higher sidelobes. Even though random arrays have been studied as early as the 1970’s, new results reveal interesting
links to compressive sensing. MIMO provides further improvements in angular resolution and MDV, while supporting waveform
diversity and additional savings in the number of sensors. Compressive sensing algorithms are designed to cope with ambiguities
introduced by undersampling. We propose an algorithm for target detection and analyze its performance for detecting, slow
ground targets.
TA6a-2
8:40 AM
Efficient Linear Time-Varying System Identification Using Chirp Waveforms
Andrew Harms, Duke University, United States; Waheed Bajwa, Rutgers University, United States; Robert
Calderbank, Duke University, United States
Linear time-varying systems (LTV) are operators that impart a time shift and frequency shift to a probing waveform and are
important models in applications such as radar and channel estimation. In this paper, we present a novel scheme for efficient
identification of LTV systems that uses linear frequency modulated (LFM) pulses. In previous work, we showed that this scheme
offers asymptotically perfect identification, even in the presence of noise, with a diverse selection of LFM pulses. In this work,
we show that we can adaptively choose the LFM pulses to efficiently eliminate ambiguity in the LTV system identification.
TA6a-3
9:05 AM
Robust Multipath Exploitation Radar Imaging in Urban Sensing Based on Bayesian
Compressive Sensing
Qisong Wu, Yimin Zhang, Moeness Amin, Fauzia Ahmad, Villanova University, United States
Radar-based urban sensing has attracted considerable attention in many civil and military applications. In particular, throughthe-wall radar imaging (TWRI) has the capability of acquiring high-resolution images of targets of interest behind an opaque
obstacle. Compressive sensing (CS) techniques provide an effective means for this purpose because target activities in TWRI
applications are typically sparse. Among different CS techniques that are currently available, those based on sparse Bayesian
learning generally demonstrate increased robustness against dictionary coherence and noise when compared their counterparts
based on greedy and basis pursuit algorithms. In particular, the multi-task Bayesian CS techniques provide effective solutions
to a large class of group sparse problems due to, for example, frequency-dependent reflectivity. An important class of group
sparsity in TWRI applications stems from the existence of multipath propagation. Improper processing of unresolved multipath
signals due to wall reflections may yield ghost targets that undesirably clutter the scene. One of the effective techniques to avoid
such problems is multipath exploitation. In CS, this can be implemented by incorporating the multipath propagation model
in the sensing dictionary. To achieve effective exploitation of unresolvable multipath signals, existing CS techniques assume
perfect knowledge of the wall positions that are embedded in the dictionary matrix. In practice, however, the assumed wall
positions generally have errors. Even a small error in the wall position, on the order of a fraction of the wavelength, would yield
considerable phase error in the wall reflected waves, generating discrepancies between the assumed sensing dictionary and the
actual scene and, thereby, reducing the imaging quality and sparse scene recoverability. The main objective of this paper is to
develop a robust multi-task Bayesian CS technique that accounts for the wall position uncertainties and provides robust highresolution target images in such scenarios.
54
TA6a-4
9:30 AM
Joint Sparse and Low-rank Model for Radio-Frequency Interference Suppression in Ultrawideband Radar Applications
Lam Nguyen, Army Research Laboratory, United States; Minh Dao, Trac Tran, Johns Hopkins University, United
States
Radio-frequency interference is the most common, and also the most challenging type of interference or noise source that has a
direct impact on the performance of ultra-wideband radar systems in various practical application settings. This paper proposes a
robust and adaptive technique for the separation and then suppression of RFI signals from ultra-wideband radar data via modeling
RFI as low-rank components in a joint optimization framework. We advocate a joint sparse-plus-low-rank recovery approach
that simultaneously solves for (i) UWB radar signals as sparse representations with respect to a dictionary containing transmitted
waveforms; and (ii) RFI signals as a low-rank structure.
Track E – Array Signal Processing
Session: TAb6 – Statistical Inference in Smart Grids
Co-Chairs: H. Vincent Poor, Princeton University and Yue Zhao, Stanford University
TA6b-1
10:15 AM
Revisiting Cyclo-Stationary Random Signal Analysis for Modeling Renewable Power
Masood Parvania, University of California, Davis, United States; Francesco Verde, Universita’ Federico II di
Napoli, Italy; Anna Scaglione, University of California, Davis, United States; Donatella Darsena, Giacinto Gelli,
Universita’ Federico II di Napoli, Italy
In this paper we revisit the theory of cyclo-stationary processing to study multivariate cyclo-stationary random processes and
of their derivatives. The objective is to model Wind and Solar power generation and of their ramping characteristics, going
beyond the essential but limiting task of forecasting. The models are are meant to quantify the risk of synchronizing demand and
generation to the random excursions or renewable power using a mutli-settlement dispatch. The also can help the compression
and communication of local sensor data in real time.
TA6b-2
10:40 AM
Integrating PMU-data-driven and Physics-based Analytics for Power Systems Operations
Yang Chen, Le Xie, P. R. Kumar, Texas A&M University, United States
This work studies the data-driven approaches using phasor measurement unit (PMU) data, and the integrations with the physicsbased analytics. The principal component analysis (PCA) based dimensionality reduction is first applied to explore the underlying
dimensionality of power systems from the massively deployed PMU data. Then the physical interpretations are analyzed for the
power engineering insight: spatial interpretation suggests the coherency of generators; temporal analysis indicates the time-scale
separations of power system operations. Numerical examples using realistic PMU data are conducted to demonstrate the results.
TA6b-3
11:05 AM
Sensor Placement for Real-Time Dynamic State Estimation in Power Systems: A Structural
Systems Approach
Pedro Rocha, University of Porto, Portugal; Sergio Pequito, Carnegie Mellon University, United States; Pedro
Aguiar, Paula Rocha, University of Porto, Portugal; Soummya Kar, Carnegie Mellon University, United States
This paper studies sensor placement design for efficient dynamic real-time state estimation in electric power networks. Given a
(linearized) dynamic physical model of the power system, efficient sensor placement strategies are proposed that minimize the
observability-index of the system. The observability-index plays a key role in determining the minimum window length of filters
that guarantee stable estimation error. Moreover, the problem addressed in the structural systems framework, i.e., the placement
strategies are obtained on the basis of the sparsity pattern of the system coupling matrix, and the design guarantees hold for
almost all numerical parametric realizations of the system.
55
TA6b-4
11:30 AM
Dynamic Joint Outage Identification and State Estimation in Power Systems
Yue Zhao, Stanford University, United States; Jianshu Chen, University of California, Los Angeles, United States;
Andrea Goldsmith, Stanford University, United States; H. Vincent Poor, Princeton University, United States
Joint outage identification and state estimation in a dynamically evolving power system is studied. A linear dynamic system
model is employed that characterizes the state evolution in a power system. The evolving joint posterior of outage hypotheses
and network states are developed in closed form. Metrics that characterize the real-time and steady state performance of optimal
joint detection and estimation are derived. Based on the developed metrics, sensor locations for optimizing the joint detection
and estimation performance are found efficiently. Simulation results demonstrate significant performance gains from using the
optimal joint detector and estimator with the optimal sensor locations.
Track G – Architecture and Implementation
Session: TAa7 – Computer Arithmetic I
Chair: Neil Burgess, ARM Inc.
TA7a-1
8:15 AM
Ultra-Light Weight Hardware Accelerator Circuits for Data Encryption in Wearable
Systems
Sanu Mathew, Sudhir Satpathy, Vikram Suresh, Ram Krishnamurthy, Intel Corporation, United States
Securing data on energy-constrained wearable systems requires a rethink on how we build hardware accelerators than can operate
in ultra-low area/power systems. This paper will describe the design and implementation of a 2090-gate nanoAES hardware
accelerator that uses Galois-field polynomial-based micro-architectural co-optimization to achieve 11x higher energy-efficiency,
while offering the full security of AES-128. This design represents the minimal hardware require to support the AES algorithm
with integrated round-key generation, achieving nominal encrypt/decrypt throughputs of 432/671Mbps, measured at 0.9V, 25C.
TA7a-2
Arithmetic Operations in the Heterogeneous System Architecture
8:40 AM
Michael Schulte, AMD Research, United States
The Heterogeneous System Architecture (HSA) provides specifications to seamlessly integrate CPUs, GPUs, and other
computing elements through shared virtual memory, user mode queuing, the HSA intermediate language (HSAIL), cache
coherence, and signaling. This paper describes HSA arithmetic operations, discusses how these operations can be efficiently
implemented, and provides ideas for future computer arithmetic research for heterogeneous systems.
TA7a-3
Low Latency is Low Energy
9:05 AM
David Lutz, Neil Burgess, ARM, United States
For an out-of-order CPU, the multi-cycle execution units (e.g., multipliers or dividers or floating-point adders) consume only a
small fraction of the area and power used by the core. By reducing the latency of these units, we can often complete a task early,
saving large amounts of total energy. We describe a simple formula to gauge the relationship between execution unit power,
core power, execution unit latency, and overall energy required for a task. We then use that formula to argue for lower latency
execution, even if the power of the execution units has to be increased.
TA7a-4
Optimizing DSP Circuits by a New Family of Arithmetic Operators
9:30 AM
Javier Hormigo, Julio Villalba, Universidad de Malaga, Spain
A new family of arithmetic operators to optimize the implementation of circuits for digital signal processing is presented. Thanks
to use of a new technique which reduces the quantification errors, the proposed operators may decrease significantly the size of
the circuits required for most applications. That means a simultaneous reduction of area, delay and power consumption.
56
Track B – MIMO Communications and Signal Processing
Session: TAb7 – MIMO Sensing
Chair: Jian Li, University of Florida
TA7b-1
Bi-Static MIMO Radar Operations for Range-Folded Clutter Mitigation
10:15 AM
Yuri Abramovich, WR Systems Ltd., United States; Gordon Frazer, DSTO, Australia; Geoffrey San Antonio, Naval
Research Laboratory, United States; Ben Johnson, Colorado School of Mines, United States
In certain radar applications range-folded clutter returns can significantly degrade radar performance if the properties of the
range-folded clutter are significantly different from the properties of the “in-range” clutter. Typical examples include skywave
over-the-horizon radar (OTHR) operating in high waveform repetition (aircraft detection) mode and sited such that either
Northern Aurora or Equatorial Anomaly reflections containing spread Doppler clutter (SDC) are present at the ranges of the
second or higher range ambiguity. There are a number of potential solutions to this problem including judicious selection of
waveform repetition frequency, the use of non-recurrent waveforms, and the method we discuss more fully in this paper, that
of exploiting the bistatic separation of the system transmitter and receiver that is typical in OTHR. No single approach to this
problem is suitable in every case so it is important that the radar designer implement a system with the scope to introduce
the appropriate technique in any given circumstance. When ionosheric propagation conditions permit it is possible to exploit
the significantly bi-static location of the transmitting (Tx) and receiving (Rx) positions. With appropriately chosen bi-static
geometry, both Tx and Rx arrays finger beams overlap at one-range ambiguity operational distances, while the range-folded
reflections from the second and higher range ambiguities SDC arrive at the Rx beampattern sidelobes that can be properly
controlled. Apart from more restrictive propagation conditions that have to be equally good to support the two geographically
different Tx and Rx sites, conventional (SIMO) implementation of this principle, has one further problem. The sidelobe level of a
typical modern OTHR Tx antenna array, consisting of 16-24 elements, is typically not low enough to completely ignore rangefolded spread clutter illuminated by these sidelobes. Within this bi-static geometry it is therefore not only required to implement
range-dependent Rx array beam steering and sidelobe mitigation, but the same range-dependent beam steering and sidelobe
mitigation is required for Tx array as well. It has been demonstrated that improved performance in this case can be achieved by
MIMO technology introduced in this paper.
TA7b-2
10:40 AM
Large Phased Array Antenna Calibration Using Radar Clutter and MIMO
Matthew Brown, Mitch Mirkin, Dan Rabideau, MIT Lincoln Laboratory, United States
Phased array radar beamforming is degraded by phase and gain deviations across antenna subarrays. Elements within an
individual subarray may be calibrated by injecting a known signal into each receiver (e.g., via a loop back path or via mutual
coupling), but this approach does not correct transmission or array deformation errors. An alternative approach for removing
these subarray errors is to leverage the digital AESA (active electronically scanned array) architecture to calibrate on radar clutter
while the radar is in flight. An antenna calibration algorithm that removes transmit and receive subarray errors using MIMO
(multiple input, multiple output) waveforms is described and simulated.
TA7b-3
11:05 AM
High Resolution Imaging for MIMO Forward Looking Ground Penetrating Radar
Jian Li, Ode Ojowu, Luzhou Xu, University of Florida, United States; John Anderson, Howard University, United
States; Lam Nguyen, Army Research Laboratory, United States
Forward-Looking Ground Penetrating Radar (FLGPR) can be used for detecting landmines. The detection process involves
generating synthetic aperture radar (SAR) images using the standard backprojection (BP). The BP approach suffers from poor
resolution and high sidelobe problems. This paper focuses on enhancing imaging resolution and reducing sidelobes using the
Sparse Iterative Covariance-based Estimation (SPICE). A MIMO FLGPR developed by the Army Research Laboratory (ARL) is
used for analysis.
57
TA7b-4
11:30 AM
Structure Health Monitoring Exploiting Mimo Ultrasonic Sensing and Group Sparse
Bayesian Learning
Qisong Wu, Yimin Zhang, Moeness Amin, Andrew Golato, Sridhar Santhanam, Fauzia Ahmad, Villanova
University, United States
Real-time imaging of defects in thin-walled structures using guided ultrasonic waves has emerged as a significant application
area in structure health monitoring (SHM). In the proposed multiple-input multiple-output (MIMO) ultrasonic sensing system,
a network of piezoelectric (PZT) transducers is affixed to the structure. On impulsive excitation of the transducers, elastic stress
waves called Lamb waves propagate through the thin-walled structure and interact with defects and plate boundaries before
arriving at the receiving transducers. The received signals at the receive transducers can be very complex with significant
overlapping of modal pulses because of the propagation dispersion of the Lamb waves. The underlying SHM sensing problem
invites application of sparse signal reconstruction techniques with the consideration of multi-dimensional group sparsity for
effective defect imaging. First, the defects have an extended spatial occupancy that is clustered in the image domain. Second,
the use of MIMO sensing methodologies forms another dimension of group sparsity and, as such, increases the overall multidimensional clustered problem. To provide reliable and high-resolution defect imaging, we propose the exploitation of sparse
Bayesian learning that accounts for the multi-dimensional group sparsity of the defects due to their clustered spatial occupancy
and the multiple-aspect MIMO observations. Sparse Bayesian learning techniques have shown to provide robustness for highresolution signal reconstruction due to its insensitivity to dictionary coherence and have the flexibility of effective exploitation of
the signal structure. As such, the proposed approach result in an effective high-resolution imaging methodology which is robust
to noise and speckles. The superiority of the proposed technique over the state-of-the-art sparse signal reconstruction techniques
will be demonstrated through simulations.
Track B – MIMO Communications and Signal Processing
Session: TAa8 – Channel Estimation and MIMO Feedback
8:15 AM–9:55 AM
Chair: Ananthanarayanan Chockalingam, Indian Institute of Science
TA8a1-1
Channel Estimation in Millimeter Wave MIMO Systems with One-Bit Quantization
Jianhua Mo, University of Texas at Austin, United States; Philip Schniter, Ohio State University, United States;
Robert W. Heath Jr., University of Texas at Austin, United States
We develop a channel estimator for millimeter wave (mmWave) multiple input multiple output (MIMO) systems with with
one-bit ADCs. Since the mmWave MIMO channel is sparse due to the propagation characteristics, the estimation problem is
formulated as a one bit compressed sensing problem. We use the generalized approximate message passing (GAMP) algorithm to
solve this optimization problem. The initial simulation results show that GAMP can exploit sparsity to reduce mean squared error
in the important low SNR region.
TA8a1-2
Maximum-Likelihood Joint Channel Estimation and Data Detection for Space Time Block
Coded MIMO Systems
Haider Alshamary, Weiyu Xu, University of Iowa, United States
This paper considers exact maximum likelihood (ML) joint channel estimation and data detection for orthogonal space time block
coded (OSTBC) multiple input multiple output (MIMO) systems. We propose an efficient algorithm which achieves exact ML
blind data detection even for non-constant-modulus constellations. To the best of our knowledge, this is the first algorithm which
efficiently achieves exact ML blind detection for OSTBC MIMO systems with non-constant-modulus constellations. Theoretical
and simulation results validate performance and complexity improvements.
TA8a1-3
Cramer-Rao Bound for Blind Channel Estimation in Cyclic Prefixed MIMO-OFDM
Borching Su, Kai-Han Tseng, National Taiwan University, Taiwan
In this paper, a Cramer-Rao bound (CRB) for blind channel estimation in cyclic prefixed MIMO-OFDM systems is derived.
The derived bound is valid for arbitrary numbers of transmit antennas, received antennas, and received symbols available for
blind channel estimation, and serve as a benchmark of performances of several recently proposed subspace-based algorithms.
Computer simulations are conducted to compare these algorithms with the derived bound. The results show that there are still
rooms for improvement for the currently available blind channel estimation algorithms.
58
TA8a1-4
Efficient MIMO Sparse Channel Estimation Using LTE Sounding Reference Signal
Jeng-Kuang Hwang, Jen-Hao Liu, Chien-Min Chen, Chuan-Shun Lin, Yuan Ze University, Taiwan
In order to make full use of various LTE MIMO transmission modes, we aim for a feasible MIMO channel estimation scheme
with high accuracy and low complexity. Exploiting the sounding reference signal and channel sparsity, an efficient MIMO
channel estimation algorithm is derived in terms of both the maximum likelihood estimation and minimum description length
criterion for tap selection. The scheme exhibits excellent performance for sparser channel. Furthermore, its complexity is
quite low due to the use of FFT and power sorting. Simulation and real-world experimental results of a 2x2 MIMO system are
demonstrated to confirm the above merits.
TA8a1-5
Impact of Received Signal on Self-interference Channel Estimation and Achievable Rates
in In-band Full-duplex Transceivers
Dani Korpi, Lauri Anttila, Mikko Valkama, Tampere University of Technology, Finland
This paper analyzes the effect of the calibration period on self-interference channel estimation in full-duplex radio transceivers.
In particular, we consider a scenario where channel estimation must be performed without a separate calibration period, which
means that the received signal of interest will act as additional noise from the estimation perspective. We will explicitly analyze
its effect, and quantify the increase in parameter estimation variance, or sample size, if similar channel estimation accuracy is to
be achieved as with a separate calibration period. In addition, we will analyze how the calibration period, or its absence, affects
the overall achievable rates.
TA8a1-6
MIMO Nullforming with RVQ Limited Feedback and Channel Estimation Errors
D. Richard Brown III, Worcester Polytechnic Institute, United States; David Love, Purdue University, United States
This paper explores limited feedback nullforming techniques based on random vector quantization (RVQ) with and without
receiver coordination. The availability of receiver coordination affects the type and amount of feedback required to select an
appropriate precoding vector. Approximate upper and lower bounds are developed for the mean received power at primary
receivers with and without receiver coordination. Numerical results show that the performance of RVQ nullforming can approach
the lower bound for moderate codebook sizes if the number of primary receivers is small and/or if the variance of the channel
estimation errors is not too small.
TA8a1-7
Limited Feedback in OFDM Systems for Combating ISI/ICI Caused by Insufﬁcient Cyclic
Preﬁx Length
Erich Zoechmann, Stefan Pratschner, Stefan Schwarz, Markus Rupp, Vienna University of Technology, Austria
In 3GPP LTE communication systems, channel state information is fed back by means of the channel quality indicator, the
precoding matrix index and the rank indicator. This limited feedback information is used to adapt the transmission scheme such
as to improve the throughput and to reach a certain block error rate target. We follow a common opinion from literature and use
the channel impulse response to calculate the interference power caused by insufﬁcient cyclic preﬁx length. By accounting for
this inter symbol and inter carrier interference power in the feedback calculation, we obtain an improved throughput in MIMO
OFDM systems.
TA8a1-8
Frugal Channel Tracking for Transmit Beamforming
Omar Mehanna, Nicholas Sidiropoulos, University of Minnesota, United States
Channel state feedback is a serious burden for transmit beamforming systems with many antennas in FDD mode. Instead of
estimating the channel at the receiver and feeding back quantized beamformer information, a different approach that exploits the
spatio-temporal correlation of the channel is proposed. The transmitter periodically sends a beamformed pilot signal, while the
receiver feeds back the quantized innovation derived from either a Kalman filtering or a MAP tracking loop. Simulations show
that close to optimum performance can be attained with only 2 bits per channel dwell, clearing a hurdle for transmit beamforming
with many antennas in FDD mode.
59
Track H – Speech, Image and Video Processing
Session: TAa8 – Image Processing I
8:15 AM–9:55 AM
Chair: Kenneth Jenkins, Pennsylvania State University
TA8a2-1
Second Order Model Deviations of Local Gabor Features for Texture Classification
David Picard, Inbar Fijalkow, ETIS - UMR 8051 / ENSEA, Université Cergy-Pontoise, CNRS, France
In this paper, we tackle the problem of texture classification with a local approach based on measuring second order deviations
with respect to a dictionary of characteristic patterns. At each pixel, we extract local signal properties thanks to several Gabor
filters that are aggregated on a small support region. Then, we compute a dictionary of such features that serves as a universal
model. The texture signature is the deviation of second order statistics between its local features and the universal model.
Experiments are made on two sets of photographic paper textures, and show the soundness of the approach.
TA8a2-2
Weighted Boundary Matching Error Concealment for HEVC Using Block Partition
Decisions
Yan-Tsung Peng, Pamela Cosman, University of California, San Diego, United States
We propose a weighted boundary matching error concealment method for HEVC. It uses block partition decisions to improve a
common block matching algorithm that finds blocks with the best matched boundaries from the previous frame to conceal the
currently corrupted blocks. The block partition decisions from the co-located block of the corrupted one are exploited. For each
partition, a summed boundary weight is computed; the one with the highest weight is chosen to be concealed next. Experimental
results show the proposed method performs better than conventional error concealment methods objectively and subjectively.
TA8a2-3
Reducing the Latency and Improving the Resolution of Vector Quantization with
Anamorphic Stretch Transform
Haochen Yuan, Mohammad H. Asghari, Bahram Jalali, University of California, Los Angeles, United States
Application of the recently introduced Discrete Anamorphic Stretch Transform to data clustering is proposed and demonstrated.
We show that the Transform enhances vector quantization in terms of its computation speed and peak SNR by increasing the
image coherence. To validate our technique’s utility in data compression, we show superior performance compared to JPEG
2000.
TA8a2-4
Supervised Facial Recognition based on Multiresolution Analysis with Radon Transform
Ahmed Aldhahab, George Atia, Wasfy Mikhael, University of Central Florida, United States
A new supervised facial recognition system based on the integration of Two Dimensional Discrete Multiwavelet Transform (2D
DMWT), 2D Radon Transform (2D RT), and 3D DWT is proposed. In the feature extraction step, 2D DMWT is used to extract
the useful information from the image. The extracted features are then aligned using 2D RT and localized in one single band by
using 3D DWT. The resulting features are fed into a Neural Network for both training and testing. The proposed algorithm is
tested on different databases. It is shown that the proposed approach can significantly improve the overall performance.
TA8a2-5
On Compensating Unknown Pixel Behaviors for Image Sensors with Embedded Processing
William Guicquero, Michele Benetti, Arnaud Peizerat, Antoine Dupret, Commissariat à l’énergie atomique et aux
énergies alternatives, France; Pierre Vandergheynst, École Polytechnique Fédérale de Lausanne, Switzerland
In recent image sensor design a trend emerges [1]: either block-based image processing operations or Compressive Sensing (CS)
[2] tend to be performed at the focal plane level. In many cases, the objective is to alleviate the limitations of analog signals,
such as limited dynamic range or power consumption. Unfortunately, this processing of analog signals often introduces artifacts.
This article deals with a generic method that compensates those artifacts by post processing operations. The proposed restoration
algorithm is composed of a three steps loop: regularize the image, fit the model parameters describing unknown pixel behaviors
and update a regularization coefficient.
60
TA8a2-6
Representative Selection for Big Data via Sparse Graph and Geodesic Grassmann
Manifold Distance
Chinh Dang, Hayder Radha, Michigan State University, United States
This paper addresses the problem of identifying a very small subset of data points that belong to a significantly larger massive
dataset (i.e., Big Data). The small number of selected data points must adequately represent and faithfully characterize the
massive Big Data. We propose a novel representative selection framework by generating a l_1 norm sparse graph, Grassmann
manifold distance, and Principal component centrality for a given Big-Data dataset. We validate the proposed framework onto the
problem of video summarization, and compare the result with the ground truth and with some state-of-the-art related methods.
TA8a2-7
A Generic Particle Filtering Approach for Multiple Polyhedral Object Tracking in a
Distributed Active Sensor Network
Benoit Fortin, Regis Lherbier, Jea-Charles Noyer, Univ. Littoral Cote d’Opale, France
This paper presents a method for multi-target tracking in a multisensor system composed of several distributed active sensors
(rangefinders). The final goal is to deliver a complete reconstruction of the environment of a vehicle. The originality of this work
relies on the joint exploitation of geometric invariance properties of objects avoiding any loss of optimality and on an efficient
management of transitions when the objects of interest move from one field-of-view to another.
TA8a2-8
Spatial Domain Synthetic Scene Statistics
Debarati Kundu, Brian Evans, University of Texas at Austin, United States
Natural Scene Statistics (NSS) has been applied to natural images obtained through optical cameras for automated visual quality
assessment. Since NSS does not need a reference image for comparison, NSS has been used to assess user quality-of-experience,
such as for streaming wireless image and video content acquired by cameras. In this paper, we take an important first step in
using NSS to automate visual quality assessment of synthetic images found in video games and animated movies. In particular,
we analyze NSS for synthetic images in the spatial domain using mean-subtracted-contrast-normalized (MSCN) pixels and
their gradients. The primary contributions of this paper are (1) creation of a publicly available ESPL Synthetic Image database,
containing 221 color images, mostly in high definition resolution of 1920x1080, and (2) analysis of the statistical distributions
of the MSCN coefficients (and their gradients) for synthetic images, obtained from the image intensities. We find that similar to
the case for natural images, the distributions of the MSCN pixels for synthetic images can be modeled closely by Generalized
Gaussian and Symmetric Alpha Stable distributions, with slightly different shape and scale parameters.
Track A – Communications Systems
Session: TAa8 – Signal Processing for Communications
8:15 AM–9:55 AM
Chair: Bhavya Kailkhura, Syracuse University
TA8a3-1
Energy-Efficient Secure Communications in MISO-SE Systems
Alessio Zappone, Pin-Hsun Lin, Eduard A. Jorswieck, TU Dresden, Germany
The problem of resource allocation for energy-efficient secure communications in MISO systems is investigated, wherein a
malicious user tries to eavesdrop the communication between two legitimate users. The legitimate transmitter has multiple
antennas, whereas the eavesdropper and the receiver have a single antenna. Unlike most papers dealing with physical layer
security, the goal is not secrecy capacity maximization, but rather secrecy energy efficiency maximization, defined as the ratio
between the secrecy capacity and the consumed power. The three scenarios of perfect, partial, and statistical channel state
information are solved and compared.
TA8a3-2
Distinguishing BFSK from QAM and PSK by Sampling Once per Symbol
Mohammad Bari, Milos Doroslovacki, George Washington University, United States
In this paper we propose a feature to distinguish FSK from QAM and PSK modulations. The feature is based on the imaginary
part of product of two consecutive signal values where every symbol is sampled only once. Conditional probability density
functions of the feature given the present modulation are determined. Central limit theorem for strictly stationary m-dependent
61
sequences is used to obtain Gaussian approximations. Then the thresholds are determined based on the minimization of total
probability of misclassification. Effects of AWGN, carrier offset and non-synchronized sampling on the performance are studied.
Proposed classifier is compared to the maximum likelihood classifier.
TA8a3-3
Koorosh Firouzbakht, Guevara Noubir, Masoud Salehi, Northeastern University, United States
In this paper we use a game-theoretic approach to model a wireless link under jamming. Instead of a zero-sum framework, we
use the bimatrix framework where it is no longer required that the sum of the players’ payoffs to be zero. Hence, a much larger
class of jamming problems could be modeled. Furthermore, we assume players’ strategies must be chosen from some hyperpolyhedron. We prove that the NE solution of this game corresponds to the global maximum of a quadratic program. We provide
an example of a jamming problem and show that the bimatrix framework can improve jammer’s payoff.
TA8a3-4
An Iterative Soft Decision Based Adaptive K-best Decoder Without SNR Estimation
Mehnaz Rahman, Ehsan Rohani, Gwan Choi, Texas A&M University, United States
This paper presents an adaptive K-best multiple-input-multiple-output (MIMO) decoding algorithm. It includes the concept
of iterative lattice reduction (LR)-aided minimum-mean-square-error (MMSE) extended K-best decoding, thereby reducing
the computational complexity to a great extent. The method adaptively changes list size, K with respect to channel condition,
considering only the ratio of first minimum distance to the second one. Hence, accurate measurement of signal-to-noise-ratio
(SNR) is not required. Using the method, we obtain similar performance compared to conventional LR-aided K-best algorithm
and 1.6 dB improvement against the iterative least-sphere-decoder (LSD) at bit error rate (BER) of 10^(-6).
TA8a3-5
MMSE Scaling Enhances Performance in Practical Lattice Codes
Nuwan Ferdinand, University of Oulu, Finland; Matthew Nokleby, Duke University, United States; Brian Kurkoski,
Japan Advanced Institute of Science and Technology, Japan; Behnaam Aazhang, Rice University, United States
We investigate the value of MMSE scaling for practical lattice codes. For ideal lattices, MMSE scaling has been shown to be a
key ingredient in achieving the capacity of the AWGN channel. We demonstrate that MMSE scaling enhances the performance,
particularly at low SNR, for practical lattice codes. For example, a dimension $n=10000$ LDLC lattice exhibits approximately
0.6 dB gain when MMSE scaling is used for a rate of 1 bit/dimension. Furthermore, we provide a novel derivation of the MMSE
scaling rule, showing that it emerges naturally from principles of belief propagation decoders which account for the transmit
power constraint.
TA8a3-6
RLS-Based Frequency-domain DFE for Uplink SC-FDMA
Naveed Iqbal, Azzedine Zerguine, King Fahd University of Petroleum and Minerals, Saudi Arabia; Naofal AlDhahir, University of Texas at Dallas, United States
In this paper, we develop a low-complexity Adaptive Frequency Domain Decision Feedback Equalizer for SC-FDMA systems.
Both the feedforward and feedback filters operate in the frequency-domain and are adapted using the block RLS algorithm. We
also show that the RLS-based AFD-DFE not only enjoys a significant reduction in computational complexity when compared
to the frequency-domain non-adaptive channel-estimate-based MMSE-DFE but its performance is also better than that of the
practical MMSE DFE (with decision errors) and close to the ideal MMSE DFE (with correct decisions)
TA8a3-7
Jeng-Kuang Hwang, Jeng-Da Li, Yu-Chang Hsu, Chuan-Shun Lin, Yuan-Ze University, Taiwan
Under the cyclic ISI channel model induced by CP-aided single-carrier transmission system, the optimum cyclic Viterbi (CV)
receiver is derived, which outperforms those suboptimal frequency-domain equalizers. To further apply the CV to localized SCFDMA system, a reduced-state CV is then devised by using decision feedback (DF) with local tentative decisions. Moreover, an
all-pass prefilter (APPF) is incorporated to suppress the DF error propagation under non-minimum phase channel. The resulting
CV-APPF-DF receiver is therefore merited in performance and complexity. Simulations results demonstrate the excellent
performance of the full-state CV receiver, and the near optimum performance of the reduced-state CV-APPF-DF receiver.
