Resource Allocation and Spectrum Sensing for 5G Networks Based on Deep Learning

Abstract

As wireless systems are increasingly utilized, the volume of data traffic in wireless networks continues to grow. To manage these demands and ensure quality service for each user, employing multiple-input multiple-output (MIMO) systems is seen as an effective approach for the future of telecommunications. Additionally, managing radio resources and controlling transmitted power are crucial in wireless systems. Various algorithms have been developed to address these power control challenges. Notably, machine learning algorithms are gaining traction for optimizing power allocation in MIMO systems because they offer lower computational complexity and faster processing times suitable for real-time applications. However, gathering training data for deep learning applications in telecommunications poses significant challenges. Deep neural networks require extensive data for training, and labeling this data is a complex process. In this paper, we explore a power allocation strategy in massive MIMO systems using a deep neural network that employs unsupervised learning, where a cost function updates the network instead of relying on labeled data. The objective is to maximize the signal-to-interference-plus-noise ratio (SINR). The findings demonstrate that unsupervised deep learning can effectively allocate power while reducing computational complexity and processing time.