Dynamic multiple access based on deep reinforcement learning for Internet of Things

With the development of wireless communication technology, the rational use of spectrum resources utilizing efficient multiple access technology has become a research hotspot. The channel access strategy in dynamic spectrum access technology will directly affect the system's spectrum utilization. Deep reinforcement learning is proposed to train users in the network to learn optimal channel access strategy to achieve reasonable allocation of spectrum resources on the channel. At the same time, the multiple access scheme adopted by users who have access to the channel will also affect the throughput of the system. Traditional multiple access schemes include Orthogonal Multiple Access scheme and Non-orthogonal Multiple Access (NOMA) scheme. In a dynamic environment, the multiple access scheme suitable for users on the channel is also dynamic. Therefore, DRL is introduced to help users learn the multiple access scheme selection strategy to maximize user throughput. Finally, we propose a dynamic multiple access algorithm based on deep reinforcement learning, which optimizes the channel access strategy and scheme selection strategy respectively, and improves the performance. The simulation results show that the proposed scheme is superior to other schemes in various scenarios, especially, the average throughput of the proposed scheme is about 0.8 Mbps and 0.5 Mbps higher than that of the FDMA scheme and NOMA scheme, respectively.