Satellite Signal Blind Detection Method Based On Time-frequency Diagrams Semantic Segmentation

In modern satellite communication systems,blind signal detection is an important function of the receiving end.Accurate blind detection of signals is of great value in both commercial and military fields.Existing detection methods are difficult to adapt to the situation of dense signals and high noise,so their performance in the actual channel environment is not ideal.In this paper,oriented to the application scenarios of satellite signal detection,the blind detection technology of signals based on deep learning is studied,and a method of blind detection of satellite signals based on semantic segmentation of time-frequency graphs is proposed,and its processing performance is improved,mainly including the following two points of innovation:(1)Satellite signal blind detection technology based on semantic segmentation of time-frequency mapA method of blind detection of satellite signals based on deep learning is proposed.First,a high-precision time-frequency diagram drawing method is designed,so that as much detail information as signal energy,signal boundary,substrate fluctuation and other details can be retained in a single image.Then,the semantic segmentation technology in deep learning is applied to divide the signals in the time-frequency graph as semantics.Finally,combine the characteristics of the signal spectrogram to perform post-processing to realize the detection task of signal center frequency and bandwidth.Using the data of two actual satellite channels for testing and verification,the results show that the method proposed in this paper can obtain better precision,accuracy and recall.(2)Improved performance of blind signal detection based on model combinationIn view of the long processing time of the blind detection method based on time-frequency map semantic segmentation,the following improved methods are proposed:firstly,the time-frequency map signal is pre-divided through the target detection model to reduce the calculation amount of semantic segmentation;secondly,the semantic segmentation The ordinary convolution of the segmentation model is replaced with a depth separable convolution to reduce the amount of calculation,and the Attention structure is added to accelerate the model's convergence and avoid unnecessary feature calculations.Compared with the original method,the signal blind detection method based on model combination has a speed increase of 7.6 times while the accuracy is only reduced by 2.26%.Experiments prove that the method proposed in this paper can effectively complete the blind detection task of signals in a complex channel environment,and provides a new idea for using deep learning technology to solve classic and complex problems in the communication field.