Identification And Separation Of Time-Frequency Mixed Signals In Complex Electromagnetic Environment

With the updating of communication mode,more and more military and civil electronic facilities put into use,the signal modulation patterns show diversity development,and the spectrum space signal is dense,which constitutes a complex electromagnetic environment.Especially for non-cooperative broadband reception,most of the observed signals are mixed signals with unknown parameters.Therefore,it is of great significance to effectively identify and separate single source signals from multiple and multi-modulation time-frequency overlapping signals in a complex electromagnetic environment.Currently,traditional signal modulation recognition algorithms mainly focus on single source signal,and there are many related recognition methods,but there are few studies on direct recognition of time-frequency overlapping signals.Therefore,the method of separation followed by recognition is usually used.Because of the large time delay,the high complexity of the algorithm and the dependence of the recognition effect on the separation effect.For the signal separation,the research is quite sufficient under the conditions of Overdetermined and positive definite,but the number of receiving channels is usually less than the number of component signals under the extremely limited conditions of reception conditions.So,in this project,we will study the identification method of time-frequency mixed signals and the estimation of mixture matrix and source signal recovery in underdetermined blind source separation.The main contents and innovations of this paper include.1.For the research content of this article,the signal mixing model and the principle of time-frequency transformation were analyzed and introduced.Define the degree of signal overlap in time and frequency domains,visualize the degree of signal mixing,and provide theoretical support for the full text discussion.The evaluation criteria of signal recognition accuracy,matrix estimation accuracy and source signal recovery and separation effect were given,which provide quantitative criteria for experimental analysis.2.A direct modulation classification method based on Seg Net was presented to solve the problem of low classification methods and low efficiency for time-frequency mixed signals in single-channel broadband receiving system.This method obtained the label image set based on the label fusion signal label method.After the Seg Net network training was completed,the high classification labels were filtered out by threshold filter method to improve the recognition accuracy and form a complete direct recognition system.The simulation results showed that when the number of sources in a single channel was not more than four,the method can identify the modulation modes of each component in any combination of time-frequency mixed signals at the same time.Although the overlap was 100%,the recognition accuracy of each hybrid model was higher than 91%.3.Aiming at the problem of source signal clustering error in mixed matrix estimation,a matrix estimation algorithm based on semantic single source point labeling was proposed.In this algorithm,single source and mixed time-frequency points were labeled by semantic segmentation network.The interval probability statistical detection method was used to filter out the interference points in the semantic single source points to enhance the accuracy of matrix estimation.The simulation results showed that the estimation accuracy of the algorithm was about 10% higher than that of the comparison algorithm at low SNR.Compared with other algorithms,the normalized mean square error of this algorithm was about 4d B lower,which meant it had better estimation accuracy.4.A source signal recovery algorithm based on multi-source point label was presented to overcome the performance degradation of single-source-dominant signal recovery algorithm when the degree of time-frequency overlap of received signal increased.The algorithm obtained the location of multiple source points and the mixed type of the location based on the prediction results of the semantic network.When the mixed matrix was known,the time-frequency coefficients of each single source signal at the multisource point were calculated accurately by solving the matrix.Finally,the recovery of each component of the mixed signal was completed with the single source point.The experimental results showed that the correlation coefficient between the recovered signal and the original signal can reach 0.9 even if the mixing signal had a high degree of overlap,and the algorithm performs well in the presence of noise when the threshold value was reasonable.