线性修正下的神经网络盲均衡算法

研究了一种利用横向滤波器对神经网络进行线性修正的盲均衡算法。在神经网络的输入层之前加入一个横向滤波器，以横向滤波器的节点输出作为神经网络盲均衡器的输入，利用常数模代价函数分别得到横向滤波器和神经网络盲均衡器的瞬时输出误差，将瞬时误差加权处理作为调节误差分别对横向滤波器和神经网络盲均衡器的权系数进行调节，算法实现了对非凸性误差曲面进行线性和非线性寻优的组合。计算机仿真证明提出的算法有效提高了神经网络盲均衡算法的收敛速度，降低了稳态剩余误差，具有更好的实用性和均衡性能。

Blind equalization based on neural network by linear correction

Blind equalization based on the neural network with a linear correction is proposed in this paper. A lateral filter adds before the input layer of neural network, then, the output signal from the node of the lateral filter is taken as the input signal of neural network. The instantaneous output error of the lateral filter and neural network blind equalization, which can be obtained by the constant modulus cost function, is used for adjusting error to update the weight coefficients of the lateral filter and neural network. This algorithm carries out the combination of linear and nonlinear optimization on non-convexity error surface. Simulation results show that the method of blind equalization in this paper provides higher convergence rate and better performance.