A Generic Formulation of the Extended Chirp Scaling Algorithm (ECS) for Phase Preserving ScanSAR and SpotSAR Processing Josef Mittermayer and Albert0 Moreira Deutsches Zentrum fiir Luft- und Raumfahrt (DLR) Institut flir Hochfrequenztechnik & Radarsysteme 82234 Oberpfaffenhofen,Germany T: +49-8153-28-2373, E-Mail:
[email protected] Abstract -- The paper first explains the common properties of The following properties of the azimuth signals are identiScanSAR and SpotSAR raw data. Then, a generalized Ex- cal in SpotSAR and a single ScanSAR burst: tended Chirp Scaling (ECS) processing approach is proposed 0 valid targets can be defined as targets with an azimuth for both, ScanSAR and SpotSAR phase preserving data procillumination length equal to the full burstlspotlight aperessing. The performance of the proposed approach is demonture length strated by interferometric ScanSAR and SpotSAR processing 0 the start and the end times of the azimuth illumination are examples. equal for all valid targets independent of range 0 I. INTRODUCTION the azimuth resolution is almost independent of the aziScanSAR and SpotSAR are two operation modes of SAR muth positioning of a target but shows noticeable desystems, which improve the standard stripmap mode of oppendency on the range position eration in two different ways. 0 the azimuth processing can be performed by stripmap In ScanSAR, the scene size in range is extended by scanmethods but this requires a large data overhead by zero ning the antenna in elevation and alternating illumination of padding, which makes the processing inefficient several subswaths. The raw data of one illumination of a 0 due to the constant illumination start and end times for all subswath are denoted by raw data burst. Like in stripmap targets, SPECAN [ 7 ] is an efficient processing apSAR, the scene extension in azimuth is principally only deproach, which can be made highly accurate and phase termined by the length of the overflight. By the scanning in preserving by azimuth scaling [ 5 I[ 2 ] elevation, the synthetic aperture length is reduced to the burst The target Doppler histories in a single ScanSAR raw data length. Due to the short burst length, the achievable azimuth resolution is reduced. ScanSAR shows a lot of interesting burst and in SpotSAR raw data are very similar as shown in Fig. 1. The left diagram shows the Doppler history of three properties, which are for example discussed in [ 5 1. In SpotSAR, the antenna is steered in azimuth direction. targets A, B and C located at different azimuth but identical During the formation of the synthetic aperture, the antenna is range position for a SpotSAR illumination. The signals in the constantly looking to scene center direction. The available raw data are shown in dashed style while the continuous lines synthetic aperture is much longer than in the stripmap mode show the unambiguous signals. The right diagram shows the and is denoted by spotlight aperture. The long spotlight aper- Doppler history of the same targets in ScanSAR. The Doppler ture provides azimuth signals with a large Doppler bandwidth frequency is denoted by fa, the azimuth time by ta and the and therefore provides a high geometric resolution in azi- frequency histories are linearly approximated. From the figure muth. In this paper, a SpotSAR mode with a straight flight it is obvious that a similar azimuth processing can be applied path and a data acquisition without Dechirp on Receive [ 1 ] to both, ScanSAR and SpotSAR if the later one is made unambiguous, i.e. by subaperture processing [ 2 1. or with Dechirp on Receive in range is considered. Due to the azimuth antenna steering in SpotSAR, the azi4 4 muth scene size is limited by the azimuth angle of the real I 4 4 aperture. The scene size in range is mainly determined by the antenna elevation pattern and secondly by the antenna steering. More information about the spotlight mode and a processing algorithm without interpolation can be found in [ 2 1.
11. SCANSAR AND SPOTSAR RAW DATA The range modulation in SpotSAR and ScanSAR raw data is given by the modulation of the transmitted radar pulse, which is mostly a linear frequency modulation. In SpotSAR, often Dechirp on Receive is used before the A/D-conversion in order to transform the range signal into a superposition of sinusoidal signals with reduced bandwidth. In case of no Dechirp on Receive, the range modulation in SpotSAR and ScanSAR raw data is identical.
