Reinforcement Learning Empowered Digital Pre-Distortion in Non-Differentiable Channel

This work proposes a novel digital pre-distortion technique based on deep reinforcement learning to mitigate instantaneous nonlinearities in communication systems stemming from nonideal electrical components. Comparative simulation results demonstrate its significant performance gains over conventional methods, including Volterra series and look-up tables, in terms of bit error rate reduction and effective SNR enhancement.