Deep Learning for Power Allocation in Massive MIMO

Wireless communication technology has become a fundamental part of our society, changing the way we communicate and overall improving the quality of our everyday lives. Since wireless based services and applications constantly arise, the demand for ubiquitous connectivity and coverage is growing exponentially, increasing the amount of required voice and data connections. These are provided by cellular wide area networks (e.g., based on the GSM, UMTS2, and LTE3 standards) among other technologies, such as WiFi based local area networks (802.11 IEEE standard) and satellite services. However, innovation in this area of research will enable the development into a fully networked society, where homes, cars and machines are connected, service quality expectations are continously rising, yet electromagnetic spectrum resources remain limited. In this context massive MIMO cellular networks represent a promising solution for satisfying the demanding needs of today’s technology. Thanks to the employment of a number of antennas much greater than the number of served users, Massive MIMO improves the spectral efficiency of the network, providing great wireless throughput. In this report, the problem of power allocation in this type of networks will be analysed, suggesting a possible solution based on machine learning technology for reduced complexity and time optimization. Chapter 1 contains an overview of cellular network and more specifically massive MIMO technology, in chapter 2 machine learning and neural networks are introduced, in order to clarify the concepts utilized in chapter 3, in which our suggested power allocation solution is explained