Improved UCB MAB-Based Algorithm for Relay Selection in Cooperative Narrowband PLC Communication

In this paper, we apply artificial intelligence to find the best relay for a two-hop cooperative narrowband Power Line Communication (NB PLC). There is no channel side information, and the channel is corrupted by the periodic impulsive noise modelled using Middleton Class-A noise. The main idea is to exploit statistical properties of the arms rewards distributions, basing on the realistic physical characteristics of the channel, to enhance the Multi-Armed Bandit (MAB) Upper-Bound Confidence (UCB)-based algorithm decisions. Two new variants of the considered machine learning method are detailed and their performances to quickly and accurately choose the best relay are discussed. Simulation results showed that the proposed algorithms outperform the conventional UCB algorithm in terms of cumulative regret, probability of good selection, and Bit Error Rate (BER).