基于动量项前馈神经网络盲均衡算法

The Blind Equalization Algorithm Based on Momentum Feed-Forward Neural Network

针对基于前馈神经网络的盲均衡算法中,BP优化算法具有收敛速度慢、易陷入局部极小的缺点,提出了一种新的盲均衡算法,该算法结合动量项前馈神经网络与传统恒模盲均衡算法的优点,将以前权值的调节量用于当前权值的修改过程,降低了算法对于误差曲面局部极值点的敏感性。仿真结果表明,该算法可有效抑制网络陷入局部极小,防止振荡,加快盲均衡器的收敛速度。

For the disadvantages of blind networks, such as slow convergence speed, equalization algorithms based on feed-forward neural etc. , this paper proposes a new algorithm which combined the advantages of the momentum feed-forward neural networks and the traditional CMA blind equalization algorithms, which adjusts the new weight value with the adjusting value used before so that the algorithm could be less sensitive to the stationary point of the error surface. The simulation results showed that the new algorithm could control the local minimum, avoid the oscillation, and quicken the convergence speed of blind equalization procession.