Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Linear Attention for Joint Power Optimization and User-Centric Clustering in Cell-Free Networks

Irched Chafaa, Giacomo Bacci, Luca Sanguinetti

TL;DR

The paper tackles dynamic, scalable resource management in user-centric cell-free MIMO by jointly predicting AP clusters and UL/DL powers using only spatial coordinates. It introduces ELU-CosFormer, a CosFormer-based transformer with ELU activation and pilot-aware constraints to achieve linear-attention complexity and pilot-contamination elimination. Through dynamic supervision and synthetic data, the approach attains near-optimal max-min spectral efficiency while adapting to varying network loads and preserving robustness to input errors. The results demonstrate substantial computational advantages over baselines and other Transformer variants, enabling real-time applicability in large-scale deployments. Overall, the method offers a practical, scalable solution for adaptive clustering and power optimization in next-generation wireless networks.

Abstract

Optimal AP clustering and power allocation are critical in user-centric cell-free massive MIMO systems. Existing deep learning models lack flexibility to handle dynamic network configurations. Furthermore, many approaches overlook pilot contamination and suffer from high computational complexity. In this paper, we propose a lightweight transformer model that overcomes these limitations by jointly predicting AP clusters and powers solely from spatial coordinates of user devices and AP. Our model is architecture-agnostic to users load, handles both clustering and power allocation without channel estimation overhead, and eliminates pilot contamination by assigning users to AP within a pilot reuse constraint. We also incorporate a customized linear attention mechanism to capture user-AP interactions efficiently and enable linear scalability with respect to the number of users. Numerical results confirm the model's effectiveness in maximizing the minimum spectral efficiency and providing near-optimal performance while ensuring adaptability and scalability in dynamic scenarios.

Linear Attention for Joint Power Optimization and User-Centric Clustering in Cell-Free Networks

TL;DR

The paper tackles dynamic, scalable resource management in user-centric cell-free MIMO by jointly predicting AP clusters and UL/DL powers using only spatial coordinates. It introduces ELU-CosFormer, a CosFormer-based transformer with ELU activation and pilot-aware constraints to achieve linear-attention complexity and pilot-contamination elimination. Through dynamic supervision and synthetic data, the approach attains near-optimal max-min spectral efficiency while adapting to varying network loads and preserving robustness to input errors. The results demonstrate substantial computational advantages over baselines and other Transformer variants, enabling real-time applicability in large-scale deployments. Overall, the method offers a practical, scalable solution for adaptive clustering and power optimization in next-generation wireless networks.

Abstract

Optimal AP clustering and power allocation are critical in user-centric cell-free massive MIMO systems. Existing deep learning models lack flexibility to handle dynamic network configurations. Furthermore, many approaches overlook pilot contamination and suffer from high computational complexity. In this paper, we propose a lightweight transformer model that overcomes these limitations by jointly predicting AP clusters and powers solely from spatial coordinates of user devices and AP. Our model is architecture-agnostic to users load, handles both clustering and power allocation without channel estimation overhead, and eliminates pilot contamination by assigning users to AP within a pilot reuse constraint. We also incorporate a customized linear attention mechanism to capture user-AP interactions efficiently and enable linear scalability with respect to the number of users. Numerical results confirm the model's effectiveness in maximizing the minimum spectral efficiency and providing near-optimal performance while ensuring adaptability and scalability in dynamic scenarios.

Paper Structure

This paper contains 24 sections, 17 equations, 11 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Transformer-based power allocation
  4. Main contributions
  5. Reproducible research
  6. Paper outline and notation
  7. Wireless Network Model
  8. Uplink and downlink spectral efficiency
  9. Problem Formulation
  10. Overview of CosFormer
  11. Proposed Learning Model for Clustering and Power Optimization
  12. Model architecture
  13. Attention mechanism
  14. Training Setup
  15. Dataset construction
  16. ...and 9 more sections

Figures (11)

  • Figure 1: Illustration of a user-centric cell-free mMIMO network. Each UE is served by its own cluster of AP demir2021foundations.
  • Figure 2: Architecture diagram of the proposed model to predict jointly AP clusters, UL and DL powers leveraging spatial information at the input.
  • Figure 3:
  • Figure 4: Comparison of ReLU and ELU activation functions. ELU smoothly transitions for negative inputs using an exponential curve, improving gradient flow and stability.
  • Figure 5: Comparison of per-UE SE using different activation functions in CosFormer for UL and DL. ELU yields the highest SE across all UE counts.
  • ...and 6 more figures