Hybrid Centralized-Distributed Precoding in Fronthaul-Constrained CF-mMIMO Systems

Basic Information

• Paper ID: 2510.12406
• Title: Hybrid Centralized-Distributed Precoding in Fronthaul-Constrained CF-mMIMO Systems
• Authors: Zahra Mobini, Hien Quoc Ngo, Ardavan Rahimian, Anvar Tukmanov, David Townend, Michail Matthaiou, Simon L. Cotton
• Classification: cs.IT (Information Theory), math.IT (Mathematics-Information Theory)
• Publication Date: October 14, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.12406

Abstract

This paper investigates fronthaul-constrained cell-free massive MIMO (CF-mMIMO) systems and proposes a hybrid centralized-distributed precoding strategy that dynamically adapts to different fronthaul and spectral efficiency (SE) requirements. The approach partitions users into two groups: one served by centralized precoding and another by distributed precoding. The authors formulate a novel optimization problem for user grouping and power control to maximize total spectral efficiency while satisfying fronthaul and per-access point (AP) power constraints. The problem is transformed into a tractable form with an efficient solution algorithm. Numerical results confirm the versatility and superior performance of the hybrid scheme, consistently outperforming purely centralized and distributed approaches across various system configurations.

Research Background and Motivation

1. Research Problem

Although CF-mMIMO systems demonstrate excellent performance in spectral efficiency, energy efficiency, and connectivity, they face challenges from limited fronthaul capacity. Traditional systems either employ fully centralized precoding (requiring substantial fronthaul resources but offering good performance) or fully distributed precoding (low fronthaul demands but limited performance).

2. Problem Significance

• Technology Trends: Aligns with the decomposed and software-defined radio access network trends promoted by the O-RAN Alliance
• Practical Deployment Needs: Requires balancing performance and fronthaul efficiency
• System Scalability: Fronthaul overhead and complexity increase dramatically with growing user numbers

3. Limitations of Existing Approaches

• Centralized Precoding (Functional Split 7b): Requires global CSI exchange with high fronthaul overhead and complexity
• Distributed Precoding (Functional Split 7c): Uses only local CSI, unable to effectively suppress interference from other APs
• Lack of Flexibility: Existing systems cannot dynamically adjust precoding strategies according to actual requirements

4. Research Motivation

Propose a hybrid approach combining advantages of both methods to achieve scalable, cost-effective, and flexible transition toward centralized architecture, providing practical solutions for future wireless networks.

Core Contributions

1. Hybrid Precoding Framework: Proposes the first hybrid centralized-distributed precoding strategy for fronthaul-constrained CF-mMIMO systems
2. Optimization Problem Formulation: Establishes a joint optimization framework for user grouping and power control to maximize total system spectral efficiency
3. Efficient Solution Algorithm: Develops power optimization algorithm based on successive convex approximation (SCA) and K-means user grouping method
4. Performance Validation: Numerical results demonstrate that the hybrid scheme outperforms traditional purely centralized and distributed methods across various system configurations

Methodology Details

Task Definition

Input: CF-mMIMO system with M APs and K users, each AP equipped with L antennas Output: User grouping strategy (Kc, Kd) and power allocation coefficients η Constraints: Fronthaul capacity limits, per-AP power limits, user QoS requirements

Model Architecture

1. System Model

• Channel Model: gmk = β^(1/2)_mk * hmk, where βmk is the large-scale fading coefficient and hmk is the small-scale fading vector
• CSI Acquisition: Uses TDD protocol and MMSE channel estimation
• Functional Split: Adopts functional split 7c architecture with BBL possessing precoding capability

2. Hybrid Precoding Strategy

Users are partitioned into two groups:

• Centralized Group Kc: Uses global CSI with precoding vectors computed at BBH
• Distributed Group Kd: Uses local CSI with precoding vectors computed at BBL

Transmitted signal expression:

sm = Σ(k∈Kc) √(ηc_mk) * wc_mk * xk + Σ(k∈Kd) √(ηd_mk) * wd_mk * xk

3. Precoding Design

• Centralized ZF: wc_mk = EmĜc((Ĝc)^H*Ĝc)^(-1)*ei
• Distributed ZF: wd_mk = Ĝd_m*((Ĝd_m)^H*Ĝd_m)^(-1)*πj / √E{||...||^2}

Technical Innovations

1. Fronthaul Constraint Modeling

Fronthaul requirements are divided into two components:

• Data Transmission: FHm,data = (Kc + Kd)α1
• Precoding Vectors: FHm,pr = Kcα2

2. SINR Derivation

• Centralized Group: SINRc_k = ηck / Σ interference terms + 1
• Distributed Group: SINRd_k = (L-Kd)(Σ√(ηd_mk*γmk))^2 / Σ interference terms + 1

3. Optimization Problem Decomposition

The original non-convex problem is decomposed into:

• User Grouping Subproblem: Solved using improved K-means algorithm
• Power Optimization Subproblem: Solved using SCA method

Experimental Setup

Dataset

• Network Topology: 2×2 km² area with wrap-around topology
• System Parameters: τ = 2000 samples, Sc_k = Sd_k = 1 bit/s/Hz
• Power Settings: AP maximum transmit power 1W (data), 500mW (training)
• Number of Users: K = 20

Evaluation Metrics

• Primary Metric: Total system spectral efficiency (Sum SE)
• Constraint Metrics: Fronthaul capacity utilization, power consumption

