Wireless Transmission Protocols Recognition Based On Deep Learning

Protocol recognition refers to the identification of network transport protocols.This paper studies application layer protocols and wireless transmission protocols Therefore,it is of great significance to accurately identify various protocols for network security,network planning and management,network traffic optimization.In the aspect of application layer protocol,the traditional identification methods,such as port based identification technology and static load feature protocol recognition technology,all have defects in varying degrees.It cannot meet the needs of current protocol identification.In the aspect of wireless transmission protocol,there are a few research efforts in the protocol identification of wireless transmission in public information.The existing recognition methods identify protocols through classifying the identified signal parameter.This method relies on the recognition accuracy of the previous part too much.When the parameter recognition rate decreases,the accuracy of protocol recognition falls very quickly.First of all,aiming at the application layer protocol recognition,a method based on depth learning is proposed,which solves the problems of low recognition rate and privacy involved in existing methods.First the original bit stream protocol data is identified using the existing four basic network frameworks including DNN,CNN,LSTM,and LSTM-DNN,and then the effect of different parameters of the network on the recognition performance are analyzed through experiments.The result shows that when the length difference of different protocols is too big,the recognition rate of the network will be greatly reduced.According to the experimental results of different networks,a modified network based on LSTM-DNN networks which preforms best among the four networks above is proposed.The modified network solves the above problems,and finally obtains a higher recognition accuracy.Secondly,based on the application layer protocol recognition,the neural networks are improved to realize the recognition of wireless transmission protocols,which provides a new method for the analysis of wireless transmission protocols.The method firstly compares the difference between the bit-stream data and the wireless signal data,then designs different data processing schemes according to the data characteristics of the wireless signal,and finally improves the best performance networks,CNN and LSTM-DNN,in the application layer identification to achieve wireless transmission protocol recognition.It is found by experiments that the recognition rate of wireless transmission protocols decreases rapidly with the decrease of the signal-to-noise ratio(SNR).In this case,the denoising auto-encoder is used to initialize the network weights,and to a certain extent the protocol recognition rate is improved when the SNR is low.Finally,according to the actual needs,we use the training method based on biased strategy to train the network,which can improve the recognition accuracy of SNR range which is set to a certain extent,and finally get a higher recognition accuracy.