基于余弦基神经网络的FIR滤波器设计

Cosine Basics Neural Networks Based FIR Filter Design

文章叙述了FIR滤波器设计的不同方法,包括窗函数和频率采样,详细研究了FIR线性相位滤波器的幅频特性与余弦基神经网络算法的关系,给出了神经网络的训练算法,并应用该方法构造了一个中心频率大于10KHz的带阻滤波器。通过MATLAB仿真实现,与常规的用窗函数法设计的滤波器进行了对比,发现基于神经网络的方法所设计出的滤波器通带阻带无过冲无波动,具有更好的性能。

The article describes the different FIR filter design methods, including the window function and frequency of sampling, a detailed study of linear phase FIR filter with the cosine amplitude-frequency characteristics of neural network algorithm-based relationship, given the training of neural network algorithm, and Application of the method structure is greater than a center frequency of band-stop filter 10KHz. Through the MATLAB simulation, the conventional design using window function of the filter were compared and found that the method based on neural network designed by the stop-band filter passband without overshoot without fluctuations, with better performance.