62
TA8a3-8
Energy Detection Using Very Large Antenna Array Receivers
Alex Oliveras Martinez, Elisabeth De Carvalho, Petar Popovski, Gert Frølund Pedersen, Aalborg University,
Denmark
We propose the use of energy detection for single stream transmission and reception by a very large number of antennas, with
primary application to millimeter wave communications. The reason for applying energy detection is low complexity, cost and
power efficiency. While both energy detection and millimeter wave communications are limited to short ranges due respectively
to noise sensitivity and propagation attenuation, processing by a large number of receive antennas overcomes those shortcomings
to provide significant reach extension. This processing is solely based on long-term statistics of the channel and noise, making it
robust to user mobility and imperfect channel knowledge.
Track D – Signal Processing and Adaptive Systems
8:15 AM–9:55 AM
Chair: Milos Doroslovacki, George Washington University
TA8a4-1
On Component-Wise Conditionally Unbiased Linear Bayesian Estimation
Mario Huemer, Oliver Lang, Johannes Kepler University Linz, Austria
The classical unbiased condition utilized e.g. by the best linear unbiased estimator (BLUE) is very stringent. By softening the
”global” unbiased condition and introducing component-wise conditional unbiased conditions instead, the number of constraints
limiting the estimator’s performance can in many cases significantly be reduced. In this work we extend the findings on
component-wise conditionally unbiased (CWCU) linear Bayesian estimation for linear data models investigated e.g. in [1]. We
discuss the requirements on the parameter vector, that allow for finding CWCU linear estimators outperforming the BLUE in
Bayesian measures, and we derive the CWCU linear minimum mean square error (LMMSE) estimator under these conditions. In
depth comparisons to the BLUE and the LMMSE estimator are followed by a well known channel estimation application which
impressively demonstrates the advantages of the CWCU LMMSE estimator over the BLUE.
TA8a4-2
Performance of Proportionate-type NLMS Algorithm with Gain Allocation Proportional to
the Mean Square Weight Deviation
Kevin Wagner, Naval Research Laboratory, United States; Milos Doroslovacki, George Washington University,
United States
The complex colored water-filling algorithm for gain allocation has been shown to provide improved mean square error
convergence performance, relative to standard complex proportionate-type normalized least mean square algorithms. This
algorithm requires sorting operations and matrix multiplication on the order of the size of the impulse response at each iteration.
In this paper, the mean square weight deviation and two suboptimal gain allocation algorithms are presented. They are motivated
by similar algorithms introduced before for real-valued signals and systems. The presented algorithms no longer require sorting.
It is shown that they provide significant computational complexity savings while maintaining comparable mean square error
convergence performance. The algorithms are also investigated in the case of unknown input correlation matrix and speech input
signals.
TA8a4-4
An Efficient Least Mean Squares Algorithm based on q-Gradient
Ubaid Al-Saggaf, Mohammad Moinuddin, King Abdulaziz University, Saudi Arabia; Azzedine Zerguine, King
Fahd University of Petroleum and Minerals, Saudi Arabia
In this work, we propose a novel LMS type algorithm by utilizing the q-gradient concept which is derived from the definition
of Jacksons derivative. The q-gradient-based LMS algorithm results in faster convergence for q > 1 because of the fact that the
q-derivative, unlike the conventional derivative which evaluates tangent, computes the secant of the cost function and hence takes
larger steps towards the optimum solution. Convergence analysis of the proposed algorithm is also presented. Simulation results
are presented to support our theoretical findings.
63
TA8a4-5
Optimal Step Size Control for Acoustic Echo Cancellation
Khosrow Lashkari, Seth Suppappola, Cirrus Logic, United States
Depth and speed of convergence are important metrics for the performance of acoustic echo cancellers (AECs). The LMS
(Least Mean Square) or NLMS (normalized LMS) algorithms constitute the core of AECs. This paper presents an optimal step
size control strategy for the NLMS algorithm when the input signal is white. This strategy gives the deepest convergence in
the shortest amount of time. The optimum step size turns out to be the solution of the Riccati equation. Simulation results are
presented to confirm the analysis. Practical considerations for colored signals and time varying echo impulse responses are also
discussed.
TA8a4-6
Stochastic Gradient Algorithm Based on an Improved Higher Order Exponentiated Error
Cost Function
Umair bin Mansoor, Syed Asad, Azzedine Zerguine, King Fahd University of Petroleum and Minerals, Saudi Arabia
We propose stochastic gradient algorithm based on exponentiated cost functions that employ higher order moments of the chosen
error. Recently, such algorithms based on exponential dependence of squared of the error have attracted a lot of attention. It has
been felt that such algorithms have only been tested in the Gaussian noise environment. Motivated by the performance of the
least-mean-fourth algorithm in sub-Gaussian environments, we make use of the same strategy to come up with a new algorithm
with superior convergence and steady-state performance. Simulation shows promising results.
TA8a4-7
Spectral Multiscale Coverage with the Feature Aided CPHD Tracker
Ramona Georgescu, Shuo Zhang, Amit Surana, Alberto Speranzon, Ozgur Erdinc, United Technologies Research
Center, United States
A closed loop approach for surveillance was developed leveraging the Spectral Multiscale Coverage (SMC) algorithm for sensor
management coupled with the Cardinalized Probability Hypothesis Density (CPHD) multitarget tracker. Additionally, the CPHD
was formulated such that it is able to ingest features, if available. Simulations with fixed and mobile sensors (the latter, tasked by
the SMC) providing data to the tracker underlined the benefits of sensor fusion with respect to standard metrics of performance.
TA8a4-8
Adaptive Sampling with Sensor Selection for Target Tracking in Wireless Sensor Networks
In this paper, we consider a target tracking problem where the time interval between adjacent sensor measurements is a decision
variable to be optimized. Therefore, we aim to sample the target very frequently when the uncertainty is large and sample the
target less frequently when the uncertainty is small. Having obtained when to sample the target, we next determine which sensors
to be selected at the intended sampling instant. Simulation results illustrate the efficiency of the proposed algorithm.
Track A – Communications Systems
Session: TAb8 – Multiuser and Cellular Systems
10:15 AM–11:55 AM
Chair: Rafael F. Schaefer, Princeton University
TA8b1-1
Average Sum MSE Minimization in the Multi-User Downlink With Multiple Power
Constraints
Andreas Gründinger, Michael Joham, Technische Universität München, Germany; Jose Pablo Gonzalez Coma, Luis
Castedo, University of A Coruna, Spain; Wolfgang Utschick, Technische Universität München, Germany
We consider a sum mean square error (MSE) transceiver design for the multi-user downlink with linear transmit power
constraints. Since the multi-antenna transmitter has only imperfect CSI, the average sum MSE (SMSE) is minimized via an
alternating optimization (AO). For fixed equalizers, the average SMSE minimizing precoders are found via an uplink-downlink
SMSE duality based on Lagrangian duality. The precoder update therewith transforms to an uplink max-min average SMSE
problem, i.e., an MMSE equalizer design and an outer worst-case noise search. This problem is optimally solved and strong
duality is shown to holds at the optimum.
64
TA8b1-2
Hierarchical Precoding for Ultra-Dense Heterogeneous Networks
Lars Thiele, Martin Kurras, Fraunhofer Institute for Telecommunications Heinrich Hertz Institute, Germany
In this work, we demonstrate the limits from joint regularized zero-forcing precoding within a cluster of several macro and
small-cell base stations. While studying the effects of heterogeneous power constraints in such a cluster of transmission nodes,
we develop a hierarchical precoding solution which mitigates the inter-cell interference from the macro-cells caused a the users
mainly served by the small-cells. Due to significantly reduced transmit power budget both, inter-small-cell interference as well
as interference from small-cells caused at the macro users is of less importance and hence is not taken into account within the
precoding algorithm.
TA8b1-3
Detection using Block QR Decomposition for MIMO HetNets
Robin Thomas, Raymond Knopp, Eurecom, France; Sunil (B.T.) Maharaj, University of Pretoria, South Africa
Interference management between uncoordinated eNBs and UEs is a crucial aspect of Heterogeneous Network (HetNet)
deployments, especially if the interference can be exploited and suppressed at the receiver. In this paper, a novel preprocessing
Block QR decomposition technique is proposed for a low complexity max-log-MAP receiver, in order to decode the dual-stream
interferer in a 4×4 HetNet scenario. The authors show that for an SNR of 20 dB, the proposed receiver detection scheme has
minimal mutual information loss for Gaussian signals, which would then enable analysis and validation (with LTE downlink
simulations) of higher order MIMO detection schemes.
TA8b1-4
On Performance Prediction for Multiuser Detection Enabled Systems in Packet Based
Asynchronous Gaussian Multiple Access Channels
Prabahan Basu, MIT Lincoln Laboratory, United States
Interference Multiple Access is a spectral sharing paradigm wherein multiuser detection capable transmitters target those
occupied spectral bands where the mutual interference between the existing and infringing nodes is tolerable. Successful
coexistence requires the coexistence seeking node to accurately gauge its own performance and to accommodate the existing
node, assumed to lack MUD. We previously proposed a polynomially approximated score to rank candidate bands occupied by
nodes employing packet based communications protocols. Here, we derive a computation of the proposed score based on more
efficient closed form approximations, thus enabling its use in practical settings.
TA8b1-5
Decentralized Target Rate Optimization for MU-MIMO Leakage Based Precoding
Tim Rüegg, Marc Kuhn, Armin Wittneben, ETH Zurich, Switzerland
In this paper we propose a decentralized target rate precoding for multi-user multiple-input multiple-output downlink setups.
The precoding is optimized for each link separately with respect to the transmit and leakage power. This allows to control the
interference and to design the precoding flexibly with respect to energy efficiency and outage minimization. A closed form
solution for the optimization is presented and thorough analysis on the interdependency of the transmit and leakage power is
provided. The proposed precoding is tested in numerical simulations.
TA8b1-6
Leveraging Interference for Increasing Throughput and Reliability of Commercial
Wireless Small Cells
Rachel Learned, Michael Pitaro, Matthew Ho, Massachusetts Institute of Technology, United States
This paper examines the potential of a new interference-leveraging scheme that relies upon interference mitigation and cognitive
algorithms to enable a case by case in-base evaluation of interfering links and, as appropriate, to recognize and leverage occupied
bands as opportunities for reliable communication. Simulation based analysis of achievable throughput and dropped call rates for
three contending solutions are provided. Preliminary results indicate cognitive interference leveraging to be unique in its ability
to provide continuous and high quality of service for both small cell and macro cell links without overt coordination with macro
or other small cell bases.
65
TA8b1-7
Throughput Analysis of LTE and WiFi in Unlicensed Band
Abhijeet Bhorkar, Christian Ibars Casas, Pingping Zong, Intel Corporation, United States
In this paper, the co-channel performance of large scale deployment of LTE in Unlicensed (LTE-U) band and WiFi is studied
using the stochastic geometry. Analytical expressions of LTE-U throughput in presence of WiFi are presented and are partly
validated by the simulation results. The LTE-U Low Power Nodes (LPNs) are deployed as Poisson Point Process (PPP), while,
the WiFi transmissions are modeled as hardcore Mat´ ern point process. Using this analytical approach the impact of various
parameters such as sensing threshold and transmission power on the co-existence of LTE-U and WiFi is studied.
TA8b1-8
Multi-User Detection for xDSL with Partial Cooperation Among Multiple Operators
Syed Hassan Raza Naqvi, Umberto Spagnolini, Politecnico di Milano, Italy
Cross-talk (FEXT) is the most limiting factor in xDSL systems. When multiple CPEs share the same cable-binder, the interoperator FEXT cannot be controlled or mitigates. We propose a cooperative method for inter-operator FEXT mitigation where
different operators mutually cooperate without sharing any sensitive information, but rather by exchanging only their mutual
interference. The iterative method based on the exchange the alien-FEXT largely improves the intra-operator MMSE multiuser
detection. The performances of this interference-only exchange are comparable with fully cooperative approach where all
operators fully share all the available data.
Track G – Architecture and Implementation
Session: TAb8 – Computer Arithmetic II
10:15 AM–11:55 AM
Chair: Sardar Muhammad Sulaman, Lund University
TA8b2-1
Improved Non-restoring Square Root Algorithm with Dual Path Calculation
Kihwan Jun, Earl Swartzlander, University of Texas at Austin, Republic of Korea
This paper focuses on reducing the delay of the non-restoring square root algorithm. Although the non-restoring square root
algorithm is the fastest of the other radix-2 digit recurrent square root algorithms, there are still some possibilities to enhance
its performance. To improve its performance, two new approaches are proposed here. For the first proposed approach, a novel
method to find a square root bit for every iteration, which hides the total delay of the multiplexer with dual path calculation is
presented. Secondly, a new method uses the modified Most Significant Carry (MSC) generator, which determines the sign of
each remainder faster than a carry lookahead adder, which reduces the total delay.
TA8b2-2
Merged Residue Number System Generation
Michael Sullivan, Earl Swartzlander, University of Texas at Austin, United States
Residue number system (RNS) implementations often restrict themselves to specialized moduli sets that are chosen for efficient
logic synthesis. Despite the paramount importance of implementation efficiency using these specialized moduli sets, such RNS
implementations typically use separate conversion logic for the generation of each RNS modulus. This paper considers a novel
parallel residue generation procedure for moderate dynamic range RNS numbers. This procedure shares the majority of logic
between selected moduli, greatly increasing efficiency over existing forward conversion schemes.
TA8b2-3
Partial Product Generation and Addition for Multiplication in FPGAs With 6-Input LUTs
George Walters, Penn State Erie, The Behrend College, United States
Multiplication is the dominant operation for many applications implemented on field-programmable gate arrays (FPGAs).
Although most current FPGA families have embedded hard multipliers, soft multipliers using lookup tables (LUTs) remain
important. This paper presents novel circuits for partial product generation and addition and their usage in two’s complement
multipliers. Synthesis results for Virtex-7 are presented. Proposed single-cycle multipliers use 35% to 45% fewer LUTs and have
9% to 22% less delay than the best alternatives. Proposed pipelined multipliers use 32% to 40% fewer LUTs than LogiCORE IP
multipliers and can be clocked 29% to 42% faster than embedded hard multipliers.
66
TA8b2-4
Milos Ercegovac, University of California, Los Angeles, United States
A low-power radix-4 quotient generator is presented. A modified digit-recurrence division algorithm is used to produce only
quotient without final remainder. A gradual bit-slice deactivation is exploited to reduce the number of active modules across
iterations and, consequently, to reduce power dissipation and energy. In this paper we report the synthesis results of radix-4
quotient generators for 16, 24, 32, and 54 bits with respect to area, delay, power dissipation, and energy. We compare these
results with the corresponding results of radix-4 SRT dividers for the same precisions.
TA8b2-5
Divya Mahajan, Matheen Musaddiq, Earl Swartzlander, University of Texas at Austin, United States
Currently memristors are being researched to offer logic and memory functions. Recently, ultra dense resistive memory arrays
built from various two terminal semiconductor or insulator thin film devices have been demonstrated [1]. This paper presents
memristor-based design of commonly used (ripple carry, conditional sum and parallel prefix) adders. The latency and area of
TA8b2-6
Canonic Real-Valued FFT Structures
Megha Parhi, Yingjie Lao, Keshab K. Parhi, University of Minnesota, Twin Cities, United States
This paper, for the first time, presents novel DIT and DIF structures for computing real FFT, referred as RFFT, that are canonic
with respect to the number signal values computed at each FFT stage. In the proposed structure, in an N-point RFFT, exactly N
signal values are computed at the output of each FFT stage and at the output. While canonic FFT structures based on decimationin-frequency were presented before, these structures were derived in an adhoc way. This paper presents a formal method to derive
canonic DIF RFFT structures. Examples of canonic structures for non-power-of-two size RFFT are also presented.
TA8b2-7
A High Throughput and Low Power Radix-4 FFT Architecture
Soumak Mookherjee, Linda S. DeBrunner, Victor DeBrunner, Florida State University, United States
An FTT architecture is proposed that is suitable for both high performance and low power applications. The architecture is based
on a Radix-4 algorithm using pipelined Multi-path Delay Commutators. Two separate datapaths are used so that the hardware
can process eight inputs in parallel. The throughput is increased by a factor of eight while achieving 100% hardware utilization.
Power consumption of this architecture is approximately 75% less than the Radix-4 MDC structure with the same throughput.
Using a Xilinx FPGA Virtex 5, only slightly more than twice the area is required for eight times higher throughput for a
256-point FFT.
TA8b2-8
A Domain Splitting Algorithm for the Mathematical Functions Code Generator
Olga Kupriianova, Christoph Lauter, UPMC, LIP6, PEQUAN team, France
The general approach to mathematical function implementation consists of three stages: range reduction, approximation and
reconstruction. The range reduction step is needed to reduce the degree of the approximation polynomial and to simplify the error
analysis. For some particular functions (e.g. exp) it is done using its algebraic properties. In general case the whole domain is
split into small subdomains to get low-degree approximation on each of them. Here we present a novel algorithm for the domain
splitting that will be integrated soon to Metalibm code generator.
67
Track E – Array Signal Processing
Session: TAb8 – Array Processing Methods
10:15 AM–11:55 AM
Chair: Piya Pal, University of Maryland
TA8b3-1
Array Self Calibration with Large Initial Errors
Benjamin Friedlander, University of California, Santa Cruz, United States
A self-calibration algorithm is presented for direction finding in the presence of unknown complex sensor gains. The method
jointly estimates the directions of arrival of all the sources as well as the unknown gains. Existing self-calibration techniques are
iterative and require sufficiently good initial estimates in order to converge to a correct solution. These techniques are ineffective
in the presence of initially large calibration errors which make it impossible to reliably initialize the algorithm. The method
presented here is non-iterative, does not require initialization, and handles large sensor gain errors. The performance of the
algorithm is illustrated by numerical examples.
TA8b3-2
Maximum Likelihood Estimation for Geolocation in the Presence of Multipath
Benjamin Friedlander, University of California, Santa Cruz, United States
We consider the problem of estimating the location of an emitter in a multipath environment from measurements by multiple
widely spaced sensors. We present a maximum likelihood approach for estimating the emitter position jointly with the positions
of the reflectors, the unknown complex gains of the propagation paths, and the unknown transmitted signal. This approach
provides unbiased emitter position estimates whether or not there is a direct line of sight. A variation of the ExpectationMaximization (EM) algorithm is used to provide an iterative, computationally feasible calculation of the maximum- likelihood
location estimate.
TA8b3-3
Enhanced Location Detection Algorithms Based on Time of Arrival Trilateration
Sajina Pradhan, Jae-young Pyun, Goo-Rak Kwon, Seokjoo Shin, Suk-seung Hwang, Chosun University, Republic of
Korea
The TOA trilateration method decides the location of MS using an intersection point of three circles with centers corresponding
three BS coordinates and radius based on the distance between MS and three BSs. Since the distance between BS and MS is
generally estimated counting the number of time delay samples, the estimated distances are slightly increased and three circles
may not intersect at a point. In this paper, we introduce the shortest distance and line intersection algorithms for improving
conventional TOA trilateration method to resolve above problem. Also, the mathematic analysis is provided to indicate the
relation between both algorithms.
TA8b3-4
Designing Radio Interferometric Positioning Systems for Indoor Localizations in
Millimeter Wave Bands
Marie Shinotsuka, Georgia Institute of Technology, United States; Yiyin Wang, Shanghai Jiao Tong University,
China; Xiaoli Ma, G. Tong Zhou, Georgia Institute of Technology, United States
Taking advantage of the spectrum characteristics of MMW bands, the indoor localization system using the radio interferometric
positioning system (RIPS) employing space-time coding is proposed in this paper. The space-time RIPS (STRIPS) is the
enhancement of the RIPS to work in MMW bands under the flat-fading channel. Two anchor nodes simultaneously transmit at
two time slots with different frequencies, and the target node samples the signal at the low-sampling frequency. The STRIPS
achieves the range-difference (RD) estimate without approximation, is immune to channel fading effects, and accommodates the
frequency offsets. The performance of the STRIPS is confirmed by Monte-Carlo simulations.
TA8b3-5
Indoor Sound Source Localization and Number Estimation Using Infinite Gaussian
Mixture Model
Longji Sun, Qi Cheng, Oklahoma State University, United States
This paper deals with sound source number estimation and localization problem in indoor environments using a circular
microphone array. The problem is approached in the time-frequency (TF) domain by performing a single source localization
algorithm at selected TF points and consequently generating one direction of arrival (DOA) estimate at each of the points.
68
The resulted DOA estimates are assumed from an infinite Gaussian mixture model. The number and the means of Gaussian
components correspond to the number and the true DOAs of the sources, respectively. The proposed algorithm is tested using
real experiments.
TA8b3-6
On the Structural Nature of Cooperation in Distributed Network Localization
Alireza Ghods, Stefano Severi, Giuseppe Abreu, Jacobs University Bremen, Germany; Samuel Van de Velde, Ghent
University, Belgium
We demonstrate analytically that the contribution of cooperation in improving the accuracy of network localization has a
fundamentally structural nature, rather then statistical as widely believed. The New approach to build Fisher Information
Matrices offers new insight onto the structure of FIMs, enabling us to easily account for both anchor and node location
uncertainties. Surprisingly, it is found that in the presence of node location uncertainty and regardless of ranging error variances,
the key contribution of cooperation in network localization is not to add statistical node-to-node information, but rather to provide
a structure over which node-to-anchor information is better exploited.
TA8b3-7
Enabling Distributed Detection with Dependent Sensors
Brian Proulx, Junshan Zhang, Douglas Cochran, Arizona State University, United States
Performing distributed detection with a large number of sensors that have correlated measurements is not computationally
feasible at the fusion center. Even storing the probability distributions of the measurements necessary to decide between
competing hypotheses is intractable for even a moderate number of remote sensors when the measurements are correlated. We
propose utilizing the t-cherry junction tree, an approach from probabilistic graphical models, to approximate the distribution
of the measurements. This approach allows for a compact representation of the distributions while maintaining a good
approximation to the true distribution. The performance of the approach is demonstrated via simulations.
TA8b3-8
Active Sonar Transmission Strategies in the Presence of Strong Direct Blast
Luzhou Xu, Jian Li, Akshay Jain, University of Florida, United States
In this paper, with the help of in-water experimentation data, we study: (1) the merits and limitations of two major classes of
Active Sonar Systems, namely, Pulse Active Sonar (PAS) and Continuous Active Sonar (CAS), with the Doppler-Tolerant Linear
Frequency Modulation (LFM) waveform; and (2) the merits and limitations of Doppler-Tolerant LFM and the Doppler- Sensitive
SHAPE waveforms with CAS. Throughout the study, Matched Filter (MF) is used as the receiver design. The strong delay and
Doppler spread direct blast and poor MF performance in noisy environments warrant us to introduce processing techniques which
enable receivers to produce accurate range estimates.
Track D – Signal Processing and Adaptive Systems
Session: TAb8 – Compressed Sensing III
10:15 AM–11:55 AM
Chair: Victor DeBrunner, Florida State University
TA8b4-1
Super-resolution Line Spectrum Estimation with Block Priors
Kumar Vijay Mishra, Myung Cho, Anton Kruger, Weiyu Xu, University of Iowa, United States
We address the problem of super-resolution line spectrum estimation of an undersampled signal with block prior information.
The component frequencies of the signal are assumed to take arbitrary continuous values in known frequency blocks. We
formulate a general semidefinite program to recover these continuous-valued frequencies using theories of positive trigonometric
polynomials. The proposed semidefinite program achieves super-resolution frequency recovery by taking advantage of known
structures of frequency blocks. Numerical experiments show great performance enhancements using our method.
69
TA8b4-3
Complexity Reduction in Compressive Sensing using Hirschman Uncertainty Structured
Random Matrices
Peng Xi, Victor DeBrunner, Florida State University, United States
Compressive Sensing (CS) increases the computational complexity of decoding while simplifying the sampling process. In this
paper, we apply our previously discussed Hirschman Optimal Transform to develop a series of sensing matrices that reduce
the computational complexity of decoding while preserving the recovery performance. In addition, this application provides us
alternative choices when we need different accuracy levels for the recovered image. Our simulation results show that with only
1/4 the computational resources of the partial DFT sensing basis, our proposed new sensing matrices achieve the best PSNR
performance, which is fully 5dB superior to other commonly used sensing bases.
TA8b4-4
A Sparse Approach for Estimation of Amplitude Modulated Sinusoids
Stefan Ingi Adalbjörnsson, Johan Swärd, Andreas Jakobsson, Ted Kronvall, Lund University, Sweden
We consider the problem of spectral analysis of signals composed of sums of multiple amplitude modulated, possibly
harmonically related, sinusoids using a sparse approach. By separating the nonlinear frequency variables using a dictionary of
possible frequency components as well as a spline basis for the amplitude modulation, results in a convex criterion which can be
efficiently solved without worrisome local minima. The resulting method makes no model order assumption and automatically
estimates both the signal parameters and their amplitude modulations.
TA8b4-5
Sparsity Order Estimation for Single Snapshot Compressed Sensing
Florian Roemer, Anastasia Lavrenko, Giovanni Del Galdo, Thomas Hotz, Technische Universitaet Ilmenau,
Germany; Orhan Arikan, Bilkent University, Turkey; Reiner Thomae, Technische Universitaet Ilmenau, Germany
In this paper we discuss the estimation of the sparsity order for a Compressed Sensing scenario where only a single snapshot is
available. We demonstrate that a specific design of the sensing matrix enables us to transform this problem into the estimation
of a matrix rank in the presence of additive noise. Thereby, we can apply existing model order selection algorithms to determine
the sparsity order. We also argue that the proposed sensing matrix design may have benefits that go beyond the estimation of the
sparsity both for the measurement as well as for the reconstruction strategy.
TA8b4-6
Streaming Signal Recovery Using Sparse Bayesian Learning
Uditha Wijewardhana, Marian Codreanu, Centre for Wireless Communications, Finland
We consider the recursive reconstruction of a streaming signal from compressive measurements. We reconstruct the streaming
signal over shifting intervals using a recovery method based on sparse Bayesian learning. The proposed method use the support
of the previously reconstructed estimate to provide a warm-start to the recovery algorithm, which improves the accuracy and the
speed of the signal estimates.
TA8b4-7
Compressed Change Detection for Structural Health Monitoring
Omid Sarayanibafghi, George Atia, Masoud Malekzadeh, Necati Catbas, University of Central Florida, United
States
The problem of detection of a sparse number of damages in a structure is considered. The idea relies on the newly developed
framework for compressed change detection, which leverages the unique covering property of identifying codes to detect
statistical changes in stochastic phenomena. Since only a small number of damage scenarios can occur simultaneously, change
detection is applied to responses of pairs of sensors that form an identifying code over a learned damage-sensing graph. An
asymptotic analysis of the detection delay and the probability of detection of the proposed approach is provided when the number
of damage scenarios is large.
70
TA8b4-8
A Sparse Semi-Parametric Chirp Estimator
Johan Swärd, Johan Brynolfsson, Andreas Jakobsson, Maria Hansson-Sandsten, Lund University, Sweden
We introduce a dictionary learning approach for estimating linear chips. A dictionary is created from a set of possible starting
frequency points and rates, and the estimates are then found by minimizing the distance between the signal and the dictionary
components together with an $\ell_1$-norm penalty which enforces sparsity. In typical solutions, one is forced to include a
vast number of dictionary elements in order to enable high resolution estimates, which frequently leads to a computationally
cumbersome optimization. Herein, we divide the coupled estimation of the starting frequency and the rate into two optimization
problems, thereby lowering the computational complexity.
Track D – Signal Processing and Adaptive Systems
Session: TPa1 – Covariance Mining Chair: Pradeep Ravikumar, University of Texas at Austin
TP1a-1
Abstract Algebraic-Geometric Subspace Clustering
1:30 PM
Manolis Tsakiris, Rene Vidal, Johns Hopkins University, United States
We consider an abstract version of the subspace clustering problem, where one is given the algebraic variety of a union of
subspaces, and the goal is to decompose it into the constituent subspaces. We propose a provably correct algorithm for addressing
the general case of unknown number of subspaces of unknown and possibly different dimensions. Using gradients of vanishing
polynomials at a point, the algorithm intersects the variety with a chain of hyperplanes until the subspace containing the point is
identified. By repeating this procedure for other points, our algorithm eventually identifies all the subspaces and their dimensions.
TP1a-2
1:55 PM
Minimum Variance Portfolio Optimization with Robust Shrinkage Covariance Estimation
Liusha Yang, Hong Kong University of Science and Technology, Hong Kong SAR of China; Romain Couillet,
Supelec, France; Matthew McKay, Hong Kong University of Science and Technology, Hong Kong SAR of China
This paper aims at designing a novel covariance estimation technique for optimizing large portfolios in the presence of heavytailed data or outliers, and under practical conditions in which the number of observed samples is of similar order to the number
of assets in the portfolio. This covariance estimation approach is based on the shrinkage Tyler’s robust M-estimator with riskminimizing shrinkage parameter. Our portfolio optimization method is shown via simulations to outperform existing methods
both for synthetic data and for real historical stock returns from the Hang Seng Index.
TP1a-3
Greedy Algorithms in Convex Optimization on Banach Spaces
2:20 PM
Vladimir Temlyakov, University of South Carolina, United States
Greedy algorithms which use only function evaluations are applied to convex optimization in a general Banach space $X$. Along
with algorithms that use exact evaluations, algorithms with approximate evaluations are treated. A priori upper bounds for the
convergence rate of the proposed algorithms are given. These bounds depend on the smoothness of the objective function and the
sparsity or compressibility (with respect to a given dictionary) of a point in $X$ where the minimum is attained.
TP1a-4
Greedy Algorithms for Learning Graphical Models
2:45 PM
Ali Jalali, Christopher Johnson, Pradeep Ravikumar, University of Texas at Austin, United States
Undirected graphical models, also known as Markov random fields, are widely used in a variety of domains. Recovering their
underlying Markov conditional independence graph structure is important for many of the applications of MRFs. Here, we revisit
a classical greedy algorithm, that iteratively adds and deletes edges, and show that when these forward and backward steps
are performed appropriately, we surprisingly obtain a state of the art method, that not only has computational advantages over
regularized convex optimization based approaches, but is also sparsistent, or consistent in sparsity pattern recovery, under weaker
conditions, and with a smaller sample complexity.
71
Track C – Networks
Session: TPb1 – Large-Scale Learning and Optimization
Chair: Alejandro Ribeiro, University of Pennsylvania
TP1b-1
3:30 PM
Jia Chen, Ioannis Schizas, University of Texas at Arlington, United States
The problem of tracking sensor clusters acquiring information for time-varying sources which are sensed via nonlinear data
models is considered. Canonical correlation analysis (CCA) is integrated with norm-one regularization along with exponential
weighing to adaptively process the acquired data. This leads to a framework that has the potential to adaptively cluster the sensors
based on their source information content which may be varying with time. A separable formulation is derived which then is
tackled using coordinate descent techniques to derive a sensor clustering scheme that processes data online and employs sparsity
to carry out the clustering.
