基于并行神经网络的二维FIR数字滤波器设计

Design of 2-D FIR Digital Filters Using Parallel Neural Networks Algorithm

通过对二维FIR线性相位滤波器的幅频响应特性的分析,提出了一种用并行神经网络算法来设计二维FIR线性相位数字滤波器的新方法,其主要思想是使幅频响应误差函数最小化.该方法避免了矩阵的求逆运算,而且因为采用了并行算法,能快速获得滤波器系数.给出了二维FIR圆对称线性相位低通数字滤波器优化设计实例.计算机仿真结果表明由该方法设计的二维数字滤波器,通带和阻带范围波动小,所需计算量非常少,稳定性强.

A new approach for the design of two-dimensional （2-D） finite-impulse response （FIR） linearphase digital filters is presented based on a parallel neural networks algorithm （PNNA） by analyzing the characteristics of 2-D FIR linear-phase filters, its main idea is to minimize the amplitude-frequency response error function. The method avoids matrix inversion, and makes a very fast calculation of the filter＇s coefficients possible because of the utilization of parallel algorithm. Optimal design example of 2-D FIR circularly-symmetric linearphase lowpass filters are given, and the results show that the ripple in passband and stopband is much small, and the PNNA-based method is of strong stability and requires considerably little amount of computations.