Research On Radar Modulation Signal Recognition Technology

Radar modulation signal recognition is the main component of radar signal reconnaissance and sorting,and it plays a key role in electronic warfare.With the explosive development of deep learning,the latest research on radar modulation signal recognition technology mainly focuses on radar modulation signal recognition technology based on deep learning,but most existing recognition algorithms cannot effectively solve the problems that they have the high demand for hardware computing power,take a long training time and cannot accurately identify unknown modulation methods and newly-added modulation methods.In order to solve these problems,this paper designs and proposes a radar modulation signal recognition algorithm based on lightweight neural networks,open set recognition and incremental recognition technologies.First of all,in terms of reducing hardware computing requirements and shortening the training time of network models,this paper inputs the time-frequency image data set generated by Choi-Williams distribution of radar signals with different modulation methods into the lightweight neural network Mobile Net V3 to realize radar signals effective classification and identification of modulation methods.In order to further solve the problem that the image feature dimension extracted by the Mobile Net V3 network is too high to affect the recognition performance of the modulation method,this paper uses a new classifier composed of PCA and SVM to replace the classification layer of the Mobile Net V3 network,and designs and proposes a radar modulation signal recognition algorithm based on improved Mobile Net V3 to achieve better classification and recognition performance and robustness.Secondly,in terms of unknown modulation methods in the test set of radar signals,this paper designs a lightweight neural network feature extractor based on Mobile Net V3 and an open set recognition classifier based on SVDD,and proposes an open set recognition algorithm for radar modulation signals based on SVDD by using the trained SVDD hypersphere corresponding to each known modulation method to realize the effective classification and recognition of the known modulation method and the unknown modulation method of the radar signal.Finally,in terms of new modulation methods appearing in the training set of radar signals,this paper designs an incremental network based on SVDD,and proposes an incremental recognition algorithm for radar modulation signals based on SVDD.The new SVDD hyperspheres obtained from the training of the new modulation method and the initial SVDD hyperspheres are combined into an incremental network to realize the efficient classification and recognition of the initial modulation method and the new modulation method of the radar signal and make the algorithm have the ability to continuously learn.