1 out of 30 Digital Processing Algorithms for Bistatic Synthetic Aperture Radar Data 4 May 2007 by Y. L. Neo Supervisor : Prof. Ian Cumming Industrial Collaborator : Dr. Frank Wong Sponsor : DSO National Labs (Singapore) T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

2 out of 30 Agenda Bistatic SAR principles A Review of Processing Algorithms Contributions 1.Point Target Spectrum 2.Relationship Between Spectra 3.Bistatic Range Doppler Algorithm 4.Non Linear Chirp Scaling Algorithm Conclusions

3 out of 30 Bistatic SAR In a Bistatic configuration, the Transmitter and Receiver are spatially separated and can move along different paths. Bistatic SAR is important as it provides many advantages –Cost savings by sharing active components –Improved observation geometries –Passive surveillance and improved survivability

4 out of 30 Imaging geometry of monostatic/bistatic SAR Target Monostatic SAR Platform RTRT Platform flight path 2 x

5 out of 30 Focusing problems for bistatic algorithms Traditional monostatic SAR algorithms based on frequency domain methods make use of 2 properties 1.Azimuth-invariance 2.Hyperbolic Range Equation Bistatic SAR data, unlike monostatic SAR data, is inherently azimuth-variant –targets having the same range of closest approach do not necessarily collapse into the same trajectory in the azimuth frequency domain. Difficult to derive the spectrum of bistatic signal due to the double square roots term (DSR).

6 out of 30 Agenda Bistatic SAR principles A Review of Processing Algorithms Contributions 1.Point Target Spectrum 2.Relationship Between Spectra 3.Bistatic Range Doppler Algorithm 4.Non Linear Chirp Scaling Algorithm Conclusions

7 out of 30 Overview of Existing Algorithms Time domain algorithms are accurate as it uses the exact replica of the point target response to do matched filtering Time based algorithms are very slow – BPA, TDC Traditional monostatic algorithms operate in the frequency domain –RDA, CSA and ωKA –Efficiency achieved in azimuth frequency domain by using azimuth-invariance properties

8 out of 30 Existing Bistatic Algorithms Frequency based bistatic algorithms differ in the way the DSR is handled. Majority of the bistatic algorithms restrict configurations to fixed baseline. Three Major Categories 1.Numerical Methods – ωKA, NuSAR – replace transfer functions with numerical ones 2.Point Target Spectrum – LBF 3.Preprocessing Methods – DMO

9 out of 30 LBF (Loffeld’s Bistatic Formulation) Solution for the stationary point is a function of azimuth time given in terms of azimuth frequency  (f  ) LBF derived an approximation solution -  b (f  ) to stationary phase Using this relation, the analytical point target spectrum can be Formulated - LBF Approximate Solution to Stationary phase Quasi-monostatic Term Bistatic Deformation Term

10 out of 30 Phase modulator Migration operator Rocca’s smile operator transforms Bistatic Trajectory to Monostatic Trajectory DMO (Dip and Move Out) Transform Bistatic data to Monostatic (using Rocca’s Smile Operator) Assumes a Leader-Follower flight geometry (azimuth-invariant)

11 out of 30 Agenda Bistatic SAR principles A Review of Processing Algorithms Contributions 1.Point Target Spectrum 2.Relationship Between Spectra 3.Bistatic Range Doppler Algorithm 4.Non Linear Chirp Scaling Algorithm Conclusions

12 out of 30 Major Contributions of the Thesis #1 Derived an accurate point target spectrum using MSR (Method of Series Reversion) #3 Derived Bistatic RDA - Applicable to Parallel flight cases with fixed baseline #2 Compare Spectrum Accuracy - MSR is more accurate than Existing point target Spectrum – LBF and DMO #4 Derived NLCS Algorithm – Applicable to Stationary Receiver & Non-parallel Flight cases Focused Real bistatic data. Collaborative work with U. of Siegen (airborne-airborne data)

13 out of 30 Derivation of the analytical bistatic point target spectrum Problem : To derive an accurate analytical Point Target Spectrum POSP: Can be used to find relationship between azimuth frequency f  and azimuth time  f  = [1/(2  )] d  (  )/d  But we have to find  = g(f  ). Difficulty: phase  (  ) is a double square root Contribution#

14 out of 30 Our Approach to solving the point target spectrum Approach to problem: –Azimuth frequency f  can be expressed as a polynomial function of azimuth time . –Using the reversion formula,  can be expressed as a polynomial function of azimuth frequency f  Journal Paper Published : Y.L.Neo., F.H. Wong. and I.G. Cumming A two-dimensional spectrum for bistatic SAR processing using Series Reversion, Geoscience and Remote Sensing Letters, Jan Contribution#

15 out of 30 Series Reversion Series reversion is the computation of the coefficients of the inverse function given those of the forward function Contribution#

