Network Selection Over 5G-Advanced Heterogeneous Networks Based on Federated Learning and Cooperative Game Theory
5G-Advanced and beyond claims a 3D ecosystem with cooperation between terrestrial and non-terrestrial networks to achieve seamless coverage, improve capacity, and support advanced applications with strict quality of service (QoS) requirements. In this complex environment, the Open Radio Access Netwo...
| Authors: | , , , , , |
|---|---|
| Format: | article |
| Publication Date: | 2024 |
| Country: | España |
| Institution: | Universidad del País Vasco |
| Repository: | Addi. Archivo Digital para la Docencia y la Investigación |
| OAI Identifier: | oai:addi.ehu.eus:10810/73947 |
| Online Access: | http://hdl.handle.net/10810/73947 |
| Access Level: | Open access |
| Keyword: | 5G-advanced and beyond federated deep reinforcement learning game theory network slicing QoS |
| Summary: | 5G-Advanced and beyond claims a 3D ecosystem with cooperation between terrestrial and non-terrestrial networks to achieve seamless coverage, improve capacity, and support advanced applications with strict quality of service (QoS) requirements. In this complex environment, the Open Radio Access Network (O-RAN) architecture supporting the network slicing paradigm is a prominent solution to guarantee flexibility, dynamism, and differentiated traffic management. Furthermore, intelligence is critical for future wireless networks to enable machine learning (ML)-based optimization for autonomous RANs deployment, handling huge and heterogeneous environments, and adapting to numerous scenarios. This paper presents an enhanced Dynamic Radio Access Network Selection (eDRANS) algorithm based on Federated Double Deep Q-Network (F-DDQN) and inserted in the novel O-RAN architecture. The proposal selects the most suitable base station (BS) to satisfy user petitions optimizing QoS and slicing resource utilization. Moreover, the solution employs a Cooperative Game Theory (CGT) approach to manage resources in overloading situations. This load-balancing process enables the acceptance of new clients without abruptly degrading the active users’ perception. eDRANS is adapted to diverse network conditions, multiple service constraints, and several types of users with different priorities and mobility behaviors. The proposal is validated through network-level simulations, recreating a heterogeneous environment composed of terrestrial/airborne nodes and using the Max-SINR criterion, a heuristic algorithm, a centralized ML, and a distributed ML solution as benchmarks. Results show that eDRANS correctly learns during multiple trial-and-error interactions with the environment and guarantees to maximize user satisfaction. |
|---|