Research On Specific Emitter Identification Based On ZYNQ

Specific emitter identification(SEI)is a technique to distinguish among different emitters using the differential unintentional modulation information.SEI plays an important role in modern electronic warfare,communication security,network security and other fields.At present,the researches of SEI are also more extensive.Although these studies have greatly improved the identification performance of SEI,they still have some deficiencies in the hardware implementation of the identification algorithm,the full utilization of the heterogeneous features of the emitter,and the full characterization of the unintentional modulation information of the pulse signal of the emitter.To this end,the SEI based on ZYNQ is studied in this dissertation.Firstly,based on the demand for experimental data when verifying each key algorithm in SEI,this dissertation takes the early warning aircraft radar APS-145 as an example to analyze the source of the unintentional modulation information in the emitter.Based on this,three simulated specific emitters are modeled,and simulated signal samples from three specific emitters are obtained under different signal-to-noise ratios.Based on the AD9361 board and the ZYNQ development board,four semi-physical emitter platforms are built,and semi-physical signal samples from four specific emitters are collected.In addition,the measured signal samples of three specific emitters of an air-based platform are collected..Secondly,in order to find individual feature extraction methods that can more fully characterize the unintentional modulation information of the emitter signal,this dissertation starts from the time domain,frequency domain and transformation domain,and is based on multiscale dispersion entropy,bispectrum and wave atoms transformation.A series of individual feature extraction methods are studied.Experiments with classical single kernel support vector machine under various signal samples show that the multiscale dispersion entropy individual features and wave atoms individual features are better than bispectrum individual features identification and less affected by noise.For the simulated signal samples and the half-physical signal samples with a signal-to-noise ratio of 10 d B,the identification rate of both types of features can reach more than 90%.For the measured signal samples,the identification rate of the two types of features can also reach more than 85%.Finally,based on the consideration of heterogeneous characteristics when multi-domain features are applied to SEI,and several common multiple kernel learning and support vector machine,this dissertation studies a method that can make full use of multi-domain features heterogeneous information.That is data combined multiple kernel learning.The comparison experiments of simple multiple kernel learning and data combined multiple kernel learning under various signal samples show that under simulated signal samples,semi-physical signal samples and measured signal samples,the data combined multiple kernel learning can improve the individual identification performance of SEI more obviously.In addition,based on the consideration of the hardware implementation of the algorithm of the SEI,this dissertation designs an FPGA program for bispectrum individual feature extraction and single kernel support vector machine decision process on the ZYNQ development board,and simulates each FPGA program through the Modelsim tool.The calculation result under the is the standard,and the calculation error of each FPGA program is analyzed.The results show that the relative error of the calculation result of the FPGA program for bispectrum individual feature extraction is less than 0.08%.The calculation result of the FPGA program for the decision process of the single kernel support vector machine is basically consistent with the calculation result under the PC.