On board bandwidth resource reservation algorithm based on Dueling DDQN

DOI:	10.11809/bqzbgcxb2023.12.036
Keywords:	satellite network; user service quality; deep reinforcement learning; bandwidth resources; dynamic reservation
Abstract:	Aiming at the problem that the lack of available bandwidth leads to the interruption of users’ connection and affects users’ QoS in the process of frequent spot beam switching of LEO satellites, research on on board bandwidth resource reservation algorithm based on Dueling DDQN is proposed. By establishing Dueling DDQN neural network as a decision evaluator, bandwidth is allocated for calls, which effectively avoids human interference. Dueling DDQN neural network structure adopts dual network, which can improve the learning performance. In addition, DDQN method is applied to construct the objective network function to solve the problem of over estimation of Q value. Different from ε greedy strategy used in action space exploration, Noisy method is used to realize the exploration process, increase the exploration ability of the model, and realize the dynamic reservation of channel bandwidth resources in LEO satellite network communication system. The simulation results show that the proposed dynamic reservation method of on board bandwidth resources can reduce the handover failure rate and new call blocking rate of users, increase bandwidth utilization rate, and improve users’ QoS satisfaction, and it has better results than the traditional heuristic method.
Issue:	Vol. 44 No. 12 (2023)
Published:	2023-12-28
PDF HTML