一种新的复值递归神经网络训练方法及其应用

一种新的基于数字滤波器理论的全互连复值递归神经网络训练方法被提出.每个递归神经元均具有复数ⅡR滤波器结构.通过优化ⅡR滤波器的系数来更新神经网络的权值,而优化过程则采用逐层优化(LBLO)技术和递归最小平方(RLS)方法.该算法的性能通过将其应用于复信道均衡来加以说明.计算机仿真结果表明,该算法具有较快的收敛速度.这为快速训练复值递归神经网络提供了一条新的途径.

A Novel Complex-Valued Recurrent Neural Network Training Method with Applications

A new training approach for the training algorithm of a fully connected recurrent neural network based on the digital filter theory is proposed. Each recurrent neuron is modeled by an IIR filter. The weights in the network are updated by optimizing IlR filter coefficients and optimization is based on the layer-by-layer optimizing procedure (LBLO) and the recurrent least squares (RLS) method. The performance of the proposed algorithm is demonstrated with application in complex communication channel equalization. Computer simulation results indicated that the proposed mathod provides fast convergence rate. This provides a new way to the fast training of complex valued recurrent neural network.