论文部分内容阅读
在二电平信号下基于多层感知器MLP(MultiLayerPerception)的均衡器(MLPE)性能远远优于传统的线性模向均衡器LTE(LinearTransversalEqualizer).但在多电平调制信号下,MLPE性能迅速下降.其主要原因在于激活函数的选择.文中提出了一种适于多电平信号均衡的神经网络模型——分段多层感知器SMLP(SegmentMultiLayerPerception),并给出其算法.模拟结果表明,基于SMLP的均衡器,比LTE和MLP收敛速度更快,最小均方误差MMSE(MinimumMeanSquareEror)也小得多,而计算复杂度则与MLP相同.
The equalizer (MLPE) performance based on Multilayer Percepitplexer MLP (MultiLayerPerception) is far superior to that of the traditional linear mode equalizer LTE (LinearTransversalEqualizer) under two-level signals. However, under the multi-level modulation signal, MLPE performance drops rapidly. The main reason is the activation function of choice. In this paper, a neural network model suitable for multi-level signal equalization (SMLP) is proposed and its algorithm is given. Simulation results show that SMLP-based equalizers have faster convergence than LTE and MLP, MMSE (Minimum Mean Square Error) is much smaller, and computational complexity is the same as that of MLP.