Research Of Acceleration For Critical Algorithms Of SAR Real-time Signal Processing

Synthetic Aperture Radar(SAR) imaging has the capacity of all-weather, all-day, long distance and high-resolution, and has wide application in military and civil field. SAR real-time signal processing is characterized by large amount of computation and high calculation complexity. So the research of acceleration for SAR real-time signal processing is of great importance.The real time processing of the SAR signal can be divided into three independent stages, namely preprocessing, imaging and imagepost-processing. This paper has a deep research on the critical algorithms in SAR real-time signal processing, such as FIR filtering, Fast Fourier Transform and image classification, and proposes optimization solutions respectively. The main work of this paper can be summarized as follows.Properly reducing the sampling rate of the echo signal, can improve the real-time performance of SAR imaging. In order to avoid the drop spectrum aliasing, that may appeared in the process of sampling, the echo signal first needs to pass a linear phase FIR low-pass filter. This paper presents a hardware design scheme based on FPGA to realize linear phase FIR filter.Fast Fourier Transform(FFT) plays a very important role in radar imaging algorithms. How fast to accomplish the large amount-float FFT directly determines the performance of SAR real-time imaging system. The paper presents a design of a double-precision floating-point FFT processor based on FPGA. By optimizing the memory mapping algorithm of the operands and the twiddle factors, the FFT processor can calculate eight radix-2 butterflies in one clock cycle in parallel. The experiment results show that comparing to the FFTW program on the platform of Intel(R) Core(TM)2 Quad CPU, our FFT processor can achieve about 2X speedup.With SAR imaging technology developing rapidly, the amount of data that need to be processed by the SAR real-time imaging system fold increase. Because of the limited computing power of single FPGA, it’s difficult to accomplish super large-scale FFT. The paper proposes a big point one-dimensional FFT design based on FPGA array. The big points FFT is decomposed into multiple small points FFT, and be processed by multiple FPGAs respectively. The paper presents a data exchange strategy between FPGAs. Finally, we carry on the performance analysis on our design.The goal of radar imaging is to object classification and recognization. Traditional SAR image classification algorithms can hardly identify the targets which scattering close signal. The paper put forward a new SAR imagine classification algorithm based on Deep Belief Network. To improve the the real-time performance, we make use of the powerful parallel processing capacity of GPU to accelerate the SAR image classification algorithm.