Chapter Summary
Chapter 13 Summary — Distributed Processing
Key Points
- 1.
Centralized vs. distributed processing defines the fundamental tradeoff in cell-free massive MIMO: centralized MMSE (Level 4) maximizes SINR by jointly processing all received samples, while distributed processing (Levels 1–3) reduces the fronthaul from complex vectors to complex scalars per channel use, at the cost of suboptimal interference suppression.
- 2.
Four cooperation levels span the continuum: Level 1 (local MRC) is simplest but ignores inter-user interference; Level 2 (local MMSE) suppresses intra-AP interference; Level 3 (LSFD) optimizes the CPU combining weights using large-scale statistics; Level 4 (centralized MMSE) is optimal but requires full signal forwarding.
- 3.
LSFD is the sweet spot. The optimal LSFD weights solve a generalized Rayleigh quotient and depend only on large-scale fading coefficients. Level 3 closes 70–90% of the gap between Level 2 and Level 4 at negligible additional fronthaul cost.
- 4.
Energy efficiency has an optimal user load. The Göttsch/Caire analysis shows that the network EE is maximized at a moderate user load – users per AP, balancing the underutilization cost (circuit power wasted on idle APs) against the interference cost (rate degradation from overloading). AP switching saves 30–50% power during low-traffic periods.
- 5.
Fronthaul quantization with 4–6 bits per dimension is sufficient for near-lossless distributed processing. The SINR loss is less than 0.5 dB, which is far smaller than the inter-level gaps. The fronthaul perspective reveals that Level 3 requires only 15–25% of Level 4's fronthaul rate while delivering most of its performance.
- 6.
The choice of cooperation level is a system-design decision, not a theoretical one. It depends on fronthaul technology (fiber vs. wireless), AP antenna count (), user density, and energy budget. Most practical deployments will operate at Level 2–3.
Looking Ahead
Chapter 14 extends this analysis to fronthaul-aware design, where the cooperation level and power control are jointly optimized subject to per-AP fronthaul constraints. We will also examine the uplink–downlink duality in cell-free networks and the O-RAN architecture that maps the cooperation levels to concrete functional splits.