Hybrid centralized-distributed precoding in fronthaul-constrained CF-mMIMO systems

TL;DR

This work tackles fronthaul-limited CF-mMIMO by introducing a hybrid centralized-distributed precoding scheme that splits users into $\mathcal{K}_{\mathrm{c}}$ (global CSI, centralized ZF) and $\mathcal{K}_{\mathrm{d}}$ (local CSI, distributed ZF). It develops a tractable optimization framework that integrates K-means-based user grouping with successive convex approximation-based power allocation to maximize sum SE under fronthaul and per-AP power constraints, providing explicit fronthaul budgeting for data and precoding vectors. Numerical results show the hybrid approach consistently outperforms fully centralized or fully distributed schemes across a range of $\mathrm{FH}_{\mathrm{max}}$, antenna counts, and network scales, with notable gains from the joint grouping and power optimization. The proposed method offers a scalable and flexible solution for future wireless networks (e.g., O-RAN) where fronthaul resources are heterogeneous and dynamic.

Abstract

We investigate a fronthaul-limited cell-free massive multiple-input multiple-output (CF-mMIMO) system and propose a hybrid centralized-distributed precoding strategy that dynamically adapts to varying fronthaul and spectral efficiency (SE) requirements. The proposed approach divides users into two groups: one served by centralized precoding and the other by distributed precoding. We formulate a novel optimization problem for user grouping and power control aimed at maximizing the sum SE, subject to fronthaul and per-access point (AP) power constraints. To tackle the problem, we transform it into a tractable form and propose efficient solution algorithms. Numerical results confirm the hybrid scheme's versatility and superior performance, consistently outperforming fully centralized and distributed approaches across diverse system configurations.

Figures (3)

• Figure 1: Average sum SE versus $\mathrm{FH_{max}}$ ($M=20$, $L=14$).
• Figure 2: Average sum SE versus $L$ ($\mathrm{FH_{max}}=8$ Gbps).
• Figure 3: Average sum SE versus $\mathrm{FH_{max}}$ ($M\!=\!20,L\!=\!21$, $M_\mathrm{{order}}=32$).