Performance Comparison of Neural Network based Equalizers for Wireless Communication Channels

In this paper, artificial neural network (ANN) architectures are used to develop equalization models which are the basic requirements to design wireless communication receivers. Because wireless channels are affected by noise and multipath fading and the transmission is affected by a critical problem of inter symbol interference(ISI). Recurrent Neural Network (RNN) and Functional Link ANN(FLANN) using different functional expansions are used to design adaptive equalizers which can minimize the effect of ISI and improve BER performance of the receiver. The neural network based equalization models are trained using LMS algorithm. Other orthogonal polynomials such as Legendre and Hermite polynomials are used for expanding the input patterns in case of FLANN based equalizer . RNN equalizer uses feedback mechanism and the order of feedback filter depends on the number of delayed input samples. The performances of the equalizers are compared by evaluating bit error rate(BER), mean square error(MSE) and constellation diagram using MATLAB simulations. From the analyzed results, it is quite apparent that equalizer using RNN model consistently outperform FLANN based equalizers in terms of BER when L-ary PSK data is passed through a noisy nonlinear channel.