Blind adaptive equalization techniques using the complex multi-layer perceptron

Abstract

Among the useful blind equalization algorithms, there are stochastic-gradient iterative equalization algorithms. Because these algorithms basically use a linear FIR filter except to adopt a memoryless nonlinear function of the equalizer output to generate the desired response, only linear channel distortion can be corrected well. To overcome nonlinear channel distortion two new blind equalization schemes using the structure of the complex multi-layer perceptron that can deal with signals of any constellation size are proposed. One is a modification of the constant modulus algorithm (CMA), the most widely used blind algorithm. The other is a modification of the radius-directed equalization (RDE) algorithm with multiple constant modulus that is more suitable for the node activation function having a multi-saturated output region. The experimental results show that the proposed schemes have very low MSE in the steady state and at the same time can recover the arbitrary phase rotation due to the channel when the major channel distortion is both linear and nonlinear.