Machine Learning based Adaptive Predistorter for High Power Amplifier Linearization

In this paper, we have developed a machine learning based adaptive predistorter for high power amplifier linearization approach under dynamically changing environment. In the “bent-pipe” transponder in satellite communication (SATCOM) system, the High Power Amplifiers (HPAs), which are similar to other amplifiers in the communication system, can cause nonlinear distortions to transmitted signals, deteriorating the system transmission performance. The traditional model based processing techniques such as the Extended Saleh's Model (ESM) based predistortion design can be applied to maximize transponder throughput along with HPA power efficiency but sensitive to dynamically changing environment. In this paper, the compensated HPA linearity characterized through Amplitude Modulation-to-Amplitude Modulation (AM-AM) and Amplitude Modulation-to-Phase Modulation (AM-PM) effects is used as the system reward, we leveraged reinforcement learning approach to dynamically optimize the parameter set for the ESM based PD to improve system performance in various environmental conditions. Finally, simulation results are provided to evaluate and verify Bit Error Rate (BER) improvement for the considered SATCOM system by applying our proposed PD technique.