Research On Frequency Hopping Signal Detection Algorithm

Frequency hopping(FH)signal detection is one of the key aspects of FH communication reconnaissance and has important military application value.Because FH signal detection is difficult,many researchers have paid close attention to it.This thesis mainly studies the FH signal detection algorithm and gives the performance tests of the proposed algorithm.In the light of the existing time-frequency spectrum-based FH signal detection algorithms do not make full use of the detailed information of the time-frequency spectrum,which is obtained by using the short-time Fourier transform(STFT),three FH signal detection algorithms are proposed.Firstly,the different characteristics of the time-frequency spectrum of FH signals,fixed-frequency interference signals and Gaussian white noise are utilized,and the time-frequency spectrum entropy is introduced as a detection statistic to reflect the the uncertainty of the fluctuation of the signal time-frequency spectrum,and a time-frequency spectrum entropy based FH signal detection algorithm(TFSE)is proposed.Since the uncertainty of the fluctuation of the time-frequency spectrum of the FH signal is smaller than that of the noise,the entropy of the time-frequency spectrum is used to achieve the detection of the FH signal.Then,in order to further improve the detection performance of FH signals under low signal-to-noise ratio,there are different dispersions of the time-frequency spectrum of FH signals,fixed-frequency interference signals and Gaussian white noise,so the time-frequency spectrum local variance is introduced,and a time-frequency spectrum local variance based FH signal detection algorithm(TFSLV)is proposed.Finally,considering the complex actual electromagnetic environment,a FH signal detection algorithm based on the variance mean value of the time-frequency spectrum(TFVM)is proposed.The variance mean values of the time-frequency spectrum in the time direction and the frequency direction of the FH signals,Gaussian white noise,fixed-frequency interference,swept frequency and burst interference signals is analyzed respectively.The detection of FH signals is achieved in complex backgrounds by using the different feature of the variance mean value.The simulation results show that the three time-frequency analysis algorithms proposed in this paper all have better detection performance compared with similar algorithms.Because the performance of the traditional time-frequency spectrum-based FH signal detection methods is limited by the time-frequency resolution trade-off and spectrum leakage,FH signal detection algorithms based on generalized S-transform and deep learning are investigated.Firstly,the FH signal detection algorithm(CNN-ST)using S-transform and convolutional neural network(CNN)is studied.The S-transform of the received signal is computed and the time-frequency spectrum is obtained,and it is normalized to make it resistant to noise power uncertainty.A convolutional neural network structure is designed to input the time-frequency spectrum into the CNN network,and the network is used to directly extract the time-frequency characteristics of the signal and Gaussian noise for FH signal detection.Then,a FH signal detection algorithm(Res NetOp GST)based on the optimized generalized S-transform(GST)and residual neural network(Res Net)is studied to address the problems that the performance of the signal time-frequency analysis based on the traditional S-transform is limited by the fixed Gaussian time-window function.Taking the time-frequency aggregation measure as the criterion,the optimization of GST parameters ? and(37)is realized by using multiple swarm genetic algorithms.The time-frequency spectrum of the FH signal is obtained by using the optimized GST.A residual neural network structure is designed to realize the detection of the FH signal.The simulation results show that the CNN-ST and Res Net-Op GST algorithms proposed in this paper have better detection performance and lower computational complexity and storage complexity.A signal reception and processing hardware platform is designed,and composed of DSP and FPGA chips to receive the FH signals sent by the signal source.The test system is constituted and consists of the hardware platform and software system to evaluate the performance of TFSE,TFSLV and TFVM algorithms.The above three algorithms are implementd with C programming language on DSP.The detection performance of the algorithms on the actual FH signals is tested online.The detection algorithms based on deep learning need pre-processing and multi-layer network calculation,which has high complexity,and the storage resource of DSP is limited,so this hardware platform receives the FH signals sent by the signal source,and the actual received samples of FH signals under various signal-to-noise ratios are obtained.The detection performance of the actual signal of the CNN-ST and Res Net-Op GST methods is tested offline on MATLAB.The performance test results of three time-frequency spectrum-based detection algorithms,TFSE,TFSLV and TFVM,show that the performance of the three algorithms is better than the detection performance of the power spectrum elimination algorithm in [34],and all of them can perform realtime detection,with the best performance of the TFVM algorithm and the second best performance of the TFSLV algorithm.The performance test results of two S-transform and deep learning based detection algorithms,CNN-ST and Res Net-Op GST,show that both algorithms outperform the detection performance of the HCRNN deep learning algorithm of [74],with the Res Net-Op GST algorithm performing best and the CNN-ST algorithm second best.