Low Complexity Neural Network Based Digital Predistortion for Memory Power Amplifiers.

Digital Predistortion (DPD) is an effective technique for Power Amplifier (PA) non-linear distortion and memory effects compensation. Different topoligies of DPD are presented in the literature. In this paper, we propose a mimetic neural network based DPD for Hammerstein power amplifier for OFDM signal with a reduction of Peak to Average Power Ration (PAPR) by Selective Mapping (SLM) method. This proposed model is compared with Real Valued Multilayer Perceptron (R-MLP). Simulation results show that the mimetic-R-MLP manifests more efficiency for PA linearization and for memory effect reduction in terms of Error Vector Magnitude (EVM) by a gain of 2 dB. It outperforms the R-MLP in terms of Mean Squared Error (MSE) for the convergence of the Neural Network (NN) and its complexity is 23 % lower. The results in terms of Power Spectral Density (DSP) show also that our model compensates efficiently the out of band distortion (OOB) of the PA.