Joint User-Panel Grouping in Cell-free Massive MIMO Systems Using Spectral Co-Clustering of Bipartite Graphs

In this paper, we consider a practical architecture for cell-free massive MIMO systems based on multi-antenna panels distributed over the coverage area and connected to a central processing unit (CPU) via optical fiber. We address the problem of scalability by grouping jointly multi-antenna panels and users into disjoint user-panel clusters. The clustering is based on the minimization of the total strength of intercluster channels. We show that this problem is equivalent to partitioning a bipartite graph and solve the optimization problem via spectral co-clustering, which is a well-known sub-optimal method for graph partitioning. Numerical results confirm the significant improvement in spectral efficiency offered by the proposed approach compared to a centralized massive MIMO system and the effectiveness of spectral co-clustering.