Synthetic aperture radar systems usually require considerable signal processing treatment in order to allow the retrieval of the desired information from the data. Spatial resolution as well as radiometric, polarimetric and interferometric calibration accuracy have direct influence on the potential to measure or infer physical parameters.
The focus of the SAR Signal Processing Group is to continuously improve, adapt and develop new algorithms for the processing of data of DLR’s airborne SAR sensors for generating products ready to use by application scientists.
Efficient algorithm implementation is one of the strong requirements for handling the increasing amount of the acquired and processed F-SAR data. Each year, about 300-400 data takes are acquired by F-SAR (see "More Links"), corresponding in average to ca. 20 GB of raw data per data take and about 60 GB per processed data product.
The STEP is used for generating offline-processed F-SAR data products at highest possible resolution and optimum calibration accuracy. The processor includes:
Linked to the requirements of high resolution and precise motion compensation together with the high number of flight campaigns abroad, the STEP processing framework has been extensively adapted to run on clustered LINUX servers, while assessing common data storage for the raw and processed data. During extensive campaigns a mobile server is used to allow on-site data processing and quality check.
Due to its modular architecture the STEP can be easily adapted to process data of other airborne SAR systems.
The processing chain for repeat-pass SAR interferometry demands extremely high complexity. The accurate processing is a precondition to obtain high quality Pol-InSAR data products as input to the model based inversion approaches (see Pol-InSAR Group in "More Links").
The performance of standard airborne SAR processing is limited by the accuracy of the navigation data available to perform the motion compensation (state-of–the art is a combination of inertial and GPS sensors). Although the relative accuracy is very good, enabling well focused data, the absolute performance is limited by the absolute precision of the differential GPS signal which is in the order of 1-5cm (ca. one interferometric phase cycle assuming SAR data at C-band and the two way propagation delay). It is obvious that this accuracy is not sufficient for repeat-pass interferometry. A robust estimation approach based on the so-called multi-squint technique is integrated into the processing flow, which estimates residual errors (in horizontal and vertical directions) between the two tracks of the interferometric acquisitions directly from the processed interferometric data pairs.
The original residual track errors are in the order of 1-2cm and are in accordance with the accuracy of the navigation system based on differential GPS and INS. Without this compensation approach, phase errors and coherence degradations would occur which have strong impact on the interferometric data quality.
Residual track errors measured and corrected by repeat-pass processing approach. Almost perfect compensation is reached after 3-4 iterations, corresponding to a correction of the residual motions errors with millimeter accuracy.
The STEP processor now provides the following operational products:
• Slant range polarimetric multilook and SLC data (Radar Geometry Images – RGI product).
• Geocoded Terrain Corrected Data (GTC product).
• Digital Elevation Models (DEM product)
• Interferometric Data (INF product, including coregistered SLC data, coherence and phase)
• Circular SAR Data (CSAR GTC product, for data takes acquired along circular trajectories)
For archiving purposes, each data product is ingested into the Data Ingestion and Management System (DIMS) of the German Remote Sensing Data Center (DLR-DFD) allowing world-wide access via EOWEB (see "More Links"). Personalized data ordering options are set up for external customers.
Polarimetric F-SAR images may be combined into large scale mosaics. For this purpose the correction of the incidence angle is applied. Border effects are hardly visible because of the precise 3D radiometric calibration of the antenna patterns.
Polarimetric L-band image mosaic of the open cast mining area close to Jülich, Germany, acquired in 2013. Simultaneous data acquisition from the same Do-228 airborne platform of PSLR radiometer and IR data.
For more informations on TERENO, see "More Links" (right).
Digital elevation models from single-pass SAR interferometry are one of the standard F-SAR products.
Digital elevation model of Findelen Glacier, Switzerland, computed from single-pass interferometric F-SAR data in X-band (WGS-84, UTM zone 32) acquired in October 2014.
However, for high resolution and high precision DEMs a fusion approach has been developed combining the excellent height precision of large baseline repeat-pass interferometry with the good absolute accuracy of single-pass acquisitions. The resulting DEMs have accuracy similar to DEMs derived from Airborne Laser Scanning. The methodology is applied to generating DEMs in the Wadden Sea with data acquired during low tide. For large scale mapping a mosaic procedure is used to calibrate and combine adjacent strips for generating large area DEMs.
Comparison of digital elevation models computed from fusion of single- and repeat-pass F-SAR data and ALS data. The area corresponds to the western part of the Jadebight, German Wadden Sea. Within the 3 months of data acquisition several changes in the deposits can be noted.
The high resolution offered by the F-SAR sensor in X-band (especially the step-frequency mode) allows imaging of man-made targets and infrastructure, including the possibility of change detection by means of image overlay, interferometric methods (coherence) or analysis of polarimetric signatures.
Polarimetric X-band image of the stage area of the Rock-am-Ring festival on June 6, 2014 (left). The audience in front of the stage is easily recognized and also the moving people along the roads towards the stage. Change detection image using a second reference image acquired on May 14, 2014. See "More Links" (right) for further results of the campaign.
For more informations on Rock am Ring, see "More Links" (right).
Repeated data acquisitions can be performed also with large temporal baselines of several days, weeks or even month. This allows applying differential interferometry techniques for the detection of subsidence and/or land slide effects or for measuring the surface velocity of glaciers. Especially in areas which decorrelate at X- or C-bands, the performance of spaceborne sensors is rather limited. To fill the gap, airborne sensors at larger wavelength, preferably at L-band or S-band, can be used. The temporal baselines between acquisitions ranges from 1day for glacier movement to several months for subsidence.
The processing of SAR data recorded along circular trajectories is conveniently performed using an adapted fast factorized backprojection algorithm. The extremely good theoretical horizontal resolution is in the order of a fraction of the wavelength (e.g. 6cm in L-band). The focusing requires consideration of a precise digital elevation model. Also tomographic processing with multiple aspect angles is supported for the mapping of vegetation or ice volumes, ideally by combining data acquired along multiple circles (spiral).
Fully polarimetric L-band circular SAR image in the Pauli basis (image diameter 1.8 km). Zooms 1 to 4 depict an agricultural area, high-voltage pylons, a cattle herd and a low voltage power line, respectively.
When airborne SAR data acquisition is repeated along parallel tracks a second synthetic aperture is generated, which allows true 3D imaging. Signal processing techniques for volumetric imaging include different beamforming techniques like MVDR and MUSIC, but also Compressive Sensing (see Multimodal Algorithms in "More Links", right).
SAR focussing techniques can also be applied to nadir looking ice sounding radar data. The increased processing gain leads to a better detection of the bedrock. In addition, the clutter signals can be suppressed when using multi-aperture antennas combined with digital beamforming techniques.
Bedrock beneath 2000m thick ice in Antarctica. Raw data of 150 MHz ice sounding radar provided by British Antarctic Survey (BAS).
Improving the surface clutter suppression: traditional beamforming (top) versus MVDR (bottom). Data acquired 2012 in P-band by ESA’s POLARIS system across Jutulstraumen glacier, Antarctica.