Distributed User-Centric Clustering Based on Local Observation and Value Decomposition

View at publisher Summarise with VIRAT
Authors: Yihui Wang, Shanzhi Chen, Shaoli Kang, Li Chen
Cited by: 0
References: 45

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The user-centric access network (UCAN) has emerged as a promising solution to address the growing demand for high-quality wireless services, but its dynamic clustering faces challenges such as high computational complexity, partial observability at access points (APs), and scalability limitations of centralized control. This paper proposes a multi-agent reinforcement learning (MARL) clustering framework that explicitly addresses the credit assignment problem in decentralized clustering using value decomposition methods. We formulate user-centric clustering as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) to jointly optimize minimum spectral efficiency (SE) for fairness and average SE for system efficiency. The proposed framework learns to decompose global Q-function estimation into local Q-functions for individual APs during training, thereby resolving the multi-agent credit assignment challenge and enabling fully distributed execution based solely on local observations. We implement two value decomposition methods: the linear VDN method, which uses additive decomposition, and the nonlinear QMIX method, which employs mixing networks with global state conditioning through hypernetworks. Extensive simulations across static and dynamic scenarios demonstrate that the proposed methods significantly outperform independent learning, MARL baseline lacking explicit credit assignment, and heuristic baselines, achieving superior user fairness while maintaining competitive average performance. The framework exhibits robust generalization across varying mobility speeds and candidate pool sizes. This work establishes value decomposition as an effective clustering solution for practical user-centric network deployments, offering scalable, distributed clustering with provable coordination guarantees.

Fields of Study & Subjects

Advanced Wireless Network Optimization Advanced MIMO Systems Optimization Cognitive Radio Networks and Spectrum Sensing

Publication details

Journal / source
IEEE Internet of Things Journal
Volume
13
Issue
12
Pages
26883–26894
Publication date
25 March 2026
Language
EN