Decentralized Learning-Based Relay Assignment for Cooperative Communications

Cooperative communication exploits spatial diversity via relay node antennas in order to increase data rates in wireless networks. Relay node selection therefore plays a critical role in system performance. This study examines the problem of relay node selection in cooperative networks, in which one relay node can be used by multiple source-destination transmission pairs, and all transmission pairs share the same set of relay nodes. Centralized approaches may exhibit higher complexity when the number of source nodes is increased. This study gives source nodes self-optimizing and self-learning abilities, and enables them to autonomously select relay nodes.

A fully distributed approach, called a ”Decentralized Learning-based Relay Assignment” (DLRA) algorithm, is proposed to achieve this goal. DLRA uses a reinforcement learning technique called stochastic learning automata, with each source node attaining the self-learning ability to find an appropriate relay node according to environmental feedback. This study shows the convergency, optimality, and performance of DLRA via mathematical analysis, and evaluates the performance of DLRA in two different network systems: a cooperative ad hoc network, and an LTEAdvanced relay network. The experimental results present some properties of DLRA: the effectiveness of DLRA in cooperative communication systems, the significant improvements made by DLRA in fairness and the capacity of each node in LTE-Advanced systems.

Share this post