Research On Phase Calibration Method Of Passive Direction-Finding Channel Based On Deep Learning

As the electromagnetic environment becomes more and more complex,cognitive electronic warfare has gradually become the mainstream direction of development,and the accuracy and precision of passive direction-finding receiving system are increasingly required.With the development of recent years,many channel correction algorithms are being developed and researched,hoping to be applied to more complex situations and achieve better correction effects.With the deepening of the research,the channel model becomes more and more close to the actual channel,and the algorithm of filter design and channel equalization becomes more and more practical.However,traditional methods such as look-up table method and equalizer design have heavy workload and unsatisfactory correction effect in complex scenes.This paper mainly studies and designs a scheme for the inconsistencies of phase characteristics between passive direction-finding channels and proposes a deep learning method to solve the problem of channel phase correction.The main work is to carry out the simulation and model of the channel,a large number of simulation data and measured data simulation and result analysis,the establishment of evaluation model,finally through the analysis of the simulation results,to prove the effectiveness of the method proposed in this paper.Firstly,four existing channel mismatch models are described,and their principles and derivation are analyzed.This paper introduces the specific parameters and internal structure of Universal Software Radio Peripheral(USRP),and after analyzing its internal structure,summarizes three main sources of phase error of receiver channel: error of the nonlinear devices,error of mixer,and time error of A/D sampling.According to the conclusion,the three parts are modeled respectively.In practice,the complex situation in the receiver channel will produce some random errors over time,so the phase error coefficient in the three-part model is processed as Gaussian variable,in order to fit the real receiver situation better.Then,After describing two existing traditional channel calibration methods,a channel phase calibration method based on Deep Neural Networks(DNN)is proposed.Firstly,the data set used in the experiment is described,including the channel model parameters of the simulated data set and the antenna signal parameters.Then,the collection experiment of measurement data set is designed,and the parameters of the equipment are introduced.Then,the structure of DNN network designed in this paper and the final evaluation model are described.Finally,by analyzing the experimental simulation results,it not only verifies the effectiveness of this method compared with the traditional algorithm,but also proves that the established channel mismatch model can adapt to the real channel well.Secondly,a phase calibration method of passive direction-finding channel based on Convolutional Neural Network(CNN)is proposed.After redefining the simulated and measured data sets,the convolution layer and pooling layer are used for better characteristics extraction and fitting.After giving the flow chart of the method and the structure of the network,the simulation experiment is carried out.Finally,after analyzing a series of simulation results,the generalization ability is better than DNN algorithm,and better calibration effect is obtained.The advantages of the deep learning algorithm and the practicability of the channel mismatch model are further proved.