一种适于多电平调制信号均衡的神经网络

A neural network model suited for multi level modulated signal equalization

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 multilayer perceptron (MLP) based equalized (MLPE) functions better than the comentional linear transversal equalizer (LTE) in case of bipolar valued data. But when dealing with multilevel modulated signals (such as 4 PAM or 16 QAM), the performance of the MLPE deteriorates rapidly. In this paper, a new kind of neural network model, named sgemented MLP (SMLP), is presented, which is suited for multilevel signal equalization. The algorithm of the SMLP is also given. The experimental results show that the SMLP based equalizer has a faster convergence speed and a smaller minimum mean square error than the LTE and the MLPE if the computational complexity is the same as that of MLP.