MIT Lincoln Laboratory C. D. Richmond-1 Monday 10th July SEA@MIT 2006 Adaptive Array Detection, Estimation and Beamforming Christ D. Richmond Workshop on Stochastic Eigen-Analysis and its Applications 3:30pm, Monday, July 10th 2006 *This work was sponsored by Defense Advanced Research Projects Agency under Air Force contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the author and are not necessarily endorsed by the United States Government.
34
Embed
Adaptive Array Detection, Estimation and Beamformingweb.mit.edu/sea06/agenda/talks/Richmond.pdf · Adaptive Array Detection, Estimation and Beamforming ... Workshop on Stochastic
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MIT Lincoln LaboratoryC. D. Richmond-1Monday 10th July
SEA@MIT 2006
Adaptive Array Detection,Estimation and Beamforming
Christ D. Richmond
Workshop on Stochastic Eigen-Analysis and its Applications
3:30pm, Monday, July 10th 2006
*This work was sponsored by Defense Advanced Research Projects Agency under Air Force contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the author and are not necessarily endorsed by the United States Government.
MIT Lincoln LaboratoryC. D. Richmond-2Monday 10th July
SEA@MIT 2006
Outline
• Introduction
– Radar/Sonar problem
• Detection algorithms
• Estimation algorithms
• Open problems
• Summary
MIT Lincoln LaboratoryC. D. Richmond-3Monday 10th July
SEA@MIT 2006
Airborne Surveillance Radars
• RAdio Detection And Ranging = RADAR• Goals / Mission:
– Long range surveillance– Airborne Moving Target Indication (AMTI)– Ground Moving Target Indication (GMTI)– Synthetic Aperture Radar (SAR) Imaging
• RAdio Detection And Ranging = RADAR• Goals / Mission:
– Long range surveillance– Airborne Moving Target Indication (AMTI)– Ground Moving Target Indication (GMTI)– Synthetic Aperture Radar (SAR) Imaging
MIT Lincoln LaboratoryC. D. Richmond-4Monday 10th July
SEA@MIT 2006
Transmit Power Pattern
Airborne Surveillance Radars: Signals and Interference
Target
v
θAzimuth
sTX t( )= Re ˜ p t( )⋅ e j 2πfc t{ }
TX/RXWaveform
sRX t( )= Re α˜ p t − τ( )⋅ e j 2π ( fc + fd )t{ }
Time Delay(Range)
Doppler(Velocity)
GroundClutter
HostileJamming Interferer
MIT Lincoln LaboratoryC. D. Richmond-5Monday 10th July
SEA@MIT 2006
Two-dimensional filtering required to cancel ground clutter
-0.5
0
0.5 -0.5
0
0.50
10
20
30
40
50
-1
1
50
0
Primary snapshot (target range gate)
NOISEJAMMERCLUTTER
Radar Data Modeland Optimum Linear Filter
*Brennan and Reed, IEEE T-AES 1973First to propose this for Radar Sig. Proc
GroundClutter
xT = Sv θT , fT( )+ nE xT{ }= Sv θT , fT( )cov xT( )= E nnH{ }= R
-1
1
50
0
Space-Time Adaptive Processing(STAP)
Space-Time Adaptive Processing(STAP)
ClutterNull Jammer
Null
Filter Response of w ∝R−1v
Sin (Azimuth) Doppler (Hz)
Pow
er (d
B)
TARGET
MIT Lincoln LaboratoryC. D. Richmond-6Monday 10th July
SEA@MIT 2006
Outline
• Introduction
• Detection algorithms
• Estimation algorithms
• Open problems
• Summary
MIT Lincoln LaboratoryC. D. Richmond-7Monday 10th July
SEA@MIT 2006
Adaptive Detection Problem
•Two Unknowns R & S
H0 : xT = nH1 : xT = SvT + n
Test Cell:
cov(xT) = R
Assumptions:
• All Data Complex Gaussian
• Training Samples
– Homogeneous with Test Cell → cov(xT) = cov(xi)
• Perfect Look ( v = vT )
