Abstract : Power line communication (PLC) exploits the existence ofinstalled infrastructure of power delivery system, in order to trans-mit data over power lines. In PLC networks, different nodes of thenetwork are interconnected via power delivery transmission lines,and the data signal is flowing between them. However, the attenu-ation and the harsh environment of the power line communicationchannels, makes it difficult to establish a reliable communicationbetween two nodes of the network which are separated by a longdistance. Relaying and cooperative communication has been usedto overcome this problem. In this paper a two-hop cooperative PLChas been studied, where the data is communicated between a trans-mitter and a receiver node, through a single array node which hasto be selected from a set of available arrays. The relay selectionproblemcan be solved by having channel state information (CSI) attransmitter and selecting the relay which results in the best perfor-mance. However, acquiring the channel state information at trans-mitter increases the complexity of the communication system andintroduces undesired overhead to the system. We propose a classof machine learning schemes, namely multi-armed bandit (MAB),to solve the relay selection problem without the knowledge of thechannel at the transmitter. Furthermore, we develop a new MABalgorithm which exploits the periodicity of the synchronous impul-sive noise of the PLC channel, in order to improve the relay selec-tion algorithm.
Index terms : Cooperative communication, multi-armed bandit, power line communication (PLC), relay selection.