Graph Attention Network for Optimal User Association in Wireless Networks

TL;DR

The paper tackles network energy efficiency in dense 5G deployments by formulating UA and cell switch-off decisions as a graph-based optimization problem. It introduces a Graph Attention Network (GAT) with attention mechanisms to learn an energy-aware UA policy on a homogeneous wireless network graph, balancing fixed and traffic-dependent power while meeting throughput constraints. The authors define an unsupervised loss incorporating total network power and regularizers that encourage near-one-hot UA and balanced PRB utilization, and they validate the approach through simulations on a 7-cell cluster with realistic channel and traffic settings. Results show significant NES gains over traditional RSRP-based UA and competitive gains against a genie-aided per-PRB SINR benchmark, with larger gains at higher bandwidths and tunable regularization guiding cell switch-off behavior. This work demonstrates the practicality of graph-based, attention-enabled learning to optimize energy efficiency in RAN operations with potential real-world impact on operators’ OpEx and sustainability.

Abstract

With increased 5G deployments, network densification is higher than ever to support the exponentially high throughput requirements. However, this has meant a significant increase in energy consumption, leading to higher operational expenditure (OpEx) for network operators creating an acute need for improvements in network energy savings (NES). A key determinant of operational efficacy in cellular networks is the user association (UA) policy, as it affects critical aspects like spectral efficiency, load balancing etc. and therefore impacts the overall energy consumption of the network directly. Furthermore, with cellular network topologies lending themselves well to graphical abstractions, use of graphs in network optimization has gained significant prominence. In this work, we propose and analyze a graphical abstraction based optimization for UA in cellular networks to improve NES by determining when energy saving features like cell switch off can be activated. A comparison with legacy approaches establishes the superiority of the proposed approach.

Figures (7)

• Figure 1: Logical connectivity graph representation between cell sites (BSs) and UEs.
• Figure 2: An example of user association for $50$ UEs across $7$ cells and $20$ (MHz) bandwidth: (a) SINR matrix, the values indicates the strength of SINR at each UE location for each BS. (b) The UE association based on the GNN technique. The $(k,n)$ entry indicates the association of $k^{\rm th}$ UE to the BS in the $n^{\rm th}$ cells. (c) The UE association based on the best SINR.
• Figure 3: The graph representation of the UE association of the example in Fig. \ref{['Net_GNN1']}. (a) The graph corresponding to the GNN technique. (b) The graph corresponding to the best SINR. Each color represents the UEs connected to the BS in the same cell.
• Figure 4: GNN performance gain ($\%$) vs $W$, Performance gain with respect to GA-SubSINR.
• Figure 5: GNN performance gain ($\%$) vs $W$, Performance gain with respect to best RSRP.