0-7803-6359-0100/$10.00 0 2000 IEEE
I
Spotlight Aperture _ _ ~
+---U
*
Burst Length
Fig. 1: Doppler history in SpotSAR (IeB) and ScanSAR (right) 111. BLOCK DIAGRAM OF THE GENERIC ECS The block diagram of the generic formulation of the ECS for phase preserving SpotSAR and ScanSAR processing is shown in Fig. 2. All parts in black letters on white background are used in ScanSAR and SpotSAR processing. Black
108
The azimuth processing is continued by azimuth scaling and the azimuth antenna pattern correction. The azimuth scaling transforms the azimuth modulation into a pure quadSpotlight Raw Data / ScanSAR Raw Data Burst ratic phase modulation which is independent of range [ 5 ] [ 21. The possibility of selecting the final azimuth sampling distance during the azimuth scaling is very useful for exact Path for ScanSAR co-registration of interferometric image pairs and for the mosaic of the single ScanSAR subswaths. The term symmetSpotSAR Raw Data ric azimuth scaling is due to an azimuth time shift during the azimuth scaling for a minimized symmetric azimuth time extension [ 2 1. The azimuth pattern correction is important since the azimuth signal of a target is weighted by different parts of the azimuth antenna pattern dependent on the target azimuth position. If not corrected, an amplitude modulation is visible in the image. In ScanSAR, the pattern can be corrected accurately in azimuth frequency domain. For SpotSAR, the subaperture approach allows an approximated azimuth antenna pattern correction, which is the more accurate the smaller the subapertures are. The azimuth antenna pattern correction in can also be performed in the final image. The apRg. Siibbe S u p p e ~ ~ SpotSAR ~( IRg. Siilobe S u d proximation there is the assumption that each target is always illuminated by the same antenna diagram angle. After transforming the data into azimuth time domain, the deramping operation is performed by using a quadratic phase function, which is constant with range. In case of SpotSAR, the subapertures are recombined next. The azimuth compression is then performed by the final azimuth FIT, which is followed by a quadratic phase correction, required for phase preserving SPECAN processing and by an additional azimuth scaling phase correction, which is due to the azimuth time shift during the symmetric azimuth scaling. After the phase corrections, the SpotSAR single look complex (SLC) image is available for further interferometric processing. In case of ScanSAR, the processed bursts have to be combined in order to form one SLC ScanSAR subswath image for each azimuth look. The number of possible azimuth looks is defined by the ScanSAR scanning strategy. From Processed SpotSAR SLC each interferometric SLC subswath image pair an interferogram can be generated and all these interferograms can be coherently added for phase noise reduction.
letters on grey background mean usage only in SpotSAR processing and white letters mean usage only for ScanSAR.
I
I
Fig, 2: Block diagram of generic ECSfor SpotSAR and ScanSAR
IV. AZIMUTH PROCESSING The azimuth processing starts with the weighting of the raw data for azimuth sidelobe suppression. This can easily be performed since in the raw data all azimuth signals are located at the same azimuth time position and the ECS is focussing in azimuth frequency domain. Next, the subaperture formation is performed in case of spotlight data for avoiding an upsampling in azimuth [ 2 1. After transforming the data into range Doppler domain, either chirp scaling or frequency scaling is performed for full range compression and range cell migration correction, as explained in more detail in section V.
V. RANGE PROCESSING As can be seen in Fig. 2, the range processing is performed by following one of the two signal paths. The left signal path is for ScanSAR data and for SpotSAR data without Dechirp on Receive. The range processing in this path is performed analog to the range processing in the ECS algorithm [ 5 ] for stripmap. In case of spotlight with Dechirp on Receive, the right signal path has to be selected using the frequency scaling approach [ 2 1. Both methods require only multiplications and FFTs and use partially equal analytical expressions. In case of chirp scaling, the weighting for range sidelobe suppression is performed in range frequency domain while the weighting in the frequency scaling approach is Performed in range time domain.
