Deep Reinforcement Learning-Based Signal Processing for Cell-Free Massive MIMO Networks in Coal Mine Power Grids

In coal mine power grids, ensuring reliable, high-capacity, and low-latency communication is critical to maintaining efficient operations.To meet these demands, combining nonorthogonal multiple access (NOMA) with cell-free massive multiple input multiple output (CF-mMIMO) networks presents a powerful solution.This paper focuses on the signal processing and resource allocation challenges inherent in the downlink of CF-mMIMO-NOMA systems, specifically tailored to the complex communication environment of coal mines.We propose a hierarchical deep deterministic policy gradient (H-DDPG) algorithm to optimize system performance, with a focus on user pairing and power allocation.The algorithm addresses signal processing tasks at both system and link levels by leveraging two-layer control networks that operate on distinct time scales.At the system level, the user pairing problem is solved, while power allocation is optimized at the link level, with dedicated DDPG agents guiding both processes.Extensive simulations demonstrate that the proposed H-DDPG method significantly enhances the system sum rate compared to benchmark approaches, making it a robust solution for improving signal processing and resource management in coal mine power grid CF-mMIMO networks.