TP1b-2
3:55 PM
Game-Theoretic Learning In A Distributed-Information Setting: Distributed Convergence
To Mean-Centric Equilibria
Brian Swenson, Soummya Kar, Carnegie Mellon University, United States; Joao Xavier, Instituto Superior Tecnico,
Portugal
The paper considers distributed learning in large-scale games via fictitious-play type algorithms. Given a preassigned
communication graph structure for information exchange among the players, this paper studies a distributed implementation
of the Empirical Centroid Fictitious Play (ECFP) algorithm that is well-suited to large-scale games in terms of complexity and
memory requirements. It is shown that the distributed algorithm converges to an equilibrium set denoted as the mean-centric
equilibria (MCE) for a reasonably large class of games.
TP1b-3
Network Newton
4:20 PM
Aryan Mokhtari, Alejandro Ribeiro, University of Pennsylvania, United States
We study the problem of minimizing a sum of convex objective functions where the components of the objective are available at
different nodes of a network. We propose a decentralized network Newton (NN) method that achieves faster convergence than
distributed gradient methods. Similar to the latter, NN solves a surrogate problem that is an approximate version of the original.
The surrogate problem can be efficiently solved using an inexact Newton method that uses an approximation to the Newton in
a decentralized manner. Theoretical analysis of NN shows a superlinear rate of convergence to a neighborhood of the optimal
solution.
TP1b-4
4:45 PM
Communication-Computation Tradeoffs in Decentralized Stochastic Optimization
Konstantinos Tsianos, Michael Rabbat, McGill University, Canada
We study dual averaging algorithms for decentralized stochastic optimization. The aim of the network is to minimize a convex
objective. Each node has access to a stochastic first-order oracle which provides noisy observations of the objective function
gradient at a query point. Existing theory focuses on how the optimization error decays as a function of the number of oracle
queries. In this paper we consider the tradeoff between computation and communication, noting that communication is often
significantly more time-consuming than computation when solving big-data problems over a compute cluster.
72
Track F – Biomedical Signal and Image Processing
Session: TPa2 – Bioinformatics and DNA Computing Co-Chairs: Olgica Milenkovic, University of Illinois at Urbana-Champaign and Farzad
Farnoud, California Institute of Technology
TP2a-1
On the Capacity of String-Duplication Systems and Genomic Duplication
1:30 PM
Farzad Farnoud, California Institute of Technology, United States; Moshe Schwartz, Ben-Gurion University of the
Negev, Israel; Jehoshua Bruck, California Institute of Technology, United States
It is known that the majority of the human genome consists of repeated sequences. Furthermore, it is believed that a significant
part of the rest of the genome also originated from repeated sequences and has mutated to its current form. We investigate the
possibility of constructing an exponentially large number of sequences from a short initial sequence and simple duplication rules,
including those resembling genomic duplication processes. Our results include the exact capacities, and bounds on the capacities,
of four fundamental string-duplication systems.
TP2a-2
Intrinsic Universality and the Computational Power of Self-Assembly
1:55 PM
Damien Woods, California Institute of Technology, United States
One of the goals of the theory of molecular programming is to study systems composed of tiny molecules that interact together
to form complicated structures and carry out sophisticated dynamics, and in particular to characterize the abilities and limitations
of such systems. The talk will introduce some of the concepts in the field with a focus on the topic of algorithmic self-assembly.
Here the idea is to design small sets of interacting square tiles, that stick together in very specific ways to form larger structures.
Tiles bind in a way that can be considered as carrying out a computation, similar to what is seen in cellular automata and Wang
tiling. This allows for the fabrication of complicated shapes and patterns in a bottom-up fashion. A number of theoretical selfassembly models have been studied with a variety of results characterizing their abilities. We will discuss the use of simulation as
a tool to compare these models, and indeed to classify and separate their computational and expressive power. Our journey begins
with the result that there is a single intrinsically universal tile set, that with proper initialization and scaling, simulates any tile
assembly system. This intrinsically universal tile set exhibits something stronger than Turing universality: it directly simulates
the geometry and dynamics of any simulated system. From there we find that there is no such tile set in the noncooperative, or
temperature 1, model, proving it weaker than the full tile assembly model. In the two-handed or hierarchical model, where large
assemblies can bind together in one step, we encounter an infinite set, of infinite hierarchies, with strictly increasing simulation
power. Towards the end of our adventure, we find one tile to rule them all: a single rotatable, flipable, polygonal tile that can
simulate any tile assembly system. It seems this could be the beginning of a much longer journey, so directions for future work
are suggested.
TP2a-3
Hybrid Rank Aggregation for Gene Prioritization
2:20 PM
Minji Kim, Farzad Farnoud, Olgica Milenkovic, University of Illinois at Urbana-Champaign, United States
Gene prioritization refers to a family of computational techniques used to infer disease genes through a set of training genes and
carefully chosen similarity criteria. Test genes are scored based on their average similarity to the training set, and the rankings of
genes under various similarity criteria are aggregated (fused) via statistical methods. The contributions of our work are threefold:
a) first, we investigate the predictive quality of a number of aggregation methods known from machine learning and social choice
theory; b) second, guided by the findings of the first stage of testing, we propose a new approach to genomic data fusion, termed
hybrid rank aggregation, which extracts the advantages of score-based and combinatorial aggregation techniques; c) third, we
propose an iterative procedure for gene discovery that operates via successful augmentation of the set of training genes by genes
discovered in previous rounds of testing and checked for consistency. We test our methods on a number of disease datasets,
including breast cancer, colorectal cancer, ischemic stroke, autism and others, and perform inference on new glioblastoma-related
genes. Our methods are conceptually simple and, in many instances, outperform state-of-the-art software tools such as ToppGene
and Endeavour.
73
TP2a-4
Rate-Independent Computation in Chemical Reaction Networks
2:45 PM
David Doty, California Institute of Technology, United States
We study the following problem: what functions f:R^k --> R can be computed by a chemical reaction network that eventually
produces the correct amount of the “output” molecule, no matter the rate at which reactions proceed? Such a network is
correct whether its evolution is governed by the standard model of mass-action kinetics or alternatives such as Hill-function or
Michaelis-Menten kinetics. We prove that f is computable in this manner if and only if it is *continuous and piecewise linear*.
Track H – Speech, Image and Video Processing
Session: TPb2 – Echo Cancellation
Chair: Steven Grant, Missouri University of Science and Technology
TP2b-1
Echo Cancellation for Bone Conduction Transducers
3:30 PM
Mohammad Behgam, Steven L. Grant, Missouri University of Science and Technology, United States
Bone conduction transducers are attractive for challenging acoustic environments. Bone vibrators (BVs) allow users to leave their
ears open, enhancing situational awareness. Bone conduction microphones (BCMs) increase transmitted SNR because they are
insensitive to air-conducted sound. In full-duplex mode, coupling between BVs and BCMs results in annoying echo. The echo
path’s linearity, stationarity, and length all affect the feasibility building an echo canceller. This paper describes those properties
and describes a proposed echo canceller design.
TP2b-2
Uncertainty Modeling in Acoustic Echo Control
3:55 PM
Gerald Enzner, Rainer Martin, Ruhr-University Bochum, Germany; Peter Vary, RWTH Aachen University,
Germany
Acoustic echo control (AEC) is a crucial component of hands-free voice interfaces. For sufficient echo suppression, the acoustic
echo canceler needs to be complemented by an adaptive echo suppression postfilter. Based on a stochastic echo path model, this
contribution derives an MMSE solution for echo canceler and postfilter jointly. The resulting postfilter utilizes the deterministic
far-end signal and employs the undermodeling error and uncertainty of the acoustic echo path in its gain computation. It thus
compensates typical deficiencies of acoustic echo cancelers in real-world applications. Another implication lies in the deep
justification of the recent Kalman filtering trend in AEC.
TP2b-3
A Kalman Filter for Stereophonic Acoustic Echo Cancellation
4:20 PM
Constantin Paleologu, University Politehnica of Bucharest, Romania; Jacob Benesty, University of Quebec, Canada;
Steven L. Grant, Missouri University of Science and Technology, United States; Silviu Ciochina, University
Politehnica of Bucharest, Romania
The stereophonic acoustic echo cancellation (SAEC) scheme was recently recast by using the widely linear (WL) model, i.e.,
as a single-input/single-output system with complex random variables. In this paper, we present a Kalman filter with individual
control factors (ICF-KF) in the context of the WL model for SAEC. As a specific feature, this algorithm uses a different level of
uncertainty for each coefficient of the filter. Simulation results indicate that the ICF-KF outperforms the recursive least-squares
(RLS) algorithm, which is usually considered as the benchmark for SAEC.
TP2b-4
Study and Design of Differential Microphone Array Beamforming
4:45 PM
Jingdong Chen, Northwestern Polytechnical University, China; Jacob Benesty, INRS-EMT, University of Quebec,
Differential microphone arrays (DMAs), a particular kind of sensor arrays that are responsive to the differential sound pressure
field, have a broad range of applications in sound recording, noise reduction, signal separation, dereverberation,etc. Traditionally,
an Nth-order DMA is formed by combining, in a linear manner, the outputs of a number of DMAs up to (including) the order
of N-1. This method, though simple and easy to implement, suffers from a number of drawbacks and limitations that prevent
DMAs from being widely deployed. This paper presents a new approach to the design of linear DMAs. The proposed technique
converts the DMA beamforming design to simple linear systems to solve. It is shown that this approach is much more flexible as
74
compared to the traditional methods in the design of different directivity patterns. Some methods are also presented to deal with
the white noise amplification problem, which used to be a big problem for DMAs, particularly the high-order ones, being used in
practice.
Track D – Signal Processing and Adaptive Systems
Session: TPa3 – Machine Learning
Chair: Vassilis Kekatos, University of Minnesota
TP3a-1
Consensus Inference with Multilayer Graphs for Multi-modal Data
1:30 PM
Karthikeyan Natesan Ramamurthy, IBM T. J. Watson Research Center, United States; Jayaraman J. Thiagarajan,
Lawrence Livermore National Laboratory, United States; Rahul Sridhar, Premnishanth Kothandaraman,
Ramanathan Nachiappan, SSN College of Engineering, India
The increasing modalities of data generation necessitates the development of machine learning techniques that can perform
efficient inference with multi-modal data. In this paper, we present an approach that can learn discriminant low-dimensional
projections from supervised multi-modal data for consensus inference. We construct intra- and inter-class similarity graphs for
each modality and optimize for consensus projections in the kernel space. We also provide methods for out-of-sample extensions
with novel test data. Classification results with standard multi-modal data sets show that the proposed consensus approach
performs better than classification using the individual modalities.
TP3a-2
Energy Price Matrix Factorization
1:55 PM
Vassilis Kekatos, University of Minnesota, United States
Statistical learning tools are applied here to study the potential risks of revealing the topology of the underlying power grid using
publicly available market data. It is first recognized that the matrix of real-time locational marginal prices admits an interesting
bilinear decomposition: It can be expressed as the product of a sparse, positive definite matrix with non-positive off-diagonal
entries times a sparse and low rank matrix. A convex optimization problem involving sparse and low-rank regularizers is
formulated to recover the constituent matrix factors. The novel scheme yielded encouraging topology recovery results on market
data generated using the IEEE 14-bus grid.
TP3a-3
A New Reduction Scheme for Gaussian Sum Filters
2:20 PM
In this paper we propose a low computational complexity reduction scheme for Gaussian Sum Filters. Our method uses an initial
state estimation to find the active noise clusters and removes all the others. Since the performance of our proposed method relies
on the accuracy of the initial state estimation, we also propose five methods for finding this estimation. We provide simulation
results showing that with suitable choice of the initial state estimation, our proposed reduction scheme provides better accuracy
and precision when compared with other reduction methods.
TP3a-4
Exploring Upper Bounds on the Number of Distinguishable Classes
2:45 PM
Catherine Keller, MIT Lincoln Laboratory, United States; Gary Whipple, Laboratory for Telecommunication
Sciences, United States
Information theoretic upper bounds on the number of distinguishable classes enable assessments of feasibility when applying
classification techniques. A goal of this paper is to examine the behavior of these upper bounds as the items being classified
becomes more complex in the sense that the number of degrees of freedom increases. We synthesize filters with different
numbers of stages to represent items with varying levels of complexity. We examine the behavior of feature scatter statistics and
the Fano upper bound for the number of distinguishable classes as a function of SNR, to make the comparisons.
75
Track D – Signal Processing and Adaptive Systems
Session: TPb3 – Sparse Signal Recovery
Co-Chairs: Daniel Palomar, Hong Kong University of Science and Technology and Gonzalo
Mateos, University of Rochester
TP3b-1
Compression Schemes for Time-Varying Sparse Signals
3:30 PM
Sundeep Prabhakar Chepuri, Geert Leus, Delft University of Technology, Netherlands
In this paper, we will investigate adaptive compression schemes for time-varying sparse signals. In particularly, we focus on
designing sparse compression matrices. Sparse sensing (i.e., a sparse compression matrix) leads to a decentralized compression
which is important for distributed sampling, and thus minimizes the number of sensors. The compression matrices at each time
step are designed based on the entire history of measurements and known dynamics. The compression matrices are determined by
evaluating a function of the a posteriori error covariance, such that the selected subset of sensors minimizes the estimation error.
TP3b-2
A Fast Algorithm for Sparse Generalized Eigenvalue Problem
3:55 PM
Junxiao Song, Prabhu Babu, Daniel Palomar, Hong Kong University of Science and Technology, Hong Kong SAR
of China
We consider an L0-norm penalized formulation of the generalized eigenvalue problem, aimed at extracting the leading sparse
generalized eigenvector of a matrix pair. To attack the problem, we first approximate the L0-norm by a continuous surrogate
function. Then an algorithm is developed via iteratively majorizing the surrogate function by a quadratic separable function,
which at each iteration reduces then to a regular generalized eigenvalue problem. An efficient specialized algorithm for finding
the leading generalized eigenvector is provided. Numerical experiments show that the proposed algorithm outperforms existing
algorithms in terms of both computational complexity by orders of magnitude and support recovery.
TP3b-3
Bootstrapped Sparse Bayesian Learning for Sparse Signal Recovery
4:20 PM
Ritwik Giri, Bhaskar Rao, University of California, San Diego, United States
In this article, the sparse signal recovery problem is studied in a hierarchical Bayesian framework and a novel Bootstrapped
Sparse Bayesian Learning method is developed. In SBL the choice of prior over the variances of the Gaussian Scale mixture
has been an interesting area of research and it still remains an open and interesting question. This motivates our use of a more
generalized maximum entropy density as the prior leading to a new variant of SBL. It is shown to perform better than traditional
SBL empirically and also found to accelerate the convergence and make the pruning procedure more robust.
TP3b-4
4:45 PM
A Fast Proximal Gradient Algorithm for Reconstructing Nonnegative Signals with Sparse
Transform Coefficients
Renliang Gu, Aleksandar Dogandžic, Iowa State University, United States
We develop a fast proximal gradient scheme for reconstructing nonnegative signals that are sparse in a transform domain from
underdetermined measurements. We adopt the unconstrained regularization framework where the objective function to be
minimized is a sum of a data fidelity (negative log-likelihood) term and a regularization term that imposes signal nonnegativity
and sparsity via an l1-norm constraint on the signal’s transform coefficients. This objective function is minimized via Nesterov’s
the numerical examples, we demonstrate the performance of the proposed method.
76
Track A – Communications Systems
Session: TPa4 – Optical Communications
Chair: Philippe Ciblat, TELECOM ParisTech
TP4a-1
1:30 PM
Fifth-Order Volterra Series Based Nonlinear Equalizer for Long-Haul High Data Rate
Optical Fiber Communications
Abdelkerim Amari, Philippe Ciblat, Yves Jaouen, Telecom ParisTech, France
We propose a fifth-order Inverse Volterra Series Transfer Function based nonlinearities compensation for ultra high data rate
optical fiber communications using OFDM. The main contribution consists of the derivations of the corresponding fifth-order
kernel. Compared to the third-order case, we significantly improve the performance in terms of BER and/or transmission
distance.
TP4a-2
Improving the Ultraviolet Scattering Channel Via Beam Reshaping
1:55 PM
Difan Zou, Shang-Bin Li, Zhengyuan Xu, School of Information Science and Technology, and Optical Wireless
Communication and Network Center, China
The beam reshaping method is adopted for improving the efficiency in the non-line of sight (NLOS) ultraviolet scattering
communication channel. By random scattering trajectory Monte Carlo simulation, the numerical results show the beam with
rectangular or elliptical photometries has significant advantage in the received signal intensity against the circular one. The
influences of the geometric configurations of both the transmitter and the receiver on the received signal intensity are discussed.
The corresponding impulse responses are also analyzed.
TP4a-3
2:20 PM
Correlation Study on the SIMO Channel Output of NLOS Optical Wireless
Communications
Boyang Huang, Chen Gong, Zhengyuan Xu, University of Science and Technology of China, China
The gain of single-input multiple-output (SIMO)/multiple-input multiple-output (MIMO) communication over single-input
single-output (SISO) communication critically depends on the correlation on the link gain between the transmitting-receiving
antenna pairs. In this work, in order to evaluate the SIMO gain of non-line of sight (NLOS) optical wireless communication, we
study such link channel correlation based on the channel generated from stochastic physics. In simulations with one transmitter
to 20m; and for transmitter-receiver distance 1000m, the correlation is larger than $0.4$ for the receiver-receiver distance up to
200m.
TP4a-4
2:45 PM
An Improved Performance Decoding Technique for Asymmetrically and Symmetrically
Clipped Optical (ASCO)-OFDM
Nan Wu, Yeheskel Bar-Ness, New Jersey Institute of Technology, United States
We propose an improved receiving technique for asymmetrically and symmetrically clipped optical (ASCO)-OFDM intensity
modulation direct detection (IM/DD) wireless communication systems. At the receiver, the ACO-OFDM symbols can be easily
obtained by extracting the data from the odd subcarriers; the SCO-OFDM symbols can be obtained by subtracting both the
estimated ACO-OFDM clipping noise and the SCO-OFDM clipping noise from the even subcarriers. The symbol error rate
performance of SCO-OFDM signals depends on the precision of ACO-OFDM signals. Thus, we apply an improved ACO-OFDM
receiving technique in our current receiver to further improve the SER performance of the whole ASCO-OFDM signal.
77
Track A – Communications Systems
Session: TPb4 – Energy Harvesting Wireless Communications
Chair: Sennur Ulukus, University of Maryland
TP4b-1
On the Capacity of the Energy Harvesting Channel with Energy Transfer
3:30 PM
Aylin Yener, Pennsylvania State University, United States
Energy harvesting wireless communication refers to communicating via wireless devices that acquire the energy needed for their
operation from nature. Such devices can cooperate not only at the signal level, but also by sharing their harvested energy in order
to improve network performance. While recent work addressed throughput optimizing policies for such channels, information
theoretic performance limits are not yet established. In this work, we consider a Gaussian energy harvesting channel with two
energy harvesting nodes and compute its capacity in the presence of the possibility of energy transfer between the two.
TP4b-2
Sum-rate Analysis for Systems with Wireless Energy Transfer
3:55 PM
Rania Morsi, Derrick Wing Kwan Ng, Robert Schober, Friedrich-Alexander University of Erlangen-Nuremberg,
Germany
Energy harvesting based mobile communication system design enables self-sustainability of energy constrained wireless devices.
This paper studies the performance of wirelessly powered communication systems. We analyze the sum-rate of a protocol
where single antenna receivers harvest energy from a multiple antenna transmitter in the downlink by wireless energy transfer to
support their wireless information transmission in the uplink. We first derive the optimal downlink energy transmission strategy
maximizing the total harvested energy at the receivers subject to a total transmit power constraint. Then, an analytical expression
for the uplink sum-rate is derived to provide valuable insights for system design. Simulation results verify our analytical results
and illustrate the trade-off between sum-rate and energy transfer.
TP4b-3
Optimal Energy Routing in Networks with Energy Cooperation
4:20 PM
B. Gurakan, O. Ozel, Sennur Ulukus, University of Maryland, United States
We consider a multi-user multi-hop wireless communication network where all nodes can harvest energy from nature and all
nodes can transfer energy from one to another. This is a model of an energy self-sufficient, energy self-sustaining autonomous
wireless network. In this network, we determine the jointly optimum data packet scheduling and wireless energy transfer policies
in order to maximize the end-to-end throughput. These policies determine the joint optimum flow of information and energy in
the network.
TP4b-4
4:45 PM
Kaibin Huang, University of Hong Kong, Hong Kong SAR of China
from mobiles to the cloud can improve their computation capability and reduce energy consumption. Nevertheless, offloading
also consumes energy for transmission. Given this tradeoff, it is critical to optimize the offloading process based on the states of
the channel, energy and computation tasks so as to cope with energy intermittence and maximize the battery life. In this paper,
mobile offloading is formulated as a Markov decision process based on multiple Markov chains modeling the random energy
arrivals, dynamic computation tasks and wireless channel. The structure of the optimal policy is analyzed using stochastic
optimization theory. Furthermore, the fundamental gains of mobile offloading in terms of computation capability and battery life
are quantified.
78
Track H – Speech, Image and Video Processing
Session: TPa5 – Speech Enhancement
Chair: Dalei Wu, Nanjing University of Posts and Telecommunications
TP5a-1
1:30 PM
Noise Power Spectral Density Matrix Estimation Based on Improved IMCRA
Qipeng Gong, Benoit Champagne, Peter Kabal, McGill University, Canada
In this paper, we present a new method for noise power spectral density (PSD) matrix estimation based on IMCRA which
consists of two parts. For the auto-PSD (diagonal) estimation, we propose a modification to IMCRA where a special level
detector is employed to improve the tracking of non-stationary noise backgrounds. For the cross-PSD (offdiagonal) estimation,
we propose to calculate a smoothed cross-periodogram by using estimated noise components derived as residuals after the
application of a speech enhancement algorithm on the individual microphone signals. Simulation results show the effectiveness
of our proposed approach in estimating the noise PSD matrix and its robustness against reverberation when used in combination
with an MVDR-based speech enhancement system.
TP5a-2
1:55 PM
BI-CosampSE: Block Identification based Compressive Sampling Matching Pursuit for
Speech Enhancement
Dalei Wu, Nanjing University of Posts and Telecommunications, China; Wei-Ping Zhu, M.N.S. Swamy, Concordia
In this paper, we propose a novel method to tackle this problem by using a block based identification strategy (BIS) to seek
the most prominent components in the observed data to update the sparse estimate of CoSaMP. The proposed method has been
found to be very effective to reduce musical noise in speech enhancement, in combination with some time-frequency smoothing
techniques.
TP5a-3
Pitch Estimation for Non-Stationary Speech
2:20 PM
Mads Græsbøll Christensen, Jesper Rindom Jensen, Aalborg University, Denmark
Recently, parametric methods based on the harmonic model have proven to be capable of overcoming the problems of
correlation-based methods. However, the argument against parametric methods is that the model is wrong, particularly for nonstationary signals like speech. To address this, we propose a new chirp model for pitch estimation in speech. This model takes the
non-stationary nature of the pitch explicitly into account, and we derive an estimator for determining the parameters of the model.
In experiments, we investigate the properties and capabilities of the model and the estimator and investigate whether it is needed
for pitch estimation.
TP5a-4
2:45 PM
Estimating the Noncircularity of Latent Components within Complex-Valued Subband
Mixtures with Applications to Speech Processing
Greg Okopal, Scott Wisdom, Les Atlas, University of Washington, United States
This paper describes an approach that estimates the circularity coefficients of multiple underlying components within complex
subbands of an additive mixture of voiced speech and noise via the strong uncorrelating transform (SUT). For the SUT to be
effective, the latent source signals must have unique nonzero circularity coefficients; this requirement is satisfied by using narrow
filters to impose a degree of noncircularity upon what would typically be circular noise. The circularity coefficient estimates are
then used for voice activity detection, pitch tracking, and enhancement.
79
Track B – MIMO Communications and Signal Processing
Session: TPb5 – Full Duplex MIMO Radio
Chair: Yingbo Hua, University of California, Riverside
TP5b-1
3:30 PM
Non-Linear Distortion Cancellation in Full Digital Domain for Full Duplex Radios
Yang-Seok Choi, Feng Xue, Roya Doostnejad, Shilpa Talwar, Intel Corporation, United States
Analogue domain cancellation of echoes for full duplex has been suggested. However, still considerable power of residual echoes
appears at receive chain. The residual echoes include not only linear term of transmitted signal but also non-linear terms as well
due to non-linear behavior of power amplifier (PA). In this presentation, we propose methods of cancelling the residual echoes
using adaptive filter in digital domain. PA can be modeled by linear combination of multiple kernels. Each kernel goes to an
adaptive filter and each adaptive filter estimates the corresponding kernel of received residual echoes and cancels the residual
echoes.
TP5b-2
Blind Digital Tuning for Interference Cancellation in Full-Duplex Radio
3:55 PM
Yingbo Hua, University of California, Riverside, United States
Interference cancellation is critical for full-duplex radio where self-interferences must be removed. Although the source of an
interference in such context can be tapped, the hardware impairments such as phase noises and IQ imbalances severely limit the
performance of cancellation by conventional adaptive methods. Also the high cost of obtaining accurate measurement of the RF
signal before each attenuator in an adaptive transversal RF filter makes the conventional methods infeasible. In this paper, we
present a recent progress in developing a blind digital tuning strategy. This strategy avoids the use of expensive hardware and is
robust against hardware impairments.
TP5b-3
4:20 PM
Daniel Bliss, Yu Rong, Arizona State University, United States
In-band (cochannel) full-duplex multiple-input multiple-output (MIMO) radios with antenna reuse employ an array of antennas
for which each antenna transmits and receives simultaneously. One of the most significant challenges is the mitigation of
the radio’s self-interference. We consider the performance constraints of these MIMO full-duplex nodes. We consider three
self-interference mitigation approaches including there interactions under the assumption of nonideal hardware. The selfinterference mitigation approaches include circulators, radio frequency (RF) active suppression, and temporal baseband selfinterference mitigation. We assume simultaneous communications and channel probing. We explore performance as a function of
parameterized hardware nonidealities.
TP5b-4
4:45 PM
Xu Du, Rice University, United States; Christopher Dick, Xilinx Incorporated, United States; Ashutosh Sabharwal,
Rice University, United States
In this paper, we study the use of full-duplex radios for continuous feedback of channel state information in MIMO broadcast
channels. The simultaneous transmission of feedback on the same frequency band as downlink transmissions causes inter-node
interference at the receiver. We quantify the impact of this inter-node interference and associated tradeoffs in the design of the
feedback channel.
80
Track E – Array Signal Processing
Session: TPa6 – Passive and Multistatic Radars
Chair: Muralidhar Rangaswamy, Air Force Research Labs
TP6a-1
Passive Multistatic Radar Based on Long-term Evolution Signals
1:30 PM
Sandeep Gogineni, Wright State Research Institute, United States; Muralidhar Rangaswamy, Wright Patterson Air
Force Base - AFRL, United States; Arye Nehorai, Washington University in St. Louis, United States
Passive multistatic radar has drawn much attention in recent years owing to the several advantages of operating in a distributed
configuration that have already been demonstrated for active radar. In a passive setup, the transmitted signal can be selected from
among several illuminators occupying the electromagnetic spectrum. In this paper, we compute the non-coherent and coherent
ambiguity function expressions using 4G long term evolution signals as the illuminators of opportunity.
TP6a-2
1:55 PM
A Correlation-Based Signal Detection Algorithm in Passive Radar with DVB-T2 Emitter
Guolong Cui, Hongbin Li, Stevens Institute of Technology, United States; Braham Himed, Air Force Research
Laboratory, United States
This paper considers target detection in passive radar that employs a digital video broadcasting-terrestrial version 2 (DVB-T2)
emitter as an illuminator of opportunity. The target detection is equivalent to identifying the presence/absence of the DVB-T2
signal in the returns. A correlation-based detection strategy is proposed by exploiting a unique C-A-B structure of the P1
symbol that is ubiquitous in all DVB-T2 transmissions. The P1 symbol, originally introduced for a DVB-T2 receiver to obtain
synchronization, is exploited here for target detection. The performance of the proposed detector is evaluated by Monte Carlo
simulations. Our results show that the proposed detector can reliably detect the target without full knowledge of the DVB-T2
signal waveform (except for the C-A-B structure).
TP6a-3
2:20 PM
Improving Multistatic MIMO Radar Performance in Data-Limited Scenarios
Tariq Qureshi, Muralidhar Rangaswamy, Air Force Research Laboratory, United States; Kristine Bell, Metron Inc.,
United States
A MIMO Multistatic radar system consists of multiple bistatic MIMO pairs working in potentially different configurations.
If a bistatic pair in a Multistatic MIMO radar system employs multiple transmit and receive elements, this increases the
dimensionality of the data received over a Coherent Processing Interval (CPI), which in turn increases the training data needed
to reliably estimate the covariance matrix. This, coupled with the non-stationarity in the received data resulting from the bistatic
geometry further degrades the quality of the covariance matrix estimate used in the adaptive detector. In [1], Kristine Bell et al.
presented a physics based MIMO clutter model, and showed that lack of training data support renders the MIMO radar unfeasible
in that the individual bistatic pairs can outperform the overall MIMO system. In this paper, we investigate techniques to improve
the performance of the multistatic MIMO radar in data limited scenarios. More specifically, we seek a parametric approximation
to the clutter as an AR process, resulting in a reduction in the amount of data that is needed to reliably estimate the AR
parameters. We compare the performance of the parametric approximation to the case where the covariance matrix is estimated
as a sample average using the same amout of training data.
TP6a-4
Market based Sensor Mobility Management for Target Localization
2:45 PM
Nianxia Cao, Swastik Brahma, Pramod Varshney, Syracuse University, United States
We propose a framework for the mobile sensor scheduling problem in target localization by designing an equilibrium based
two-sided market model where the fusion center (FC) is modeled as the consumer and the mobile sensors are modeled as the
producers. The FC motivates the sensors to optimally relocate themselves that maximizes its information gain for estimation.
On the other hand, the sensors calculate their own best moving distances that maximize their profits. Price adjustment rules are
designed to compute the equilibrium prices and moving distances, so that a stable solution is reached. Simulation experiments
show the effectiveness of our model.
81
Track G – Architecture and Implementation
Session: TPb6 – Many-Core Platforms
TP6b-1
3:30 PM
Towards Modeling and Analyzing Performance of LTE Base Station Software
Konstantin Popov, SICS, Sweden; Mats Brorsson, KTH Royal Institute of Technology, Sweden
We present a software model of LTE uplink data processing in 3GPP Radio Base Stations (eNodeB). The model is developed in
ArchiDeS, a a lean yet expressive framework for developing message-passing component-based applications. ArchiDeS enables
application- and platform-specific scheduling of parallel execution on multicore architectures. ArchiDeS’ implementation as a
C++ library has low and predictable overhead. The LTE model captures all available concurrency at the component level, which
enables parallel execution on large-scale multicore systems such as Tilera. Still, executing the model on tens of cores requires
careful scheduling and synchronization. We present and analyze our experiences with the model and ArhiDeS.
TP6b-2
3:55 PM
REPLICA T7-16-128 - A 2048-threaded 16-core 7-FU Chained VLIW Chip Multiprocessor
Martti Forsell, Jussi Roivainen, VTT, Finland
Processor-based solutions are getting increasingly popular over dedicated logic/accelerators among embedded system designers
due to their flexibility. The drawbacks - weaker performance and higher power consumption - are usually compensated with
application-specific multicore technologies. Unfortunately, these make programming difficult and result to less flexible
designs. REPLICA is VTT’s effort to solve the performance and programmability problems of current multicore processors
without tampering flexibility. In this paper we introduce T7-16-128 - a 2048-threaded 16-core prototype of the REPLICA chip
multiprocessor. The main principles of the architecture and structure of the prototype are explained. Preliminary comparison to
current alternatives is given.
