改进的扩展互信息分离算法

摘要

扩展互信息分离算法采用单隐层神经网络近似算法代价函数中的非线性函数，可调节的参数有限，需要多次迭代才能收敛，从而导致收敛速度较慢。针对这一问题，采用双隐层神经网络近似非线性函数，以分离结果的互信息最小化作为代价函数，采用梯度下降方法对代价函数进行优化，增加了可调节参数数量。仿真实验结果表明，改进后的算法相对原算法收敛速度更快，误差更小。%Extended mutual information separation(EMISEP)algorithm uses a single hidden layer neural network to approximate nonlinear function of cost function,so the adjustable parameter is limited and it needs more itera-tion times to converge,which leads to relatively slow convergence speed. To overcome this problem,this paper uses double hidden layer perceptions to approximate nonlinear function of cost function,and uses mutual infor-mation minimum of separation signals as cost function,which is optimized by gradient descent method. This in-creases the number of adjustable parameters. The simulation results prove that the improved algorithm has faster convergence speed and smaller error comparing with the original algorithm.