Energy efficiency optimization for Massive MIMO network: A Neural Network-based Approach

Massive MIMO is widely recognized as a fundamental technology for 5G networks and an essential component for future generations of networks. The integration of massive MIMO offers substantial advantages, encompassing enhanced data transmission speed, increased reliability, reduced transmission time, and improved spectral efficiency. However, the utilization of a large number of antennas in a massive MIMO system may lead to energy wastage and the emission of greenhouse gases detrimental to the environment. This study aims to employ deep learning in power allocation to optimize the energy efficiency of the massive MIMO system while meeting the real-time channel requirements. To this end, we first use convex optimization methods to solve the energy efficiency problem. Then a data set including the positions of users, each with its corresponding solved allocated power is used as the input of a neural network. Subsequently, this trained network is utilized to allocate the optimal power for a novel set of UEs’ positions. The energy efficiency of the neural network-based approach is verified by comparing with that of the traditional convex optimization.