Research On Intelligent Recognition Methods For Wideband Modulated Electromagnetic Signals

Intelligent recognition of broadband modulated electromagnetic signals is an important topic in the security of electromagnetic environments.Wideband modulated electromagnetic signal recognition is indispensable and plays an important role in the judgment and early warning of abnormal behavior of electromagnetic target equipment(systems)in combat environments and complex electromagnetic environments.At the same time,it is of great significance in the fields of spatial spectrum monitoring,malicious electromagnetic attacks,and electromagnetic interference identification,which is conducive to communication reconstruction and electromagnetic environment analysis.In recent years,with the rapid development of deep learning,automatic modulation recognition technology can help solve the problems of high computational complexity and model mismatch in traditional electromagnetic signal recognition research.Electromagnetic signal recognition based on deep learning can learn deeper information from different electromagnetic signal data.Therefore,this article applies deep learning to electromagnetic signal recognition,studies the recognition effect of neural networks on electromagnetic signals,and improves the network and recognition rate.The main work of this article is as follows:(1)Electromagnetic signal recognition based on time-frequency feature pairs.Introducing Transformer into electromagnetic signal recognition and selecting Swim Transformer network,in order to give Swim Transformer network the advantages of convolutional neural network,the RCSTNet network is constructed by combining convolutional neural network and Swim Transformer.Tested with 8 signals with a mixed signal-to-noise ratio,the average recognition rate is 93.12%.The recognition effect is not good at-10 d B,and the recognition rate of most types of signals is around 60%.At 0d B,the recognition rate of signals can reach 94%.(2)Electromagnetic signal recognition based on feature dimensionality reduction.Firstly,the feature dimensionality reduction method was introduced,using a hybrid feature dimensionality reduction method of filtered feature selection and unified manifold approximation and projection feature dimensionality reduction.Obtain appropriate feature subsets and train them into the improved Res Net50 network Res Net50-A.At a signal-to-noise ratio of-10 d B,the average recognition rate of 8 signals is 90%,and at 0d B,the average recognition rate is 95%.(3)Build a data generation platform and create a dataset of actual signals,and use the previously recognized electromagnetic signal recognition methods for signal recognition.Firstly,build a signal generation platform,use AD9361+Zedboard to generate different electromagnetic signals with variable parameters,and use NI acquisition equipment to collect them.Build a dataset and process it into RCSTNet and Res Net50-A networks for training and recognition.Finally,an electromagnetic signal recognition system interface was constructed to visualize the results of the proposed method for easy prediction.In the actual dataset,timefrequency feature features and RCSTNet network were used for recognition.The recognition rate reached 94% when the transmission power was-10 d Bm,and 99% when the transmission power was 0d Bm.In the electromagnetic signal recognition method based on feature dimensionality reduction,using mixed data for prediction,an average accuracy of 88% and an average recall rate of 85.125% were obtained.When the transmission power was-10 d Bm,the recognition rate reached 95%.