Research On Communication Signals Analysis Method Based On Generalized S Transform

In the field of signal processing,time-frequency analysis has always been a hot research topic.Generalized S transform,as a flexible and excellent time-frequency analysis method,is widely used in image processing,signal processing,seismology,medical image processing and other fields.Based on the theory of generalized S transform,this paper focuses on the detection,filtering,instantaneous frequency extraction and recognition problems in the analysis of communication signals,including the following aspects:Aiming at the optimization problem of generalized S transform window function,combined with the improved genetic algorithm and taking the time-frequency aggregation measure as the criterion,a parameter optimization method of generalized S transform is proposed.Simulation results show that this algorithm has greatly improved the performance of standard S transform.Compared with other kinds of classical time-frequency analysis method,the improved generalized S transform has obvious advantage both in time-frequency concentration and Rényi entropy.In time-frequency detection,through detailed analysis and deduction of the statistical properties of Gaussian white noise in generalized S spectral domain,a time-frequency detection model based on generalized S transform is proposed.By eliminating the effect of frequency on the threshold,a constant false alarm rate(CFAR)detection model based on generalized S transform is proposed.The simulation results show that,compared with other linear frequency analysis methods,this model has higher detection rate when the false alarm rate are 0.01,0.005 and 0.001.The detection probability of GST algorithm can reach over 80% when signal to noise ratio is greater than 3dB.Compared with short time Fourier transform and S transform,it has the best detection results,and it can better adapt to signal detection in low SNR environment.Besides,this model also can be applied in time-frequency filtering.In time-frequency filtering,using generalized S transform algorithm and other two kinds of linear time-frequency analysis method for comparison,in addition to CFAR detection filtering method,a filtering method based on image processing is used.Compared with the singular value decomposition and fuzzy C-means(SVD-FCM)filtering method,the reconstruction using filtering method of image processing based on generalized S transform can get higher signal-to-noise ratio(SNR).In these three filtering methods,the filtering effect based on image processing is superior to the other two when the SNR is greater than-7dB.When the SNR is 10 dB,the image processing method improves the filtering effect of 1dB and 6dB than the constant false alarm rate filter and the singular value decomposition filtering method respectively.Besides,the instantaneous frequency extraction method based on image processing can reduce the mean square error of the extracted result by 0.015 than the traditional ridge extraction method at the SNR of-10 dB,which can extract the instantaneous frequency of the signal more effectively.Aiming at the signal type recognition problem,the instantaneous frequency can be considered as a kind of characteristics.This paper uses a back propagation neural network(BPNN)classifier to classify four kinds of different signals,and make a contrast with other time-frequency analysis methods.The experimental results show that the recognition rate of BP neural network classifier based on generalized S transform is more than 90% when SNR is-2dB.The effective classification of signals can be achieved at low SNR,and the classification effect is better than that of short time Fourier transform and S transform.