Chapter Summary

Chapter Summary

Key Points

• 1.

  The scalability bottleneck. The original cell-free massive MIMO formulation requires every AP to process every user, leading to $O(MK)$ computational and fronthaul cost that grows without bound as the network densifies.

• 2.

  User-centric clustering. Each user $k$ is served by a dynamically selected cluster $\mathcal{M}_k$ of nearby APs, chosen based on large-scale fading coefficients. The cluster indicator matrix $\mathbf{D}$ is the key design variable, and its sparsity ensures scalability: $O(M L_{\max})$ complexity with bounded per-AP load.

• 3.

  Negligible SINR loss. Restricting service to the $N_{\text{cl}}$ strongest APs typically costs only 1–3 dB in signal power while reducing computation and fronthaul by factors of 20–100$\times$. Path loss decay makes distant APs essentially useless.

• 4.

  Fog Massive MIMO. The architecture proposed by Bursalioglu, Caire, and collaborators combines user-centric clustering with the master AP concept. The master AP for each user handles pilot assignment, power control, and data routing — distributing the CPU's coordination role to the network edge.

• 5.

  Pilot assignment exploits cluster structure. In user-centric systems, users with non-overlapping serving clusters can safely share pilots. The interference graph is sparse, and graph-coloring-based pilot assignment reduces the required number of orthogonal pilots from $O(K)$ to $O(\chi(G)) \ll K$, significantly reducing pilot overhead.

• 6.

  Multiple cooperation levels. The Fog architecture supports Levels 1–4 of AP cooperation, from non-coherent combining to cluster-wide MMSE. The choice of level trades performance for fronthaul and computational cost.