Sharpening Techniques - Design of Sequential Circuit - Lecture Slides, Slides of Electronic Circuits Design

Download Sharpening Techniques - Design of Sequential Circuit - Lecture Slides and more Slides Electronic Circuits Design in PDF only on Docsity! SHARPENING TECHNIQUES FOR SENSOR FEATURE ENHANCEMENT Docsity.com EMERGING SENSOR EXPLOITATION OPPORTUNITY • INCREASED DIMENSIONALITY & CHANNELS (multiple sensors/platform and multiple platforms) • 3-D IMAGERY DATA • Non-Computed Imaging: Hyperspectral Imaging (HSI) • Computed Imaging: Interferometric Synthetic Aperture Radar (IFSAR) • Scanning LADAR • Video Frame Sequence • 3-D NON-IMAGERY DATA • STAP (arrays or synthetic aperture) • Micro-Doppler (more general: nonstationary micro-motion effects) • MULTI-CHANNEL • Fusion of Coherent RF Sensors of Different Operational Frequencies • Waveform Diversity MIMO of Multiple Platform Scenarios • CREATES LIKELIHOOD OF CORRELATIVE RELATIONSHIPS AMONG DIMENSIONS AND CHANNELS • SHARPEN 3-D Multi-Channel DETAIL WITHOUT DECREASING TRANSMISSION RATE • RESTORE DETAIL WHILE DECREASING TRANSMISSION RATE • MOTIVATION EXAMPLE FOR A 2-D MICRO-DOPPLER CASE & A 2-D,2-Ch RADAR FUSION CASE Docsity.com GMTI MICRO-MOTION FEATURES 4.3 m 5.8 m 7.0 m 9.0 m Back Rim Tire Treads No Doppler Tire Treads Micro-Temporal (localized time variations) Micro-Spectral (localized frequency variations) Micro-Spatial (localized range cell variations) Docsity.com Generic Sharpening / Enhancement Concept 1-D, 2-D, 3-D, 4-D 1-C, M-C (MIMO) Domain Change Sharpening Algorithms 1-D, 2-D, 3-D, 4-D 1-C, M-C (MIMO) Reverse Domain Change Signal Entity In Sharpened / Enhanced Signal Entity Out Special Transform Inverse Transform Estimate/Predict Missing High “Frequency’ Content • CURRENT STATE: 1-D, 2-D, and 1-D/Multi-Ch Sharpening Algorithms • SHARPENING PROCEDURE REQUIRES TWO FUNDAMENTAL COMPONENTS: • Selection of Appropriate Transform (may not be the same in all dimensions) • Predictive Transform Extrapolation Techniques in 1/2/3-D & MIMO Versions Docsity.com NOTIONAL EXAMPLES OF 1-D,2-D,3=D SHARPENING RESULTS SIGINT Application 4-D, 2-D, or 3-D i 1D ANS! => (Fourier, Wavelet, KLT, Or Other) SAR Imaging 1-D, 2-D, or 3-D PREDICTIVE SHARPENING & WAVENUMBER TECHNIQUES Hsl Imaging Cube INVERSE TRANSFORM (Fourier, Wavelet, > KLT, Or Other Before Sharpening TRANSFORM 1-D 3-D 1-D, 2-D, or 3-D 3-D | After Sharpening Docsity.com AMTI Vs GMTI COLLECTION CONDITIONS Plane, Helicopter Radar Radar Platform Truck, Tank, Vehicle PROCESSING COMPARISONS: • Short-Time Fourier Transform (FFT-based baseline) • Wigner-Ville T-vs-F Quadratic Representation/Distribution • Predictive Time-Bandwidth Extrapolation (Sharpening) Docsity.com AMTI: AIRBORNE RADAR TARGETS • X-band ( ~ 10 GHz ) homodyne CW radar returns off helicopters in flight CREATES ONLY A A 2-D RESPONSE RATHER THAN A 3-D RESPONSE ( no movement through range bins ) • Doppler signatures in baseband ( 0 Hz IF ) I/Q signals after complex demodulation; + frequencies toward radar and – frequencies away from radar • Baseband sampled at 48000 sps with 16-bit A/D conversion precision ( 96 dB DNR ) • Up to 70 dB signal component level range may be observed in data, since negligible clutter • Nonstationary components contributing doppler signatures: Fuselage skin line Hub signature Alias of JEM Main rotor modulations & blade flash Multi-path bounces Cross feed Tail rotor modulations & blade flash Stabilizer bar (Huey) • Note that –500 to +500 Hz region replaced with time code signal for accessing taped data German UH – 1D Huey Eurocopter BO - 105 Docsity.com STFT T-vs-F GRAM of 4-BLADE BO-105 HELICOPTER European BO-105 Helicopter NOTE THIS Docsity.com WINDOWED - CAF* WIGNER