采用DSP实现的神经网络实时仿真系统

Real-Time Simulational System Using Neural Network Based on DSP

采用 DSP设计完成了神经网络实时仿真系统。从神经网络协处理器的硬件结构、协处理器中的神经网络算法 ,神经网络协处理器及其宿主机间数值交换等方面系统地描述了该仿真系统。在神经网络在线算法的实现中 ,利用 DSP能与其宿主机实现并行工作的特点 ,采用 DSP和计算机并行工作 :在宿主机中实现 BP网络对受控对象的辨识 ,在神经网络协处理器中完成模糊神经网络在线控制算法 ,从而加快了神经网络运算速度 ,使其能达到在线控制的要求。文末给出了一个直升机总矩通道在本仿真系统的仿真实例 。

The simulational system using neural network based on DSP is designed and realized. This paper presents a real time simulational system using neural network based on DSP. The hardware structure of neural network (NN) co processor, the algorithm used in the neural network co processor and the communication of NN co processor with computer are discussed in detail. In the implementation of the online neural network algorithm the host and DSP are worked in parallel and thus the operation speed of neural network is accelerated. The requirement of online control is naturally satisfied. The practicability of this simulational platform is verified by simulation of a helicopter.