基于DSP的神经网络信号处理机的研究

Study of Neurocomputer Based on DSP for Signal Processing

人工神经网络实现技术的研究具有重大意义，它可推动神经网络研究的进一步深化并加快其实用化步伐。本文重点研究了基于ＤＳＰ（数字信号处理）的神经网络虚拟实现技术，设计并实现了以ＴＭＳ３２０Ｃ３０作为处理器的并行处理系统，峰值速度为１亿ＦＬＯＰＳ，并行效率可达到９０％（ＢＰ回忆阶段）。该系统可以方便地扩展为更大规模的系统，用于水声信号处理，以满足实际应用的要求。

Implementation of Artificial Neural Network(ANN)is very important to theoretical study andapplications of ANN. Based on studying existing niethods,the paper concentrates on the DSP-based virtualimplementation of ANN. A parallel processing system based on TMS 320 C30 is designed and configured.This systein can provide a peak speed as high as 100M FLOPS and the parallel efficiency is 9 0%(duringthe forward phase of BP).The expansibility of the system into large-scale systems is also shown.