基于线性神经网络的余弦调制滤波器组设计  

Efficient Design of M-channel Cosine-modulated Filter Banks Using the Structure of Linear Neural Networks

作　　者：徐微[1,2] 李怡 缪竟鸿 赵加祥[3] Xu Wei;Li Yi;Miao Jinghong;Zhao Jiaxiang(School of Electronics and Information Engineering,Tianjin Polytechnic University,Tianjin 300387,China;Tianjin Key Laboratory of Optoelectronic Detection Technology and System,Tianjin 300387,China;College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300071,China)

机构地区：[1]天津工业大学,电子与信息工程学院,天津300387 [2]天津市光电检测技术与系统重点实验室,天津300387 [3]南开大学,电子信息与光学工程学院,天津300071

出　　处：《南开大学学报（自然科学版）》2020年第4期52-56,共5页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：国家自然科学基金(61501324);天津市自然科学基金(14JCYBJC16100)。

摘　　要：通过拟合满足准确重建条件的理想频率响应训练网络,提取训练完成的网络权值和阈值,构造原型滤波器,经过余弦调制得到满足要求的余弦调制滤波器组.仿真表明原型滤波器阶数较少,近似准确重建余弦调制滤波器组.High-order of cosine-modulated filter bank results in excessive adder and multiplier,causes macrooperation and high power dissipation for cosine-modulated filter bank.Solution for this problem is using the powerful fitting ability of linear neural networks.Firstly,analyze relationships between the in-out equation of linear neural networks and the frequency response equation of type I linear-phase FIR filter.Secondly,train the network by fitting the ideal frequency response meets the perfect reconstruction conditions.Then a prototype filter designed by the weights and thresholds extracted from trained network.Finally,a cosine-modulated filter bank meeting the requirements obtained with cosine modulation.The proposed algorithm is simulated and compared with others.Simulation and comparison results show that the proposed method can design a nearly perfect reconstruction cosine-modulated filter bank with very few orders.

关 键 词：余弦调制滤波器组 有限长单位冲激响应 线性神经网络 

分 类 号：TN713[电子电信—电路与系统]