基于深度学习的HARQ辅助空天地融合网络时延受限容量预测

Delay-limited Throughput Prediction of HARQ-assisted Satellite-Aerial-Terrestrial Integrated Network:A Deep Learning Approach

空天地融合网络(Satellite-Aerial-Terrestrial Integrated Network,SATIN)可以满足未来网络对全时全域全空通信和网络互联互通的需求。为了降低用户端传输时延并满足高频谱利用效率的要求,研究了基于深度学习的混合自动重复请求(Hybrid Automatic Repeat reQuest,HARQ)辅助的SATIN的时延受限容量(Delay-Limited Throughput,DLT)。为了提升性能预测效率和实时性,提出了基于卷积神经网络(Convolutional Neural Network,CNN)的性能预测方法,采用了一种去除池化层的改进CNN模型。预测结果表明,所提出的CNN预测结果较优,较Elman、BP等传统机器学习方法有更好的预测性能,其误差在10~(-3)浮动,且预测时间较其他方法大幅度减少。

The Satellite-Aerial-Terrestrial Integrated Network(SATIN)can provide seamless information services for users,and meets the needs of future networks for full-time,all-domain,all-space communication and network interconnection.In order to reduce the transmission delay at the user terminal and satisfy the requirements of high-frequency spectrum utilization efficiency,the authors investigate the Delay-limited Throughput(DLT)of SATIN assisted by Hybrid Automatic Repeat reQuest(HARQ)based on deep learning.In order to improve the efficiency and real-time of performance prediction,a performance prediction method based on Convolutional Neural Network(CNN)is proposed,which adopts an improved CNN model with removing the pooling layer.The prediction results show that the proposed CNN prediction results are better and have enhanced prediction performance than other machine learning methods such as Elman and BP,and the error fluctuates at 10-3.Also,the prediction time is much less.