Recurrent Neural Network Based BER Prediction for OFDMA Channel

The prediction of Bit Error Rate (BER) in OFDMA Channel (IEEE 802.16e Mobile WirlessMAN) network is investigated here. The state of the channel is estimated on symbol by symbol basis on a realistic fading environment. The state of a channel is modeled as nonlinear and temporal system. Neural network method is the best system to predict and analyze the behaviors of such nonlinear and temporal system. In this context, BER prediction by k symbol ahead is investigated by two different recurrent neural network architectures such as Recurrent Radial Basis Function Network(RRBFN) and Echo State Network (ESN). The Prediction accuracy RRBFN and ESN is in the range of 92.92 % to 96.69% and 94.31% to 96.75% respectively.