Online Energy-Efficient Power Control in Wireless Networks by Deep Neural Networks

Abstract

The work describes how deep learning by artificial neural networks (ANNs) enables online power allocation for energy efficiency maximization in wireless interference networks. A deep ANN architecture is proposed and trained to take as input the network communication channels and to output suitable power allocations. It is shown that this approach requires a much lower computational complexity compared to traditional optimization-oriented approaches, dispensing with the need of solving the optimization problem anew in each channel coherence time. Despite the lower complexity, numerical results show that a properly trained ANN achieves similar performance as more traditional optimization-oriented methods.