Analogy to 1-DCFAR StatisticAnalogy to 1-DCFAR Statistic
2
Cells#
2 ˆ|| Nt ση ⋅<≥
⋅= ∑cov(xi) = R[ ]LxxxX ||| 21=
•Use Noise Only Training Set
MIT Lincoln LaboratoryC. D. Richmond-8Monday 10th July
MIT Lincoln LaboratoryC. D. Richmond-9Monday 10th July
SEA@MIT 2006
10 12 14 16 18 200
0.2
0.4
0.6
0.8
1
Adaptive Detection Performance:An Example
•Random matrix theory predicts performance loss due tocovariance estimation
N=10, L=2N,PFA=1e-6
Output Array SINR (dB)
PD
GLRTASB, Fixed Thr.ACEAMFMax ASB PD
PD vs SINR
Optimal MF
R unknownR unknown
R knownR known
Loss Due toCovarianceEstimation
Loss Due toCovarianceEstimation
MIT Lincoln LaboratoryC. D. Richmond-10Monday 10th July
SEA@MIT 2006
Outline
• Introduction
• Detection algorithms
• Estimation algorithms
• Open problems
• Summary
MIT Lincoln LaboratoryC. D. Richmond-11Monday 10th July
SEA@MIT 2006
No Information
Threshold
Asymptotic
Mea
n Sq
uare
d Er
ror (
dB)
SNR (dB)SNRTH
Cram�r-Rao Bound
Mean-Squared Error Performance:No Mismatch vs Mismatch
θT
No MismatchArray Element
Positions
θ θT ˆ θ ML
Low SNR
LargeErrors
Driven by GlobalAmbiguity/Sidelobe
Errors
Driven by GlobalAmbiguity/Sidelobe
Errors
θ θT ˆ θ ML
High SNRSmallErrors
Driven by LocalMainlobe ErrorsDriven by LocalMainlobe Errors
ˆ θ ML = argmax
θ tML θ,data( )
Scan Angle
θ θT
Noise FreeAmbiguityFunction
MIT Lincoln LaboratoryC. D. Richmond-12Monday 10th July
SEA@MIT 2006
Mean-Squared Error Performance:No Mismatch vs Mismatch
Array Element Positions
Array ElementPositions
No Information
Threshold
Asymptotic
CRB
SNR (dB)SNRTH
Mea
n Sq
uare
d Er
ror (
dB)
MismatchSNRTH
No Mismatch Signal Mismatch
TrueAssumed
θT θT
No Information
Threshold
Asymptotic
Mea
n Sq
uare
d Er
ror (
dB)
SNR (dB)SNRTH
Cram�r-Rao Bound
SidelobeTarget
SidelobeTarget
Mismatch affects threshold and asymptotic region leading to atypical performance curves
Mismatch affects threshold and asymptotic region leading to atypical performance curves
MIT Lincoln LaboratoryC. D. Richmond-13Monday 10th July
SEA@MIT 2006
Maximum-Likelihood Signal Parameter Estimation
• Complex Gaussian data model: All snapshots N x 1– Arbitrary N x N Colored Covariance– Deterministic Signal (“Conditional”)
• Colored noise only training samples available
π −N R −1 exp − x − Sv θ( )[ ]H R−1 x − Sv θ( )[ ]{ }
θML = argmax tMF θ( ) tMF θ( )=
vH θ( )R−1x2
vH θ( )R−1v θ( )
Data Model:
ML Estimator*:
S unknown
ClairvoyantMatched Filter
π −N (L +1) R −(L +1) exp − x − Sv θ( )[ ]H R−1 x − Sv θ( )[ ]− tr R−1XXH( ){ }
θML = argmax tAMF θ( ) tAMF θ( )=
vH θ( )ˆ R −1x2
vH θ( )ˆ R −1v θ( ) ˆ R ≡ 1
LXXH
Data Model:
ML Estimator*:
R unknown
S unknown
AdaptiveMatched Filter
*See Swindlehurst & Stoica Proc. IEEE 1998
Test Cell Training Data
MIT Lincoln LaboratoryC. D. Richmond-14Monday 10th July
SEA@MIT 2006
Approximating MSE Performance:Based on Interval Errors
• MSE given by E ˆ θ −θ1( )2⎧ ⎨ ⎩
⎫ ⎬ ⎭ ≡ ω −θ1( )2 p ˆ θ
ω( )dω∫
≈ ω −θ1( )2 × dω∫NIEIE IE
“Global Errors”
E ˆ θ ML −θ1( )2
θ1⎧ ⎨ ⎩
⎫ ⎬ ⎭
≅ 1− p ˆ θ ML = θk θ1( )k= 2
K
∑⎡
⎣ ⎢
⎤
⎦ ⎥ ⋅σ ML
2 θ1( )+ p ˆ θ ML = θk θ1( )k= 2
K
∑ ⋅ θk −θ1( )2
“Local Errors”
and asymptotic MSE:
σ ML2 θ1( )= ?
Both are functions of the estimated covarianceBoth are functions of the estimated covariance
p ˆ θ ML = θk θ1( )= ?
