Neural Network-Based Mitigation of Nonlinear Inband Distortion in Coded OFDM System

Orthogonal frequency-division multiplexing (OFDM) technique suffers from the high peak-to-average power ratio (PAPR) drawback due to its multicarrier modulation nature. Clipping techniques can reduce the PAPR. However, they generate a clipping noise that distorts the in-band spectrum of OFDM and increases the error vector magnitude (EVM). Considering that nonlinear distortion impacts some subcarriers more than others, this work proposes an approach in which an artificial neural network (ANN) is trained to predict the highly distorted subcarriers in the clipped OFDM signal. Information in the frequency-domain is first encoded by Reed Solomon (RS) encoder, and then an ANN is used to forecast the indices of severely distorted subcarriers and punctures their corresponding symbols in RS codewords before OFDM multiplexing. Simulation results demonstrate the accuracy of our method in deactivating the erroneous subcarriers in OFDM with nonlinear noise and improving the bit error rate (BER) performance of the system while reducing complexity.