Machine learning based adaptive modulation scheme for energy harvesting cooperative relay networks

Abstract

Energy harvesting has emerged as a promising technique in next general wireless communication. Although plenty of studies have already focused on the energy management in energy harvesting relay network, the problem is still open when we consider using different modulation method for the nodes in the network. In this paper, we propose an offline transmission scheme to improve the total throughput of the relay cooperative system. Both the source node and the relay node adjust M-ary quadrature amplitude modulation level and choose the corresponding transmitted power. We first determine the QAM level for the source node according to the data buffer state. Following this, we further use machine learning method to adaptively choose its QAM level based on whole view of the system information including channel state information and energy storage information. We have also proposed an online transmission scheme for the relay to choose QAM level under causal system information based on classification method. Taking the result of the offline scheme as training data, we use QAM level to divide channels into different classes and using the channel information to get the threshold of different class. We also need to dynamic adjust the threshold considering the energy storage information. With the class, we can determine the QAM level under present channel information. The simulation result show that the performance of the offline scheme is much better compared to the best effort scheme and the online scheme also improve the performance but still litter worse than the offline scheme.