Conjugate Beamforming Variants for Multicasting in Cell-Free Massive MIMO Systems

1. Problem Statement

The paper addresses the challenge of efficient physical-layer multicasting in Cell-Free Massive MIMO (CF-mMIMO) systems. While CF-mMIMO offers uniform coverage and high spectral efficiency (SE) by using distributed Access Points (APs), traditional multicast strategies face significant bottlenecks:

• Single Multicast Stream: Serving all users with one stream limits the rate to the user with the worst channel conditions (the "weakest link" problem).
• Unicast Transmission: Sending individual streams to every user leads to severe pilot contamination and excessive resource consumption as the user count grows.
• Scalability: Centralized precoding schemes often suffer from high fronthaul signaling overhead and computational complexity, making them difficult to scale.

The authors aim to identify the optimal distributed precoding strategy and multicast transmission mode (unicast vs. subgroup-based multicast) under varying spatial user geometries (uniform, clustered, and heterogeneous).

2. Methodology

The study proposes a subgroup-centric framework where users are partitioned based on the similarity of their large-scale fading statistics.

System Model

• Architecture: A TDD-based CF-mMIMO system with $L$ distributed APs (each with $N$ antennas) serving $K$ single-antenna User Equipments (UEs).
• User Grouping: Users are clustered into $G$ disjoint subgroups using a K-means algorithm based on their large-scale fading vectors ($\beta_{lk}$). Users in the same subgroup share an uplink pilot and a downlink precoder.
• Channel Estimation: APs estimate a composite channel for each subgroup using Minimum Mean Square Error (MMSE) estimation. This approach mitigates pilot contamination within the subgroup while acknowledging interference from other subgroups sharing the same pilot.
• Precoding Schemes: The paper evaluates three fully distributed Conjugate Beamforming (CB) variants, which require no inter-AP channel state information (CSI) exchange:
  1. Classical CB (CB): Uses the MMSE channel estimate directly as the beamforming vector.
  2. Normalized CB (NCB): Normalizes the beamforming vector by its instantaneous norm ($\|\hat{h}\|$) to reduce effective channel gain fluctuations.
  3. Enhanced CB (ECB): Normalizes by the squared instantaneous norm ($\|\hat{h}\|^2$) to further stabilize the channel gain, requiring offline Monte Carlo averaging for power normalization.

Simulation Setup

• Scenarios: Monte Carlo simulations were conducted across three distinct spatial geometries:
  1. Uniform: Users distributed evenly across the coverage area.
  2. Clustered: Users grouped into spatial clusters (moderate and extreme density).
  3. Heterogeneous: A mix of uniform users and dense hotspots (simulating realistic 6G environments).
• Metrics: Performance was evaluated using Aggregated Spectral Efficiency (ASE).

3. Key Contributions

1. Unified Framework: Developed a simulation framework integrating composite channel estimation, distributed precoding, and adaptive AP cooperation clustering for subgroup-centric multicasting.
2. Comparative Analysis: Provided the first in-depth comparison of CB, NCB, and ECB specifically within the context of multicast CF-mMIMO under diverse spatial distributions.
3. Geometry-Dependent Insights: Quantified how spatial geometry dictates the optimal transmission mode (unicast vs. multicast) and the optimal number of subgroups ($G$).
4. Practical Guidelines: Established trade-offs between complexity and performance, identifying when advanced normalization (NCB/ECB) is beneficial versus when simple CB suffices.

4. Key Results

The performance of the precoders is heavily dependent on the spatial distribution of users:

• Uniform Deployments:

  • Unicast ($G=K$) is the superior mode, fully exploiting spatial multiplexing gains.
  • Among precoders, NCB offers a robust improvement over classical CB by reducing gain variability. ECB provides gains only when the number of AP antennas ($N$) is large.

• Clustered Deployments:

  • Unicast fails due to severe pilot contamination and intra-cluster interference.
  • Subgroup-based Multicasting becomes essential.
  • Classical CB emerges as the most robust and effective strategy. The benefits of NCB and ECB diminish significantly because the channel hardening conditions required for their normalization to be effective are not met in highly correlated clustered environments.

• Heterogeneous Deployments:

  • The optimal number of subgroups ($G$) follows a unimodal behavior: it is neither 1 (pure multicast) nor $K$ (unicast), but an intermediate value determined by the degree of clustering.
  • NCB and ECB do not systematically outperform CB. ECB gains are only visible in scenarios with large antenna arrays and weak clustering.
  • Conclusion: The spatial geometry dictates the optimal subgrouping strategy, while the precoding choice primarily affects the absolute ASE levels.

5. Significance and Impact

• Scalability: The study validates that fully distributed precoding (requiring only local CSI) is a viable and scalable solution for future 6G CF-mMIMO networks, avoiding the fronthaul bottlenecks of centralized schemes.
• Operational Guidelines: The paper provides clear rules for network operators:
  • Use Unicast + NCB/ECB for uniform user distributions.
  • Use Subgroup Multicast + Classical CB for clustered or hotspot-heavy environments.
• Resource Efficiency: By demonstrating that complex normalization (ECB) is often unnecessary in clustered scenarios, the work suggests that simpler, lower-complexity algorithms can achieve near-optimal performance in realistic, heterogeneous deployments.
• Future Directions: The authors suggest extending this work to heterogeneous APs (different antenna counts) and adaptive schemes where APs dynamically select the precoding variant based on local subgroup characteristics.

In summary, this paper demonstrates that spatial geometry is the primary driver for performance in CF-mMIMO multicasting. It argues against a "one-size-fits-all" precoding approach, advocating instead for a dynamic selection of transmission modes and precoding variants based on the specific user distribution of the network.