TFA GRAM OF BO-105 *CAF = complex ambiguity function Docsity.com Traditional STFT X(t,t) H(t,t) MOTIVATION FOR SHARPENING ALGORITHM: TRADITIONAL STFT TFA PROCESSING FLOW DIAGRAM Alternative Path Motivated by Quadratic TFAs Docsity.com Traditional STFT X(t,t) H(t,t) ALTERNATIVE STFT TFA PROCESSING FLOW CHART Alternative Path Motivated by Quadratic TFAs CREATES 2-D DATA ARRAY AND 2-D COMPLEX TRANSFORM ARRAY MAPPINGS FOR FINITE DATA Docsity.com GMTI : GROUND-BASED TANK TARGET • X-BAND SAR SYSTEM PHASE HISTORY DATA (DARPA project) • DPCA RECEIVE • CLUTTER CANCELLATION BIG INGREDIENT IN PROCESSING, but not discussed here Docsity.com GMTI BEFORE/AFTER SHARPENING OF GROUND-BASE TARGET (TANK) Note extra detail now revealed and lack of sidelobes due to Fourier aperture artifacts 8 Center Time of Anabysis Window (mms) 3 of 08 0 oz of of OB Fomueney (3 Hi Predictive Sharpening-Based Moving Target Features a 2 urer-Based Moving Target Features @)- Docsity.com EXAMPLE OF 2-D/2-CHANNEL SHARPENING Bandwidth-Sharpened Founer-Based (2-D FFT} Minimum-Variance-Based Parametric Sharpened (right) IS.AR Imagery, 3.42 GHz center frequency, S00 MH banteridth, 40 dB cok - © 100 a SHiamd Founer-Based em) and = i 6) E 4 i coded range, VV polarization bend Fourter-Based Jef] and Parametic Sharpened (ngit] ISAR Imagery, 10.67 GHz center irequency, $00 Mi bandwidth, Onder 6 x 6 Parameter sharpenmng, 40 d6 color- Coded range, ¥¥ polarization 200 . au) _ 0.200 a 300 i ” . py sce (cen) [pine wine: 4 BOM: ates tron se a ie 300 a Fine A885 e tes toe be SUPPLEMENTARY SLIDES Docsity.com BASELINE STFT LINEAR TFA (SPECTRO)GRAM ( ) τ τ π τ τ − = − − = ∫ 2 2 2 ( , ) ( ) ( ) exp( 2 ) STFT , ( , ) T T X f t x t h j f d f t X f t where t is the analysis window center time t τ T x(τ) Docsity.com THIRD TARGET: ARBEITSGEMEINSCHAFT TRANSALL C-160 TWIN-ENGINE TURBOPROP TRANSPORT a = aie od Docsity.com Agorithm A: FLOW DIAGRAM OF STFT WITH 2X SIGNAL SUBSPACE EXTRAPOLATION (1-D solution) Forward LP Extrapolation Backward LP Extrapolation Original Data Within Analyis Window Nonstationary Signal FFT OF EACH ROW OF WDF SQUARED MAGNITUDE OF STFT Time - Frequency Representation by STFT with Linear Prediction or Signal Subspace Data Extrapolations APPLY WINDOW TO EXTENDED DATA AT CENTER TIME WINDOWED DATA FUNCTION (WDF) OF EXT. DATA or Linear Prediction Technique Signal Subspace Technique APPLY UNIFORM ANALYSIS WINDOW AT EACH CTR. TIME COMBINE ORIGINAL DATA + FOR. & BACK. EXTENDED DATA AR Model Order; No. of Estimated Signals CALCULATE FORWARD & BACKWARD LIN. PRED. DATA EXTENSIONS ESTIMATE FOR.& BACK. LIN. PRED. PARAMS. BY COVARIANCE LP METHOD ESTIMATE FOR.& BACK. LIN. PRED. PARAMS. BY TRUNCATED SVD 1X 2X Docsity.com • Rectangular Toeplitz Data Matrix of Covariance Case of Linear Prediction for N Data Samples • Least Squares Normal Equations for Forward f and Backward b Linear Prediction Filters of Order p (note that is the squared error) • SVD of Data Matrix • Excise (delete) “Noise” Eigenvectors Leaving M Dominant Eigenvectors (assume singular values are ordered by magnitude ) • Use in lieu of to Compute Linear Prediction Parameters +       = − +      −          [ 1] [1] [ ] [ 1] [ ] [ ] x p x X x N p x p x N x N p ρ                =     −                1 [1] 0 [ 1] 0 [ ] 0 f p f p H f p f p a X X a p a p ρ           −          =                     0[ ] 0[ 1] 0[1] 1 b p b p H b p b p a p a p X X a σ − = = Σ = ∑ 0 N p H H i i i i X U V u v σ − = = ∑ 1 0 M H excised i i i i X u v HX XHexcised excisedX X and ρp σ σ σ −≥ ≥ ≥0 1 N p Reduced Rank Data Matrix LINEAR PREDICTION SOLUTION BY NOISE EXCISION (De-Noising) Docsity.com STFT + Noise Excision (via SVD) + Extrapolation TFA GRAM OF BO-105 LINEAR Docsity.com