Optimization of Delta-Sigma Modulator Based on Reinforcement Learning for Mobile Fronthaul

Abstract

Delta-sigma modulator (DSM) offers a high-fidelity solution for mobile fronthaul. It achieves a high quantization signal-to-noise ratio through oversampling and noise shaping. However, the DSM is a nonlinear system that lacks of comprehensive design method for the loop filter. The traditional loop filter design methods established on linear filter systems have suboptimal performance. In this work, we propose a reinforcement learning (RL)-based method to optimize the design of the loop filter, which utilizes the historical information. The RL method achieves a 14-dB and a >2.4-dB SNR gain compared to traditional methods and polynomial fitting-based approach, respectively.