• Challenge is calculation of error probabilities
MIT Lincoln LaboratoryC. D. Richmond-15Monday 10th July
SEA@MIT 2006
Broadside Planewave Signal in White Noise: No Mismatch, R known, ULA
• N=18 element uniform linear array (ULA), (λ/2.25) element spacing– 3dB Beamwidth ≈ 7.2 degs, search space [60 120] degs– 0dB white noise, True Signal @ 90 degs (broadside)
• Asymptotic ML MSE agrees with CRB above threshold SNR• MIE MSE predictions very accurate above and below threshold
Element Level SNR (dB)
RM
SE in
Bea
mw
idth
s (d
B)
From 4000Monte CarloSimulations
ThresholdSNR
Var. UniformCRBAsympt. MSEMSE PredictionMonte Carlo
Dis
tanc
e (in
uni
ts o
f λ)ULA
ElementPositions
zn
MIT Lincoln LaboratoryC. D. Richmond-16Monday 10th July
SEA@MIT 2006
Signal in White Noise: Perturbed ULA,R unknown, L = 3N
• N=18 element ULA positions perturbed by 3-D Gaussian noise– Zero mean with stand. dev. 0.1λ; use single realization – Estimated colored noise covariance from L = 3N samples
• Note @ ~15dB SNR adaptivity loss limits beam split ratio to 16:1 as opposed to 22:1 when R is known
RM
SE in
Bea
mw
idth
s (d
B)
Element Level SNR (dB)
ThresholdSNR
From 4000Monte CarloSimulations
AdaptivityLoss
MC Known RMC Unknown RAsympt. MSEMSE PredictionCRB
( )233 , RMSn N σI0z +
λσ 1.0=RMS
Dis
tanc
e (in
uni
ts o
f λ)
ULAElement
Positions
MIT Lincoln LaboratoryC. D. Richmond-17Monday 10th July
SEA@MIT 2006
The Capon-MVDR Algorithm
R = E x l( )xH l( ){ } l =1,2,…,L
• Capon proposed filterbank approach to spectral estimation that designs linear filters optimally:
Given N x 1 vector snapshots with covariance
for x l
choose filter weights w according to
( )
minw HRw wHv θ( )=1
Minimum Variance Distortionless Response
w θ( )=
R−1v
such that
θ( )vH θ( )R−1v θ( )
E wH θ( )x l( )2{ }=
1vH θ( )R−1v θ( )
Solution well-known: Average Output Power of Optimal Filter:
PCapon θ( )≡
1vH θ( )ˆ R −1v θ( )
Capon’s Spectrum:
ˆ R = 1
Lx l( )xH l( )
l=1
L
∑where is sample covariance matrix
• Parameter estimate given by location of maximum power
θ θ θTˆ θ θT
AmbiguityFunction
Scan Angle
Estimation Error
Scan Angle
ˆ θ
1vH θ( )ˆ R −1v θ( )
1vH θ( )ˆ R −1v θ( )
1vH θ( )R−1v θ( )
1vH θ( )R−1v θ( )
Capon1969
θT
MIT Lincoln LaboratoryC. D. Richmond-18Monday 10th July
SEA@MIT 2006
Diagonally Loaded Capon Algorithm
ˆ R α = α ⋅ I +
1L
x l( )xH l( )l=1
L
∑ PCapon
I θ,α( )=1
vH θ( )ˆ R α−1v θ( )
PCaponI θ,α( )=
1vH θ( )ˆ R α
−1v θ( )
• In practice it is common to diagonally load the sample covariance:
– Slow convergence of small/noise eigenvalues (DL compresses)– High sidelobes (DL provides sidelobe [white noise gain] control)– Excessive loading can degrade performance
*Cox, IEEE T-SP 1987, Carlson, IEEE T-AES 1988
• Diagonal loading is necessary to invert matrix in snapshot deficient aacase, i.e. L ≤ N
• Eigenvectors of sample covariance remain unaffected by diagonal aaloading
• Featherstone et al. showed diagonally loaded Capon to be a robust aadirection finding algorithm
*“Robustify”Processing
MIT Lincoln LaboratoryC. D. Richmond-19Monday 10th July
SEA@MIT 2006
Single Signal Broadside to Arrayin Spatially White Noise, L = 0.5N
Output Array SNR (dB) Output Array SNR (dB)
L = 0.5Nα = -10dB
L = 0.5Nα = +10dB
RM
SE in
Bea
mw
idth
s (d
B)
RM
SE in
Bea
mw
idth
s (d
B)
• N=18 element uniform linear array (ULA), (λ/2.25) element spacing– 3dB Beamwidth ≈ 7.2 degs– 0dB white noise, True Signal @ 90 degs (broadside)– 4000 Monte Carlo simulations
• VB MSE prediction not applicable for L < N
Threshold SNRs
MSE Prediction
Monte CarloCRB
MIT Lincoln LaboratoryC. D. Richmond-20Monday 10th July
SEA@MIT 2006
Mismatch Example:Perturbed Array Positions
8dB Error in VB Prediction of15:1 Beamsplit Ratio SNR
Output Array SNR (dB)
RM
SE in
Bea
mw
idth
s (d
B)
L = 1.5Nα = +10dB
MSE PredictionMonte Carlo
VB MSE Prediction zn zn + en
en ~ N 3 0,I3σ RMS2( ),
AssumedNominal
Array Position
ActualPerturbed
Array Position
Based on Single Realization of Gaussian Perturbation:
• N=18 element ULA with perturbed positions but assumed straight• VB MSE prediction can lead to large errors in required SNRs• DL Capon is more robust DF approach: 18:1 vs 28:1 @ 40dB ASNR
10dB Error for 17:1
σ RMS = 0.04λ
MIT Lincoln LaboratoryC. D. Richmond-21Monday 10th July
SEA@MIT 2006
Outline
• Introduction
• Detection algorithms
• Estimation algorithms
• Open problems
• Summary
MIT Lincoln LaboratoryC. D. Richmond-22Monday 10th July