Ground surface imaging is an important component in a variety of environmental and geophysical applications. The Synthetic Aperture Radar (SAR) technique uses aircraft- or satellite-mounted radar to bounce microwave signals off the ground to produce image data. Both phase and amplitude of the signal are used, in contrast with conventional radar, which uses only amplitude. This allows a larger aperture to be synthesized, thus permitting high-resolution images. There are a total of eight radar datasets addressed in this application: two like-polarized and two cross-polarized datasets for each of two radar frequencies (L-band at 23 cm. wavelength and C-band at 5 cm. wavelength). This results in a very large amount of data. In a recent space shuttle mission, using the Spaceborne Imaging Radar-C/X-Band Synthetic Aperture Radar (SIR-C/X-SAR) instrument to study the Earth's surface, the data generated was approximately 32 terabits.
Processing of the captured data is done in three stages: preprocessing, correlation, and postprocessing, with the correlation phase being the most compute-intensive. Parallelism is achieved both through data distribution for processing a given channel by multiple processors and by processing up to eight channels simultaneously. The data is formed into overlapping frames, each containing approximately 3,000 samples in range, which consist of samples taken as the SAR sensor moves along the azimuth direction. The resulting matrix of range and azimuth samples are divided among the processors. Focusing along the range lines is done using a Fast Fourier Transform (FFT), performing multiplication operations, and then doing an inverse FFT. The matrix is then transposed and divided and focused along the azimuth lines. Consecutive frames have a great deal of overlap, and this redundancy is taken into account to position the data for the most efficient transpose operation. A specialized transpose is used to accommodate the decomposition scheme. Output may be destined either for display on a graphics workstation or for storage on disk.Platforms
SAR is specifically tuned for the Intel Paragon.Acknowledgements
This application was provided by Thanh Phung of Intel Scalable Systems Division. Trace data was obtained using the Pablo trace capture library.
This research is supported in part by the National Science Foundation.