16 out of 30 New Point Target Spectrum Method of Series Reversion (MSR) An accurate point target spectrum based on power series is derived Solution for the point of stationary phase is given by The accuracy is controlled by the degree of the power series MSR can be used to adapt Monostatic algorithms to process bistatic data - RDA and NLCS Contribution#

17 out of 30 Linking - MSR, LBF and DMO We established the Link between MSR, LBF and DMO Using the MSR, we derived a new form of the point target spectrum using two stationary points. Similar to LBF, the phase of MSR can be split into quasi-monostatic and bistatic deformation terms. Journal Paper Submitted : Y.L.Neo., F.H. Wong. And I.G. Cumming A Comparison of Point Target Spectra Derived for Bistatic SAR Processing, Transactions for Geoscience and Remote Sensing, submitted for publication,14 Dec Contribution#

18 out of 30 Link between MSR and LBF Stationary point solutions Split phase into quasi monostatic and bistatic components MSR LBF Contribution#

19 out of 30 LBF and DMO Rocca’s smile operator can be shown to be LBF’s deformation term if the approximation below is used Contribution# Approximation is valid when baseline is short when compared to bistatic Range

20 out of 30 Alternative method to derive the Rocca’s Smile Operator Contribution#

21 out of 30 Link between MSR, LBF and DMO Contribution# Monostatic Term Leader - Follower Flight configuration Quasi-Monostatic Term Bistatic Deformation Term Expand about Tx and Rx stationary points Consider up to Quadratic Phase only 2D Point Target Spectrum Rocca’s Smile Operator Baseline is short Compared to Range LBF MSR DMO

22 out of 30 Bistatic RDA Developed from the MSR 2D point target spectrum Monostatic algorithms like RDA, CSA achieve efficiency by using the azimuth-invariant property Bistatic range histories are azimuth-invariant by if baseline is constant The MSR is required as range equation is not hyperbolic Journal Paper Reviewed and Re-submitted: Y.L.Neo., F.H. Wong. And I.G. Cumming Processing of Azimuth-Invariant Bistatic SAR Data Using the Range Doppler Algorithm, IEEE Transactions for Geoscience and Remote Sensing, re-submitted for publication, 12 Apr Contribution#

23 out of 30 Main Processing steps of bistatic RDA Range FT Azimuth FT SRC RCMC Azimuth IFT Baseband Signal Focused Image Range Compression Range IFT Azimuth Compression Contribution#

24 out of 30 Real Bistatic Data Focused using Bistatic RDA Copyright © FGAN FHR Contribution#

25 out of 30 Non-Linear Chirp Scaling Existing Non-Linear Chirp Scaling –Based on paper F. H. Wong, and T. S. Yeo, “New Applications of Nonlinear Chirp Scaling in SAR Data Processing," in IEEE Trans. Geosci. Remote Sensing, May –Assumes negligible QRCM (i.e., for SAR with short wavelength) –Shown to work on Monostatic cases and the Bistatic cases where receiver is stationary We have extended NLCS to handle Bistatic non- parallel tracks cases Contribution#

26 out of 30 Extended NLCS Able to handle higher resolutions, longer wavelength cases Corrects range curvature, higher order phase terms and SRC Develop fast frequency domain matched filter using MSR Solve registration to Ground Plane Journal Paper written: F.H. Wong., I.G. Cumming and Y.L. Neo, Focusing Bistatic SAR Data using Non-Linear Chirp Scaling Algorithm, IEEE Transactions for Geoscience and Remote Sensing, to be submitted for publication, May Contribution#

27 out of 30 Main Processing steps of Extended NLCS Range compression LRCMC / Linear phase removal Azimuth compression Baseband Signal Focused Image Non-Linear Chirp Scaling Range Curvature Correction The NLCS scaling function is a polynomial function of azimuth time NLCS applied in the time domain SRC and Range Curvature Correction --- range Doppler/2D freq domain Azimuth matched filtering --- range Doppler domain Range Curvature Correction and SRC Non-Linear Chirp Scaling Contribution#

28 out of 30 Agenda Bistatic SAR principles A Review of Processing Algorithms Contributions 1.Point Target Spectrum 2.Relationship Between Spectra 3.Bistatic Range Doppler Algorithm 4.Non Linear Chirp Scaling Algorithm Conclusions

29 out of 30 Concluding Remarks With our four contributions, a more general and accurate form of bistatic point target spectrum was derived. Using this result, we were able to focus more general bistatic cases using several algorithms that we have developed. We plan to work on future projects that make use of the results from this thesis – Interest express from several agencies – DRDC (Ottawa), DSO National Labs (Singapore) and U. of Siegen (Germany). –Satellite – Airborne (TerraSAR-X and PAMIR) –Satellite/Airborne – stationary receiver (X and C band) using RADARSAT-2 or TerraSAR-X

30 out of 30 Q&A Thank You