Specific Emitter Identification (SEI) Research Aardvark AOC/21... · • Electronic Intelligence...
Transcript of Specific Emitter Identification (SEI) Research Aardvark AOC/21... · • Electronic Intelligence...
Specific Emitter Identification (SEI) Research
April 2009
Mariëtte Conning
Slide 2 © CSIR 2009 www.csir.co.za
Electronic Support SEI Research• Context of SEI within EW• Why do we do SEI?• Research project• Proof of concept• Proof of SEI• Issues related to SEI• Future work
Slide 3 © CSIR 2009 www.csir.co.za
Context of SEI within EW
Electronic Warfare (EW)
Electronic Support (ES) [ESM]
Electronic Attack (EA) [ECM]
Electronic Protection (EP) [ECCM]
•ECCM
•EMCON
•COMSEC
•SIGINT
•ELINT
•COMINT
“search for, intercept, locate, classify sources of intentional and unintentional radiated EM
energy”
“use EM and directed energy to attack facilities and equipment in order to degrade, neutralise
or destroy enemy activity”
“use EM and directed energy to control the EM spectrum”
“actions taken to protect facilities and equipment from
any effects of friendly or enemy EW”
•ECM
•DEWs
•ARMs
Adopted from Electronic Warfare in the Information Age, (1999), D.C. Schleher
Slide 4 © CSIR 2009 www.csir.co.za
Signals Intelligence
Electronic Warfare (EW)
Electronic Support (ES) [ESM]
•SIGINT
•ELINT
•COMINT
“search for, intercept, locate, classify sources of intentional and unintentional radiated EM
energy”
“use EM and directed energy to control the EM spectrum”
Signals Intelligence (SIGINT)• Electronic Intelligence (ELINT). Concerned with
the characterisation of emitter signal parameters e.g. pulse width and carrier frequency
• Communications Intelligence (COMINT). Concerned with information content e.g. sms
“ES Techniques refer to the development of signal processing algorithms in order to develop an ES
capability for the EDERI”
Slide 5 © CSIR 2009 www.csir.co.za
Context of SEI within EW
ES
Detection Tracking Locate Classify
Emitter identification SEI
Intra classExtra class
Slide 6 © CSIR 2009 www.csir.co.za
Why do SEI?• Identity a particular Radio Frequency (RF) emitter
• Cooperative or non- cooperative• Adds a further dimension to emitter classification
• Useful in Radar, EW and Communications• Situational awareness of friend or foe• Emitter performance management over time
Slide 7 © CSIR 2009 www.csir.co.za
Research project (1)• WdP studied possible techniques for SEI on radars• MC – similar research project at GEW on radio
transmitters• Research done is not to develop a product• Development of signal processing algorithms (detection
and feature extraction)• Classification algorithm
Slide 8 © CSIR 2009 www.csir.co.za
Research project (2)• Technical aspects:
• Detection – determine presence of signals• Feature extraction – calculation of values representing a specific
feature• Features are calculated from waveform statistics and amplitude-
scale transform (DWT)• Effectively, process performs data rate reduction on received
signals
Slide 9 © CSIR 2009 www.csir.co.za
Proof of concept – Emitter Classification• “Can we classify emitters based on their SIGINT
description?”• “Can we do this automatically?”• “Can we reuse emitter data already captured?”• “Can we identify EA?”
Slide 10 © CSIR 2009 www.csir.co.za
Proof of concept – Emitter Classification
8800 8900 9000 9100 9200 9300 9400 9500 9600 9700 9800-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Pulse time dataPulse start/endPhase derivative
3000 4000 5000 6000 7000 8000 9000 10000 11000-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Pulse time dataPulse start/endPhase derivative
0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04 1.05-1.5
-1
-0.5
0
0.5
1
Pulse time dataPulse start/endPhase derivative
• Analysis of measured data• Pulse detection
Slide 11 © CSIR 2009 www.csir.co.za
Proof of concept – Emitter Classification
8
• Overlapped features of units• High correlation between features
Slide 12 © CSIR 2009 www.csir.co.za
Proof of concept – Emitter Classification• Results show feasibility of SEI
Normalisation Percentage
Training Data [%] CCD [%]
5 78.91 Global
10 79.22
5 83.03 Individual
10 83.33
Slide 13 © CSIR 2009 www.csir.co.za
Proof of SEI (1)
• Extends emitter classification to identify emitters of the same type
• Here we explore the subtle differences in emitter parameters• Usually originates from analogue emitter subsystems
Slide 14 © CSIR 2009 www.csir.co.za
Proof of SEI (2)• Emitter parameters tend to be severely overlapped
• Emitters are fundamentally the same• It is usually required to fully characterise an emitter for
accurate classification• Many parameters require an intensive characterisation effort
• Need for an adaptable emitter database• Own systems are easily characterised• Enemy systems are generally quite difficult or impossible to
characterise
Slide 15 © CSIR 2009 www.csir.co.za
Proof of SEI (3)• Time and frequency measurement setup• Antenna patterns, polarisation, high sample rate
Slide 16 © CSIR 2009 www.csir.co.za
Proof of SEI (4)• Results from feature calculation show distinction between
radars• Scatter plot of 2 features – Droop and Pulse width
Dro
op
Pulse width
TMR2TMR3TMR4
Radar 1Radar 2Radar 3
Slide 17 © CSIR 2009 www.csir.co.za
TMR2TMR3TMR4
Radar 1Radar 2Radar 3
Dimensionality reduction• Redundant features clutter the feature vector, has an
influence on the classification• Three different methods were used to reduce the
dimensionality of the input feature vector (Correlation method, LDA, Gamma test)
• Successful reduction of features
Slide 18 © CSIR 2009 www.csir.co.za
Spectral information (1)• Compare spectrum of unknown transmitter with set of
reference spectra (computed through averaging)• Either only the magnitude response or the magnitude and
phase responses• Only magnitude response comparison was done• Included in future work, design of a matched filter is a
possibility – this will consider both magnitude and phase responses
• Exact alignment (in time and frequency) is very important for accurate calculation of the average
Slide 19 © CSIR 2009 www.csir.co.za
Spectral information (2)• Results of comparing an unknown radar pulse magnitude
response with 3 reference magnitude responses
0 500 1000 1500 2000 25000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
FFT bin
Dis
tanc
e m
easu
re
TMR2 RefTMR3 RefTMR4 Ref
Radar 1Radar 2Radar 3
Slide 20 © CSIR 2009 www.csir.co.za
Classification• A fuzzy ARTmap classifier was used• Results are shown for different sets of features
73.958 features as selected using LDA
74.93 6 features as selected using LDA
77.74 4 features as selected using LDA
84.97 7 features as selected using the Gamma test
92.15 2 features (transmitted frequency and droop)
85.42 13 features as selected using the correlation method
82.32 All 19 features
CCD (%)Combination of features
Slide 21 © CSIR 2009 www.csir.co.za
Issues related to SEI• Incomplete data• Complete characterisation required of own equipment for
effective identification of friendly forces
Slide 22 © CSIR 2009 www.csir.co.za
Conclusion and Future workConclusions• Current research shows that classification is possible on
radars of the same make and model• There are many methods that can be implemented to do
SEI (feature vector, spectral information, other possible methods yet to be discovered)
Future work• High sampling rate measurements for more detailed
features• Polarisation measurements and analysis• Design of matched filter for magnitude and phase
response comparison• Further trials to record more of the same, including radio
transmitters
Slide 23 © CSIR 2009 www.csir.co.za
Questions?