109
shown in Fig. 5 after flat earth subtraction and an additional VI. PROCESSING EXAMPLES In Fig. 3, a spotlight interferogram is shown, which was phase noise reduction by a moving average in range. processed by the ECS of Fig. 2. Interferometric E-SAR stripmap raw data were used and two SLC spotlight images were extracted for the interferogram generation [ 3 3. The interferogram is shown after flat earth subtraction. The height variation was smaller than the 2x-height ambiguity of 40 m and thus no phase un-wrapping was required. The azimuth resolution is 0.40 m in near range and 0.44 m in far range. The range resolution is 2.16 m and the maximum height variation in the image is about 20 m. The interferometric baseline. was 1.6 m. The SpotSAR interferogram was validated with a stripmap interferogram of the same scene. Both interferograms provide the same height information [ 3 1.
Fig. 3: Spotlight interferogram Fig. 4 shows a result of ECS ScanSAR processing. With courtesy from CSA, Radarsat ScanSAR data within the CCRS ADRO Proposal #500 were used for the processing. The image on the left in the figure was processed from a part of one subswath and shows a part of the Barthust Island in Canada, including land surface, ocean and ice areas.
Fig, 4: ScanSAR image ('le@) and coherency map (right) On the right in Fig. 4, there is the coherency map of one SLC azimuth image pair. Two azimuth SLC image pairs were available due to the scanning and the image on the left is the multilook image with 6 looks in total. The scene dimensions are 181 km in azimuth and 78 km in slant range (196 km ground range). The azimuth resolution is 68 m in near and 74 m in far range. The slant range resolution is about 35 m. The multilook interferogram resulting from the coherent addition of the two azimuth SLC subswath interferograms is
110
VII. CONCLUSIONS This paper shows the similarities between ScanSAR and SpotSAR and demonstrates how these similarities can be exploited by using a generic formulation of the ECS for the processing. The generic formulation provides advantages for the processor implementation in hard- or software and off-line or real-time since equal or very similar processor modules can be used for both, ScanSAR and SpotSAR data processing.
VIII. REFERENCES G.Carrara, R S. Goodman, R.M. Majewski: "Spotlight Synthetic Aperture Radar", Artech House Boston, London, 1995. [ 2 ] J. Mittermayer, A. Moreira, 0. Lofeld: "Spotlight S A R Data procesSing Using the Frequency Scaling Algorithm", IEEE Trans. on Geosc. and Remote Sensing, Vol. 37, No. 5, September 1999. [ 3 I Mittemayer, J., A. Moreim and 0. Loffeld "Gmparison of Stripmap and Spotlight Interfmmtric SAR procesSing using E-SAR Raw Data". Proc. of EUSAR 2000. [ 4 ] Mittermayer, J., A. M m k "The Extended chvp scaling Alge rithm for ScanSAR Interfmmetry".Proc. of EUSAR 2000. [ 5 3 A. Moreira, J. Mittemayer and R Scheiber, " Extended chvp Scaling Algorithm for Air- and Spacebome SAR Data procesSing in Stripmap and ScanSAR Imaging Modes", lEEE Trans. on Geosci. and Remote Sensing, Vol. 34, NO. 5, September 1996. [ 6 ] R K. b e y , Runge, H, Bamler, R Cumming, I. and Wong, F.: "precisionSAR processing without Interpoldonfor Range Cell Migration correction". IEEE Trans. on Geosci. and Remote Sensing, Vo1.32, No.4, July 1994. [ 7 ] Sa& M., Ito, M.R, Cumming, I.G.: "Application of Efficient Linear FM Marched Filtering Algorithms to S A R Processing". IEEproC.-F, V01.132, N0.1, 1985.p ~45-57. .
[ 1]
