Intelligent Resource Management with Deep Reinforcement Learning in Device-to-Device Communication
- Publication Type:
- Thesis
- Issue Date:
- 2022
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Radio resource management in device-to-device cellular offload can be optimised to increase network capacity, quality of service, energy efficiency, lower latency and provide more resilient networks. However, this resource optimisation problem is both NP-Hard and required to operate at a millisecond timescale, limiting feasible solutions.
In this thesis, we investigate how deep reinforcement learning can be applied to improve resource allocation. To empirically demonstrate our approach, we develop a network simulator for device-to-device cellular offload research. We also introduce an improved self-play algorithm for training reinforcement learning without expert guidance.
We apply our self-play training algorithm to the game Connect Four. Leveraging the competitive pressures of coevolution, we improve the performance of agents trained with our method, achieving a 15% higher win rate. Furthermore, agents exhibit more stable training dynamics and suffer fewer performance regressions.
We evaluate our network simulator and demonstrate deep reinforcement learning can significantly increase network capacity. Our network simulator reduces research friction and provides an evaluation platform to compare, share and build upon results. Our toolkit is provided to other researchers as open-source software.
Please use this identifier to cite or link to this item: