Research On Direction Finding Algorithm Of Interference Sources In Airport Area Based On Deep Learning

Nowadays,most radio interference monitoring systems deployed in airdrome area are based on Direction-Finding theories of radio signal of model to determine the direction of interference source arrival.However,conventional theories based on direction-finding(DF)of radio signal of model are not able to meet current requirements of high-precision DF missions of interference source in airdrome area.Therefore,on the basis of technical studies of conventional DF of interference source,theories based on deep learning have been adopted to further study algorithm of DF of interference source.The results of simulation experiments revealed that the network construction designed in this dissertation can be capable of distinguishing directions very well in the aspect of DF of multi-objective interference source.In addition,proposed validation scheme and network construction for DF have both proved to be correct and effective according to real-world measured result.Major works that was done are as follows:1)One dimensional interferometer,correlation interferometer,multiple signal classification(MUSIC)algorithm have been selected as the conventional theories based on DF of radio signal,which compare and analyze DF performance towards interference source of single signal and multiple signals respectively in the scene of ideal DF array and existed channel amplitude-phase error;The advantages and limitations of related algorithms existed in the domain of DF of interference source have been summarized,which further provides basis for the research of DF algorithms of interference sources.2)In view of the limitations of traditional methods in DF of interference sources,this dissertation proposes an intelligent neural network architecture consisted of Self encoder and deep residual network for DF of interference source based on deep learning theory.The data samples of interference source are generated by simulation,and the designed network model is trained.In the process of DF of interference source in the proposed network architecture,first of all,the output data of DF array is sent to the auto-encoder for noise reduction pretreatment.Next,the output data of the auto-encoder is sent to the depth residual neural network to extract the spatial spectrum features of the signal.Finally,according to the normalized spectrum peak position,directions of interference source signal are estimated,and the feasibility of the proposed DF network of interference source is verified theoretically.3)According to the network architecture and intelligent Direction-Finding platform proposed in this paper,the verification work of Direction-Finding was carried out based on the real-world measured data of interference source.Firstly,data acquisition scheme of interference source in airdrome area was designed.And a portable Direction-Finding system was used to obtain the data of interference source at different incident direction.Secondly,real-world measured data was processed to generate the set of data sample for Direction-Finding performance verification of network.Next,the sample set was used to further train and optimize the network architecture that finally outputs the predicted incident direction of the interference source.The results of real-world measured data show that in terms of generalization ability of the network,error of Direction-Finding for the unknown azimuth interference source is less than 2 degrees,the accuracy is more than 90%,and the prediction time of single sample is less than 0.1s,which further verifies the advantages of the network architecture proposed in this paper and gives the direction for further research.