Matching Heterogeneous User Demands: Localized Self-organization Game and MARL Based Network Selection

Abstract

This chapter focuses on network selection for multiple user cases. Since users’ network selection decision determines the load distribution of networks, users’ decision-making is interacted. In particular, when heterogeneous user demand is considered, the solution of the optimal match between users and networks becomes a challenge. Centralized solutions could achieve a fair performance at a high optimization cost. Distributed solutions incur less cost but commonly result in low efficiency due to user competition. Different from centralized approaches or distributed approaches, we propose a local improvement algorithm, where networks that share users, called coupled network pairs (CNPs), cooperatively re-associate users with user demand awareness. Under a novel localized self-organization game formulation, we proved that the local improvement algorithm can achieve promising performance. To speed up the convergence of the algorithm, we further exploit the spatial independence among CNPs and propose an enhanced local improvement algorithm. Finally, simulation results indicate that the proposed algorithms achieve much better performance with relatively short convergence time, compared with three distributed algorithms.