TP6b-3
4:20 PM
Improving Image Quality by SSIM Based Increase of Run-Length Zeros in GPGPU JPEG
Encoding
Stefan Petersson, Håkan Grahn, Blekinge Institute of Technology, Sweden
This paper proposes an algorithm to improve the experienced quality in JPEG encoded images. The algorithm improves the
quality in detailed areas while reducing the quality in less detailed areas of the image, thereby increasing the overall experienced
quality without increasing the image data size. The algorithm is based on the SSIM metric and an efficient GPU implementation
is presented.
TP6b-4
4:45 PM
Kickstarting High-Performing Energy-Efficient Manycore Architectures with Epiphany
Tomas Nordström, Zain ul-Abdin, Halmstad University, Sweden; Andreas Olofsson, Adapteva, United States
In this paper we introduce Epiphany as a high-performing energy-efficient manycore architecture suitable for high-end
embedded systems. The outstanding performance per Watt (50 GFlops/W) makes this architecture a very strong candidate for
all applications that do significant signal processing in embedded and mobile environments. We have exemplified the use of
Epiphany in two such applications, radar applications and video processing. We have furthermore looked at various development
environments and languages for this architecture. Finally we will discuss what additional architectural features can be expected in
future generations of Epiphany.
82
Track G – Architecture and Implementation
Session: TPa7 – Design Methodologies for Signal Processing
Chair: Chris Lee, NCKU
TP7a-1
Finding Fast Action Selectors for Dataflow Actors
1:30 PM
Gustav Cedersjö, Jörn W. Janneck, Jonas Skeppstedt, Lund University, Sweden
Recent shift towards more parallel computing platforms and the popularization of stream applications such as signal processing,
video encoding an cryptography has renewed the interest in dataflow programming. This paper builds on previous work on
efficient implementations the basic elements of a dataflow program, the actors, and investigates heuristics for making the process
of selecting what to do in an actor faster.
TP7a-2
1:55 PM
Automatic Generation of Application Specific FPGA Multicore Accelerators
Pascal Schleuniger, Andreas Hindborg, Nicklas Bo Jensen, Maxwell Walter, Laust Brock-Nannestad, Lars
Bonnichsen, Christian W. Probst, Sven Karlsson, Technical University of Denmark, Denmark
High performance computing systems increasingly make use of hardware accelerators to improve performance and power
properties. For large high-performance FPGAs to be successfully integrated in such computing systems, methods to raise the
abstraction level of FPGAs programming are required. In this paper we propose a tool flow, which automatically generates highly
optimized hardware multicore systems based on parameters. Profiling feedback is used to adjust these parameters to improve
performance and lower the power consumption. For image and video processing applications, we show that our tools are able to
optimize the hardware to deliver competitive performance at a low power budget.
TP7a-3
2:20 PM
Dataflow Toolset for Soft-Core Processors on FPGA for Image Processing Applications
Burak Bardak, Fahad Manzoor Siddiqui, Roger Woods, Queen’s University Belfast, United Kingdom
This paper propose a design tool chain that uses dataflow language CAL[2] as a starting point and targets to custom design softcore processor on FPGA. The main purpose for the design tool is exploiting the task and data parallelism in order to achieve the
same parallelism as HDL implementation without dealing with the required design, verification and debugging steps of HDL
design, which increases the time to market, and design effort.
TP7a-4
An Enhanced and Embedded GNU Radio Flow
2:45 PM
Ryan Marlow, Peter Athanas, Virginia Polytechnic Institute and State University, United States
This paper presents a Zynq capable version of GNU Radio -- an open-source rapid radio deployment tool -- with an enhanced
flow that utilizes the processing capability of FPGAs. This work features TFlow -- an FPGA back-end compilation accelerator
for instant FPGA assembly. The Xilinx Zynq FPGA architecture integrates the FPGA fabric and CPU onto a single chip, which
eliminates the need for a controlling host computer; thus, providing a single, portable, low-power, embedded platform. By
exploiting the computational advantages of FPGAs in the GNU Radio flow, a larger class of software defined radios can be
implemented.
Track A – Communications Systems
Session: TPb7 – Optical Wireless Communications
Chair: Zhengyuan (Daniel) Xu, University of Science and Technology of China
TP7b-1
Multiuser MISO Indoor Visible Light Communications
3:30 PM
Visible light communications using LED fixtures simultaneously for indoor illumination and for transmission offer the promise
of high throughput data connectivity in an energy and cost efficient manner. In this paper we explore algorithms for supporting
many users concurrently by optimizing the use of individual LEDs in each luminary. The directivity and nonlinearity of each
LED is considered when assigning multiple LEDs to a user, forming a multiple input single output (MISO) system. Exploiting
83
the spatial separation of detectors, MISO techniques using CDMA and MMSE detection can offer high performance to many
simultaneous users while preserving the properties of the lighting system, such as spatially and temporally continuous dimmable
illumination.
TP7b-2
Optical Spatial Modulation OFDM using Micro LEDs
3:55 PM
Muhammad Ijaz, Dobroslav Tsonev, Abdelhamid Younis, University of Edinburgh, United Kingdom; Jonathan J.
D. McKendry, Erdan Gu, Martin Dawson, University of Strathclyde, United Kingdom; Harald Haas, University of
Edinburgh, United Kingdom
This paper investigates the performance of optical spatial modulation (OSM) with orthogonal frequency division multiplexing
(OFDM) in a micro multiple-input multiple-output (µMIMO) based visible light communication (VLC) system. The micro
light emitting diode (µLED) based cluster devices are considered in the current investigations. The simulation results show that
a maximum achievable data rate using OSM-OFDM is 4.6 Gb/s using adaptive bit loading for 2×2 µMIMO. The results also
indicate that due to the highly correlated channels, the system performance is largely dependent on the spatial separation and the
light emission profile of the µLEDs in the clusters.
TP7b-3
4:20 PM
Adaptation of OFDM under Visible Light Communications and Illumination Constraints
Thomas Little, Hany Elgala, Boston University, United States
OFDM is increasingly studied and adopted as a modulation technique for RF and OW communication systems. In this paper
we investigate challenges to the adoption of OFDM for use in lighting systems that support both intensity control and data
communication. In particular, we survey the requirements for energy efficiency, intensity control (dimming), and LED driver
integration in lighting systems. These requirements are mapped to contemporary and novel OFDM adaptations to show how
both the lighting and communications needs can be met in dual-use scenarios while preserving both missions with reasonable
performance.
TP7b-4
4:45 PM
Hybrid Dimmable Visible Light -with Infra-Red Optical Wireless Communications
Andrew Burton, Z Ghassemlooy, Edward Bently, Hoa LeMinh, Northumbria University, United Kingdom; S K
Laiw, National Taiwan University of Science and Technology, Taiwan; Chung Ghiu Lee, Chosun University,
Republic of Korea
This paper presents a new dimming technique for LEDs using a transparent pulse width modulation (PWM) scheme. A
combination of white visible light and infrared (IR) LEDs are used to ensure data link availability at all times. When the
visible LEDs are off the IR LEDs will be on and and vice versa. This hybrid lighting and data communication scheme ensure
data communication even when visible light switched off. Since PWM signal is made transparent to the receiving electronics
we drastically reduce the inter modulation interference (IMI) between the PWM and the data channel, and the need for
synchronization at the transmitter between the two signals. Results show a bit error rate of ≤ 1e-6 for all data within the system
bandwidth for all dimming levels.
Track A – Communications Systems
Session: TPa8 – Cognitive Radio II
1:30 PM–3:10 PM
TP8a1-1
Characterization of Outage Performance for Cognitive Relay Networks with Mixed Fading
Efthymios Stathakis, Lars K. Rasmussen, Mikael Skoglund, Royal Institute of Technology (KTH), Sweden
We consider a dual-hop underlay cognitive radio network with a single transceiver pair, which utilizes an amplify-and-forward
relay to establish end-to-end communication. The secondary nodes, i.e., the transmitter and the relay, obey transmit power
constraints which guarantee that the instantaneous peak interference at the primary receiver will not exceed a certain threshold.
Each of the secondary communication links contains a line-of-sight component whereas the external links, to the primary
receiver, are subject to Rayleigh fading. For this system model, we analyze the outage probability and demonstrate the accuracy
of the obtained mathematical expressions via numerical simulations.
84
TP8a1-2
Restless Multi-Armed Bandits under Time-Varying Activation Constraints
Kobi Cohen, Qing Zhao, Anna Scaglione, University of California, Davis, United States
We consider a class of restless multi-armed bandit (RMAB) problems, in which a player chooses K(t) out of N arms to play
at each given time t. Each arm evolves according to a two-state Markov chain, independent of the player’s action. While the
problem is in general PSPACE-hard, we focus on the optimality of the myopic policy under a time-varying activation constraint
K(t). The problem under study finds applications in various communication networks and also applies to the compressive
spectrum sensing problem in cognitive radio networks.
TP8a1-3
On the Optimal Relay Design for Multi-Antenna Cognitive Two-Way AF Relay Networks
Maksym Girnyk, KTH Royal Institute of Technology, Sweden; Mikko Vehkaperä, Sergiy Vorobyov, Aalto
University, Finland
Cognitive two-way relaying is an efficient method for tackling the problem of spectrum scarcity by serving new (secondary)
users, while keeping the existing (primary) users satisfied with their service. Moreover, additional gains can be attained from
employment of the two-way relaying with multipleantenna relays within the secondary network. In this paper, we consider an
underlay two-way cognitive network and propose an efficient algorithm for computation of a (nearly) optimal relay precoder
matrix subject to the interference constraint towards the primary network. The efficiency of the proposed solution will be
highlighted by means of numerical simulations in the full version of the paper
TP8a1-4
Network Aware Spectrum Efficiency Metric for Heterogeneous and Dynamic Radio
Environments
Aditya Padaki, Ravi Tandon, Jeffrey Reed, Virginia Polytechnic Institute and State University, United States
In this paper, we formalize a new definition for spectrum efficiency, with the specific goal of addressing the diverse needs and
requirements of various technologies and users. Existing metrics for spectrum efficiency are insufficient for future systems which
employ dynamic allocation schemes. We introduce a parameterized definition for spectrum efficiency dependent on the network
dynamics, radio environment and diverse requirements of technologies. This metric accounts for the frequency use and reuse,
interference footprint of a user, and has a parameter to specify priority/importance for users/bits (e.g. public safety). We then
evaluate the spectrum efficiency regions for three different network architectures.
TP8a1-5
A Unified Framework for Robust Cooperative Spectrum Sensing
Qi Cheng, Eric Chan-Tin, Oklahoma State University, United States
In cognitive radio, spectrum sensing performance may be degraded by various sensor faults and/or security threats, including
device malfunctions and Byzantine attacks. We propose a robust spectrum sensing framework including two steps of faulty node
detection followed by faulty node elimination or correction before decision fusion. The first step explores the decision statistics
over time to identify faulty nodes. The second step relies on the mutual behavior check among the remaining nodes. Clustering is
applied to decision sequences to distinguish faulty from normal nodes, and faulty model estimation, which is then used for data
correction.
TP8a1-6
Receiver Configuration and Testbed Development for Underwater Cognitive
Channelization
George Sklivanitis, Emrecan Demirors, Stella N. Batalama, Tommaso Melodia, Dimitris A. Pados, State University
of New York at Buffalo, United States
We propose a receiver configuration and we develop a software-defined-radio testbed for real-time cognitive underwater
multiple-access communications. In particular, the proposed receiver is fully reconfigurable and executes (i) all-spectrum
cognitive channelization and (ii) combined synchronization, channel estimation, and demodulation. Real-time experimental
results with in-house built modems demonstrate our theoretical developments and show that cognitive channelization is a
powerful proposition for underwater communications and leads to significant improvement of the spectrum utilization. Even in
the absence of interference, due to the noise characteristics of the acoustic channel, cognitive channelization offers significant
performance improvements in terms of receiver pre-detection signal-to-interference-plus-noise-ratio and bit-error-rate.
85
TP8a1-7
Estimation of Subspace Occupancy
Kaitlyn Beaudet, Douglas Cochran, Arizona State University, United States
The ability to identify unoccupied resources in the radio spectrum is a key capability for opportunistic user in a cognitive radio
environment. This paper draws upon and extends geometrically based ideas in statistical signal processing to develop estimators
for the rank and the occupied subspace in a multi-user environment from multiple temporal samples of the signal received at a
single antenna. These estimators enable identification of resources (i.e., the orthogonal complement of the occupied subspace)
that may be exploitable by an opportunistic user.
TP8a1-8
Performance Analysis: DF Cognitive Network with Transceiver Imperfections
Dang Khoa Nguyen, Kyushu Institute of Technology, Japan; Tu Thanh Lam, Post and Telecommunications Institute
of Technology, Viet Nam; Hiroshi Ochi, Kyushu Institute of Technology, Japan
We comprehensively analyze the outage performance of dual-hop decode-and-forward (DF) cognitive network subject to
independent non-identical distributed Rayleigh fading with the presence of hardware impairment level in the model of transceiver
nodes of cognitive network.Closed-form expressions of exact and asymptotic of outage probability of the DF cognitive network
are derived. A numerical simulation study is showed to corroborate our analysis results. Thereby, we found that hardware
impairment level sharply impact outage performance of DF cognitive network. The influence of this factor is small at the low
transmit power but the bad effect is higher when transmission power is increased.
Track D – Signal Processing and Adaptive Systems
Session: TPa8 – Signal Processing Methods
1:30 PM–3:10 PM
Chair: Seung Jun Kim, University of Maryland, Baltimore County
TP8a2-1
Blind Equalization Based On Blind Separation with Toeplitz Constraint
Zhengwei Wu, Saleem Kassam, University of Pennsylvania, United States
Blind equalization (BE) has been modeled as a blind source separation (BSS) problem and achieved using BSS algorithms.
We show that the Toeplitz structure of the mixing matrix in the BSS model for BE can be exploited for faster convergence and
better performance. A length constraint on the equalizer impulse response provides further improvement. We use the equivariant
adaptive separation via independence (EASI) algorithm to illustrate the ideas, although the approach is generally applicable.
Simulation results and comparisons are given. The method can be extended for multiple channels and fractional sampling.
TP8a2-2
Piecewise-Constant Recovery via Spike-and-Slab Approximate Message-Passing using a
Scalarwise Denoiser
Jaewook Kang, Heung-No Lee, Kiseon Kim, Gwangju Institute of Science and Technology (GIST), Republic of
Korea
This paper proposes a novel AMP algorithm for recovery of piecewise-constant signals under the compressed sensing framework,
called ssAMP. The ssAMP solver includes a low-complex scalarwise denoiser; therefore, its overall complexity is significantly
reduced in a high-dimensional setting compared to an existing AMP for the piecewise-constant recovery, TV-AMP. In addition,
the ssAMP iteration consists of fully scalarwise operations. Hence, the ssAMP is further accelerated via parallelization, whereas
TV-AMP cannot be parallelized. We provide a summary of the algorithm construction, discussing the superiority of ssAMP
compared to the conventional total variation approaches through experimental simulations.
TP8a2-3
Resource Allocation Optimization for Distributed Vector Estimation with Digital
Transmission
Alireza Sani, Azadeh Vosoughi, University of Central Florida, United States
We consider the problem of distributed estimation of an unknown random vector with a known covariance matrix in a wireless
sensor network. Sensors transmit their binary modulated quantized observations to a fusion center(FC), over orthogonal MAC
channels subject to fading and additive noise. Assuming the FC employs the linear minimum mean-square error (MMSE)
86
estimator, we obtain an upper bound on MSE distortion. We investigate optimal resource allocation strategies that minimize the
MSE bound, subject to total bandwidth (measured in quantization bits) and total transmit power constraints. Our simulation show
that the proposed scheme outperforms uniform bit and power allocation scheme.
TP8a2-4
Exploiting the Cramér-Rao Bound for Optimised Sampling and Quantisation of FRI
Signals
Andre Angierski, Volker Kuehn, University of Rostock, Germany
This contribution considers the sampling process for Finite Rate of Innovation signals including quantisation errors and additive
white Gaussian noise. For some applications the total amount of bits available for the sampling process is constrained, e.g. due to
memory limitations. Thus, the sharing of these bits between sampling rate and the quantisation accuracy has to be optimised. In
particular, the Cramér-Rao bound is determined and the bit allocation is optimised w.r.t. the CRB. Finally, the analytical results
are compared with simulation results obtained by spectral estimation methods.
TP8a2-5
Adaptive Waveform for Integrated Detection and Identification of Moving Extended
Target
Jo-Yen Nieh, Ric Romero, Naval Postgraduate School, United States
We propose an improvement to the maximum a posteriori probability weighted eigenwaveform (MAP-PWE) adaptive waveform
design used in target recognition with a cognitive radar platform for which we call match-filtered PWE (MF-PWE). Our interest
however is to include moving targets in the identification problem. Combining range-Doppler map (RDM) technique with the the
PWE-based adaptive waveform techniques, we propose an integrated detection and identification scheme for moving extended
targets. Target detection performance comparison between wideband, MAP-PWE, and MF-PWE techniques are shown. It is
noted the MF-PWE performs better than the wideband and MAP-PWE.
TP8a2-6
Channel Gain Cartography Via Low Rank and Sparsity
Donghoon Lee, Seung-Jun Kim, University of Minnesota, United States
Channel gain cartography aims at inferring shadow fading between arbitrary points in space based on measurements (samples)
of channel gains taken from finite pairs of transceivers. Channel gain maps are useful for various sensing and resource allocation
tasks essential for the operation of cognitive radio networks. In this work, the channel gain samples are modeled as compressive
tomographic measurements of an underlying spatial loss field (SLF), which is postulated to have a low-rank structure corrupted
by sparse errors. Efficient algorithms to reconstruct the SLF are developed, from which arbitrary channel gains can be
interpolated.
TP8a2-7
Bayesian Cramér-Rao Bound for Distributed Estimation of Correlated Data with Nonlinear Observation Model
Mojtaba Shirazi, Azadeh Vosoughi, University of Central Florida, United States
In this paper we study the problem of distributed estimation of a random vector in wireless sensor networks (WSNs) with nonlinear observation models. Sensors transmit their modulated quantized observations over orthogonal erroneous wireless channels
(subject to fading and noise) to a fusion center, which estimates the unknown vector. We derive the Bayesian Cramer-Rao
Bound (CRB) matrix and study the behavior of its trace (through analysis and simulations), with respect to the observation and
communication channel signal-to noise ratios (SNRs). The derived CRB serves as a benchmark for performance comparison of
different Bayesian estimators, including linear MMSE estimator.
TP8a2-8
Multirate Processing Using Nested Sampling
Peter Vouras, Naval Research Laboratory, United States
This paper describes the multirate processing of random signals sampled on nested intervals. Nested sampling intervals consist of
nonuniformly spaced samples formed by concatenating two or more smaller intervals each with uniform sampling. By filtering
a vectorized version of the power spectral density matrix of the signal, the input-output behavior of conventional filter banks can
be replicated. Simulated examples are presented which describe the application of the proposed technique to adaptive doppler
87
Track H – Speech, Image and Video Processing
Session: TPa8 – Image Processing II
1:30 PM–3:10 PM
Chair: Ashkan Ashrafi, San Diego State University
TP8a3-1
Smoothed Rank Approximation Algorithms for Matrix Completion
Mohammed Al-Qizwini, Hayder Radha, Michigan State University, United States
We consider using smooth rank approximation functions to solve the matrix completion problem. Our main contribution in this
paper is deriving two robust algorithms using the Accelerated Proximal Gradient (APG) and the Alternating Direction Method
of Multipliers (ADM). Further, we compare both algorithms against each other and against the iterative reweighted least squares
(IRLS-1) algorithm using a variety of noisy images. The experiments show that using ADM achieves approximately 1.5 dB SNR
improvement over IRLS-1, while it needs comparable execution time to IRLS-1. Meanwhile, using APG saves about 50% of
IRLS-1’s computation time with lower SNR than ADM.
TP8a3-2
Visibility Prediction of Flicker Distortions on Naturalistic Videos
Lark Kwon Choi, Lawrence Cormack, Alan Bovik, University of Texas at Austin, United States
We conducted human studies where we found that the visibility of flicker distortions on naturalistic videos is strongly reduced in
the presence of large coherent object motions. Based on this finding, we propose a model of flicker visibility. The model predicts
target-related activation levels of neurons corresponding to the displayed video using spatiotemporal backward masking, then
applies flicker adaptation. Results show that predictions of flicker visibility using the model are highly consistent with human
perception of flicker distortions on naturalistic videos. We believe that these results are important for understanding temporal
perceptual distortions, and how to predict and ameliorate them.
TP8a3-3
Image Compression via Wavelets and Row Compression
Mary HudachekBuswell, Georgia Institute of Technology, United States; Michael Stewart, Saied Belkasim, Georgia
State University, United States
This work exploits a stable row compression algorithm for decomposing a hierarchically or sequentially structured matrix
to compress an n x n image represented by a wavelet transform. The multiresolution discrete wavelet transform is used to
decompose an image. The row compression algorithm builds up a low rank approximation of the wavelet transform by applying
orthogonal transformations and updating techniques. The cost is O(n^2) operations.
TP8a3-4
Low Complexity Dimensionality Reduction for Hyperspectral Images
Seda Senay, Hector Erives, New Mexico Institute of Mining and Technology, United States
Although optimal, due to its computational complexity and of its being data dependent Karhunen Loeve Transform (KLT) is
not practical to be used for data compression in resource constrained hyperspectral sensing platform. Based on their relationship
with the KLT, we propose using discrete prolate sheroidal sequences (DPSSs) in hyperspectral imaging such that DPSSs
decomposition can be applied as a suitable transform for compression. The performance of the proposed method is promising
and open to improvements for further accuracy and dimensionality reduction such as for detection of certain targets for which the
spectral signature is known.
TP8a3-5
Improving Image Clustering using Sparse Text and the Wisdom of the Crowds
Anna Ma, Claremont Graduate University, United States; Arjuna Flenner, Naval Air Warfare Center, United States;
Deanna Needell, Claremont McKenna College, United States; Allon Percus, Claremont Graduate University, United
States
We propose a method to improve image clustering using sparse text and the wisdom of the crowds. In particular, we present a
method to fuse two different kinds of document features, image and text features, and use a common dictionary or wisdom of
the crowds’’ as the connection between the two different kinds of documents. With the proposed fusion matrix, we use topic
modeling via non-negative matrix factorization to cluster documents.
88
TP8a3-6
Color Image Watermarking Using Quaternion Wavelets
Lahouari Ghouti, King Fahd University of Petroleum and Minerals, Saudi Arabia
In this paper, we propose a new color image watermarking algorithm using quaternion wavelets and semi-random low density
parity check (SRLDPC) codes. Qauternion wavelets enable efficient watermark embedding in the hypercomplex domain without
incurring additional computational complexity. The watermark detection is based on statistical maximum likelihood approaches.
The efficiency and data hiding capacity of the proposed watermark embedding scheme are found to be greatly enhanced by the
use of SR-LDPC codes.
TP8a3-7
Immersion Ultrasonic Array Imaging Using a New Array Spatial Signature in Different
Imaging Algorithms
Nasim Moallemi, Shahram Shahbazpanahi, University of Ontario Institute of technology, Canada
In this paper, we investigate the performance of a new array spatial signature for imaging the material under immersion ultrasonic
test. We have used this new array spatial signature in imaging algorithms including the conventional beamforming, MUSIC, and
Capon algorithms. These three methods traditionally proposed for a homogeneous medium where the sound velocity is constant
in the material under test. Note however that, in immersion ultrasonic test, the sound wave propagates with different speeds
in water and in solid test sample. The new array spatial signature has been developed using distributed source modeling of the
interface between water and solid.
TP8a3-8
A Proof on the Invariance of the Hirschman Uncertainty to the Rényi Entropy Parameter
and an Observation on its Relevance in the Image Texture Classification Problem
Kirandeep Ghuman, Victor DeBrunner, Florida State University, United States
In [1] we developed a new uncertainty measure which incorporates Rényi entropy instead of Shannon entropy. This new
uncertainty measure was conjectured to be invariant to the Rényi order alpha > 0 . We prove this invariance, and test whether this
invariance is predictive in the problem of texture classification for digital images. In preliminary results, we find that it does, in
that the recognition performance does not depend significantly on the Rényi parameter, as compared to the texture classification
performance without using entropy. We hope that these results will be extended to other problems where Rényi entropy is used.
Track C – Networks
Session: TPa8 – Sensor and Wireless Networks
1:30 PM–3:10 PM
Chair: Usman Khan, Tufts University
TP8a4-1
Design of Orthogonal Golomb Rulers with Applications in Wireless Localization.
Omotayo Oshiga, Giuseppe Abreu, Jacobs University Bremen, Germany
Golomb rulers are useful in wide applications in engineering. Yet, the design of multiple mutually orthogonal GRs finds no
solution in current literature. We present an algorithm to solve this problem. Our solution is based on modification of a classic
algorithm, which allows the construction of GRs out of constrained sets of marks, such that orthogonal rulers can be obtained. A
new algorithm is offered, which solves the intended problem and which indicates a gain when applied to generate optimal GRs.
Wireless localization is used to illustrate the gains achievable when employing orthogonal GRs to perform multipoint ranging.
TP8a4-2
Secrecy Outage Analysis of Cognitive Wireless Sensor Networks
Satyanarayana Vuppala, Jacobs University Bremen, Germany; Weigang Liu, Tharmalingam Ratnarajah, University
of Edinburgh, United Kingdom; Giuseppe Abreu, Jacobs University Bremen, Germany
We examine the secrecy outage of primary links in cognitive wireless sensor networks with interference from secondary users,
offering original and highly accurate expressions for the aggregate interference with fading and shadowing. It is found that the
presence of shadowing has a significant impact which can swiftly increase secrecy outage. The expressions derived can also be
used to obtain other analytical results such as secrecy rate and secrecy transmission capacity.
89
TP8a4-3
On the Convergence Rate of Swap-Collide Algorithm for Simple Task Assignment
Sam Safavi, Usman A. Khan, Tufts University, United States
This paper provides a convergence rate analysis of the swap-collide algorithm for simple assignment problems. Swap-collide
is a distributed algorithm that assigns a unique task to each agent assuming that the cost of each assignment is identical and has
applications in resource-constrained multi-agent systems; prior work has shown that this assignment procedure converges in
finite-time. In this paper, we provide an analytical framework to establish the convergence rate of swap-collide, and show that for
a network of size $N$, the lower and upper bounds for the convergence rate are $O({N}^{3})$.
TP8a4-4
On the Impact of Low-Rank Interference on Distributed Multi-Agent Optimization
Chenguang Xi, Usman A. Khan, Tufts University, United States
We study the impact of low-rank interference on the problem of optimizing a sum of convex functions corresponding to
multiple agents. We prove that the impact of interference can mathematically be regarded as additional constraints to original
unconstrained optimization. The proposed analysis uses the notion of interference alignment where the agent transmissions are
aligned in either the null space or range space of interference. We consider two cases:~(i) when the interference is uniquely
determined by the transmitter; and,~(ii) when the interference is only determined by the receiver. Experiments on distributed
source localization demonstrate good performance of our strategy.
TP8a4-5
Multipath-Aided Cooperative Network Localization Using Convex Optimization
Hassan Naseri, Mario Pereira da Costa, Visa Koivunen, Aalto University, Finland
Localization in the face of multipath propagation is a challenging task in sensor networks using radio, acoustic or underwater
signals for distance measurement. Multipath-aided network localization exploits multipath propagation to improve the
identifiability and performance of cooperative localization. In this paper the problem of multipath-aided network localization is
formulated as an optimization problem and a semidefinite relaxation is proposed for it.
TP8a4-6
Mobile Sensor Mapping via Semi-Definite Programming
Giuseppe Destino, Davide Macagnano, University of Oulu, Finland
We consider the problem of mapping the locations of a mobile device into the Euclidean space utilizing its perception of
the environment through sensors, e.g. WiFi. Based on the graph-based Simulatenous Localization and Mapping (SLAM)
formulation, a semi-definite programming approach is derived in order to ensure convergence. To obtain a semi-definite program
we exploit a convex likelihood model to constrain near mobile locations to similar environment perceptions as well as the
Euclidean distance matrix properties for the resulting trajectory. Comparison with the state-of-the-art, i.e Gaussian Process and
classic graph-SLAM methods will be provided in the final version of the paper.
TP8a4-7
Indoor Node Localization using Geometric Dilution of Precision in Ad-Hoc Sensor
Networks
Sudhir Kumar, Rajesh M. Hegde, Indian Institute of Technology Kanpur, India
In this paper, a new method for sensor node localization using geometric dilution of precision (GDOP) is described. In contrast
to the existing algorithms, the proposed algorithm is not constrained by fixed geometry of sensor node placement. Additionally,
the GDOP can be used for effective localization under both line-of-sight and non-line-of-sight communication between sensor
nodes in an ad-hoc sensor network. The robustness of the algorithm is due to the fair utilization of all measurements obtained
under NLOS conditions. Location estimates are obtained using the method of GDOP which has hitherto been used for optimal
placement of satellites. Algorithms using minimum and weighted-GDOP are discussed in the context of indoor sensor node
localization. Extensive simulations and real field deployments are used to evaluate the performance of the proposed algorithm.
The localization accuracy of the proposed algorithms is reasonably better when compared to similar methods in literature.
90
TP8a4-8
Efficient Consensus Synchronization via Implicit Acknowledgment
Andrew G. Klein, Western Washington University, United States; D. Richard Brown III, Worcester Polytechnic
Institute, United States
A technique for achieving synchronization in wireless networks using only existing traffic is developed. Prior work has either
ignored propagation delay, or has required bidirectional messages consisting of explicitly acknowledged unicast transmissions.
We develop an approach using “implicit acknowledgment” that achieves precise consensus synchronization by exploiting the
broadcast nature of the wireless medium. This significantly reduces the number of transmissions needed for synchronization
throughout the network, and is applicable to networks with unacknowledged multicast and broadcast traffic. Results suggest the
technique is effective for precise, low-overhead network synchronization, and numerical results are presented for two particular
network configurations.
