硬件实现基于BP神经网络设计的带阻FIR滤波器

Implementation of band-block FIR filters designed based on BP neural network

针对带阻FIR滤波器传统设计方法的不足和缺陷,在改进BP神经网络设计带阻FIR滤波器方法的基础上,提出了BP神经网络设计FIR滤波器的硬件实现方法.通过神经网络训练使得实际滤波器的幅度响应逼近理想滤波器的幅度响应,训练得到带阻滤波器系数,再利用数字信号处理芯片(DSP),设计适当的FIR滤波器结构以及硬件结构,从硬件上实现带阻FIR滤波器.试验结果显示,用该方法设计的带阻FIR滤波器克服了传统方法的主要缺陷,具有良好的幅频特性及衰耗特性,并且边界频率控制精确.

A new method of implementing the band-block FIR filters was presented to solve the main problems of the conventional methods, based on the improved method of band-block FIR filter design using BP neural network. In the first step, the coefficient of the filters was obtained through the training of the BP neural network. Then based on digital signal process CMOS chip, the architectures of both the block FIR filters and the hardware were designed to implement the band-block finite impulse response （FIR） hatters. The experimental results showed that it overcame the main defects of the conventional methods, and that the filter had good amplitude, frequency and attenuation performance, and that the edge frequency could be set exactly.