基于卡尔曼滤波的极限学习机在线盲均衡算法

Online blind equalization algorithm using extreme learning machine based on Kalman filter

针对正交振幅调制(quadrature amplitude modulation,QAM)信号,在预测方法的盲均衡框架下,基于卡尔曼滤波(Kalman filter,KF)提出了一种新的神经网络在线盲均衡算法。采用复数型极限学习机(complex extreme learning machine,C-ELM)作为非线性预测滤波器(prediction filter,PF),用KF实时更新C-ELM的输出权值以使预测误差达到最小,再通过自动增益装置调整信号的幅度变化,最后引入相位调整因子纠正信号的相位旋转。仿真结果表明,所提算法实现了良好的实时均衡效果,具有较快的收敛速率和较小的稳态均方误差,不仅适用于方形,同时也适用于十字形QAM信号的盲均衡。

For quadrature amplitude modulation(QAM)signals,a new neural network online blind equalization algorithm based on the Kalman filter(KF)is proposed under the blind equalization framework of the prediction method.For the purpose of minimizing the prediction error,this paper adopts the complex extreme learning machine(C-ELM)as the nonlinear prediction filter(PF)and sequentially updates the output weights of C-ELM using the KF.The amplitude of the signal is then adjusted by an automatic gain control device,and finally the phase rotation problem is corrected through a phase tuning factor.Simulation results show that the proposed algorithm achieves a good real-time equalization effect and has a faster convergence speed and lower steady mean square error.Moreover,the algorithm is suitable for blind equalization of the square,as well as the cross QAM signals.