A COMPREHENSIVE REVIEW OF DEEP LEARNING-BASED (DL) APPROACHES FOR INTERFERENCE MANAGEMENT IN 5G NETWORKS

The deployment of fifth-generation (5G) mobile networks is characterized by the increased level of infrastructure complexity, which aggravates the problem of interference distortion. Interference significantly degrades signal quality, decreases network capacity, reduces coverage, and deteriorates the Quality of Service (QoS) and user experience. The application of Deep Learning (DL) offers a promising approach to addressing interference management problems in 5G networks, owing to the ability of DL algorithms to learn hidden patterns underlying interference generation. Nevertheless, majority of this studies concentrate on particular aspects of interference management, and separate research groups investigate diverse problems, using different deep learning models, categorizing interference in diverse ways, and implementing various management functions. The purpose of this study is to synthesize the current knowledge on interference management based on deep learning approaches in 5G networks. The classifications covers neural network architectures, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Deep Reinforcement Learning (DRL), and Autoencoders (AEs); interference categories, including co-channel, adjacent-channel, cross-tier, and multipath interference; management functions, such as interference detection, prediction, and control; and deployment scenarios, including massive Multiple-Input Multiple-Output (MIMO) systems, millimeter-wave (mmWave) networks, heterogeneous networks, Device-to-Device (D2D) communication, Unmanned Aerial Vehicle (UAV)-assisted networks, and Intelligent Reflecting Surface (IRS)-assisted communication systems. In addition, the review examines the current solutions and their limitations in detail. The key finding of this study is that the majority of the papers address isolated functions of interference management, missing the concept of closed-loop interference management systems. Most of the reviewed articles do not consider the mobility of users and networks, have a high computational complexity, are not generalized enough to operate in multiple scenarios, and do not consider the problem of multiple types of interference. Based on the findings of the review, several directions for future research on deep learning-based interference management in 5G networks were proposed.