基于神经网络方法的可变分数延迟FIR滤波器设计

A Neural Network-based Design for Variable Fractional Delay FIR Filters

本文提出了设计可变分数延迟FIR滤波器的一种神经网络优化方法。首先建立一个连续Hopfield神经网络模型,通过选择网络的Lyapunov能量函数与优化目标问题的最小均方误差函数相一致的关系得到系统的最优解。仿真结果表明,在计算代价略有增加的情况下,其性能优于加权最小二乘方法。

This paper proposed a neural network method to design variable-fractional-delay finite-impulse response filters. This design is achieved by choosing a Continuous Hopfield Neural Network （CHNN） model and establishing the relation between the MSE criterion and the Lyapunov energy Function. In an illustrative design example, the proposed method has improved performance of the filters, with slightly increased computation cost, as compared with improved weighted least-squares design method.