Research On Key Technologies Of Communication Signal Recognition Based On Deep Learning

The identification of communication signals is becoming more and more extensive.It is an important ield of research in the fields of spectrum management,electronic counter-measures,and Communication identification.Past algorithms of communication signal recognition requires much prior knowledge of the signal,which means high complex-ity,time-consuming,and limited application scenarios.In addition,in order to transmit more information in a complex frequency spectrum environment,modern communication equipment often uses a variety of complex working modes to send information,which brings great challenges to the past algorithms of communication signal identification.What's more,in real scenarios,there are always new unknown communication individuals participate in the transmission.Past algorithms can only retrain the entire recognition pro-cess,which is very time-consuming and labor-intensive,and severely limits the applicable scenarios.In view of the above challenges,this paper uses the advantages of deep learning to efficiently and accurately identify communication individuals with the same,multiple and complex working modes.It also implements accurate detection and and overall iden-tification of unlabeled data samples of unknown communication individuals,Dynamicly updates the system.The main work of this paper includes two parts:(1)In the paper,for each communication individual with the same model and the same multiple working mode scenarios,an algorithm based on deep learning to identify the communication individual using the time-frequency domain information of the com-munication signal is build.First,the method transforms the signal into time-frequency domain,and then uses deep learning to automatically extract the subtle features of the communication individual in the time-frequency domain.A deep learning network model is build to efficiently and accurately identify the communication individual.In addition,in the case of a small number of overall data samples or an uneven number of data sam-ples under different labels,the past algorithms always take a long time to train,frequently oscillates during convergence,and the model is prone to collapse.The method of data en-hancement enhances the robustness and stability of the identification system and improves the performance of the overall system.(2)Aiming at situations where unknown communication individuals participate in the transmission of information in practical applications,this paper build an algorithm based on semi-supervised learning and transfer learning,which can automatically detects data samples of unknown communication individuals,and automatically provides tags,then iteratively updates the signal recognition network model through model migration.It avoids the disadvantages of past algorithms that need to manually label new samples and completely retrain the algorithm,so that the recognition algorithm proposed in this paper can adapt to changes in the data set faster than traditional algorithms while maintaining high recognition accuracy.The algorithm proposed achieves the dynamic update of the overall identification system.