Customer Contact TeamQinetiQCody Technology ParkIvely Road, FarnboroughHampshire GU14 0LXUnited KingdomTel +44 (0) 8700 100942www.QinetiQ.com

APPLICATION TO SONAR

Experimental sonobuoy array

APPLICATION TO COMMUNICATIONS

HF antenna array

•Co-channel interference•Successful signal separation

Wireless multimedia networks

•Multiple antennas provide spatial diversity•Exploit multipath propagation

GENE EXPRESSION ANALYSIS

DNA Microarray

Microarray signal processing

FOETAL HEARTBEAT ANALYSIS

Non-invasive diagnostic technique•BSS applied to multiple ECG signals•Successful clinical trials

Separation of triplet heartbeat signals

•QinetiQ and Queen Charlotte’s hospital – joint research•World first – results highlighted in The Lancet (editorial)

PREDICTION OF EPILEPTIC FOCAL SEIZURES

Signal processing applied to scalp EEG•BSS used to estimate original sources

•Better prediction from scalp EEG than intracranial EEG

ACKNOWLEDGEMENTSResearch sponsored by the MoD corporate research programme (QinetiQ) and EPSRC (Cardiff) .

SISTER PARTNERSHIP

Signal Separation: Theory & Engineering Relevance

•Cardiff: Prof. Jonathon Chambers •Centre for Digital Signal Processing (CDSP)•School of Engineering

•QinetiQ: Prof. John G McWhirter FRS FREng•Centre for Signal and Information Processing (CSIP)•Malvern Technology Centre•Associate Prof. Cardiff University •Visiting Prof. Queen’s University Belfast

Scope•Advanced algorithms for signal separation

•broadband signals•convolutive mixing•non-stationary environments

•Practical applications•Wireless communications•Phased array radar and sonar•Bio-medical signal processing

Current projects•EPSRC CASE Studentship

• “Video Assisted Audio Source Separation”• Mr Andrew Aubrey • based in Cardiff

•EPSRC ICASE Studentship• “Broadband Blind Signal Separation”• Ms Joanne Foster• based in Malvern

•EPSRC ICASE Studentship•“The SBR2 Algorithm for Communications Signal Processing

•Mr Martin Davies•based in Malvern

•Participants in EPSRC funded network for ICA

BLIND SIGNAL SEPARATION (BSS)

Typical scenario

•Signal model

•Data matrix

•Unknown mixture matrix A•Unknown signals S•Input signal properties

•non-Gaussian • statistically independent

Independent component analysis (ICA)•Higher Order Statistics•QinetiQ “BLISS” algorithm

1. QinetiQ ltd, Malvern

2. School of Engineering, Cardiff University

John G McWhirter1 FRS FREng and Jonathon Chambers2

© Copyright QinetiQ ltd 2005QinetiQ/?/?/PUB??????

……

s t1( )

s t2( )

s t3( )

x t1( ) x t2( ) x t3( ) x tp ( )

)()()( ttt nAsx

NASX

Input Data

Separated sources

Amplitude(micro volts)

-200 -100 0 100 200 300-505

10152025

Time (milliseconds)

Application to triplets

-200 -100 0 100 200 300-6-4-202468

1012

Time (milliseconds)

14

Time (milliseconds)-200 -100 0 100 200 300

-505

10152025

Averaged foetal ECG Triplet 1

Triplet 2

Triplet 3

From estimated EEG source: From intracranial EEG: A DNA microarray. Each dot represents the activity level of a gene. One microarray can measure thousands of gene activity levels.

Image processing

Gene 2

Gene P

t1 t2 tN

Data matrixP >> N

Finding underlying structure

Principle component analysis.Independent component

analysis.

Cardiff – QinetiQ SISTER Partnership