Research On The Algorithm Of Spectrum Analysis Based On GPU

Considering the requirement of spectrum analysis,we designed the parallel spectrum analysis algorithm.The software section is combined with the GPU processor,making it possible that the data required massive calculation in the RF collector is transferred to the GPU to process,which has the ability of powerful parallel computing.The key part of this paper is the parallel optimization of mixed-radix FFT algorithm and STFT algorithm,in which the data memory and memory allocation are also improved.Results show that the simulation via CUDA programming is superior than CPU processing.The main work of this paper is as follows:1?The test environment is configured by building a heterogeneous software platform CPU + GPU to process data from the software radio platform on the spectrum analyzer.The spectrum analyzer finishes the work that the original signals received from the RF front-end are transferred to baseband signals by the basic processing,and the software section is combined with GPU processor to make the foundation of the optimization of the frequency spectrum analysis algorithm.2?Spectrum Analysis Algorithm Optimization Program.The existing algorithm is optimized,choosing the classical algorithm Cooley-Tukey FFT algorithm as the theoretical basis.First of all,analyzing serial characteristics of the FFT algorithm to find the part used for parallel computation.And then,specific analysis and discussion on different number of points is given,and it is found that the smaller number of points is divided,the higher calculation efficiency the algorithm has.At last,a parallel algorithm scheme based on mixed-based FFT is proposed.The non-stationary signal parallel algorithm is optimized,mainly discussing the optimization of the short-time Fourier transform-STFT algorithm.The STFT algorithm is so convenient and quick to solve the non-stationary signal that the optimization based on the mixed-base FFT algorithm is desired,in which the mixed radix FFT is used as the STFT algorithmThe peak search algorithm is discussed and studied by GPU.The purpose of the peak search algorithm is to find out which peaks have a large peak after the spectral conversion,and this operation does not change the value of each point.The experiment not only completes the peak search algorithm parallel processing but also optimizes the algorithm on the number of threads,shared memory access optimization,follow-up cycle and other aspects,improving the operation efficiency finally.3 ? CUDA Programming Optimization Program.Based on the Pascal architecture-based GTX1070,on the one hand,parallel optimization is firstly performed at the global memory,shared memory and register level to improve the operation efficiency of the algorithm.On the other hand,optimization analysis has been conducted on the efficiency of visits,the number of thread blocks and on-chip storage.After the above content analysis,the results of processing data using CPU,CPU + GPU,and CUFFT(FFT library on CUDA)are compared and analyzed.