Dynamic Resource Allocation Using Deep Reinforcement Learning for 6G Metaverse
Haesik Kim
Abstract
The 6th Generation (6G) networks are now under development. The 6G will revolutionize the cellular networks more intelligently. In 6G era, we expect much higher network requirements including massive network traffics, huge numbers of devices, extremely low latency, low energy consumption and so on. A metaverse is one of key applications in 6G. Wireless techniques are directly related to performance of a metaverse application. The metaverse application requires both a high throughput and a low latency. The condition is more challenging than 5G. In this paper, we investigate dynamic resource allocation for 6G metaverse. We formulate the resource allocation problem for metaverse as the Markov decision processes (MDP) and solve the resource allocation problem using deep reinforcement learning (DRL). The main contributions of this paper are summarized as follows: We optimize the resource allocation for both a high throughput and a low latency. We adopt a sparse reward function of the reinforcement learning in the system model. It is more realistic because we can check whether or not the resource allocation scheme satisfies the requirements after completing the packet transmission.