Track A – Communications Systems
Session: TPb8 – Topics in Communication Systems
3:30 PM–5:10 PM
Chair: Alexios Balatsoukas-Stimming, EPFL
TP8b1-1
Performance Analysis of a MMSE Turbo Equalizer with LDPC in a FTN Channel with
Ghassan Maalouli, Brian A. Banister, Comtech EF Data, United States
The advent of digital wireless communications of the past two decades has created an unprecedented spectral demand. One of
the most demanding applications is Digital Video Broadcast (DVB). DVB bandwidth requirements have motivated academic as
well as practicing researchers to find more efficient schemes that can increase spectral efficiency. This led the DVB industry to
adopt higher order modulations and efficient coding techniques, such as LDPC, which resulted in performance that approaches
the Shannon limit to within a few fractions of a decibel. More recently, the DVB community has focused its attention on FTN
signaling as a method that may achieve higher capacity. This is attained by transmitting signals at a rate that is faster than the
Nyquist rate into a band-limited channel. In his seminal work, Harry Nyquist established the ISI-free limit at which a signal
can be transmitted through a channel. Emitting the signal at a faster rate incurs inter-symbol interference (ISI). If ISI is unmitigated, it will degrade system performance beyond the capacity improvement that is attained by FTN signaling, rendering the
approach useless. However, if ISI is mitigated, it is possible under certain scenarios to completely eliminate ISI or at least reduce
it such that there is a net capacity gain through the channel. Several researchers have studied the problem of eliminating ISI in
a multipath channel using turbo-equalization techniques. It was well established that the optimal equalizer comprises a trellis
that combines the channel’s memory as well as the decoder’s. However such architecture is suitable only for short channels and
lower order modulations. Otherwise, the size of the trellis will be computationally prohibitive for real-time applications even
with modern day technology. Therefore attention shifted towards architectures that are practically amicable. In this work, we
investigate the performance of computationally efficient, MMSE-based turbo-equalizers with a LDPC decoder and study their
ability to eliminate ISI in a FTN channel. We analyze the performance of MMSE with and without feedback in low and high
SNR regimes. We measure the SNR degradation that the system incurs after ISI mitigation. We quantify the net gain in capacity
that the system can potentially attain. We demonstrate that the MMSE suffers high degradation at low SNR but converges to
a few tenth of a dB from the zero ISI condition at higher SNR. On the contrary, a Soft-feedback-equalizer (SFE) suffers very
little degradation at low SNR but converges to a BER rate that is higher than the MMSE for the same SNR. Selecting the proper
structure is therefore dependent on the desired operating point of the receiver.
TP8b1-2
Characteristics of Optical Scattering and Turbulence Communication Channels
Weihao Liu, Zhengyuan Xu, University of Science and Technology of China, China
The optical scattering and turbulence channels are modeled by the semi-analytic and semi-numerical (SASN) method, in which
the ray tracing model based on Monte-Carlo method is used to track the multiple scattering ray and analytic method is to get
the irradiance distribution of each individual ray. Characteristics of the scattering and turbulence channels are uncovered as
follows: 1) mean path loss increase with turbulence strength; 2) scintillation index is much reduced because of the smooth effect
of multiple scattering; 3) the multiple scattering and turbulence channel can be well represented by the lognormal distribution
function of small variance.
91
TP8b1-3
Comparison of SNR and Peak-SNR (PSNR) Performance Measures and Signals for Peaklimited Two-Dimensional (2D) Pixelated Optical Wireless Communication
Eyal Katz, Yeheskel Bar-Ness, New Jersey Institute of Technology, United States
Two-dimensional(2D) pixelated-Optical-Wireless-Communication-Systems (OWCS), with Intensity-Modulation, DirectDetection (IM/DD), commonly use computer-display (transmit-side), and camera (receive-side). One-dimensional(1D) IM/DD
OWCS and channels, are average-power-limited, due to eye-safety rules. 2D pixelated-channels are peak-limited. However,
in-the-literature, both 1D and 2D systems-evaluations use Average-power Signal-to-Noise Ratio (Average-power SNR). We
first show, for different 2D-signals, passing through fixed-2D-channels, that same-Average-power-SNR results, coincide with
different input-noise levels, thus may-considered as biased. This bias found-to-be related to the signal’s Peak-to-Average
Power-Ratio (PAPR). We approximate the performance using Peak-power Signal-to-Noise-Ratio (PSNR). Optimizing signal
for maximum variance, given a peak-limited-channel, under PSNR; Concluding with such signal-example showing superiorperformance when-compared to known-method.
TP8b1-4
I.I.D. Stochastic Analysis of PWM Signals
Noyan Sevuktekin, Andrew Singer, University of Illinois at Urbana-Champaign, United States
A stochastic analysis for pulse width modulation (PWM) is given. Under a discrete random generator process with independent
identically distributed (i.i.d.), continuous, range-limited samples, stochastic models for different PWM signals are proposed.
Using randomization of the signal starting point, wide sense stationarity (WSS) of proposed PWM signals are shown.
Autocorrelation functions and their corresponding power spectrum densities (PSD) are proposed in terms of the modified
complementary cumulative distribution function of i.i.d samples. For the case where the samples are uniformly i.i.d. the proposed
autocorrelation functions are tested with simulations.
TP8b1-5
Statistical Data Correction for Unreliable Memories
Christoph Roth, ETH-Zurich, Switzerland; Christoph Struder, Cornell University, United States; Georgios
Karakonstantis, Andreas Burg, École Polytechnique Fédérale de Lausanne, Switzerland
In this paper, we introduce a statistical data- correction framework that aims at improving the DSP system performance in
presence of unreliable memories. The proposed signal processing framework implements best-effort error mitigation for
signals that are corrupted by defects in unreliable storage arrays using a statistical correction function extracted from the signal
statistics, a data-corruption model, and an application-specific cost function. An application example to communication systems
demonstrates the efficacy of the proposed approach.
TP8b1-6
Sonar Data Compression using Non-Uniform Quantization and Noise Shaping
Lok Wong, Gregory Allen, Brian Evans, University of Texas at Austin, United States
Sonar arrays potentially produce huge amounts of data to be recorded or transmitted over a telemetry system. Compression can
reduce the required storage or transmission bandwidth, or allow larger or higher fidelity arrays. We use a dataset of acoustic
communication signals received in a lake test and compress it to evaluate the effect of compression on performance. Based on
analysis of the dataset, we use non-uniform quantization with a Laplace distribution along with noise-shaped feedback coding.
We demonstrate that this sonar data can be compressed from 16-bit to five-bit values with little or no change in performance
using our technique.
TP8b1-7
Stephan Pfletschinger, Monica Navarro, Centre Tecnologic de Telecomunicacions de Catalunya, Spain; Christian
Ibars, Intel Corporation, United States
This paper defines an information-theoretical framework for non-orthogonal broadcast systems with multilevel coding and gives
design guidelines for the rate selection of multiple broadcast streams. This description includes hierarchical modulation and
superposition coding with codes defined in a finite field as a special case. We show how multilevel coding can be applied to
multiple antennas where, in contrast to most space-time coding and hierarchical modulation schemes, no capacity loss occurs.
92
TP8b1-8
Dynamic Target Identification and Classification Based on Resonance Topography
Grouping
Ananya Sen Gupta, Daniel Schupp, University of Iowa, United States; Ivars Kirsteins, Naval Undersea Warfare
Center, United States
We address the long-standing challenge of sonar target identification against weak ground truths and interfering scatter
components by harnessing the inherent topographic elements of acoustic scatter. Inherent robustness against ground truth
uncertainty allows unknown target discovery when supervised learning is not practical. Specifically, we employ adaptive
subspace tracking techniques to localize scatter components, discover topographic signatures within the target scatter response,
and thus classify different targets. The false alarm rate is naturally lowered as each target class has a unique scattering response
for each wave profile that is ultimately separable against environmental effects and other interference using its topographic
signature.
Track C – Networks
Session: TPb8 – Relays, Cognitive, Cooperative, and Heterogeneous Networks
3:30 PM–5:10 PM
Chair: Andrew G. Klein, Worcester Polytechnic Institute
TP8b2-1
A Distributed Algorithm for Energy Saving in Nomadic Relaying Networks
Zhe Ren, BMW Group Research and Technology, Germany; Mahdy Shabeeb, Munich University of Technology,
Germany; Slawomir Stanczak, Fraunhofer Institute for Telecommunications Heinrich Hertz Institute, Germany;
Peter Fertl, BMW Group Research and Technology, Germany
This extended abstract presents a distributed cell selection algorithm for energy savings in nomadic relaying networks where
randomly distributed devices (e.g., parked vehicles) serve as potential relay nodes. Based on broadcasted load information and
estimated link quality, the nomadic relays and subsequently the users select access points so to minimize the energy consumption
in the network. Furthermore, admission control mechanisms are incorporated at the base stations and nomadic relay nodes to
avoid overloading. We prove the convergence of our algorithm and simulation results confirm that the proposed algorithm
significantly reduces the energy consumption compared with traditional cell selection algorithms.
TP8b2-2
Instantaneous Relaying for the 3-Way Relay Channel with Circular Message Exchanges
Bho Matthiesen, Eduard A. Jorswieck, Technische Universität Dresden, Germany
The 3-user discrete memoryless multi-way relay channel with circular message exchange and instantaneous relaying is
investigated. We first show that this channel is effectively a 3-user interference channel with receiver message side information
for every fixed (and instantaneous) relay mapping. Then, we extend the Han-Kobayashi coding scheme to this channel. Finally,
we apply these results to Gaussian channels with amplify-and-forward relaying and present numerical results showing the gain of
the proposed scheme compared to the state of the art.
TP8b2-3
On the Performance of Hybrid Satellite-Terrestrial Cooperative Networks with
Interferences
Min Lin, PLA University of Science and Technology, China; Jian Ouyang, Nanjing University of Posts and
Telecommunications, China; Zhu Wei-Ping, Concordia University, Canada
The paper investigates the performance of a hybrid satellite-terrestrial cooperative network (HSTCN), where some terrestrial
AF relays are employed to assist the signal transmission from a satellite to a destination, which is corrupted by multiple CCIs.
We derive the MGF of the output SINR and present the analytical ASER expression for the considered cooperative system.
Moreover, the asymptotic ASER analysis in terms of the diversity order and array gain is also developed. Finally, numerical
results are given to demonstrate the validity of the performance analysis and the impacts of shadowing parameters, relay number
and CCIs on the considered HSTCN.
93
TP8b2-4
An Online Parallel Algorithm for Spectrum Sensing in Cognitive Radio Networks
Yang Yang, Technische Universitaet Darmstadt, Germany; Mengyi Zhang, Chinese University of Hong Kong, Hong
Kong SAR of China; Marius Pesavento, Technische Universitaet Darmstadt, Germany; Daniel Palomar, Hong Kong
University of Science and Technology, Hong Kong SAR of China
We consider in cognitive radio the estimation of the position and transmit power of primary users based on a l1-regularized
recursive least-square problem. The power vector is possibly sparse and measurements are only sequentially available. We
propose for the first time an online parallel algorithm that is novel in three aspects: i) all elements of the unknown vector variable
are updated in parallel; ii) the update of each element has a closed-form expression; and iii) the stepsize is designed to boost the
convergence yet it still has a closed-form expression. The convergence property is both theoretically analyzed and numerically
consolidated.
TP8b2-5
On the Spatial Spectral Efficiency of ITLinQ
Ratheesh Mungara, Universitat Pompeu Fabra, Spain; Xinchen Zhang, University of Texas at Austin, United States;
Angel Lozano, Universitat Pompeu Fabra, Spain; Robert W. Heath Jr., University of Texas at Austin, United States
Device-to-device (D2D) communication has been considered as a potential ingredient of 5G cellular networks. In this paper,
we consider the so-called ITLinQ (information-theoretic independent link) scheduling scheme [1] for D2D users operating on a
dedicated spectrum with respect to the cellular users and analytically characterize the spectral efficiency achievable by ITLinQ.
The analysis relies on a stochastic geometry formulation, which facilitates obtaining compact expressions and provide means to
optimally choose system parameters.
TP8b2-6
Time and Frequency Self-Synchronization in Dense Cooperative Networks
Maria Antonieta Alvarez, Bahar Azari, Umberto Spagnolini, Politecnico di Milano, Italy
Dense cooperative network involves communication and coding among multiples uncoordinated nodes, time and frequency
synchronization is mandatory to guarantee network operations. Here we propose a novel method to perform time and frequency
synchronization in presence of large carrier frequency offsets (CFOs) based on weighted consensus algorithm to reach
synchronization in a connected network. Peculiarity is the synchronization frame structure based on a common CAZAC sequence
but arranged to decouple CFO from time error in symbol and frame synchronization. Time and frequency synchronization is
guaranteed in multi-node interference scenario without the need to assign every node an independent CAZAC sequence.
TP8b2-7
Effect of Cluster Rotation Speed in Coordinated Heterogeneous MIMO Cellular Networks
with Proportionally Fair User Scheduling
Hakimeh Purmehdi, Robert Elliott, Witold Krzymien, University of Alberta, Canada; Jordan Melzer, TELUS
The effect of how often clustering patterns change within a previously proposed rotating clustering scheme on the average
achievable downlink rates of a coordinated heterogeneous multicell multiple-input multiple-output (MIMO) system is
investigated. Rotating the base station cluster patterns allows users to be nearer the cluster center in one of the patterns. The
performance of the system with different cluster rotation rates is evaluated, using a simulated annealing user scheduling algorithm
with a proportionally fair metric. Simulations demonstrate there is a maximum speed of rotation, above which negligible further
gains in performance are achieved compared to fixed clustering.
TP8b2-8
Relay Selection for AF Wireless Relay Networks in Adverse Communication Environments
Kanghee Lee, Republic of Korea Air Force, Republic of Korea; Visvakumar Aravinthan, Sunghoon Moon, Wichita
State University, United States; Jongbum Ryou, Changki Moon, Inha Hyun, Republic of Korea Air Force, Republic
of Korea; Sun Jo, Defense Acquisition Program Administrtion of ROK, Republic of Korea
This paper addresses wireless relay networks consisting of a one-source-one-destination pair and N noncooperative relays. An
objective of this paper is to analytically derive a closed form of an optimal relay amplifying vector (or matrix) for an amplifyand-forward (AF) wireless relay network under channel uncertainty (CU), jamming, and transmission power constraints at the
relays, using the minimum mean square error (MMSE) criterion. In addition, this paper presents an efficient relay-selection
strategy using the maximum SNR and minimum MMSE cost function criterions under an adverse wireless communication
environment with transmission power constraint at the relays.
94
Track G – Architecture and Implementation
Session: TPb8 – Signal Processing Architectures
3:30 PM–5:10 PM
TP8b3-1
Hybrid Floating-Point Modules with Low Area Overhead on a Fine-Grained Processing
Core
Jon Pimentel, Bevan Baas, University of California, Davis, United States
This paper proposes Hybrid Floating-Point Modules (HFPMs) as a method to improve software floating-point (FP) throughput
without incurring the area overhead of hardware floating-point units (FPUs). The proposed HFPMS were synthesized in 65
nm CMOS. They increase throughput over a fixed-point software FP implementation by 3.6x for addition/subtraction, 2.3x
for multiplication, and require less area than hardware modules. Nine functionally equivalent FPU implementations using
combinations of software, hardware, and hybrid modules are synthesized and provide 1.07-3.34x higher throughput than a
software FPU implementation, while requiring 1.08-12.5x less area than a hardware FPU for multiply-add operations.
TP8b3-2
Scalable Hardware-Based Power Management for Many-Core Systems
Bin Liu, Brent Bohnenstiehl, Bevan Baas, University of California, Davis, United States
Due to high levels integration, the design of many-core systems becomes increasingly challenging. Runtime dynamic voltage and
frequency scaling (DVFS) is an effective method in managing the power based on performance requirement in the presence of
workload variations. This paper presents an on-line scalable hardware-based dynamic voltage frequency selection algorithm, by
using both FIFO occupancy and stall information between processors. To demonstrate the proposed solution, two real application
benchmarks are tested on a many-core globally asynchronous locally synchronous (GALS) platform. The experimental results
shows that the proposed approach can achieve near-optimal power saving under performance constraint.
TP8b3-3
Optimized FPGA Based Implementation of Discrete Wavelet Transform
Amin Jarrah, Mohsin M. Jamali, University of Toledo, United States
Discrete Wavelet Transformation (DWT) has widespread usage in many vital applications. It is used to represent real-life nonstationary signals with high efficiency and also used for de-noising the signal. However, the DWT is computationally intensive.
Therefore, Haar Wavelet Transform (HWT) has been implemented on FPGAs by exploiting parallel and pipelining approaches.
All dimensions (1-D, 2-D, and 3-D) architectures are implemented and optimized. High level synthesizer from Xilinx used to
implement HWT on FPGAs. The throughput of our optimized implementation shows considerable improvement on an unoptimized version.
TP8b3-4
Mapping and Scheduling of Dataflow Graphs - A Systematic Map
Usman Mazhar Mirza, Mehmet Ali Arslan, Gustav Cedersjö, Sardar Muhammad Sulaman, Jörn W. Janneck, Lund
University, Sweden
Dataflow is a natural way of modelling streaming applications, such as multimedia, networking and other signal processing
applications. In order to cope with the computational and parallelism demands of such streaming applications, multiprocessor
systems are replacing uniprocessor systems. Mapping and scheduling these applications on multiprocessor systems are crucial
elements for efficient implementation in terms of latency, throughput, power and energy consumption etc. Performance of
streaming applications running on multiprocessor systems may widely vary with mapping and scheduling strategy. This paper
performs a systematic literature review of available research carried out in the area of mapping and scheduling of dataflow
graphs.
TP8b3-5
Dataflow Machines
Jörn W. Janneck, Gustav Cedersjö, Lund University, Sweden; Endri Bezati, Simone Casale Brunet, École
Polytechnique Fédérale de Lausanne, Switzerland
This paper presents a model for stream programs aimed at capturing their essential logical structure in a way that is amenable to
analysis, composition, and hardware and software code generation. We discuss the properties of the model, and its relationship
to a previous effort, actor machines, which it generalizes and which can be viewed as a step in the implementation flow of
95
dataflow machines. Dataflow machines are motivated by the shortcomings of actor machines when generating highly parallel
implementations (such as in hardware), and when composing machines. The paper compares and contrasts the two models with
emphasis on these two topics.
TP8b3-6
Replacement Techniques for Improving Performance in Sub-Block Caches
Oluleye Olorode, Mehrdad Nourani, University of Texas at Dallas, United States
Recent advances in processor architecture have led to the introduction of sub-blocking to cache architectures. Sub-blocking
reduces the tag area and power overhead in caches without reducing the effective cache size, by using fewer tags to index the
full data RAM array. But they suffer from performance degradation due to cache pollution. We propose intelligent sub-block
cache replacement policies that use the valid state of individual sub-blocks in replacement decisions at the super-block level.
Performance evaluations using Simplescalar toolset show improvement of up to 4.17% in SPEC2006 benchmarks.
TP8b3-7
Dynamic Reconfiguration of FPGA-based Multi-Processor Arrays
James Glenn-Anderson, Supercomputer Systems, Inc., United States
In this paper, the Multi-Processor Array (‘MPA’) architectural form is augmented with hardware partial reconfiguration on I/Dspace memory components. Three major advantages are thus derived; (1) resource constrained extension of the MPA functional
envelope, (2) improved performance scaling across processor array order, and (3) maximal parallel processing gain. A supporting
analysis reveals the partially reconfigurable MPA (‘pr-MPA’) exhibits substantial performance benefit when compared with
standard SMP architectural forms.
TP8b3-8
Coprime Processing for the Elba Island Sonar Data Set
Vaibhav Chavali, Kathleen Wage, George Mason University, United States; John Buck, University of Massachusetts
Dartmouth, United States
Coprime sensor arrays (CSAs) use interleaved uniform line arrays (ULAs) containing a relatively small number of sensors to
obtain resolution comparable to a single densely populated ULA. For narrowband CSA processing, each interleaved subarray is
beamformed independently, and the resulting outputs are multiplied and averaged over time to obtain the CSA power spectrum.
Although the individual subarrays are undersampled, the overall CSA output is not aliased. This paper considers the problem of
designing coprime arrays for passive sonar and applies CSA processing to analyze the existing Elba Island data set.
Track D – Signal Processing and Adaptive Systems
Session: TPb8 – Signal Processing Theory and Applications
3:30 PM–5:10 PM
Chair: Yue Lu, Harvard University
TP8b4-1
Prediction of a Bed-Exit Motion: Multi-Modal Sensing Approach and Incorporation of
Biomechanical Knowledge
Jun Hao, Xiaoxiao Dai, Amy Stroder, Jun Zhang, Bradley Davidson, Mohammad Mahoor, University of Denver,
United States; Neil McClure, OKT Enterprises, United States
This paper aims to answer the following questions: 1) How to detect and predict a bed-exit motion, and 2) How early a bed-exit
motion can be predicted before it actually happens.To achieve the above goals we consider the following sensing modalities
for observing the human motion during a bed-exit: RGB images, depth images and radio frequency (RF) sensing. Using the
measurements from the aforementioned sensing modalities, we investigate different approached to infer information on the
human motion. The combination of RGB and depth images significantly enhances the performance of motion recognition.
96
TP8b4-2
Ultra-Wideband Radar based Human Body Landmark Detection and Tracking with
Biomedical Constraints for Human Motion Measuring
Xiaoxiao Dai, Zhichong Zhou, Jun Zhang, Bradley Davidson, University of Denver, United States
In this manuscript, we propose and investigate a methodology for detecting and tracking human body landmarks using ultrawideband (UWB) radars. The detection of multiple human body landmarks (HBLs) is achieved by motion target indication
techniques, and the multi-HBL tracking is accomplished by a novel iterative convex optimization based approach with
considerations of biomechanics constraints. It is demonstrated that detection and track of the moving trajectories of two markers
are feasible and successfully achieved, and thus, the human arm motion is accurately measured using one UWB radar.
TP8b4-3
Separation of Interleaved Markov Chains
Ariana Minot, Yue Lu, Harvard University, United States
We study the problem of separating interleaved sequences from discrete-time finite Markov chains. Previous work has considered
the setting where the Markov chains participating in the interleaving have disjoint alphabets. In this work, we consider the
more general setting where the component chains’ alphabets can overlap. We formulate the problem as a hidden Markov model
(HMM) and develop a deinterleaving algorithm by modifying classical HMM estimation techniques to take advantage of the
special structure of our deinterleaving problem. Numerical results verify the effectiveness of the proposed method.
TP8b4-4
Ramanujan Subspaces and Digital Signal Processing
P. P. Vaidyanathan, California Institute of Technology, United States
Ramanujan-sums have in the past been used to extract hidden periods in signals. In a recent paper [13] it was shown that for
finite duration (FIR) sequences, the traditional representation is not suitable. Two new types of Ramanujan-sum expansions
were proposed for the FIR case, each offering an integer basis, and applications in the extraction of hidden periodicities were
developed. Crucial to these developments was the introduction of Ramanujan spaces in [13]. The aim of this paper is to develop
some properties of these subspaces in the context of signal processing. This includes periodicity properties, autocorrelation
properties, and development of an integer-based projection operator for these spaces. An application in the denoising of periodic
signals is also demonstrated.
TP8b4-5
Asynchronous Discrete-time Signal Processing with Molecular Reactions
Sayed Ahmad Salehi, Marc Riedel, Keshab K. Parhi, University of Minnesota, United States
We present a new methodology to synthesize molecular reactions for DSP computations that produce time-varying quantities of
molecules as a function of time-varying input quantities. DSP structures include delay elements which need to be synchronized
by a clock signal. This paper demonstrates an approach to synthesize molecular reactions to implement DSP operations without
requiring a clock signal. In the proposed approach, each delay and output variables are mapped to two types of molecules. The
scheduling of the reactions is controlled by absence indicators, i.e., signals transfer according to the absence of other signals. All
computations are scheduled in four phases.
TP8b4-6
Sequential Prediction of Individual Sequences in the Presence of Computational Errors
Mehmet Donmez, Andrew Singer, University of Illinois at Urbana Champaign, United States
We study the performance of a sequential linear prediction system built on nanoscale beyond-CMOS circuit fabric that may
introduce in computation. We propose a new sequential linear prediction algorithm under a mixture-of-experts framework that
performs satisfactorily in the presence of computational errors. We introduce a worst-case approach to model the computational
errors, where we view erroneous circuit fabric as an adversary that perturbs the prediction algorithm to heavily deteriorate its
performance. We demonstrate that our algorithm achieves uniformly good performance under the worst-case error approach in an
individual sequence manner
97
TP8b4-7
A Scalable Feature Learning and Tag Prediction Framework for Natural Environment
Sounds
Prasanna Sattigeri, Arizona State University, United States; Jayaraman Thiagarajan, Lawrence Livermore National
Laboratory, United States; Mohit Shah, Arizona State University, United States; Karthikeyan Ramamurthy, IBM
Research, United States; Andreas Spanias, Arizona State University, United States
Building feature extraction approaches that can effectively characterize natural environment sounds is challenging due to the
dynamic nature. In this paper, we develop a framework for feature extraction and obtaining semantic inferences from such
data. In particular, we propose a new pooling strategy for deep architectures, that can preserve the temporal dynamics in the
resulting representation. By constructing an ensemble of semantic embeddings, we employ an 11-reconstruction based prediction
algorithm for estimating the relevant tags. We evaluate our approach on challenging environmental sound recognition datasets,
and show that the proposed features outperform traditional spectral features.
TP8b4-8
Extending Coherence for Optimal Detection of Nonstationary Harmonic Signals
Scott Wisdom, University of Washington, United States; James Pitton, Applied Physics Laboratory and University
of Washington, United States; Les Atlas, University of Washington, United States
This paper describes an optimal detector for nonstationary harmonic signals that unifies several classic approaches. The
detector’s performance is further improved by using a novel method for extending the coherence time of such signals. The
method applies a transformation to a noisy signal that attempts to fit a simple model to the signals’s slowly changing fundamental
frequency over the analysis duration. By matching the change in the signal’s fundamental frequency, analysis is more coherent
with the signal over longer durations, which allows the use of longer windows and thus improves detection performance.
Preliminary results show performance improvements on synthetic data.
Track B – MIMO Communications and Signal Processing
Session: WAa1 – MIMO Design for mmWave Systems
Chair: Zhouyue Pi, Samsung
WA1a-1
A Tractable Model for Rate in Noise Limited mmWave Cellular Networks
8:15 AM
Sarabjot Singh, Mandar Kulkarni, Jeffrey Andrews, University of Texas at Austin, United States
The use of millimeter wave (mmWave) spectrum for future cellular systems can be made possible with the use of highly
directional beamforming (massive MIMO) and dense base station deployments. Due to the higher frequencies in use, however,
the mmWave broadband networks would exhibit fundamentally different behaviors compared to conventional microwave
cellular systems. Prominently, interference and path loss models and the corresponding effect on SINR and rate need to be
re-examined. We propose a general and tractable model to capture and analyze the key distinguishing features of a mmWave
cellular broadband system, and characterize the SINR and rate distribution in such networks. The analytical insights are validated
by simulations using real building locations in major metropolitan areas in conjunction with empirically supported mmWave path
loss models. Using both the proposed model and simulations, it is shown that unlike the interference limited nature of 4G cellular
networks, mmWave cellular networks would tend to be noise limited and the coverage heavily relies on a user being able to
received sufficient power from the serving BS. Further, the cell edge rates are shown to be limited by the base station density and
are not necessarily improved by increasing the downlink bandwidth of the system.
WA1a-2
MIMO Designs for mmWave Wireless LAN Systems
8:40 AM
Sridhar Rajagopal, Shadi Abu-Surra, Sudhir Ramakrishna, Rakesh Taori, Samsung Research America, United States
In this paper, we explore unique aspects of MIMO designs for mmWave wireless LAN systems, focusing on MIMO feasibility
and protocol implications. We study the spectral efficiency gains achievable by using 2x2 and 4x4 MIMO with dual polarization
in an indoor environment. We show that MIMO gains (for SU-case) are limited beyond 2x2 MIMO and that fine beamforming
is essential for capacity gains under MIMO. Finally, we provide efficient association and beamforming techniques for SU/MUMIMO enabling greater number of STAs to associate and refine beams in a given amount of time compared to IEEE 802.11ad.
98
WA1a-3
9:05 AM
Analysis of Millimeter Wave Cellular Networks with Overlaid Microwave Base Stations
Tianyang Bai, Robert W. Heath Jr., University of Texas at Austin, United States
The use of millimeter wave (mmWave) spectrum for the cellular access channels is promising for 5G networks. Cellular systems
that support mmWave will likely support microwave frequencies as well to achieve the complementary benefits from both bands.
This paper proposes a stochastic geometry framework for the coverage and rate analysis for multi-band cellular networks, where
mmWave networks are overlaid with macro-microwave base stations. The system model incorporates important differentiating
characteristics in mmWave and microwave systems. Distributions of the signal-to-interference-and-noise ratio (SINR) and
achievable rate are derived under certain association rules and compared with the performance of single band microwave and
mmWave systems. Compared with prior work, the proposed model can be used to evaluate the performance of indoor users as
well. The results show that overlaid microwave base stations is useful to avoid coverage holes in mmWave networks and provide
good performance at indoor users.
WA1a-4
Initial Beamforming for mmWave Communications
9:30 AM
Vip Desai, Philippe Sartori, Weimin Xiao, Anthony Soong, Lukasz Krzymien, Huawei Technologies Co., Ltd.,
United States; Ahmed Alkhateeb, University of Texas at Austin, United States
Cellular systems were designed for frequencies in the microwave band but will operate up to 6 GHz. To meet the increasing
demands, deployments above 6 GHz are envisioned. As these systems migrate, channel characteristics impact coverage range.
To increase coverage, beamforming can be used. Because cellular procedures enable beamforming after a user establishes access,
new procedures are needed to enable beamforming during discovery. This paper discusses several issues that to resolve for access
at mmWave frequencies, and presents solutions which are verified by computer simulations. It is shown that reliable network
access and satisfactory coverage can be achieved.
Track B – MIMO Communications and Signal Processing
Session: WAb1 – Massive MIMO II
Chair: David J. Love, Purdue University
WA1b-1
10:15 AM
A Multistage Linear Receiver Approach for MMSE Detection in Massive MIMO
Ting Li, Sujeet Patole, Murat Torlak, University of Texas at Dallas, United States
A key property of Massive MIMO is the orthogonality among channels when the number of antennas at the base station becomes
large. When using MMSE detection method, it is suggested that instead of dealing with the true matrix inversion operation, we
can simply approximate the matrix inverse by the inversion of its diagonal elements. However, we show in this paper that this
diagonal inversion will not perform well and we propose a low-complexity detector based on Multistage Linear Receiver which
performs well even with low number of stages and accounts for lower computation complexity.
WA1b-2
Beamforming-Based Spatial Precoding in FDD Massive MIMO Systems
10:40 AM
Ming-Fu Tang, Meng-Ying Lee, Borching Su, National Taiwan University, Taiwan; Chia-Pang Yen, Industrial
Technology Research Institute, Taiwan
In this paper, we proposed a new method for downlink precoding in massive MIMO systems using frequency division duplex
(FDD). By taking advantage of beamforming, the proposed method not only reduces the downlink training signal overhead but
also preserves the multiplexing gain. Preliminary simulation results show that the proposed method has the competitive bit error
rate performance comparing with the conventional method in MIMO systems under certain channels.
99
WA1b-3
11:05 AM
Asymmetric Distributed Space Frequency Coded Cooperative Network for Large Scale
MIMO
Bhagyashri Honrao, Chirag Warty, Shikha Nema, SNDT University, India
In this paper, the design of distributed space frequency codes (DSFCs) implementing the decode-and-forward (DF) and Amplify
and Forward (AF) protocol for asymmetric case is considered. It designed to achieve the frequency and cooperative diversities.
To achieve the maximal diversity order the source and relay node coding is considered. For DF protocol, a two-stage coding
scheme, with source and relay nodes coding, is proposed. We derive sufficient conditions at the source and relay nodes to achieve
full diversity of order NL. For AF protocol, a structure for DSF coding is proposed.
