Distributed learning based energy-efficient operations in small cell networks

Mughees, Amna (2023) Distributed learning based energy-efficient operations in small cell networks. Doctoral thesis, Sunway University.

Preview

Text Final Hardnound Thesis_Amna.pdf - Accepted Version Download (5MB) | Preview

Abstract

Meeting the requirements of varying data rates, ultra-reliable low latency, massive machine-type communication, and other use cases involves the introduction of several new technologies in future mobile networks. Collaboration and autonomous operation among network devices are imperative in addressing these use cases. Several solutions proposed in the literature are limited to addressing the cooperation problem and resource management issues. The joint optimisation problem of user association and power allocation has been studied extensively; however, conventional optimisation techniques still have room for improvement in distributed resource management strategies that evolve based on network dynamics. This research contributes to resolving these challenges in small cell networks by optimizing cell selection and resource allocation to enhance energy efficiency in ultra-dense networks. The thesis proposed a solution that employs unique characteristics of machine learning and game-theoretic framework to enable a model-free and energy-efficient small cell network. The dense deployment issues regarding conflicts and load imbalance is also addressed while reducing energy consumption. Also, the proposed algorithms focus on the need for cooperative learning that maintains the quality of service, adapts to the network dynamics and achieves energy efficiency in dense small-cell networks. To achieve these goals, a cell selection algorithm is proposed that overcomes the issues of conflicts and load imbalance while reducing energy consumption. Additionally, an intelligent distributed cooperative power allocation learning algorithm is proposed to enhance energy efficiency in small cell networks while satisfying the scalability, power constraint, user association requirements, and QoS constraints. This research focuses on separate action spaces, often neglected for user association and power allocation. Both user association and power allocation are solved in their respective action spaces. Simulation results demonstrate improved performance in power consumption, load, sum rate, utility, learning rate, convergence, and energy efficiency for small base stations (SBSs) and user equipment (UEs) compared to four benchmarked algorithms, including WMMSE, game theory, Q-learning, and DRL. The simulation results show improved results of our proposed technique is 33% as compared to DRL. The proposed framework paves the way for energy-efficient intelligent networks beyond 5G. The proposed solution possesses the potential for future expansion across multiple user associations, up-link considerations, seamless integration of renewable energy sources, and innovative energy harvesting techniques.

Actions (login required)

View Item