机构地区: 华南理工大学电子与信息学院
出 处: 《数据采集与处理》 2005年第1期54-58,共5页
摘 要: 给出了一个基于自然梯度的后非线性多信道盲解卷算法。混合系统由线性卷积混合后接一个可逆非线性失真组成。分离系统由多层感知器 (非线性部分 )后接一个线性盲解卷过程 (线性部分 )组成。分离系统的线性部分和非线性部分参数学习都采用自然梯度算法。仿真结果显示 ,自然梯度算法比传统梯度算法收敛速度更快 。 A natural gradient based algorithm for multichannel blind deconvolution of post-nonlinear mixtures is proposed. The mixing system is composed of a linear convolution followed by an invertible nonlinear distortion. The separating system consists of a multilayer perceptron (nonlinear part) followed by a linear blind deconvolution (linear part). The natural gradient method is applied for parameter learning of the linear and nonlinear parts of the separating system. Simulation results show that the natural gradient approach has faster convergence speed and better separation performance than the conventional gradient based algorithm.