Track A – Communications Systems
Session: WAa2 – 5G and Energy Efficient Cellular Networks
Chair: Jinkang Zhu, University of Science and Technology of China
WA2a-1
8:15 AM
Shan Zhang, Sheng Zhou, Zhisheng Niu, Tsinghua University, China
Due to the rapid development of wireless technology, a long-term coexistent of different cellular systems (namely GSM, 3G and
LTE) is expected, which has made the cellular networks heterogeneous. Heterogeneous networks (HetNet) are promising to boost
the network capacity, but also bring huge power consumption. Base station (BS) sleeping is considered to be an effective way
to solve this problem. In the multi-layer HetNets, idle BSs can be freely turned off to save energy since their original coverage
can be guaranteed by other layers. Besides, more BSs can go into idle state to sleep when traffic offloading is conducted. In this
paper, we explore how much energy can be saved through BS sleeping under time varying traffic load, where GSM and OFDMA
systems coexist and differential services are offered. A tractable location-based traffic offloading and BS sleeping mechanism
is adopted for theoretical analysis. The analytical results of energy saving gain are obtained, which are evaluated by extensive
simulations. Numerical results reveal the effectiveness of BS sleeping.
WA2a-2
A Survey on 5G New Waveform: From Energy Efficiency Aspects
8:40 AM
Shunqing Zhang, Xiuqiang Xu, Yiqun Wu, Lei Lu, Yan Chen, Huawei Technologies Co., Ltd., China
With the aim of delivering any information in anytime and anywhere, 5G wireless communication networks become a fashion
topic in the wireless research areas and new waveform, as one of the key enabling technologies in 5G physical layer, attracts
growing research attentions in recent years. In this paper, we mainly focus on surveying the waveforms from the energy
efficiency point of view. Two categories of waveforms are analyzed and the related implementation issues are discussed.
Moreover, we implement the above waveforms using software-defined radio based prototype platform and generate the
measurement results for the energy efficiency comparison.
WA2a-3
9:05 AM
Evolution of LTE and new Radio Access Technologies for FRA (Future Radio Access)
Hidetoshi Kayama, Huiling Jiang, DOCOMO Beijing Communications Laboratories Co. Ltd., China
To meet the requirements for mobile traffic increase, LTE and LTE-A services have been launched by many mobile operators
in worldwide. However, according to the ongoing development of terminals and mobile cloud services, further enhancement
of channel capacity is required. Thus the discussion toward 5G technologies is becoming hot and hot in academe and mobile
communication industry now. In this talk, current situation of LTE deployment by NTT DOCOMO Japan, including area
deployment and spectrum assignment, will be introduced first. Then new technical trend for FRA (Future Radio Access) will be
presented. Here, small cell deployment and its interwork with macro-cell is regarded as one of promising ways for increasing
channel capacity while maintaining mobility support. From radio access technologies’ points of view, advanced interference
mitigation and non-orthogonal multiple access (NOMA) are likely to be key issues for future systems. As to massive-MIMO,
some technical issues such as overhead reduction are left for open issues.
100
WA2a-4
9:30 AM
A Novel Cell-Interference Model and Performance Analysis of the Future Wireless
Networks
Jinkang Zhu, Haibao Ren, University of Science and Technology of China, China
A novel quantified cell-interference depth model is proposed in this paper, to study the interference properties and networking
performance of the future wireless networks. The proposed model can be used to describe precisely the interference varies with
the cell depth entered. And then we derive the calculation formulas of the cell spectral efficiency and the energy efficiency, and
analyze numerically the achievable performance of the cellular networks. This research will provide the theoretical basis for the
architecture design of the future wireless networks.
Track F – Biomedical Signal and Image Processing
Session: WAb2 – Mobile Health
Chair: Mi Zhang, Cornell University
WA2b-1
10:15 AM
On Outlier Detection in R-R Intervals from ECG Data Collected in the Natural Field
Environment
Rummana Bari, Santosh Kumar, University of Memphis, United States
ECG is useful in inferring several health states, e.g., detection of stress or illicit drug use. Detecting outliers in R-R intervals is
critical to reliably inferring health status in field. Existing methods for outlier detection are based on data collected from lab. This
paper presents a new outlier detection method for the field environment. Evaluation on real-life data shows that this new method
detects outliers in R-R intervals with an accuracy of 99.04% in lab and 97.8% in field.
WA2b-2
Patient-Centric On-Body Sensor Localization in Smart Health Systems
10:40 AM
Ramyar Saeedi, Hassan Ghasemzadeh, Washington State University, United States
Abstract – In this paper, we introduce a localization algorithm to continuously detect location of on-body sensors. Our approach
allows patients to wear sensors on different body segments they are most comfortable with. The algorithm identifies sensor
location automatically as the patient uses the system in a normal setting. The aim of our algorithm is finding the location of
sensors with the minimum amount of intrusion, and dependence for sensor installation.
WA2b-3
Making Sense of Personal Data in Clinical Settings
11:05 AM
Harinath Garudadri, University of California, San Diego, United States
In this presentation, we make a distinction between the data collected inside hospital walls for clinical use and the data generated
by personal and wearable devices in free-living conditions aimed at promoting lifestyle and behavioural changes. We observe
that adoption of personal data in clinical settings has been slow compared to the vision put forward by thought leaders. Based on
our interactions with clinical community, there are many opportunities to improve the quality of care and/or reduce the cost of
healthcare delivery by extending the care-giver’s reach beyond the hospital walls, provided (i) such care is comparable to current
standard of care and (ii) does not overly burden the system from delivering current standard of care. Regarding (i), wireless
Electrocardiograph (ECG) is an excellent example to illustrate the technical innovations required to meet the current standard
of care. We will describe low power signal processing techniques to mitigate channel errors and motion artifacts in wireless
ECG. We will then present a platform we used to demonstrate “wired” quality in the presence of channel errors and motion
artifacts using industry standards adopted by the Food and Drugs Administration (FDA) for current ECG devices. Regarding
(ii), we are working closely with the clinical community to incorporate such innovations in their workflows with minimal impact
to current practices, and enable care beyond hospital walls. Our initial use-cases include at-home care during coronary disease
convalescence and remote monitoring to reduce readmission rates by enabling timely, and less intensive clinical interventions.
101
Track D – Signal Processing and Adaptive Systems
Session: WAa3 – Sparse Learning and Estimation
Chair: Ali Pezeshki, Colorado State University
WA3a-1
Sparse Bayesian Learning Using Approximate Message Passing
8:15 AM
Maher Al-Shoukairi, Bhaskar Rao, University of California, San Diego, United States
We use the approximate message passing framework (AMP)[1] to address the problem of recovering a sparse vector from
undersampled noisy measurements. We propose an algorithm based on Sparse Bayesian learning (SBL)[2]. Unlike the original
EM based SBL that requires matrix inversions, the proposed algorithm has linear complexity, which makes it perfect for
large scale problems. Compared to other message passing techniques, the algorithm requires fewer approximations, due to the
Gaussian prior assumption on the original vector. Numerical results show that the proposed algorithm has comparable and in
many cases better performance than existing algorithms despite significant reduction in complexity.
WA3a-2
8:40 AM
Hierarchical Bayesian Approach for Jointly-Sparse Solution of Multiple-Measurement
Vectors
Mohammad Shekaramiz, Todd K. Moon, Jacob H. Gunther, Information Dynamics Laboratory / Utah State
University, United States
Many signals can be well-estimated via a few supports under some basis. Previous work for finding such sparse representations is
mostly based on greedy Orthogonal-Matching-Pursuit and Basis-Pursuit algorithms. Though they work pretty well for SingleMeasurement-Vectors, the sparse solution exactness reduces when having Multiple-Measurement-Vectors. This problem has
applications such as Xampling‘s support recovery problem. Here, rather than using such algorithms we propose a hierarchical
Bayesian model which provides more exact solutions. Furthermore, we modify the model to account for clumps of the neighbor
supports in the solution. Several examples are considered to illustrate the merit of proposed model compared to OMP.
WA3a-3
Dictionary Approaches For Identifying Periodicities in Data
9:05 AM
Srikanth Venkata Tenneti, P. P Vaidyanathan, California Institute of Technology, United States
In this paper, we propose a number of high dimensional representations for periodic signals and use them for identifying their
periodic properties. Apart from estimating the unknown period of a signal, we target the problem of periodic decomposition that is to express the given signal as a sum of signals with periods as small as possible. Our high dimensional representations
are inspired from the DFT based Farey dictionary that was introduced in [1], where the problem of periodic decomposition was
looked at in terms of finding sparse representations for periodic signals. We take an alternate view point in this paper by showing
that periodic decomposition can instead be framed as a data-fitting problem. This allows us to design a simple $l_2$ norm
minimization framework with closed form solutions and several orders of magnitude faster computations than finding the sparse
representations with the Farey dictionary. We also generalize the Farey dictionary to construct other dictionaries with much
simpler structures that are an order of magnitude faster even for the sparsity based $l_1$ techniques. We find that dictionaries
constructed using the recently proposed Ramanujan Periodicity Transforms [2] provide the best trade-off between complexity and
noise immunity, both for the $l_1$ and $l_2$ methods.
WA3a-4
9:30 AM
An Asymptotic Maximum Likelihood Estimator for the Period of a Cyclostationary Process
David Ramírez, Peter J. Schreier, University of Paderborn, Germany; Javier Vía, Ignacio Santamaría, University of
Cantabria, Spain; Louis L. Scharf, Colorado State University, United States
We derive the maximum likelihood (ML) estimator of the cycle period of a univariate cyclostationary process. Transforming
the univariate cyclostationary process into a vector-valued wide sense stationary process allows us to obtain the structure of the
covariance matrix, which is required for the likelihood. This covariance matrix is block-Toeplitz, but the block size depends on
the unknown cycle period. Therefore, we sweep the block size and obtain the ML estimate of the covariance matrix. Since there
are no closed-form ML estimates of block-Toeplitz matrices, we resort to the frequency-domain likelihood. Finally, a numerical
example shows the utility of the estimator.
102
Track D – Signal Processing and Adaptive Systems
Session: WAb3 – Advances in Statistical Learning
Chair: Kobi Cohen, University of Illinois at Urbana-Champaign
WA3b-1
10:15 AM
Quasicontinuous State Hidden Markov Models Incorporating State Histories
Todd K. Moon, Jacob H. Gunther, Utah State University, United States
The Markovity intrinsic in conventional hidden Markov models (HMMs) does not necessarily match the statistical structure of
many real signals. Even though many signals have long-term dependencies which may not be represented by the Markovity,
HMMs are used because they provide a well-known trainable model. In this paper, we generalize the concept of the HMM state
to include the history of states leading to a state, while still limiting the number of basic states to a finite number. This expanded
view of the state is efficiently represented using real-numbered states, where the fractional portion provides a variable which
represents state histories and which can govern path-dependent model parameters, and the integer portion is the conventional
state label. State sequence estimation is accomplished using a straightforward extension of the Viterbi algorithm. Parameters
estimation for state transition probabilities and output distributions is presented.
WA3b-2
10:40 AM
A Classification Centric Quantizer for Efficient Encoding of Predictive Feature Errors
Scott Deeann Chen, Pierre Moulin, University of Illinois at Urbana-Champaign, United States
A joint compression and classification system optimizes visual fidelity and classification accuracy under a bit rate constraint.
Previous work however does not fully utilize the knowledge of the target classification task while encoding. Therefore, we
propose a classification centric quantizer (CCQ), which is tailored to preserve classification-related information in a joint
compression and classification system, and its learning algorithm. We apply and evaluate the CCQ on a scene classification
problem and compare results to previous work. We also studied the performance of using gradient descent and stochastic gradient
descent in the learning algorithm.
WA3b-3
Time-Varying Stochastic Multi-Armed Bandit
11:05 AM
Sattar Vakili, Qing Zhao, Yuan Zhou, University of California, Davis, United States
In the classic stochastic multi-armed bandit (MAB) problem, there is a given set of arms, each generating i.i.d. rewards according
to a fixed unknown distribution. The objective is an online learning algorithm for sequential arm selection that minimizes regret
defined as the total reward loss over a time horizon compared with the ideal scenario of known reward models. In this paper, we
consider a time-varying MAB problem where the unknown reward distribution of each arm can change arbitrarily over time. We
obtain a lower bound on the regret order and demonstrate that an online learning algorithm achieves this lower bound.
Track A – Communications Systems
Session: WAa4 – Physical Layer Security II
Chair: Pin-Hsun Lin, TU Dresden
WA4a-1
8:15 AM
Investigation of Secure Wireless Regions Using Configurable Beamforming on WARP
platform
Yuanrui Zhang, Queen’s University Belfast, United Kingdom; Bei Yin, Rice University, United States; Roger
Woods, Queen’s University Belfast, United Kingdom; Joseph R. Cavallaro, Rice University, United States; Alan
Marshall, University of Liverpool, United Kingdom; Youngwook Ko, Queen’s University Belfast, United Kingdom
This paper presents a novel approach to network security against passive eavesdroppers. By configuring antenna array beam
patterns to transmit the data to specific regions, it is possible to create defined regions of coverage for targeted users. By adapting
antenna configuration according to the intended user’s channel state information, the vulnerability of eavesdropping is reduced.
In this paper, we present the application of our concept to 802.11n networks where an antenna array is employed at the access
point. A range of antenna configurations (from small-scale to large-scale) are investigated by simulation and realized using the
Wireless Open-Access Research Platform.
103
WA4a-2
Wiretap-Channels with Constrained Active Attacks
8:40 AM
Carsten Rudolf Janda, Christian Scheunert, Eduard A. Jorswieck, Dresden University of Technology, Germany
We calculate an achievable secrecy rate for the Wiretap Channel with an active eavesdropper. We consider the replacement and
jamming attack explicitly, when imposing different constraints on the jamming sequence. The eavesdropper’s optimal strategy is
to disturb each symbol equiprobable in the former case, or to jam each symbol with the same jamming power in the latter case.
The eavesdropper’s replacement attack can be modeled as an additional Binary Symmetric Channel. If the attacker is able to
induce a channel corruption which corresponds to his own channel’s degradedness or which is even worse, no positive secrecy
rate is achievable.
WA4a-3
9:05 AM
Secrecy Rate Maximization for Information and Energy Transfer in MIMO Beamforming
Networks
Jens Steinwandt, Ilmenau University of Technology, Germany; Sergiy Vorobyov, Aalto University, Finland; Martin
Haardt, Ilmenau University of Technology, Germany
Consider a MIMO broadcast system, where a multi-antenna base station transmits information and energy simultaneously to
a multi-antenna information receiver (IR) and a number of multi-antenna energy receivers (ERs). In this paper, we address
the beamforming design problem that maximizes the secrecy rate subject to an energy harvesting constraint and a total power
constraint. The corresponding optimization problem is a difference of convex functions programming problem (DC), which is
generally non-convex. However, based on semidefinite-relaxation, we propose an alternating optimization strategy to tackle this
problem and provide simulation results.
WA4a-4
9:30 AM
Everlasting Secrecy in Disadvantaged Wireless Environments against Sophisticated
Eavesdroppers
Azadeh Sheikholeslami, Dennis Goeckel, Hossein Pishro-nik, UMASS-Amherst, United States
Secure communication over a wireless channel in the presence of a passive eavesdropper is considered. We present a method
to exploit inherent vulnerabilities of the eavesdropper’s receiver through the use of “cheap” cryptographically-secure key-bits
for jamming, which only need be kept secret from Eve for the (short) transmission period, to obtain information-theoretic (i.e.
everlasting) secret bits at Bob. The achievable secrecy rates for different settings are evaluated. Among other results, it is shown
that, even when the eavesdropper has perfect access to the output of the transmitter, the method can still achieve a positive
secrecy rate.
Track A – Communications Systems
Session: WAb4 – Coding and Decoding
Chair: James A. Ritcey, University of Washington
WA4b-1
Noisy Belief Propagation Decoder
10:15 AM
Chu-Hsiang Huang, Yao Li, Lara Dolecek, University of California, Los Angeles, United States
This paper analyzes an LDPC Belief Propagation (BP) decoder on noisy hardware and proposes a robust decoder implementation.
We develop a Gaussian approximate density evolution for noisy BP decoders, and find that perfect decoding is achievable for
noisy BP decoders if the message representations are of arbitrarily high precision. Noisy BP decoding thresholds are derived for
various regular LDPC codes. We propose an averaging BP decoder by averaging over the messages in all iterations. Simulation
results demonstrate that the averaging BP decoder significantly reduces the residual error rates when compared with the nominal
BP decoder.
WA4b-2
10:40 AM
A Low-Complexity Improved Successive Cancellation Decoder for Polar Codes
Orion Afisiadis, Alexios Balatsoukas-Stimming, Andreas Burg, École Polytechnique Fédérale de Lausanne,
Switzerland
In this extended abstract, we describe a new SC-based decoding algorithm for polar codes, called ﬂip SC. Flip SC can provide
signiﬁcant improvements in terms of frame error rate with respect to SC decoding, while preserving its memory complexity.
Moreover, the computational complexity of ﬂip SC is practically equal to that of SC decoding in the waterfall region.
104
WA4b-3
Differential Trellis Coded Modulation with State Dependent Mappings
11:05 AM
Ruey-Yi Wei, National Central University, Taiwan; James Ritcey, University of Washington, United States
Trellis Coded Modulation is an important bandwidth-efficient coded modulation for wireless channels. To apply this to
noncoherent channels in which a phase reference is not available, we use differential encoding (DE). This allows non-coherent
detection at the receiver. We propose a novel trellis coding scheme for DE, or differential trellis coded modulation (DTCM).
DTCM is trellis coded modulation (TCM) with DE defined by states, where distinct trellis states will usually have distinct DE
functions. We propose design methods for DE functions for use with noncoherently non-catastrophic DTCM. Further, for 8PSK
signals, set partitioning is proposed and trellis diagrams of DTCM are designed. Their advantage over DPSK is confirmed by our
simulation results.
Track C – Networks
Session: WAa5 – Information Processing for Social and Sensor Networks
Chair: Nadya Bliss, Arizona State University
WA5a-1
Fourier Transform for Signals on Dynamic Graphs
8:15 AM
Arash Golibagh Mahyari, Selin Aviyente, Michigan State University, United States
Signal processing on graphs offers a new way of analyzing multivariate signals. In most applications involving multiple signals
from different sources, the relationships among the sources generating the multivariate signals are not uniform. These different
configurations of sources can be captured by weighted graphs where the nodes are the sources and the edges indicate the
relationships. Classical signal processing concepts need to be adapted to these signals on graphs. The current work assumes the
stationarity of these relationships across time. In this paper, we propose a graph Fourier transform for signals on dynamic graphs,
where the relationships vary over time.
WA5a-2
8:40 AM
Anomalous Subgraph Detection in Publication Networks: Leveraging Truth
Nadya Bliss, Manfred Laubichler, Arizona State University, United States
Analysis of social networks has potential to provide insight to wide range of applications. As datasets continue to grow, a key
challenge is lack of existing truth models. Unlike traditional signal processing, where models of truth and background data
exist and are often well defined, these models are commonly lacking in social networks. This paper presents a transdisciplinary
approach of mitigating this challenge by leveraging research in emergence of innovation together with a novel signal processing
for graphs algorithmic framework, allowing rigorous study of innovation patterns in publication networks.
WA5a-3
Identifying Congestion in Software-Defined Networks
9:05 AM
Thomas Parker, Jamie Johnson, Murali Tummala, John McEachen, James Scrofani, Naval Postgraduate School,
United States
Software-defined networks (SDN) are an emerging technology that offers to simplify networking devices by centralizing the
network layer functions and allowing adaptively programmable traffic flows. We propose using spectral graph theory methods to
identify and locate congestion in a network. The analysis of the balanced traffic case yields an efficient solution for congestion
identification. The unbalanced case demonstrates a distinct drop in connectivity that can be used to determine the onset of
congestion. The eigenvectors of the Laplacian matrix are used to locate the congestion and achieve effective graph partitioning.
WA5a-4
Vulnerability of CPS inference to DoS attacks
9:30 AM
We study distributed inference of Cyber Physical Systems (CPS) subject to Denial of Service (DoS) attacks. For the purposes
of inference, we assume the physical-layer in the CPS is monitored by a cyber-layer. Under a DoS attack, an adversary may
disrupt the sensor network monitoring the system either by attacking the underlying communication or sensors. We investigate
countermeasures and CPS resiliency to such attacks and show that the rank-deficiency of the physical system increases the
prevalence of hubs in the cyber-layer, and consequently, the vulnerability to adversary attacks. We provide a real-world power
system monitoring application to illustrate our results.
105
Track H – Speech, Image and Video Processing
Session: WAb5 – Document Processing and Synchronization Chair: Olgica Milenkovic, University of Illinois at Urbana-Champaign
WA5b-1
Synchronizing Ordinal Data over Noisy Channels
10:15 AM
Han Mao Kiah, Lili Su, Olgica Milenkovic, University of Illinois at Urbana-Champaign, United States
We consider the novel problem of synchronizing rankings at remote locations connected by a noisy two-way channel. Such
synchronization problems arise when items in the data are distinguishable, as is the case for playlists, tasklists, crowdvotes
and recommender systems rankings. In our model, we assume data edits in the form of deletions and translocations, and
communication errors introduced by symmetric q-ary channels. Our protocols are order-optimal with respect to genie-aided
methods.
WA5b-2
Efficient Synchronization of Files in Distributed Storage Systems
10:40 AM
Salim El Rouayheb, Illinois Institute of Technology, United States; Sreechakra Goparaju, Princeton University,
United States; Han Mao Kiah, Olgica Milenkovic, University of Illinois at Urbana-Champaign, United States
We consider the problem of synchronizing data in distributed storage systems under an edit model that includes deletions and
insertions. We present two modifications of MDS and regenerating codes that allow updates in the parity-check values to be
performed with low communication complexity and with low storage overhead. Our main contributions are novel protocols that
work for both hot and semi-static data, and novel update methods that rely on permutation, Vandermonde and Cauchy matrices.
WA5b-3
Efficient File Synchronization: Extensions and Simulations
11:05 AM
Clayton Schoeny, Nicolas Bitouze, Frederic Sala, Lara Dolecek, University of California, Los Angeles, United
States
We study the synchronization of two files X and Y at two distant nodes A and B that are connected through a two-way
communication channel. We previously proposed a synchronization protocol for reconstructing X at node B with exponentially
low probability of error. We have proven the order-wise optimality of the protocol where the binary file Y is the original binary
file X modified through iid insertion and deletion edits. In this paper, we expand on previous results by presenting experimental
results from numerous scenarios including different types of files and a variety of realistic error patterns. In addition, we
introduce novel improvements to the synchronization protocol to further increase efficiency.
Track D – Signal Processing and Adaptive Systems
Session: WAa6 – Adaptive Signal Design and Analysis
Chair: Antonia Papandreou-Suppappola, Arizona State University
WA6a-1
8:15 AM
Eigen-Basis Analysis of Expected Cumulative Modulus for Constrained Signal Design
Aaron Jones, Air Force Research Laboratory, United States; Brian Rigling, Wright State University, United States;
Muralidhar Rangaswamy, Air Force Research Laboratory, United States
Radar waveforms require a constant modulus (constant amplitude) transmit signal to exploit the available transmit power.
However, recent hardware advances have forced a re-examination of this assumption to quantify the impact of modulus
perturbation from phase only signals. In this paper, we express signal modulus in terms of an eigen-spectrum obtained from an
eigenvalue distribution that mimics, in the limit of large data, the eigen-spectrum of an interference and noise covariance matrix
106
WA6a-2
Characterization of Information in Phase of Radar Range Profiles
8:40 AM
Linda Moore, Air Force Research Laboratory / University of Dayton, United States; Brian Rigling, Wright State
University, United States; Robert Penno, University of Dayton, United States
This work characterizes the information in the phase of radar range profiles with respect to the estimation of features of an
unknown target present in the measured signal. A physics-based high-frequency parametric model is employed to describe the
radar backscatter. Information is quantified by the error standard deviation of target parameter estimates from noisy radar signals
with phase either included or discarded. Information in phase is shown to provide a factor of two increase in achievable target
position estimation for X-band signals. In addition, the inclusion of phase for target parameter estimation enables improved
discrimination of frequency-dependent scattering characteristics.
WA6a-3
9:05 AM
Differential Evolution
Antonia Papandreou-Suppappola, Bryan Paul, Daniel Bliss, Arizona State University, United States
Waveform design that allows for a wide variety of chirps has proven benefits. However, dictionary based optimization is limited
and gradient search methods are often intractable. A new method is proposed using differential evolution to design cubic chirps
with coefficients constrained to the 3D unit sphere. Nonlinear functions sufficiently approximated by a third order Maclaurin
series can be represented in this chirp space. Cascaded integrator methods for generating polynomial chirps allow for practical
implementation in real world systems. While simplified tracking models and finite waveform dictionaries have information
theoretic results, we explore 2D tracking continuous waveform design in cluttered environments.
WA6a-4
Reduced Rank Adaptive Filtering in Impulsive Noise Environments
9:30 AM
Hamza Soury, King Abdullah University of Science and Technology (KAUST), Saudi Arabia; Karim Abed-Meraim,
Polytech Orleans, France; Mohamed-Slim Alouini, King Abdullah University of Science and Technology (KAUST),
Saudi Arabia
An impulsive noise environment is considered in this paper. A new aspect of signal truncation is deployed to reduce the harmful
effect of the impulsive noise to the signal. A full rank direct solution is derived followed by an iterative solution. The reduced
rank adaptive filter is presented in this environment by using two methods for rank reduction, while the minimized objective
function is defined using the Lp norm. The results are presented and the efficiency of each method is discussed.
Track C – Networks
Session: WAb6 – Distributed Detection and Optimization
Chair: Andrea Simonetto, Delft University of Technology
WA6b-1
10:15 AM
Distributed Detection for Wireless Sensor Networks with Fusion Center under Correlated
Noise
Alireza S. Behbahani, Ahmed M. Eltawil, Hamid Jafarkhani, University of California, Irvine, United States
In this paper, we study a binary distributed detection problem under correlated noise by using wireless sensors and a fusion
center (FC) with one antenna where the channel is a coherent multiple access. In order to decide between the two hypotheses, we
design sensors to maximize the error exponent derived based on minimizing probability of error for Bayesian detection subject
to network power constraint. We provide a closed form solution for the sensor encoders under correlated noise at the sensors.
Furthermore, the effect of noise correlation at the sensors is investigated. Finally, simulations are provided to verify the analysis.
WA6b-2
Distributed Asynchronous Time-Varying Constrained Optimization
10:40 AM
Andrea Simonetto, Geert Leus, Delft University of Technology, Netherlands
We devise a distributed asynchronous gradient-based algorithm to enable a network of computing and communicating nodes
to solve a constrained discrete-time time-varying convex optimization problem. Each node updates its own decision variable
only once every discrete time step. Under some assumptions (strong convexity, Lipschitz continuity of the gradient, persistent
excitation), we prove the algorithm’s asymptotic convergence in expectation to an error bound whose size is related to the
variability in time of the optimization problem. Moreover, the convergence rate is linear.
107
WA6b-3
M-ary Distributed Detection in the Presence of Channel Estimation Error
11:05 AM
Zahra Hajibabaei, Azadeh Vosoughi, University of Central Florida, United States
We consider a wireless sensor network, consisting of N sensors and a FC, tasked with distributed classification of M Gaussian
sources. Each sensor makes an M-ary decision and maps it to binary symbols. These symbols are transmitted over erroneous
channels to FC and are proceeded by a training symbol, to facilitate channel estimation. We derive the optimal fusion rules, given
the channel estimates. We show for binary PSK modulation error probability is minimized when each sensor allocates its power
equally between training and data. For binary FSK error is minimized when power is allotted to data symbols only.
Track G – Architecture and Implementation
Session: WAa7 – Implementation of Wireless Systems
Chair: Roger Woods, Queens University
WA7a-1
8:15 AM
Field-Order Based Hardware Cost Analysis of Non-Binary LDPC Decoders
Yuta Toriyama, Behzad Amiri, Lara Dolecek, Dejan Markovic, University of California, Los Angeles, United States
Non-binary low-density parity-check codes exhibit excellent coding gain at the cost of high decoding complexity. Furthermore,
while increasing the Galois field order improves the error rate, its effects on the hardware implementation cost have not
been established. We propose a modification to the Min-Max algorithm to simplify calculations while maintaining decoding
performance. In addition, a hardware area efficiency analysis is proposed, allowing a quantified exploration of the decoder design
space. This hardware estimation model is utilized to reveal 1dB coding gain or 2x implementation efficiency gain of the proposed
algorithmic simplifications, relative to the original algorithm.
WA7a-2
Algorithm and Architecture for Hybrid Decoding of Polar Codes
8:40 AM
Bo Yuan, Keshab K. Parhi, University of Minnesota, Twin Cities, United States
Polar codes are the first provable capacity-achieving forward error correction (FEC) codes. However, their error-correcting
performance under successive cancellation (SC) or belief propagation (BP) decoding algorithm is limited and need to be
improved. In this work, we propose a BP-SC hybrid decoding scheme to improve performance of polar codes. Simulation results
show that for (1024, 512) polar codes the proposed approach can lead to 0.2dB coding gain over SC or BP algorithm. In addition,
we also propose the low-complexity hardware architecture of the hybrid polar decoder.
WA7a-3
9:05 AM
A Signal Processing Approach Towards Ultra-Low Power Transceiver Design
Vijay Venkateswaran, Pawel Rulikowski, Howard Huang, Bell Labs, Ireland
This work explores the design of ultra-low power transceivers from a signal processing approach. We propose an ultra-low power
wake-up radio based on super-regenerative receiver, which is always turned on and is used to detect the beacon signal coming
from the access network, and to subsequently enable the rest of the transceiver. However, such low-power radios suffer from
poor sensitivity.The objective of this paper is to use efficient signal shaping techniques used in combination with ultra-low power
receivers in order to achieve significant power savings as well as improving its receiver sensitivity.
WA7a-4
A High Performance GPU-based Software-defined Basestation
9:30 AM
Kaipeng Li, Michael Wu, Guohui Wang, Joseph R. Cavallaro, Rice University, United States
In this paper, we present the implementation of a real-time software-defined radio(SDR) system based on graphics processing
unit(GPU) and WARP radio platform.Both sides of the transceiver consist of two components: a software component running on
general purpose processor for baseband processing, and a hardware component running on WARP FPGA for radio configuration
and signal transmission.Our major work is focused on improving the capability of the bridge module between software and
hardware components and fully utilizing computational resources on GPU for accelerating baseband processing algorithms.Our
final target is to explore the implementation of a high performance OFDM SDR system on GPU.
