神经网络在4型FIR高阶多通带滤波器的自适应优化设计研究

提出了一种新的基于神经网络的FIR线性相位数字滤波器的自适应优化设计方法。根据4型FIR滤波器的幅频响应特性，构造出一个相应的神经网络模型，并建立了FIR线性相位数字滤波器的神经网络算法，该算法通过训练神经网络权值，使设计的数字滤波器与希望得到的FIR线性相位滤波器的幅频响应之间的误差平方和最小化，从而获得FIR线性相位数字滤波器的脉冲响应．提出并证明了该算法的收敛定理，给出了FIR高阶多通带滤波器自适应优化设计实例、计算机仿真结果表明，该算法计算精度高，收敛速度快；用该算法设计的高阶多通带滤波器，其幅频响应的阻带衰减很大，而通带波动很小。

Adaptive Optimal Design Study of the Type-Four FIR High-Order Mutli-Band-Pass Filter Based on Neural Networks

Based on neural-network, this paper describes a new adaptive scheme for linear-phase finite-impulse response (FIR) digital filters. According to the amplitude-frequency response characteristic of type-4 FIR filter, the paper presents a corresponding neural-network model, and establishes an algorithm for linear-phrase FIR digital filter. The algorithm, through the training of the weights of neural-networks, can minimize the square errors between the amplitude response of the designed FIR digital filter and the desired FIR digital filter. We can get the impulse response of the designed FIR digital filter. The paper further proposes the convergence theorem of this algorithm, and offers several design schemes of high-order multi-band-pass FIR filter. Computer simulations prove the high precision and high speed of attenuation of this algorithm. The high-order multi-band-pass FIR filter designed with this algorithm is featured by big attenuation in stop-band and small ripple in pass-bands.