The integration of Distributed Access Points (APs) in cell-free massive MIMO networks poses challenges for efficient routing and parent selection, particularly in scenarios involving dense AP deployment. This paper introduces a reinforcement learning-based approach for optimal parent selection in Destination Oriented Directed Acyclic Graph (DODAG) based routing within cell-free massive MIMO networks. Unlike traditional routing protocols that rely heavily on static or computationally intensive methods, our framework enables each AP to act as an autonomous agent, dynamically selecting parent nodes based on throughput, congestion, hop count, and link stability. We model the problem using Q-learning, where each AP learns an optimal parent selection policy through interactions with the network. Simulation results demonstrate the proposed scheme's ability to maintain high throughput and packet delivery ratios even in dense network environments. The framework achieves an average throughput of 14.9 Mbps and a packet delivery ratio of 79% for a 50-node network, effectively addressing the challenges of interference and congestion. This work highlights the potential of reinforcement learning to enhance the scalability and performance of cell-free networks, paving the way for their application in future 6G communication systems.
Reinforcement Learning-Based Backhaul Routing for APs in Cell-Free Massive MIMO Networks
2025-02-18
486723 byte
Conference paper
Electronic Resource
English