108
Track G – Architecture and Implementation
Session: WAb7 – Video Coding Architecture and Design
Chair: Jorn Janneck, Lund University
WA7b-1
10:15 AM
Development and Optimization of High Level Dataflow Programs: the HEVC Decoder
Design Case
Khaled Jerbi, INSA of Rennes / IETR, France; Daniele Renzi, Damien De Saint-Jorre, École Polytechnique Fédérale
de Lausanne, Switzerland; Hervé Yviquel, INSA of Rennes / IETR, France; Claudio Alberti, École Polytechnique
Fédérale de Lausanne, Switzerland; Mickaël Raulet, INSA of Rennes / IETR, France; Marco Mattavelli, École
Polytechnique Fédérale de Lausanne, Switzerland
The availability of high resolution screens supporting 4K and 8K Ultra High Definition TV formats, has raised the requirements
for better performing video compression algorithms. With this objective MPEG has recently finalized the development of the
new High Efficiency Video Coding (HEVC) video compression standard successfully addressing these demands in terms of
higher compression and increased potential parallelism when compared to previous standards. So as to guarantee real-time
processing for such extremely high data rates, exploiting the parallel capabilities of recent many/multi-core processing platforms
is in most of the cases an obliged implementation option for both encoders and decoders. In this context dataflow programming is
a particularly attractive approach because its intrinsic properties provides the portability of the potential parallelism on different
processing platform. The MPEG-RVC framework is an ISO/IEC standard conceived to address these needs. It is essentially
constituted by the RVC-CAL actor dataflow language and a network language, and aims at replacing the traditional monolithic
standard specification of video codecs with a dataflow specification that better satisfies the implementation challenges. The
library of actors is written in RVC-CAL and provides the components that are configured using the network language to build
a dataflow program implementing an MPEG decoder. This work describes the current development and optimization of the
dataflow RVC library of the HEVC standard decoder. The RVC dataflow specification of a HEVC standard decoder is composed
by four main part: the “parser”, the “residual”, the “prediction”and the “filter”. Moreover, the current specification is conformant
with most of the JCT-VC conformance streams. However, the first implementations have revealed poor performance if compared
to the optimized sequential specifications and implementations such as the standard MPEG HEVC Model HM and an open
source implementation called OpenHEVC. This work describes the analysis methodology, the transformations, the generic
and platform specific optimizations applied to the initial fully working HEVC dataflow program. It reports the performance
increases achieved for both single core and many/multi-core platforms resulting from the implementations synthesized from the
high level dataflow program and applying different configuration (i.e. parallelization) options. Beside the possibility of using
different dataflow network structures, the standard RVC dataflow program may also be instantiated by including platformspecific optimizations. In particular, the paper presents the results of applying Intel SSE kernels to accelerate the actors sequential
processing (i.e. actions) and of providing cache-efficient FIFO channels implementations that speed-up the data communication
between processor cores. These optimizations yielded an average gain of 400% in performance compared to the implementation
not using SSE extensions of the standard specification. All described refactoring and optimizations generate a dataflow program
implementation that decode HDTV resolution streams beyond real-time on standard PC platforms.
WA7b-2
10:40 AM
A Low-Power Hybrid Video Recording System with H.264/AVC and Light-Weight
Compression
Hyun Kim, Seoul National University, Republic of Korea; Chae Eun Rhee, Inha University, Republic of Korea;
Hyuk-Jae Lee, Seoul National University, Republic of Korea
To reduce the power consumption of mobile video recording systems is important to extend the lifetime of the battery. This
paper proposes a low-power video recording system that combines both H.264/AVC with high compression efficiency and lightweight compression (LWC) with low power consumption. LWC compresses video data temporarily. When the temporal data are
determined to be meaningful, they are compressed through H.264/AVC to be stored permanently. For further power reduction,
down-sampling method is utilized for the permanent storage. The proposed video recording system achieves a power reduction of
74.4% compared to the conventional video recording system which uses only H.264/AVC.
WA7b-3
11:05 AM
Design of View Synthesis Prediction in 3D-HEVC via Algorithmic Complexity Analysis
Gwo Giun (Chris) Lee, Bo-Syun Li, Chun-Fu Chen, National Cheng Kung University, Taiwan
This paper presents a systematical approach to evaluate a system from both perspectives of algorithmic performance and
complexity. The complexity metrics in this paper have the merits that are transparent to either algorithm or architecture. A case
study of coding tool, backward view synthesis prediction in 3D-HEVC, is provided to demonstrate the evidence of the proposed
109
approach. Consequently, in comparison to HTM-7.0r1, the experimental result did not reduce the coding performance on average
and the complexity of proposed method shows that the data transfer rate and the number of storage accessing could be reduced up
to 28.85% and 93.63%, respectively.
110
AUTHOR LIST
NAMESESSION
Aazhang, Behnaam.............................. TA8a3-5
Abed-Meraim, Karim............................WA6a-4
Abramovich, Yuri..................................TA7b-1
Abreu, Giuseppe..................................TA8b3-6
Abreu, Giuseppe...................................TP8a4-1
Abreu, Giuseppe...................................TP8a4-2
Abry, Patrice..........................................TA5b-4
Acton, Scott...........................................MA5b-1
Aguiar, Pedro.........................................TA6b-3
Ahmed, Rameez.....................................MP4a-2
Aiello, Katherine..................................MP8a2-1
Aiello, Katherine..................................MP8a2-2
Akcakaya, Murat...................................MA2b-4
Alberti, Claudio.....................................WA7b-1
Aldhahab, Ahmed................................ TA8a2-4
Al-Dhahir, Naofal................................ TA8a3-6
Alkhateeb, Ahmed................................. TA4a-3
Alkhateeb, Ahmed................................WA1a-4
Allen, Gregory......................................TP8b1-6
Alouini, Mohamed-Slim.......................WA6a-4
Al-Qizwini, Mohammed.......................TP8a3-1
Al-Saggaf, Ubaid................................. TA8a4-4
Alshamary, Haider............................... TA8a1-2
Al-Shoukairi, Maher.............................WA3a-1
Alter, Orly............................................MP8a2-1
Alter, Orly............................................MP8a2-2
Alter, Orly..............................................TA1b-2
Alvarez, Maria Antonieta......................TP8b2-6
Amari, Abdelkerim.................................TP4a-1
Amin, Moeness...................................... TA6a-3
Amin, Moeness......................................TA7b-4
Amiri Eliasi, Parisa..............................MP8a2-6
An, Kang.............................................MA8b2-1
Anderson, John......................................TA7b-3
Andrews, Jeffrey...................................WA1a-1
Angierski, Andre...................................TP8a2-4
NAMESESSION
Anticevic, Alan......................................TA2b-1
Anttila, Lauri........................................ TA8a1-5
Aravinthan, Visvakumar.....................MA8b2-5
Aravinthan, Visvakumar.......................TP8b2-8
Arbabian, Amin....................................MP8a4-4
Arge, Charles........................................MA5b-4
Arikan, Orhan....................................... MP3b-3
Arikan, Orhan......................................TA8b4-5
Arslan, Mehmet Ali..............................TP8b3-4
Ashrafi, Ashkan...................................... TA5a-4
Astely, David......................................... TA4a-2
Athanas, Peter.........................................TP7a-4
Atia, George........................................MA8b4-6
Atia, George.......................................... MP3b-2
Atia, George......................................... TA8a2-4
Atia, George.........................................TA8b4-7
Atlas, Les................................................TP5a-4
Atlas, Les..............................................TP8b4-8
Aviyente, Selin....................................MA8b4-5
Aviyente, Selin.....................................MP8a5-5
Aviyente, Selin......................................WA5a-1
Azari, Bahar..........................................TP8b2-6
Azizyan, Martin.....................................MP3a-3
Ba, Demba.............................................. TA2a-1
Baas, Bevan...........................................TP8b3-1
Baas, Bevan...........................................TP8b3-2
Babu, Prabhu...........................................TP3b-2
Bai, Tianyang........................................WA1a-3
Bajwa, Waheed...................................... TA6a-2
Balatsoukas-Stimming, Alexios............WA4b-2
Banister, Brian A..................................TP8b1-1
Bardak, Burak.........................................TP7a-3
Bari, Rummana.....................................WA2b-1
Bar-Ness, Yeheskel.................................TP4a-4
Bar-Ness, Yeheskel...............................TP8b1-3
Bartels, Randy........................................MP3a-4
Basiri, Shahab.......................................MA1b-2
Basten, Twan..........................................MP7a-2
Basu, Prabahan.....................................TA8b1-4
Batalama, Stella N................................TP8a1-6
Beaudet, Kaitlyn...................................TP8a1-7
Behbahani, Alireza S............................WA6b-1
111
AUTHOR LIST
NAMESESSION
Belkasim, Saied.....................................TP8a3-3
Bell, Kristine...........................................TP6a-3
Bell, Mark R........................................MA8b3-4
Benesty, Jacob.........................................TP2b-3
Benesty, Jacob.........................................TP2b-4
Benetti, Michele................................... TA8a2-5
Bently, Edward.......................................TP7b-4
Berardinelli, Gilberto...........................MP8a4-6
Berberidis, Dimitrios.............................MA1b-4
Bezati, Endri.........................................TP8b3-5
Bhattacharyya, Shuvra...........................MP7a-1
Bhorkar, Abhijeet.................................TA8b1-7
Billings, Jacob......................................MP8a2-4
bin Mansoor, Umair............................. TA8a4-6
Bingman, Verner..................................MP8a5-3
Biswal, Bharat........................................TA2b-4
Biswas, Sampurna................................. MP3b-4
Bitouze, Nicolas....................................WA5b-3
Bliss, Daniel........................................MA8b4-4
Bliss, Daniel............................................TP5b-3
Bliss, Daniel..........................................WA6a-3
Bo Jensen, Nicklas..................................TP7a-2
Bohnenstiehl, Brent...............................TP8b3-2
Bolic, Miodrag...................................... MP6b-4
Bolucek, Muhsin Alperen....................MP8a4-7
Bonnichsen, Lars.....................................TP7a-2
Borisch, Eric........................................MP8a2-3
Bourennane, Salah................................ MP1b-2
Bovik, Alan...........................................MA5b-3
Bovik, Alan...........................................TP8a3-2
Brahma, Swastik.....................................TP6a-4
Brandt-Pearce, Maite..............................TP7b-1
Brisk, Philip...........................................MP7a-4
Brooks, Dana H.....................................MA2b-4
Brown, Christopher................................TA5b-3
Brown, Donald.....................................MP8a1-3
Brown, Emery........................................ TA2a-1
Brown, Matthew....................................TA7b-2
Brown III, D. Richard.......................... TA8a1-6
Brown III, D. Richard...........................TP8a4-8
Bruck, Jehoshua......................................TP2a-1
Brumberg, Jonathan..............................MA2b-2
Brynolfsson, Johan...............................TA8b4-8
NAMESESSION
Buck, John..........................................MA8b3-2
Buck, John............................................TP8b3-8
Bucklew, James....................................MP8a2-5
Burg, Andreas......................................MP8a4-2
Burg, Andreas.......................................TP8b1-5
Burg, Andreas.......................................WA4b-2
Burgess, Neil.......................................... TA7a-3
Burnison, Jeremy..................................MA2b-2
Burton, Andrew.......................................TP7b-4
Buthler, Jakob L...................................MP8a4-6
Caire, Giuseppe..................................... MP4b-1
Calderbank, Robert................................ TA6a-2
Calhoun, Vince......................................TA2b-3
Campagnaro, Filippo.............................MA3b-1
Cao, Nianxia...........................................TP6a-4
Casale Brunet, Simone..........................TP8b3-5
Casari, Paolo.........................................MA3b-1
Casas, Christian Ibars...........................TA8b1-7
Castedo, Luis........................................TA8b1-1
Castrillon, Gabriel................................. MP2b-2
Castro-Arvizu, Juan Manuel................. MP6b-3
Catbas, Necati......................................TA8b4-7
Caulfield, John......................................MA5b-2
Cavallaro, Joseph R.............................MP8a4-1
Cavallaro, Joseph R.............................MP8a4-2
Cavallaro, Joseph R..............................WA4a-1
Cavallaro, Joseph R..............................WA7a-4
Cedersjö, Gustav.....................................TP7a-1
Cedersjö, Gustav...................................TP8b3-4
Cedersjö, Gustav...................................TP8b3-5
Champagne, Benoit.................................TP5a-1
Chang, Yueh-Lun.................................. MP5b-2
Chan-Tin, Eric.......................................TP8a1-5
Chavali, Vaibhav...................................TP8b3-8
Che, Tiben.............................................MA7b-4
Chen, Chien-Min.................................. TA8a1-4
Chen, Chun-Fu......................................WA7b-3
Chen, Jia..................................................TP1b-1
Chen, Jianshu.........................................MP5a-2
Chen, Jianshu.........................................TA6b-4
Chen, Jie................................................. TA1a-2
Chen, Jingdong.......................................TP2b-4
Chen, Scott Deeann...............................WA3b-2
Chen, Yan.............................................WA2a-2
Chen, Yang............................................TA6b-2
Chen, Yejian.......................................MA8b1-3
Cheney, Margaret................................MA8b3-6
112
AUTHOR LIST
NAMESESSION
Cheng, Qi.............................................TA8b3-5
Cheng, Qi..............................................TP8a1-5
Cheng, Xiang.........................................MP4a-4
Cheng, Xilin...........................................MP4a-4
Chepuri, Sundeep Prabhakar...................TP3b-1
Chiba, Hironobu..................................... TA5a-3
Chin, Sang (Peter).................................MA6b-3
Chitre, Mandar......................................MA3b-2
Chklovskii, Dmitri................................ MP2b-3
Chklovskii, Dmitri................................. TA2a-2
Cho, Myung.........................................TA8b4-1
Chockalingam, Ananthanarayanan........TA3b-3
Choi, Gwan...........................................MA7b-4
Choi, Gwan.......................................... TA8a3-4
Choi, Inyong.........................................MA2b-1
Choi, Junil..............................................TA3b-2
Choi, Lark Kwon..................................TP8a3-2
Choi, Yang-Seok.....................................TP5b-1
Chua, Gabriel........................................MA3b-2
Ciblat, Philippe.......................................TP4a-1
Ciochina, Silviu......................................TP2b-3
Closas, Pau............................................ MP6b-3
Cochran, Douglas.................................MP8a3-7
Cochran, Douglas.................................TA8b3-7
Cochran, Douglas..................................TP8a1-7
Codreanu, Marian................................TA8b4-6
Cohen, Kobi..........................................TP8a1-2
Cole, Michael.........................................TA2b-1
Cormack, Lawrence..............................TP8a3-2
Corr, Jamie...........................................MP8a3-8
Cosman, Pamela.................................... MP5b-2
Cosman, Pamela................................... TA8a2-2
Cottatellucci, Laura............................... MP4b-2
Couillet, Romain.....................................TP1a-2
Coulon, Martial..................................... MP6b-1
Cousseau, Juan....................................MA8b2-6
Creusere, Charles................................MA8b4-1
Creusere, Charles................................MA8b4-2
Crider, Lauren......................................MP8a3-7
Cui, Guolong...........................................TP6a-2
Curran, Tim...........................................MA2b-3
Dabin, Jason.......................................... MP6b-1
Dahlman, Erik........................................ TA4a-2
Dai, Xiaoxiao........................................TP8b4-1
Dai, Xiaoxiao........................................TP8b4-2
Dang, Chinh......................................... TA8a2-6
NAMESESSION
Dang, Wenbing......................................MP3a-4
Dao, Minh.............................................MA6b-3
Dao, Minh.............................................. TA6a-4
Dardari, Davide..................................... MP6b-2
Darsena, Donatella.................................TA6b-1
Dasgupta, Soura.................................... MP3b-4
Dauphin, Stephen................................MA8b3-6
Davis, Philip........................................MA8b4-1
Davis, Philip........................................MA8b4-2
Dawson, Martin.......................................TP7b-2
De Carvalho, Elisabeth........................ TA8a3-8
de Kerret, Paul.......................................TA4b-1
de Sa, Virginia......................................MA2b-3
De Saint-Jorre, Damien.........................WA7b-1
DeBrunner, Linda S.............................TA8b2-7
DeBrunner, Victor................................TA8b2-7
DeBrunner, Victor................................TA8b4-3
DeBrunner, Victor.................................TP8a3-8
Declercq, David....................................MA7b-3
Dehghannasiri, Roozbeh....................... MP5b-4
Del Galdo, Giovanni............................TA8b4-5
Demirors, Emrecan...............................TP8a1-6
Desai, Vip.............................................WA1a-4
Destino, Giuseppe.................................TP8a4-6
Dick, Christopher.................................MP8a4-1
Dick, Christopher....................................TP5b-4
Ding, Eric Wei-Jhong............................MP6a-3
Djuric, Petar.......................................... MP6b-4
Do, Anh..................................................TA5b-3
Dogandžić, Aleksandar...........................TP3b-4
Dolecek, Lara........................................WA4b-1
Dolecek, Lara........................................WA5b-3
Dolecek, Lara........................................WA7a-1
Donmez, Mehmet..................................TP8b4-6
Doroslovacki, Milos............................. TA8a3-2
Doroslovacki, Milos............................. TA8a4-2
Doty, David.............................................TP2a-4
Douglas, Scott........................................MP6a-4
Du, Xu.....................................................TP5b-4
Duffy, Ken...........................................MP8a2-7
Dupret, Antoine.................................... TA8a2-5
Dutta, Arindam...................................MA8b4-7
Edfors, Ove........................................... MP4b-4
El Rouayheb, Salim..............................WA5b-2
113
AUTHOR LIST
NAMESESSION
Elgala, Hany............................................TP7b-3
El-Keyi, Amr.......................................MA8b2-4
Elliott, Robert........................................TP8b2-7
Eltawil, Ahmed M.................................WA6b-1
Enzner, Gerald........................................TP2b-2
Ercegovac, Milos.................................TA8b2-4
Erdinc, Ozgur....................................... TA8a4-7
Erdogan, Alper Tunga..........................MP8a3-2
Erdogmus, Deniz...................................MA2b-4
Erives, Hector.......................................TP8a3-4
Eslami Rasekh, Maryam.....................MA8b1-1
Evans, Brian.......................................... MP5b-1
Evans, Brian......................................... TA8a2-8
Evans, Brian..........................................TP8b1-6
Facchinei, Francisco.............................MA1b-1
Falcao, Gabriel.....................................MP8a4-2
Falk, Joachim.........................................MP7a-2
Falk, Tiago.............................................MP2a-1
Fan, Guoliang........................................MA5b-2
Favaro, Federico...................................MA3b-1
Feng, Li................................................MP8a2-6
Ferdinand, Nuwan................................ TA8a3-5
Fernandez-Canellas, Delia....................MA2b-4
Fernández-Rubio, Juan......................... MP6b-3
Ferrari, André......................................... TA3a-2
Fertl, Peter.............................................TP8b2-1
Fijalkow, Inbar..................................... TA8a2-1
Firouzbakht, Koorosh.......................... TA8a3-3
Fischione, Carlo..................................... TA3a-3
Flenner, Arjuna.....................................TP8a3-5
Ford, Russell........................................MP8a1-2
Forsell, Martti.........................................TP6b-2
Fortin, Benoit....................................... TA8a2-7
Frazer, Gordon.......................................TA7b-1
Friedlander, Benjamin............................MP6a-1
Friedlander, Benjamin..........................TA8b3-1
Friedlander, Benjamin..........................TA8b3-2
Frølund Pedersen, Gert........................ TA8a3-8
Fruth, Frank...........................................MP7a-1
Fry, Alexandra....................................... TA1a-1
Gao, David Wenzhong.........................MP8a1-7
Gao, David Wenzhong........................... TA1a-3
Gao, Xiang............................................ MP4b-4
Garcia, Nil............................................. MP6b-1
NAMESESSION
Geilen, Marc..........................................MP7a-2
Gelli, Giacinto........................................TA6b-1
Georgescu, Ramona............................. TA8a4-7
Gerges, Ramez L..................................MP8a1-6
Gesbert, David...................................... MP4b-2
Gesbert, David.......................................TA4b-1
Ghassemlooy, Z......................................TP7b-4
Ghods, Alireza.....................................TA8b3-6
Ghouti, Lahouari...................................TP8a3-6
Ghuman, Kirandeep..............................TP8a3-8
Giannakis, Georgios..............................MA1b-4
Giannakis, Georgios...............................MP5a-3
Giannakis, Georgios...............................TA1b-3
Giannakis, Georgios...............................TA1b-4
Gilbert, Keith.......................................MP8a3-4
Giri, Ritwik.............................................TP3b-3
Girnyk, Maksym...................................TP8a1-3
Glenn-Anderson, James........................TP8b3-7
Goeckel, Dennis....................................WA4a-4
Gogineni, Sandeep..................................TP6a-1
Golato, Andrew......................................TA7b-4
Goldsmith, Andrea.................................MP5a-2
Goldsmith, Andrea.................................TA6b-4
Golibagh Mahyari, Arash......................WA5a-1
Gong, Chen.........................................MA8b1-6
Gong, Chen.............................................TP4a-3
Gong, Qipeng..........................................TP5a-1
Gonzalez, Gustavo..............................MA8b2-6
Gonzalez Coma, Jose Pablo.................TA8b1-1
Goparaju, Sreechakra............................WA5b-2
Gorsevski, Peter...................................MP8a5-3
Grahn, Håkan..........................................TP6b-3
Grant, Steven L.......................................TP2b-1
Grant, Steven L.......................................TP2b-3
Gregorio, Fernando.............................MA8b2-6
Grenard, Jerry........................................TA1b-1
Grgicak, Catherine...............................MP8a2-7
Grover, Pulkit.........................................MP1a-3
Gründinger, Andreas............................TA8b1-1
Gu, Erdan................................................TP7b-2
Gu, Renliang...........................................TP3b-4
Gu, Yi...................................................MP8a1-7
Guerra, Anna......................................... MP6b-2
Guicquero, William.............................. TA8a2-5
114
AUTHOR LIST
NAMESESSION
Guidi, Francesco................................... MP6b-2
Gunther, Jacob H..................................WA3a-2
Gunther, Jacob H..................................WA3b-1
Guo, Jun...............................................MP8a5-6
Gurakan, B..............................................TP4b-3
Gurbuz, Ali Cafer.................................. MP3b-3
Gurbuz, Sevgi Zubeyde.......................MP8a4-7
Haardt, Martin....................................... MP1b-3
Haardt, Martin.......................................WA4a-3
Haas, Harald............................................TP7b-2
Hague, David......................................MA8b3-2
Haimovich, Alexander.......................... MP6b-1
Haimovich, Alexander........................... TA6a-1
Hajibabaei, Zahra..................................WA6b-3
Hakhamaneshi, Farhood......................MP8a4-6
Hall, Eric................................................ TA2a-3
Han, Keyong..........................................MP6a-2
Hannig, Frank........................................MP7a-3
Hanrahan, Sara....................................MA8b4-7
Hansen, Martin Weiss..........................MP8a5-4
Hansson-Sandsten, Maria....................TA8b4-8
Hao, Jun................................................TP8b4-1
Harati, Amir...........................................MP2a-2
Harms, Andrew...................................... TA6a-2
Hassan, Yahia......................................MP8a4-8
Haubelt, Christian..................................MP7a-2
Havlicek, Joseph...................................MA5b-2
Hayat, Majeed.....................................MA8b3-8
Heath Jr., Robert W............................... TA4a-3
Heath Jr., Robert W............................. TA8a1-1
Heath Jr., Robert W..............................TP8b2-5
Heath Jr., Robert W..............................WA1a-3
Hegde, Rajesh M..................................MP8a5-1
Hegde, Rajesh M.................................... TA5a-1
Hegde, Rajesh M...................................TP8a4-7
Hellings, Christoph.............................MA8b2-2
Henney, Carl.........................................MA5b-4
Himed, Braham.......................................TP6a-2
Hindborg, Andreas..................................TP7a-2
Ho, Chung-Cheng..................................MP6a-4
Ho, Matthew........................................TA8b1-6
Hochwald, Bertrand...............................TA3b-4
Hock, Rachel.........................................MA5b-4
Honrao, Bhagyashri..............................WA1b-3
Hormigo, Javier...................................... TA7a-4
Hotz, Thomas.......................................TA8b4-5
NAMESESSION
Hsu, Yu-Chang.................................... TA8a3-7
Hua, Yingbo............................................TP5b-2
Huang, Boyang.......................................TP4a-3
Huang, Chu-Hsiang...............................WA4b-1
Huang, Howard.....................................WA7a-3
Huang, Kaibin.........................................TP4b-4
Huang, Lei.............................................. TA1a-3
Huang, Yi...............................................MP4a-1
HudachekBuswell, Mary.......................TP8a3-3
Huemer, Mario..................................... TA8a4-1
Hui, Dennis.........................................MA8b2-3
Hwang, Jeng-Kuang............................. TA8a1-4
Hwang, Jeng-Kuang............................. TA8a3-7
Hwang, Suk-seung..............................MA8b1-2
Hwang, Suk-seung...............................TA8b3-3
Hyun, Inha...........................................MA8b2-5
Hyun, Inha.............................................TP8b2-8
Ibars, Christian......................................TP8b1-7
Inan, Huseyin Atahan...........................MP8a3-2
Ingle, Atul............................................MP8a2-5
Ingle, Atul..............................................TA5b-2
Iqbal, Naveed....................................... TA8a3-6
J. Thiagarajan, Jayaraman.......................TP3a-1
Jacob, Mathews..................................... MP3b-4
Jafarkhani, Hamid.................................WA6b-1
Jaffard, Stephane....................................TA5b-4
Jahja, Rico...........................................MA8b1-2
Jain, Akshay.........................................TA8b3-8
Jain, Ayush...........................................MP8a5-1
Jakobsson, Andreas..............................TA8b4-4
Jakobsson, Andreas..............................TA8b4-8
Jalali, Ali.................................................TP1a-4
Jalali, Bahram...................................... TA8a2-3
Jamalabdollahi, Mohsen........................MP4a-3
Jamali, Mohsin M................................MP8a5-3
Jamali, Mohsin M.................................TP8b3-3
Janda, Carsten Rudolf...........................WA4a-2
Janneck, Jörn W......................................TP7a-1
Janneck, Jörn W....................................TP8b3-4
Janneck, Jörn W....................................TP8b3-5
Jaouen, Yves...........................................TP4a-1
Jarrah, Amin..........................................TP8b3-3
Jatla, Venkatesh....................................MA5b-4
Jensen, Jesper Rindom.........................MP8a5-4
Jensen, Jesper Rindom............................TP5a-3
Jerbi, Khaled.........................................WA7b-1
115
AUTHOR LIST
NAMESESSION
Jia, Chao................................................ MP5b-1
Jiang, Feng............................................. TA1a-2
Jiang, Huaiguang..................................MP8a1-7
Jiang, Huaiguang.................................... TA1a-3
Jiang, Huiling........................................WA2a-3
Jo, Sun...................................................TP8b2-8
Joham, Michael....................................TA8b1-1
Johansen, Christopher............................TA1b-1
Johansson, Mikael.................................. TA3a-1
Johnson, Ben..........................................TA7b-1
Johnson, Christopher..............................TP1a-4
Johnson, Jamie......................................WA5a-3
Johnson, Richard....................................TA5b-1
Jones, Aaron..........................................WA6a-1
Jorswieck, Eduard A............................ TA8a3-1
Jorswieck, Eduard A.............................TP8b2-2
Jorswieck, Eduard A.............................WA4a-2
Jun, Kihwan.........................................TA8b2-1
Kabal, Peter.............................................TP5a-1
Kailkhura, Bhavya................................MA4b-1
Kamamoto, Yutaka................................ TA5a-3
Kang, Jaewook......................................TP8a2-2
Kar, Soummya...................................... MP7b-2
Kar, Soummya.......................................TA6b-3
Kar, Soummya........................................TP1b-2
Karakonstantis, Georgios.....................MP8a4-2
Karakonstantis, Georgios......................TP8b1-5
Karlsson, Marcus.................................. MP4b-3
Karlsson, Sven........................................TP7a-2
Karnick, Harish...................................... TA5a-1
Karypis, George.................................... MP1b-1
Kassam, Saleem....................................TP8a2-1
Katz, Eyal..............................................TP8b1-3
Kayama, Hidetoshi................................WA2a-3
Kaynak, Unver.....................................MP8a4-7
Keilholz, Shella....................................MP8a2-4
Kekatos, Vassilis...................................MA1b-4
Kekatos, Vassilis.....................................TP3a-2
Keller, Catherine.....................................TP3a-4
Keogh, Eamonn......................................MP7a-4
Khan, Usman A.....................................TP8a4-3
Khan, Usman A.....................................TP8a4-4
Khan, Usman A.....................................WA5a-4
Khayambashi, Misagh........................... MP3b-1
Kiah, Han Mao......................................WA5b-1
Kiah, Han Mao......................................WA5b-2
Kim, Changkyu....................................MP8a1-2
Kim, Haley............................................. TA6a-1
NAMESESSION
Kim, Hyun............................................WA7b-2
Kim, Jinsub............................................MP5a-1
Kim, Kiseon..........................................TP8a2-2
Kim, Minji..............................................TP2a-3
Kim, Seung-Jun.....................................MP5a-3
Kim, Seung-Jun....................................TP8a2-6
Kim, Sungo.........................................MA8b2-5
Kirilmaz, Tunahan...............................MP8a4-7
Kirsteins, Ivars......................................TP8b1-8
Klausmeyer, Philip.................................TA5b-3
Klein, Andrew G....................................TA5b-3
Klein, Andrew G...................................TP8a4-8
Knopp, Raymond.................................TA8b1-3
Ko, Youngwook....................................WA4a-1
Koivunen, Visa.....................................MA1b-2
Koivunen, Visa.....................................TP8a4-5
Korpi, Dani.......................................... TA8a1-5
Kothandaraman, Premnishanth...............TP3a-1
Kovvali, Narayan................................MA8b4-7
Krc, Tomas.............................................TA5b-2
Krishnamurthy, Akshay.........................MP3a-3
Krishnamurthy, Ram.............................. TA7a-1
Kroger, Jim.........................................MA8b4-2
Kronvall, Ted.......................................TA8b4-4
Kruger, Anton......................................TA8b4-1
Krzymien, Lukasz.................................WA1a-4
Krzymien, Witold.................................TP8b2-7
Kuehn, Volker.......................................TP8a2-4
Kuhn, Marc..........................................TA8b1-5
Kulkarni, Mandar..................................WA1a-1
Kumar, P. R...........................................TA6b-2
Kumar, Santosh.....................................WA2b-1
Kumar, Sudhir.......................................TP8a4-7
Kundu, Debarati................................... TA8a2-8
Kupriianova, Olga................................TA8b2-8
Kurkoski, Brian.................................... TA8a3-5
Kurras, Martin......................................TA8b1-2
Kwon, Goo-Rak..................................MA8b1-2
Kwon, Goo-Rak...................................TA8b3-3
Labeau, Fabrice.......................................TP3a-3
Lai, Lifeng.............................................MA4b-4
Lai, Lifeng..............................................MP4a-1
Laiw, S K................................................TP7b-4
Lakshmi Narasimhan, Theagarajan.......TA3b-3
Lam, Tu Thanh......................................TP8a1-8
Lameiro, Cristian...................................TA4b-3
Lang, Oliver......................................... TA8a4-1
116
AUTHOR LIST
NAMESESSION
Lanterman, Aaron...............................MA8b3-5
Lao, Yingjie.........................................TA8b2-6
Lari, Vahid.............................................MP7a-3
Lashkari, Khosrow............................... TA8a4-5
Laubichler, Manfred.............................WA5a-2
Lauter, Christoph.................................TA8b2-8
Lavrenko, Anastasia.............................TA8b4-5
Lawlor, Sean......................................... MP7b-1
Learned, Rachel...................................TA8b1-6
Lee, Chung Ghiu.....................................TP7b-4
Lee, Donghoon......................................TP8a2-6
Lee, Gwo Giun (Chris).........................WA7b-3
Lee, Heung-No......................................TP8a2-2
Lee, Hyuk-Jae.......................................WA7b-2
Lee, Kanghee......................................MA8b2-5
Lee, Kanghee........................................TP8b2-8
