Enhancing Energy Efficiency of D-MIMO Networks: Scalable Clustering and Deep Learning-Based Power Control

Distributed multiple-input and multiple-output (D-MIMO) technology is a promising candidate to be integrated in beyond fifth generation networks offering uniform quality of service along the network coverage and higher overall system capacity. It leverages the cooperation of many antenna arrays spread over the coverage area that jointly and coherently serve user equipment devices. While the spectrum efficiency benefits of D-MIMO (sometimes also referred to as cell-free technology) are well documented, the corresponding energy efficiency (EE) of such networks has received more attention in recent years as concerns about sustainability become central in future 6 G systems. In this context, this paper proposes and investigates the combined effect of intelligent access point clustering and power control to enhance D-MIMO's EE. While many works on D-MIMO assume that the whole network serves every user, we consider scalability and massive MIMO constraints to address the practical issues of a fully connected network in terms of high computational complexity, elevated signaling bandwidth, and limitations of MIMO's spatial degrees of freedom. To this end, we propose a resource-aware graph-based clustering method combined with a deep-learning-based power control. Comprehensive computer simulations demonstrate that the proposed strategy significantly enhances the network's EE, while producing comparable spectral efficiency performance to a fully connected scenario, while also outperforming a state-of-the-art existing clustering approach.