Research on Random Access Control Strategy and Optimization Algorithm of Multi-type Terminals Based on Deep Reinforcement Learning

With the further development of 5G technology, large-scale machine-type communication technology has become the key to realize the interconnection of massive terminals. However, when massive terminals initiate the random access process at the same time, it will cause serious network congestion, especially in the application scenario where multiple types of terminals coexist. Severe network congestion will definitely affect the access delay and packet loss rate of delay-sensitive terminals. It is necessary to design a reasonable competition resolution mechanism to alleviate network congestion. Therefore, this paper proposes a random access optimization algorithm for multi-type terminals based on deep reinforcement learning. By introducing a priority design into the distributed queue access mechanism, the access opportunities of delay-sensitive terminals are expanded and the probability of collisions is reduced. An optimization algorithm based on the deep Q-learning network is proposed to dynamically adjust the number of preambles exclusively used by high-priority terminals, so as to reduce the influence of resource monopoly on the delay-tolerant terminals and minimizes conflicts as much as possible. In different load scenarios, the proposed algorithm is compared with existing competition resolution mechanisms and methods, and the practicability and effectiveness of the proposed method in solving the key problem of massive multi-type terminal coexistence are proved.