Federated Reinforcement Learning for Wireless Networks: Fundamentals, Challenges and Future Research Trends

Das, SK, Mudi, R, Rahman, MS ORCID: https://orcid.org/0000-0001-7308-8961, Rabie, KM ORCID: https://orcid.org/0000-0002-9784-3703 and Li, X ORCID: https://orcid.org/0000-0002-0907-6517 (2024) Federated Reinforcement Learning for Wireless Networks: Fundamentals, Challenges and Future Research Trends. IEEE Open Journal of Vehicular Technology, 5. pp. 1400-1440.

Preview

Published Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (13MB) | Preview

Abstract

The increasing popularity of Internet of Things (IoT)-based wireless services highlights the urgent need to upgrade fifth-generation (5G) wireless networks and beyond to accommodate these services. Although 5G networks currently support a variety of wireless services, they might not fully meet the high computational and communication resource demands of new applications. Issues such as latency, energy consumption, network congestion, signaling overhead, and potential privacy breaches contribute to this limitation. Machine learning (ML) frequently offers solutions to these problems. As a result, sixth-generation (6G) wireless technologies are being developed to address the deficiencies of 5G networks. Traditional ML methods are generally centralized. However, the vast amount of wireless data generated, growing privacy concerns, and the increasing computational capabilities of edge devices have led to a shift towards optimizing system performance in a distributed manner. This paper provides a thorough analysis of distributed learning techniques, including federated learning (FL), multi-agent reinforcement learning (MARL), and the multi-agent federated reinforcement learning (FRL) framework. It explains how these techniques can be effectively and efficiently implemented in wireless networks. These methods offer potential solutions to the challenges faced by current wireless networks, promising to create a more robust, capable, and versatile network that meets the growing demands of IoT and other emerging applications. Implementing the FRL framework can significantly improve the learning efficiency of wireless networks. To tackle the challenges posed by rapidly changing radio channels, we propose a robust FRL framework that enables local users to perform distributed power allocation, bandwidth allocation, interference mitigation, and communication mode selection. Finally, the paper outlines several future research directions aimed at effectively integrating the FRL framework into wireless networks.