Research On Packet Loss Classification Based On TCP Congestion Control Mechanism In LEO Satellite Network

The distinction of packet loss has always been a hot and thorny issue in TCP congestion control.Due to the high dynamics of the LEO satellite network,the probability of packet loss has increased dramatically.If the type of data packet loss cannot be accurately determined,and inappropriate window adjustment strategies are adopted,the data packet transmission rate will be seriously affected and the network bandwidth cannot be used efficiently.In addition,compared with terrestrial networks,the problem of data flow concept drift caused by such high dynamics is more serious,which greatly affects the accuracy of the packet loss discrimination model.Based on the TCP congestion control mechanism,this paper solves the influence of sample distribution changes on the packet loss discrimination model by constructing a conceptual drift detection model,and proposes a packet loss discrimination mechanism suitable for LEO satellite networks.The work of this article is summarized as follows:1)Aiming at the problem of concept drift in the data sample,this paper constructs a concept vector model to map the individual features in the data sample into a joint concept vector.Starting from the vector level,the vector clusters are clustered according to the correlation of the sample features between the concept vectors,and the concept drift detection model is trained accordingly.In order to detect the time point of concept drift more sensitively,this paper proposes a dual-threshold concept drift decision criterion,which judges the drift state based on the varying degrees of sample distribution changes.2)Compared with the concept drift,the sample size imbalance also affects the accuracy of the classification model.In determining the type of data packet loss,it is mainly divided into two situations:congestion packet loss and wireless error packet loss.In reality,the probability of packet loss due to congestion is much greater than that of wireless error packet loss.Therefore,this paper proposes a data imbalance algorithm for this problem.By setting a penalty factor for unbalanced samples,the problem of imbalance between samples is eliminated based on the down-sampling of most samples,and a sample de-imbalance model is constructed.3)After solving the problem of concept drift and imbalance in the data,this paper constructs an integrated classification model for distinguishing packet loss types.Through online and offline learning,train an ensemble classifier containing multiple base classifiers and a naive Bayes classifier.An adaptive update algorithm for the weights of the base classifier is proposed,and a new classifier is trained when the concept drift occurs.Finally,the ELN and CE flag bits in the ACK header are designed to complete the feedback mechanism of the result of packet loss distinction,and complete the end-to-end packet loss distinction and congestion control.In the ns-2 experimental simulation,compared with the single-model machine learning classification algorithm,the classification model proposed in this paper has a better AUC,up to 0.9885;compared with the traditional TCP protocol,the packet loss classification accuracy exceeds 98%;Compared with the existing protocols for satellite networks,the TCP protocol based on packet loss distinction proposed in this paper not only has good fairness,but also has a great improvement in throughput.