Artificial Neural Network Design for Enabling Relay Selection by Supervised Machine Learning

Our recent work has proved the effectiveness of utilizing supervised machine learning to perform relay selection in cooperative networks. However, the structure of artificial neural network provided is rather heuristic and simple, where a symmetric architecture is adopted with only a single hidden layer and the number of neurons in the hidden layer is the same as the number of inputs/outputs. However, there is not a generalized design guideline for the structure of artificial neural network. In this paper, we thereby investigate the structure related issues for the artificial neural network designed for enabling relay selection by supervised machine learning and test a series of network structures from the very simple one to the multi-layer one. Finally, numerical results generated by Monte Carlo methods are shown to reflect the key findings and a number of crucial features of the proposed artificial neural network in this paper.