Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DNN-Enabled Multi-User Beamforming for Throughput Maximization under Adjustable Fairness

Kaifeng Lu, Markus Rupp, Stefan Schwarz

TL;DR

The paper tackles throughput fairness in downlink multi-user MIMO by learning CSI-driven beamformers with a Transformer (WiT) under an explicit fairness constraint. It formulates a convex-looking trade-off via a Lagrangian hinge loss, where a dual variable $\lambda$ is updated by dual ascent to enforce a target Jain’s index $J_{LB}$ while maximizing the sum rate $\overline{S}$. By training the model with multiple $J_{LB}$ values, the method traces the Pareto front between throughput and fairness, yielding near-optimal trade-offs in a scalable, deployment-friendly manner. Experimental results in a simulated single-cell setup show the DNN achieving higher mean sum rates at higher fairness targets than conventional baselines (including wSLNR) and provide per-user improvements, highlighting practical impact for flexible QoS provisioning in wireless networks.

Abstract

Ensuring user fairness in wireless communications is a fundamental challenge, as balancing the trade-off between fairness and sum rate leads to a non-convex, multi-objective optimization whose complexity grows with network scale. To alleviate this conflict, we propose an optimization-based unsupervised learning approach based on the wireless transformer (WiT) architecture that learns from channel state information (CSI) features. We reformulate the trade-off by combining the sum rate and fairness objectives through a Lagrangian multiplier, which is updated automatically via a dual-ascent algorithm. This mechanism allows for a controllable fairness constraint while simultaneously maximizing the sum rate, effectively realizing a trace on the Pareto front between two conflicting objectives. Our findings show that the proposed approach offers a flexible solution for managing the trade-off optimization under prescribed fairness.

DNN-Enabled Multi-User Beamforming for Throughput Maximization under Adjustable Fairness

TL;DR

The paper tackles throughput fairness in downlink multi-user MIMO by learning CSI-driven beamformers with a Transformer (WiT) under an explicit fairness constraint. It formulates a convex-looking trade-off via a Lagrangian hinge loss, where a dual variable is updated by dual ascent to enforce a target Jain’s index while maximizing the sum rate . By training the model with multiple values, the method traces the Pareto front between throughput and fairness, yielding near-optimal trade-offs in a scalable, deployment-friendly manner. Experimental results in a simulated single-cell setup show the DNN achieving higher mean sum rates at higher fairness targets than conventional baselines (including wSLNR) and provide per-user improvements, highlighting practical impact for flexible QoS provisioning in wireless networks.

Abstract

Ensuring user fairness in wireless communications is a fundamental challenge, as balancing the trade-off between fairness and sum rate leads to a non-convex, multi-objective optimization whose complexity grows with network scale. To alleviate this conflict, we propose an optimization-based unsupervised learning approach based on the wireless transformer (WiT) architecture that learns from channel state information (CSI) features. We reformulate the trade-off by combining the sum rate and fairness objectives through a Lagrangian multiplier, which is updated automatically via a dual-ascent algorithm. This mechanism allows for a controllable fairness constraint while simultaneously maximizing the sum rate, effectively realizing a trace on the Pareto front between two conflicting objectives. Our findings show that the proposed approach offers a flexible solution for managing the trade-off optimization under prescribed fairness.
Paper Structure (17 sections, 11 equations, 6 figures, 2 tables, 1 algorithm)

This paper contains 17 sections, 11 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Downlink Input-Output Relationship
  4. Fairness Index
  5. Weighted SLNR Beamforming
  6. Problem Formulation
  7. Transformer-Based Sum Rate Fairness Optimization
  8. Training Data
  9. Model Architecture
  10. Feature Input and Embedding Block
  11. Transformer Encoder
  12. Output Regression Block
  13. Loss Function and Training Algorithm
  14. Simulations
  15. Simulation Setup
  16. ...and 2 more sections

Figures (6)

  • Figure 1: Transformer-DNN architecture for throughput maximization under adjustable fairness
  • Figure 2: Cross comparison on mean achievable rate and fairness between wSLNR $(\alpha_\ell)$ and DNN trained with target $J_{LB}$.
  • Figure 3: User-rate ECDF at selected target fairness values from \ref{['tab:key_statistics']}; DNN (dashed) vs. wSLNR (solid).
  • Figure 4: Sum-rate ECDF at selected target fairness values from \ref{['tab:key_statistics']}; DNN (dashed) vs. wSLNR (solid).
  • Figure 5: Per-user rate comparison with users ordered by index (weakest to strongest) at selected matched-fairness pairs from \ref{['tab:key_statistics']}: DNN (darker bars) vs. wSLNR (lighter bars).
  • ...and 1 more figures