Analysis of Levenberg-Marquardt and Scaled Conjugate gradient training algorithms for artificial neural network based LS and MMSE estimated channel equalizers

Multilayer perceptron (MLP) based artificial neural network (ANN) equalizers, deploying back propagation (BP) training algorithm, have been profusely used for equalization earlier. However this algorithm suffers from slow convergence rate, depending on the size of network. In this paper, Levenberg-Marquardt and Scaled Conjugate algorithms are proposed to train an MLP based ANN for least square (LS) and minimum mean square (MMSE) estimated channel coefficients using MPSK and MQAM modulation techniques. The key analytical performance measures are comprehended in terms of three parameters i.e regression, validation and training state. Based on the regression parameter, Scaled Conjugate method outpaces Levenberg-Marquardt and on the basis of Mean Squared Error (MSE), it is seen that the Levenberg-Marquardt has better accuracy than Scaled Conjugate.