Joint Power and Flexible Numerology Allocation in 5G Networks Using Deep Reinforcement Learning

This study presents a novel power allocation optimisation strategy that does not rely on traditional power constraints. Unlike previous works in the literature, we introduce a ratio between the allocated and requested data rates and incorporate this ratio into the reward function of our deep reinforcement learning algorithm. The highest reward of 1 is achieved when the allocated and requested data rates are equal. Additionally, we jointly optimise power and numerology allocation, considering the users' delay and data rate requirements. Any numerology can be allocated to users as long as their requirements are satisfied. This approach enables users to be allocated optimum numerology and transmit power. By addressing the challenge posed by greedy users, our approach enhances the flexibility and performance of the power and numerology allocation process.