Comparison Methods

1. Centralized: All users employ centralized ZF precoding
2. Distributed: All users employ distributed ZF precoding
3. EPA: Equal power allocation scheme
4. Random: Random user grouping
5. LSF: Heuristic grouping based on channel gain

Implementation Details

• Fronthaul Parameters: Nsubcarrier = 3264, Nbits = 16, Morder = 64
• Algorithm: K-means user grouping + SCA power optimization
• Convergence Criterion: Based on objective function improvement threshold

Experimental Results

Main Results

1. Fronthaul Capacity Impact (Figure 1)

• Centralized: Performance improves with increasing FHmax but limited by user count
• Distributed: Performance saturates after FHmax = 6 Gbps due to inability to suppress inter-AP interference
• Hybrid Scheme: Achieves optimal performance at all FHmax values

2. Antenna Number Impact (Figure 2)

• Distributed: Significant improvement with increasing L
• Centralized: Trade-off exists; surpassed by distributed when L > 18
• Hybrid Scheme: Maintains best performance across all configurations

3. Performance Improvement Quantification

• K-means grouping vs. random grouping: 34% improvement
• K-means + power optimization: 83% improvement (FHmax = 12 Gbps)
• K-means vs. LSF heuristic: 3% improvement

Ablation Studies

1. User Grouping Algorithm: K-means vs. Random vs. LSF
2. Power Allocation: OPA vs. EPA
3. Fronthaul Constraints: With constraints vs. without constraints

Complexity Analysis

Experimental Findings

1. Adaptability: Hybrid scheme automatically adjusts centralized/distributed user ratio based on fronthaul capacity
2. Robustness: Maintains advantages across different antenna counts and AP configurations
3. Practicality: K-means grouping method is simple and effective, suitable for irregular user distributions

Related Work

1. CF-mMIMO Foundational Research

• Pioneering work by Ngo et al. established CF-mMIMO theoretical foundations
• Traditional research primarily focuses on purely centralized or distributed schemes

2. Fronthaul-Constrained Systems

• Existing work mainly addresses fronthaul limitations for data transmission
• Few consider fronthaul overhead for precoding vector transmission

3. User Grouping Methods

• Clustering methods based on geographic location
• Grouping strategies based on channel correlation
• First application of K-means to user grouping in hybrid precoding

4. Power Optimization

• SCA method applications in wireless communication optimization
• Novel adaptations for hybrid precoding scenarios in this work

Conclusions and Discussion

Main Conclusions

1. Hybrid Precoding Effectiveness: Outperforms single approaches across various system configurations
2. Optimization Algorithm Feasibility: SCA+K-means method effectively solves complex joint optimization problems
3. Practical Deployment Value: Provides practical solutions for emerging network architectures like O-RAN

Limitations

1. Channel Model: Assumes frequency-flat, slow-fading channels
2. Precoding Schemes: Considers only ZF precoding; other strategies unexplored
3. User Mobility: Does not account for user movement effects on grouping strategy
4. Quantization Error: Does not thoroughly analyze CSI quantization impact on performance

Future Directions

1. Dynamic Grouping: Research real-time channel condition-based dynamic user regrouping
2. Multiple Precoding Schemes: Explore hybrid applications of MRT, MMSE, and other precoding methods
3. Machine Learning: Leverage deep learning for user grouping and resource allocation optimization
4. Practical Deployment: Validate scheme practicality in real network environments

In-Depth Evaluation

Strengths

1. Strong Innovation: First to propose hybrid precoding concept for CF-mMIMO, filling research gap
2. Theoretical Rigor: Complete mathematical modeling with clear optimization problem formulation
3. Algorithm Practicality: SCA+K-means method with moderate computational complexity, easy to implement
4. Comprehensive Validation: Multi-scenario numerical experiments verify scheme effectiveness and robustness
5. Engineering Value: Closely integrated with O-RAN functional split architecture with practical application prospects

Weaknesses

1. Idealized Assumptions: Perfect CSI assumption difficult to achieve in practice
2. Grouping Strategy: K-means method relatively simple; potentially superior strategies exist
3. Dynamic Adaptability: Does not address adaptive adjustment mechanisms for changing network conditions
4. Experimental Limitations: Validation based only on simulation; lacks real system testing

Impact

1. Academic Value: Provides new solution approach for fronthaul-constrained CF-mMIMO problems
2. Industrial Significance: Offers practical technical solutions for 5G/6G network deployment
3. Follow-up Research: Likely to inspire further research on hybrid precoding strategies

Applicable Scenarios

1. Dense Urban Areas: Scenarios with limited fronthaul capacity but high spectral efficiency requirements
2. Edge Computing: Scenarios requiring load balancing between cloud and edge processing
3. Heterogeneous Networks: Networks with APs having different processing capabilities and fronthaul capacity
4. Dynamic Networks: Networks requiring flexible adjustment based on load and demand variations

References

1 H. Q. Ngo et al., "Ultradense cell-free massive MIMO for 6G," Proc. IEEE, 2024. 2 E. Nayebi et al., "Precoding and power optimization in cell-free massive MIMO systems," IEEE Trans. Wireless Commun., 2017. 3 O. T. Demir et al., "Cell-free massive MIMO in O-RAN," IEEE J. Sel. Areas Commun., 2024.

Overall Assessment: This is a high-quality technical paper proposing an innovative solution to the fronthaul-constrained problem in CF-mMIMO systems. It features rigorous theoretical analysis, reasonable algorithm design, and comprehensive experimental validation. While improvements are possible in certain assumptions and practical deployment aspects, the paper makes important contributions to the field with high academic and practical value.