Lee, Meng-Ying....................................WA1b-2
LeMinh, Hoa...........................................TP7b-4
Leonardi, Nora...................................... MP2b-4
Lerman, G.............................................MA1b-3
Leus, Geert............................................MA3b-3
Leus, Geert............................................. TA4a-3
Leus, Geert..............................................TP3b-1
Leus, Geert............................................WA6b-2
Lev-Ari, Hanoch..................................MP8a4-5
Lherbier, Regis..................................... TA8a2-7
Li, Bo-Syun...........................................WA7b-3
Li, Hongbin.............................................TP6a-2
Li, Jeng-Da........................................... TA8a3-7
Li, Jian....................................................TA7b-3
Li, Jian..................................................TA8b3-8
Li, Jichuan............................................MP8a1-5
Li, Juane................................................MA7b-1
Li, Kaipeng...........................................WA7a-4
Li, Min................................................MA8b2-1
Li, Minyue............................................MP8a5-6
Li, Shang-Bin..........................................TP4a-2
Li, Shuo................................................MP8a3-3
Li, Ting.................................................WA1b-1
Li, Xin................................................... MP5b-3
Li, Yang................................................. TA4a-1
Li, Yao..................................................WA4b-1
Li, Yun...................................................MP1a-1
Lian, Jie...................................................TP7b-1
Liang, Yingbin......................................MA4b-4
Lin, Chuan-Shun.................................. TA8a1-4
Lin, Chuan-Shun.................................. TA8a3-7
NAMESESSION
Lin, Min..............................................MA8b2-1
Lin, Min................................................TP8b2-3
Lin, Pin-Hsun....................................... TA8a3-1
Lin, Shu.................................................MA7b-1
Lin, Xuehong......................................MA8b3-3
Lin, Yuan-Pei......................................MA8b1-4
Little, Thomas.........................................TP7b-3
Liu, Bin.................................................TP8b3-2
Liu, Brian............................................MA8b4-3
Liu, Chun-Lin......................................MP8a3-5
Liu, Jen-Hao......................................... TA8a1-4
Liu, Keke...............................................MA7b-1
Liu, Weigang.........................................TP8a4-2
Liu, Weihao...........................................TP8b1-2
Lops, Marco.......................................... MP6b-1
Love, David............................................TA3b-2
Love, David.......................................... TA8a1-6
Low, Steven...........................................MP5a-4
Lozano, Angel.......................................TP8b2-5
Lu, Lei...................................................WA2a-2
Lu, Yue.................................................TP8b4-3
Lu, Yue M............................................. MP7b-4
Lutz, David............................................ TA7a-3
Ma, Anna...............................................TP8a3-5
Ma, Shuoxin........................................MA8b4-4
Ma, Xiaoli............................................TA8b3-4
Ma, Zhanyu..........................................MP8a5-6
Maalouli, Ghassan.................................TP8b1-1
Macagnano, Davide..............................TP8a4-6
Magnússon, Sindri................................. TA3a-3
Mahajan, Divya....................................TA8b2-5
Maharaj, Sunil (B.T.)...........................TA8b1-3
Mahmood, Mir H................................MA8b3-4
Mahzoon, Majid.....................................MP1a-3
Makino, Shoji......................................... TA5a-3
Malysa, Greg........................................MP8a4-4
Mamandipoor, Babak...........................MP8a4-4
Manduca, Armando..............................MP8a2-3
Mansukhani, Jyoti..................................TA4b-2
Mardani, Davood.................................. MP3b-2
Mardani, Morteza...................................TA1b-4
Maric, Ivana........................................MA8b2-3
Markovic, Dejan...................................WA7a-1
117
AUTHOR LIST
NAMESESSION
Marlow, Ryan.........................................TP7a-4
Marot, Julien......................................... MP1b-2
Marshall, Alan.......................................WA4a-1
Martin, Rainer.........................................TP2b-2
Mathew, Sanu........................................ TA7a-1
Mattavelli, Marco..................................WA7b-1
Matthiesen, Bho....................................TP8b2-2
Maurandi, Victor................................... MP1b-4
Maurer, Alexander..............................MA8b4-7
McClure, Neil.......................................TP8b4-1
McEachen, John....................................WA5a-3
McKay, Matthew....................................TP1a-2
McKendry, Jonathan J. D........................TP7b-2
McRae, Nathan...................................MA8b4-1
McWhirter, John..................................MP8a3-8
Médard, Muriel....................................MP8a2-7
Medda, Alessio...................................MA8b4-3
Medda, Alessio....................................MP8a2-4
Mehanna, Omar.................................... TA8a1-8
Melodia, Tommaso...............................TP8a1-6
Melvin, William..................................MA8b3-5
Melzer, Jordan.......................................TP8b2-7
Memarian, Negar...................................MP2a-4
Messier, Paul..........................................TA5b-1
Mikhael, Wasfy.................................... TA8a2-4
Milenkovic, Olgica.................................TP2a-3
Milenkovic, Olgica...............................WA5b-1
Milenkovic, Olgica...............................WA5b-2
Minot, Ariana........................................TP8b4-3
Mirkin, Mitch.........................................TA7b-2
Mirza, Usman Mazhar..........................TP8b3-4
Mirzaei, Golrokh..................................MP8a5-3
Mishra, Kumar Vijay...........................TA8b4-1
Miyabe, Shigeki..................................... TA5a-3
Mo, Jianhua.......................................... TA8a1-1
Moallemi, Nasim...................................TP8a3-7
Mogensen, Preben................................MP8a4-6
Mokhtari, Aryan......................................TP1b-3
Mollison, Matthew................................MA2b-3
Mönich, Ullrich....................................MP8a2-7
Mookherjee, Soumak...........................TA8b2-7
Moon, Changki...................................MA8b2-5
Moon, Changki.....................................TP8b2-8
Moon, Sunghoon.................................MA8b2-5
Moon, Sunghoon...................................TP8b2-8
Moon, Todd K.......................................WA3a-2
NAMESESSION
Moon, Todd K.......................................WA3b-1
Moore, Linda.........................................WA6a-2
Moreau, Eric......................................... MP1b-4
Moriya, Takehiro................................... TA5a-3
Morsi, Rania............................................TP4b-2
Moulin, Pierre.......................................WA3b-2
Mudumbai, Raghuraman.....................MA8b1-1
Mukherjee, Amitav..............................MP8a1-4
Mungara, Ratheesh...............................TP8b2-5
Nachiappan, Ramanathan.......................TP3a-1
Nafie, Mohammed...............................MA8b2-4
Nam, Young-Han................................... TA4a-1
Naqvi, Syed Hassan Raza....................TA8b1-8
Naseri, Hassan.......................................TP8a4-5
Nassif, Roula.......................................... TA3a-2
Natesan Ramamurthy, Karthikeyan........TP3a-1
Nathwani, Karan..................................MP8a5-1
Navab, Nassir........................................ MP2b-2
Navarro, Monica...................................TP8b1-7
Navasca, Carmeliza................................ TA1a-1
Nayar, Himanshu................................... TA1a-4
Needell, Deanna.....................................TA1b-1
Needell, Deanna....................................TP8a3-5
Nehorai, Arye.........................................MP6a-2
Nehorai, Arye.......................................MP8a1-5
Nehorai, Arye..........................................TP6a-1
Nema, Shikha........................................WA1b-3
Ng, Derrick Wing Kwan.........................TP4b-2
Nguyen, Chuong...................................MA5b-2
Nguyen, Dang Khoa..............................TP8a1-8
Nguyen, Lam.......................................... TA6a-4
Nguyen, Lam..........................................TA7b-3
Nguyen, PhuongBang..........................MP8a1-1
Nie, Ding................................................TA3b-4
Nieh, Jo-Yen.........................................TP8a2-5
Nitinawarat, Sirin...................................MP1a-1
Niu, Zhisheng........................................WA2a-1
Noh, Eunho...........................................MA2b-3
Nokleby, Matthew................................ TA8a3-5
Nordström, Tomas..................................TP6b-4
Norman, Mark.....................................MA8b4-1
Noubir, Guevara................................... TA8a3-3
Noujeim, Karam...................................MP8a4-4
118
AUTHOR LIST
NAMESESSION
Noyer, Jea-Charles............................... TA8a2-7
Ochi, Hiroshi.........................................TP8a1-8
Ogunfunmi, Tokunbo............................. TA5a-2
Ojowu, Ode............................................TA7b-3
Okopal, Greg...........................................TP5a-4
Oliveras Martinez, Alex....................... TA8a3-8
Ollila, Esa..............................................MA1b-2
Olofsson, Andreas...................................TP6b-4
Olorode, Oluleye...................................TP8b3-6
Orhan, Umut.........................................MA2b-4
Oshiga, Omotayo..................................TP8a4-1
Otazo, Ricardo.....................................MP8a2-6
Ouyang, Jian.........................................TP8b2-3
Oweiss, Karim........................................ TA2a-4
Ozdemir, Alp........................................MP8a5-5
Ozel, O....................................................TP4b-3
Ozer, Sedat............................................MA5b-1
Pacheco, Courtney................................MA2b-1
Pakrooh, Pooria.....................................MA6b-1
Pal, Piya................................................MA6b-4
Paleologu, Constantin.............................TP2b-3
Palka, Thomas......................................MP8a3-6
Palomar, Daniel.......................................TP3b-2
Palomar, Daniel.....................................TP8b2-4
Pan, Yen-Chang..................................MA8b1-4
Papandreou-Suppappola, Antonia.......MA8b4-7
Papandreou-Suppappola, Antonia.........WA6a-3
Parhi, Keshab K...................................MP8a4-3
Parhi, Keshab K...................................TA8b2-6
Parhi, Keshab K....................................TP8b4-5
Parhi, Keshab K....................................WA7a-2
Parhi, Megha........................................TA8b2-6
Paris, Alan...........................................MA8b4-6
Parker, Thomas.....................................WA5a-3
Parkvall, Stefan...................................... TA4a-2
Parvania, Masood...................................TA6b-1
Patole, Sujeet.........................................WA1b-1
Pattichis, Marios...................................MA5b-2
Pattichis, Marios...................................MA5b-4
Paul, Bryan............................................WA6a-3
Payton, Karen.......................................MP8a3-4
Peizerat, Arnaud................................... TA8a2-5
Peng, Yan-Tsung.................................. TA8a2-2
Penno, Robert........................................WA6a-2
Pequito, Sergio.......................................TA6b-3
NAMESESSION
Percus, Allon.........................................TP8a3-5
Pereira da Costa, Mario.........................TP8a4-5
Pesavento, Marius.................................TP8b2-4
Petropulu, Athina...................................MP3a-2
Pezeshki, Ali.........................................MA6b-1
Pezeshki, Ali..........................................MP3a-4
Pfletschinger, Stephan...........................TP8b1-7
Phelps, Shean......................................MA8b4-3
Phoong, See-May................................MA8b1-4
Picard, David........................................ TA8a2-1
Picone, Joseph........................................MP2a-2
Pimentel, Jon.........................................TP8b3-1
Pishro-nik, Hossein...............................WA4a-4
Pitaro, Michael.....................................TA8b1-6
Pitton, James.........................................TP8b4-8
Planjery, Shiva......................................MA7b-3
Plishker, William...................................MP7a-1
Poor, H. Vincent...................................MA4b-3
Poor, H. Vincent....................................MP5a-2
Poor, H. Vincent....................................TA6b-4
Popov, Konstantin...................................TP6b-1
Popovski, Petar.................................... TA8a3-8
Pratschner, Stefan................................ TA8a1-7
Probst, Christian W.................................TP7a-2
Proudler, Ian.........................................MP8a3-8
Proulx, Brian........................................TA8b3-7
Purmehdi, Hakimeh..............................TP8b2-7
Pyun, Jae-young..................................MA8b1-2
Pyun, Jae-young...................................TA8b3-3
Qureshi, Tariq.........................................TP6a-3
Rabbat, Michael.................................... MP7b-1
Rabbat, Michael..................................... TA3a-1
Rabbat, Michael..................................... TA3a-3
Rabbat, Michael......................................TP1b-4
Rabideau, Dan........................................TA7b-2
Rahman, Mehnaz................................. TA8a3-4
Rajagopal, Sridhar.................................WA1a-2
Rajaram, Siddharth...............................MA2b-1
Ramakrishna, Sudhir.............................WA1a-2
Ramamurthy, Karthikeyan....................TP8b4-7
Ramezani, Hamid..................................MA3b-3
Ramírez, David.....................................WA3a-4
119
AUTHOR LIST
NAMESESSION
Ramlall, Rohan...................................MA8b1-5
Rangan, Sundeep..................................MP8a1-2
Rangan, Sundeep..................................MP8a2-6
Rangaswamy, Muralidhar...................MA8b3-1
Rangaswamy, Muralidhar.......................TP6a-1
Rangaswamy, Muralidhar.......................TP6a-3
Rangaswamy, Muralidhar.....................WA6a-1
Rani, Ruchi..........................................MP8a5-1
Rao, Nikhil...........................................MP8a3-1
Rasmussen, Lars K................................TP8a1-1
Ratnarajah, Tharmalingam....................TP8a4-2
Raulet, Mickaël.....................................WA7b-1
Ravindran, Niranjay.............................. MP1b-1
Raviteja, Patchava..................................TA3b-3
Recht, Benjamin.....................................MP3a-1
Reed, Jeffrey.........................................TP8a1-4
Ren, Haibao...........................................WA2a-4
Ren, Zhe................................................TP8b2-1
Renzi, Daniele.......................................WA7b-1
Repovš, Grega........................................TA2b-1
Reynolds, Daryl.................................... MP5b-3
Rhee, Chae Eun.....................................WA7b-2
Ribeiro, Alejandro...................................TP1b-3
Richard, Cédric...................................... TA3a-2
Richiardi, Jonas..................................... MP2b-2
Riedel, Marc..........................................TP8b4-5
Riederer, Stephen.................................MP8a2-3
Riedl, Thomas.......................................MA3b-4
Rigling, Brian........................................WA6a-1
Rigling, Brian........................................WA6a-2
Riley, Robert.......................................MA8b3-6
Rish, Irina...............................................TA2b-2
Ritcey, James........................................WA4b-3
Rocha, Paula..........................................TA6b-3
Rocha, Pedro..........................................TA6b-3
Roemer, Florian...................................TA8b4-5
Rohani, Ehsan.......................................MA7b-4
Rohani, Ehsan...................................... TA8a3-4
Roivainen, Jussi......................................TP6b-2
Romero, Ric..........................................TP8a2-5
Rong, Yu.................................................TP5b-3
Ross, Jeremy........................................MP8a5-3
NAMESESSION
Rostamian, Majed................................. MP6b-4
Roth, Christoph.....................................TP8b1-5
Roux, Stephane......................................TA5b-4
Rüegg, Tim..........................................TA8b1-5
Rulikowski, Pawel................................WA7a-3
Rupp, Markus....................................... TA8a1-7
Rusek, Fredrik....................................... MP4b-4
Ryou, Jongbum...................................MA8b2-5
Ryou, Jongbum.....................................TP8b2-8
Sabharwal, Ashutosh...............................TP5b-4
Saeedi, Ramyar.....................................WA2b-2
Safavi, Sam...........................................TP8a4-3
Sagratella, Simone................................MA1b-1
Sahu, Anit............................................. MP7b-2
Sala, Frederic........................................WA5b-3
Salah, Aya...........................................MA8b2-4
Salehi, Masoud..................................... TA8a3-3
San Antonio, Geoffrey...........................TA7b-1
Sangari, Arash......................................MP8a5-2
Sani, Alireza..........................................TP8a2-3
Sankaranarayanan, Preethi...................MP8a2-2
Santamaria, Ignacio................................TA4b-3
Santamaría, Ignacio...............................WA3a-4
Santhanam, Balu.................................MA8b3-8
Santhanam, Sridhar................................TA7b-4
Sarayanibafghi, Omid..........................TA8b4-7
Sarkar, Rituparna..................................MA5b-1
Sartori, Philippe....................................WA1a-4
Satpathy, Sudhir..................................... TA7a-1
Sattigeri, Prasanna.................................TP8b4-7
Sayed, Ali H........................................... TA3a-2
Sayeed, Akbar.....................................MA8b1-7
Scaglione, Anna..................................... TA3a-4
Scaglione, Anna.....................................TA6b-1
Scaglione, Anna....................................TP8a1-2
Schaefer, Rafael F.................................MA4b-3
Scharf, Louis L......................................MA6b-1
Scharf, Louis L......................................WA3a-4
Scheunert, Christian..............................WA4a-2
Schizas, Ioannis.......................................TP1b-1
Schleuniger, Pascal.................................TP7a-2
Schniter, Philip....................................MA8b1-7
Schniter, Philip..................................... TA8a1-1
Schober, Robert.......................................TP4b-2
Schoeny, Clayton..................................WA5b-3
Schomay, Theodore.............................MP8a2-2
Schreier, Peter J....................................WA3a-4
120
AUTHOR LIST
NAMESESSION
Schulte, Michael.................................... TA7a-2
Schupp, Daniel......................................TP8b1-8
Schwartz, Moshe.....................................TP2a-1
Schwarz, Stefan.................................... TA8a1-7
Scrofani, James.....................................WA5a-3
Scutari, Gesualdo..................................MA1b-1
Sen Gupta, Ananya...............................TP8b1-8
Senay, Seda...........................................TP8a3-4
Sethares, William.................................MP8a2-5
Sethares, William.................................MP8a5-2
Sethares, William...................................TA5b-2
Setlur, Pawan......................................MA8b3-1
Seto, Koji............................................... TA5a-2
Severi, Stefano.....................................TA8b3-6
Sevuktekin, Noyan................................TP8b1-4
Shabeeb, Mahdy....................................TP8b2-1
Shah, Mohit...........................................TP8b4-7
Shah, Parikshit.......................................MP3a-1
Shah, Parikshit.....................................MP8a3-1
Shahbazpanahi, Shahram......................TP8a3-7
Sheikholeslami, Fatemeh.......................TA1b-4
Shi, Zhijie...............................................MP4a-1
Shin, Seokjoo......................................MA8b1-2
Shin, Seokjoo.......................................TA8b3-3
Shinn-Cunningham, Barbara.................MA2b-1
Shinotsuka, Marie................................TA8b3-4
Shirazi, Mojtaba....................................TP8a2-7
Shynk, John J.......................................MP8a1-6
Sidiropoulos, Nicholas.......................... MP1b-1
Sidiropoulos, Nicholas......................... TA8a1-8
Silva, Vitor...........................................MP8a4-2
Simonetto, Andrea................................WA6b-2
Singer, Andrew.....................................MA3b-4
Singer, Andrew.....................................TP8b1-4
Singer, Andrew.....................................TP8b4-6
Singh, Aarti............................................MP3a-3
Singh, Sarabjot......................................WA1a-1
Sinno, Zeina.......................................... MP5b-1
Skeppstedt, Jonas....................................TP7a-1
Sklivanitis, George................................TP8a1-6
Skoglund, Mikael..................................TP8a1-1
Slavakis, Konstantinos..........................MA1b-3
Slavakis, Konstantinos...........................TA1b-3
Song, Junxiao..........................................TP3b-2
NAMESESSION
Soong, Anthony....................................WA1a-4
Sørensen, Troels B...............................MP8a4-6
Soury, Hamza........................................WA6a-4
Sousa, Ericles.........................................MP7a-3
Spagnolini, Umberto............................TA8b1-8
Spagnolini, Umberto.............................TP8b2-6
Spanias, Andreas...................................TP8b4-7
Speranzon, Alberto.............................. TA8a4-7
Sridhar, Rahul.........................................TP3a-1
Stanacevic, Milutin..............................MP8a3-3
Stanczak, Slawomir...............................TP8b2-1
Stathakis, Efthymios.............................TP8a1-1
Steinwandt, Jens....................................WA4a-3
Stewart, Michael...................................TP8a3-3
Stojanovic, Milica.................................MA3b-3
Stojanovic, Milica..................................MP4a-2
Stroder, Amy.........................................TP8b4-1
Strohmer, Thomas..................................MP6a-1
Strother, Stephen................................... MP2b-1
Struder, Christoph.................................TP8b1-5
Stuijk, Sander.........................................MP7a-2
Su, Borching..........................................MP6a-3
Su, Borching........................................ TA8a1-3
Su, Borching.........................................WA1b-2
Su, Lili..................................................WA5b-1
Sullivan, Michael.................................TA8b2-2
Sun, Longji...........................................TA8b3-5
Sun, Shunqiao........................................MP3a-2
Sun, Wensheng......................................MP4a-3
Suo, Yuanming.....................................MA6b-3
Suppappola, Seth.................................. TA8a4-5
Surana, Amit........................................ TA8a4-7
Suresh, Vikram...................................... TA7a-1
Swamy, M.N.S........................................TP5a-2
Swärd, Johan........................................TA8b4-4
Swärd, Johan........................................TA8b4-8
Swartzlander, Earl................................TA8b2-1
Swartzlander, Earl................................TA8b2-2
Swartzlander, Earl................................TA8b2-5
Swenson, Brian.......................................TP1b-2
Swindlehurst, A. Lee.............................. TA1a-2
Swindlehurst, Lee................................. MP3b-1
Tajer, Ali................................................MP1a-4
Talwar, Shilpa.........................................TP5b-1
Tanan, Subhash....................................MP8a5-1
Tanchuk, Oleg.......................................MA6b-2
Tandon, Ravi.........................................TP8a1-4
121
AUTHOR LIST
NAMESESSION
Tang, Gongguo......................................MP3a-1
Tang, Ming-Fu......................................WA1b-2
Taori, Rakesh........................................WA1a-2
Tarango, Joseph.....................................MP7a-4
Tavares, Fernando M. L.......................MP8a4-6
Teich, Juergen........................................MP7a-3
Teixeira, Andr´e..................................... TA3a-1
Teke, Oguzhan...................................... MP3b-3
Tenneti, Srikanth Venkata....................WA3a-3
Theelen, Bart..........................................MP7a-2
Thiagarajan, Jayaraman........................TP8b4-7
Thiele, Lars..........................................TA8b1-2
Thomae, Reiner....................................TA8b4-5
Thomas, Robert......................................MP5a-1
Thomas, Robin.....................................TA8b1-3
Thomas, Timothy................................... TA4a-4
Thompson, Keith..................................MP8a3-8
Tonelli, Oscar.......................................MP8a4-6
Tong, Lang.............................................MP5a-1
Toriyama, Yuta.....................................WA7a-1
Torlak, Murat........................................WA1b-1
Traganitis, Panagiotis.............................TA1b-3
Tran, Trac..............................................MA6b-3
Tran, Trac............................................... TA6a-4
Tripathy, Abhijit..................................MP8a5-1
Tsakiris, Manolis.....................................TP1a-1
Tseng, Kai-Han.................................... TA8a1-3
Tsianos, Konstantinos.............................TP1b-4
Tsonev, Dobroslav..................................TP7b-2
Tufvesson, Fredrik................................ MP4b-4
Tullberg, Hugo....................................... TA4a-2
Tummala, Murali..................................WA5a-3
Tyagi, Himanshu...................................MA4b-2
ul-Abdin, Zain.........................................TP6b-4
Ulukus, Sennur........................................TP4b-3
Utschick, Wolfgang............................MA8b2-2
Utschick, Wolfgang...............................TA4b-3
Utschick, Wolfgang.............................TA8b1-1
Vaccaro, Richard..................................MP8a3-6
Vaidyanathan, P. P................................WA3a-3
Vaidyanathan, P. P................................MA6b-4
Vaidyanathan, P. P...............................MP8a3-5
Vaidyanathan, P. P................................TP8b4-4
Vakili, Sattar.........................................WA3b-3
Valdivia, Nicolas.................................MA8b3-7
Valkama, Mikko.................................. TA8a1-5
NAMESESSION
Van de Velde, Samuel..........................TA8b3-6
Van De Ville, Dimitri........................... MP2b-4
Vandergheynst, Pierre.......................... TA8a2-5
Varghese, Lenny...................................MA2b-1
Varghese, Tomy...................................MP8a2-5
Varshney, Pramod.................................MA4b-1
Varshney, Pramod..................................TA4b-2
Varshney, Pramod...................................TP6a-4
Vary, Peter..............................................TP2b-2
Vasic, Bane...........................................MA7b-3
Vaughan, Andrew...............................MA8b4-3
Veeravalli, Venugopal...........................MP1a-1
Vehkaperä, Mikko.................................TP8a1-3
Venkateswaran, Vijay...........................WA7a-3
Verde, Francesco....................................TA6b-1
Vía, Javier.............................................WA3a-4
Vidal, Rene.............................................TP1a-1
Vilà-Valls, Jordi.................................... MP6b-3
Villalba, Julio......................................... TA7a-4
Vook, Frederick..................................... TA4a-4
Vorobyov, Sergiy..................................TP8a1-3
Vorobyov, Sergiy..................................WA4a-3
Vosoughi, Aida....................................MP8a4-2
Vouras, Peter.........................................TP8a2-8
Vuppala, Satyanarayana........................TP8a4-2
Wage, Kathleen.....................................TP8b3-8
Wagner, Kevin..................................... TA8a4-2
Wai, Hoi To........................................... TA3a-4
Walter, Maxwell.....................................TP7a-2
Walters, George...................................TA8b2-3
Wang, Gang..........................................MA1b-4
Wang, Guohui......................................MP8a4-2
Wang, Guohui.......................................WA7a-4
Wang, Rui............................................MP8a1-3
Wang, X................................................MA1b-3
Wang, Xin...........................................MA8b3-3
Wang, Yiyin.........................................TA8b3-4
Wang, Zhaohui.......................................MP4a-3
Wang, Zhongfeng.................................MA7b-2
Warty, Chirag........................................WA1b-3
Wassie, Dereje A.................................MP8a4-6
Watanabe, Shun....................................MA4b-2
Weavers, Paul......................................MP8a2-3
122
AUTHOR LIST
NAMESESSION
Wei, Ruey-Yi........................................WA4b-3
Wei-Ping, Zhu.......................................TP8b2-3
Weiss, Stephan.....................................MP8a3-8
Wellner, Genevieve..............................MP8a2-7
Wen, Miaowen.......................................MP4a-4
Wendt, Herwig.......................................TA5b-4
Wenndt, Stanley..................................... TA5a-4
West, Derek.........................................MA8b3-6
Whipple, Gary.........................................TP3a-4
Wijewardhana, Uditha.........................TA8b4-6
Wilcher, John......................................MA8b3-5
Willett, Rebecca..................................... TA2a-3
Wimalajeewa, Thakshila.......................MA4b-1
Wisdom, Scott.........................................TP5a-4
Wisdom, Scott.......................................TP8b4-8
Wittneben, Armin................................MP8a4-8
Wittneben, Armin................................TA8b1-5
Wong, Lok............................................TP8b1-6
Wood, Sally............................................TA5b-2
Woods, Damien.......................................TP2a-2
Woods, Roger.........................................TP7a-3
Woods, Roger.......................................WA4a-1
Wright, Stephen...................................MP8a3-1
Wu, Dalei................................................TP5a-2
Wu, Michael.........................................MP8a4-1
Wu, Michael..........................................WA7a-4
Wu, Nan..................................................TP4a-4
Wu, Qisong............................................ TA6a-3
Wu, Qisong............................................TA7b-4
Wu, Yiqun.............................................WA2a-2
Wu, Yonglin.........................................MP8a2-7
Wu, Zhengwei.......................................TP8a2-1
Xavier, Joao............................................TP1b-2
Xi, Chenguang......................................TP8a4-4
Xi, Peng...............................................TA8b4-3
Xia, Xiang-Gen......................................TA3b-1
Xiao, Weimin........................................WA1a-4
Xie, Le....................................................TA6b-2
Xu, Jingwei...........................................MA7b-4
Xu, Luzhou............................................TA7b-3
Xu, Luzhou..........................................TA8b3-8
Xu, Tianyi..............................................TA3b-1
Xu, Weiyu..............................................MP1a-2
Xu, Weiyu............................................ TA8a1-2
Xu, Weiyu............................................TA8b4-1
Xu, Xiuqiang.........................................WA2a-2
Xu, Zhengyuan....................................MA8b1-6
NAMESESSION
Xu, Zhengyuan........................................TP4a-2
Xu, Zhengyuan........................................TP4a-3
Xu, Zhengyuan......................................TP8b1-2
Xue, Feng................................................TP5b-1
Yang, Liuqing........................................MP4a-4
Yang, Liusha...........................................TP1a-2
Yang, Peng.............................................MP6a-2
Yang, Shuo..........................................MA8b3-3
Yang, Yang...........................................TP8b2-4
Yen, Chia-Pang.....................................WA1b-2
Yener, Aylin............................................TP4b-1
Yin, Bei................................................MP8a4-1
Yin, Bei.................................................WA4a-1
Yin, Haifan............................................ MP4b-2
You, Xiaohu..........................................MA7b-2
Young, Phillip......................................MP8a2-3
Younis, Abdelhamid...............................TP7b-2
Yu, Hong..............................................MP8a5-6
Yuan, Bo..............................................MP8a4-3
Yuan, Bo...............................................WA7a-2
Yuan, Haochen..................................... TA8a2-3
Yviquel, Hervé......................................WA7b-1
Zaker, Nazanin....................................MA8b4-7
Zaki, George..........................................MP7a-1
Zappone, Alessio.................................. TA8a3-1
Zariffa, Jose............................................MP2a-3
Zekavat, Seyed.......................................MP4a-3
Zerguine, Azzedine.............................. TA8a3-6
Zerguine, Azzedine.............................. TA8a4-4
Zerguine, Azzedine.............................. TA8a4-6
Zhai, Yixuan..........................................TA4b-4
Zhang, Chuan........................................MA7b-2
Zhang, Huishuai....................................MA4b-4
Zhang, Huishuai....................................MA4b-4
Zhang, Jianshu...................................... MP1b-3
Zhang, Jianzhong (Charlie).................... TA4a-1
Zhang, Jun...........................................MA8b4-7
Zhang, Jun.............................................. TA1a-3
Zhang, Jun.............................................TP8b4-1
Zhang, Jun.............................................TP8b4-2
Zhang, Junshan....................................TA8b3-7
Zhang, Mengyi......................................TP8b2-4
Zhang, Shan..........................................WA2a-1
Zhang, Shunqing...................................WA2a-2
Zhang, Shuo......................................... TA8a4-7
Zhang, Xiaoke.....................................MA8b1-6
Zhang, Xinchen.....................................TP8b2-5
123
AUTHOR LIST
NAMESESSION
Zhang, Yimin......................................... TA6a-3
Zhang, Yimin.........................................TA7b-4
Zhang, Yingchen..................................MP8a1-7
Zhang, Yingchen.................................... TA1a-3
Zhang, Yuan.......................................... MP5b-2
Zhang, Yuanrui.....................................WA4a-1
Zhao, Changhong...................................MP5a-4
Zhao, Qing.............................................TA4b-4
Zhao, Qing............................................TP8a1-2
Zhao, Qing............................................WA3b-3
Zhao, Ran...............................................TA1b-1
Zhao, Yue...............................................MP5a-2
Zhao, Yue...............................................TA6b-4
Zhou, G. Tong......................................TA8b3-4
Zhou, Sheng..........................................WA2a-1
Zhou, Shengli.........................................MP4a-1
Zhou, Wentian....................................... MP5b-3
Zhou, Yuan...........................................WA3b-3
Zhou, Zhichong.....................................TP8b4-2
Zhu, Jinkang..........................................WA2a-4
Zhu, Meifang......................................... MP4b-4
Zhu, Wei-Ping.....................................MA8b2-1
Zhu, Wei-Ping.........................................TP5a-2
Zoechmann, Erich................................ TA8a1-7
Zong, Pingping.....................................TA8b1-7
Zorzi, Michele.......................................MA3b-1
Zou, Difan...............................................TP4a-2
NAMESESSION
124
125
126