Identification Technology For Wireless Ad Hoc Network's MAC Protocols Based On Neural Network

Network confrontation is a research focus in cognitive electronic warfare field recently,which includes network identification,network interference and antiinterference,etc.Wireless network with self-organization and changeable links is commonly used in battlefield environment.How to obtain MAC protocol information of Ad hoc network under noncooperative conditions is not only the key content of network confrontation,but also the basic work of network interference.Aiming at this problem,this thesis studies MAC protocol identification technology based on neural network.The main work is as follows:1.Set up experiment environment.Firstly,the programmable open source network simulator NS-3 is used to simulate the Ad hoc network,a detection node is set outside the network to gather the network communication timing in a non-cooperative way;Secondly,the network communication timing data obtained by detection is simulated in physical layer,which including modulation,noise addition,filtering and energy detection;The data recovered from energy detection is used as support data of this thesis;This step truly simulate the actual confrontation scene.2.Aiming at the inter-class identification problem of three types' MAC protocols,a new link layer's feature extraction scheme and identification algorithm based on RNN are proposed.The indoor and outdoor tests are implemented.When the received signalto-noise ratio is greater than 3d B and the Ad hoc network has a certain throughput,simulation results indicate that this method can reach more than 98% accuracy for interclass identification;Indoor test results indicate that this method can reach more than 95%accuracy for allocation protocols;Outdoor measurement results indicate that this method can reach more than 90% accuracy for inter-class identification problem.3.Aiming at the internal identification problem of competitive MAC protocols,traditional identification algorithm and identification algorithm based on multiple neural networks are proposed.This scheme firstly design a new link layer's feature extraction scheme for the internal identification of competitive protocols,then design identification algorithm based on LSTM and Dense Net.When the received signal-to-noise ratio is greater than 3d B and the Ad hoc network has a certain throughput,simulation results indicate that this method can reach more than 95% accuracy for internal identification of competitive MAC protocols.4.Aiming at the problem of long inference time of Dense Net,the FPGA deployment scheme of model is proposed.The scheme takes ZCU102 development board as the hardware basis,completes the deployment based on the new model hardware development environment Vitis-AI.Test results indicate that this method reduces model's inference time consumption by 90% and power consumption by 97% at the cost of tiny loss of model accuracy.This thesis identifies MAC protocols based on link layer features and neural network,optimizes it with FPGA.It can obtain MAC protocol information of Ad hoc network quickly and accurately in battlefield environment,has high application value in network confrontation.