针对DSP平台算法移植时遇到的超长点FFT实现和运算效率问题,本文结合TI公司的TMS320C6678的DSP,利用FFT的分解算法和L2内存段高效的访存效率,将DSP内存数据EDMA搬移与FFT分解计算相并行,设计出一种超长点FFT计算并行处理方法,通过262 144点FFT计算描述了该方法的具体实现过程,将DSP计算结果与MATLAB结果比对,统计了该方法的计算误差为10^-4。最后,在DSP平台上,在相同FFT点数时将直接调用FFT库函数计算,与使用该方法的计算时间进行比较,说明了其在超长点FFT计算具有更好的运算性能。

Aiming at the problems of ultra-long points FFT implementation and computational efficiency encountered during the migration of DSP platform algorithms, this article combines TI's TMS320C6678 DSP, we use the FFT decomposition algorithm and the efficient memory access efficiency of the L2 memory segment to design a parallel processing method for ultra-long points FFT calculation by paralleling the EDMA movement of the DSP memory data and the FFT decomposition calculation.The specific implementation process of the method is described by 262144-points FFT calculation, and the calculation result of DSP is compared with the result of MATLAB, and the calculation error of the method is 10^-4.Finally, on the DSP platform, when the number of FFT points is the same, the calculation time of directly calling the FFT library function is compared with the calculation time using this method, which shows that it has better calculation performance in the ultra-long points FFT calculation.