The Research Of Multi-core FFT Algorithm In Radar Signal Processing

In order to meet the growing real-time performance requirement of radarsignal processing, algorithms of radar signal processing need to be applied to multi-core platform. Fast Fourier Transform(FFT) is commonly used for fastimplementing the key computations of radar signal processing, so the research ofmulti-core FFT algorithms has more important meaning for improving theperformance of radar signal processing. The implementation of FFT algorithms willpromote further application for radar signal processing algorithms on multi-coreplatform.This paper mainly studies the theory of multi-core FFT algorithm as well as itsimplementation and application in radar signal processing.In the aspect of theoretical research, the butterfly matrix which plays adecisive role in FFT algorithm was chosen as research object. Firstly, theoriesrelated to sparse factorization of butterfly matrix were studied. Secondly, how tochoose the algorithm that enable to take advantage of the structure of computerproperly was discussed. Finally, extended to universal level, the parallelizationfeasibility of algorithm was researched and analyzed with mathematical methods.In the aspect of algorithm implementation, the multi-core implementation ofradix-2FFT was taken as an example. The parallelization analysis of radix-2FFTalgorithm was done firstly, a feasible parallel scheme was obtained as a result.Secondly, a MEX file was used to implement C++and Matlab mixed programming,solved the problem that multiple threads can’t be created in Matlab. In programdesigning, OpenMP and SSE instructions were adopted to create multiple threadsand implement vector operation separately. Because the storage way of complexarray is different between Matlab and C++, it was rearranged to improve cache hitrate. FSINCOS function was used to eliminate the redundancy of the twiddlefactors computation.In the aspect of application, multi-core FFT algorithm was applied in pulsecompression and doppler filtering of doppler beam sharpening. Compared with thesame application of single thread FFT algorithm, application with multi-core FFThas a better performance.