Fast Assessment of Requests Schedulability in Data Relay Satellite Systems: A Multi-layer Neural Network Method

In data relay satellite (DRS) systems, the widely adopted centralized scheduling mode usually results in varying degrees of feedback delay to users, which is detrimental to the efficiency of the whole system and users'QoE (quality of experience). To solve this problem, we investigate the periodicity of such system's mission planning and the actual characteristics of relay missions, and then construct a multi-layer neural network model to simulate the mapping relationship between the inputs (i.e., the mission requests) and outputs (i.e., the scheduling results). The model has a simple and practical structure, composed of the input layer, the hidden layer and the output layer, which benefits from the accurate identification and extraction of relay mission’ characteristic parameters. Last, we utilize a back propagation (BP) algorithm to train the neural network. The proposed method can quickly assess the schedulability of users' requests with the average accuracy 85% in multiple scenarios, which effectively improve the efficiency of the DRS system and the QoE.