Unknown Signal Recognition And Parameter Estimation Based On Deep Learning

With the rapid development of communication technology,various new types of modulated signals are emerging,and the receiver may receive some unknown modulated signals under non cooperative conditions.In addition,accurate estimation of signal parameters is a prerequisite for achieving correct signal demodulation.Therefore,unknown modulation identification and parameter estimation of signals are key technologies in non cooperative communication,and play a very important role in both military and civilian fields.With the gradual emergence of deep learning in various research fields,signal modulation recognition and parameter estimation based on deep learning has become the latest mainstream research content in this field.Compared to traditional methods,signal modulation recognition and parameter estimation based on deep learning networks can not only reduce the huge workload of manual feature extraction,but also achieve the same or even better results as manual feature extraction,making signal recognition and parameter estimation methods more intelligent and accurate.In this paper,the recognition and parameter estimation of unknown modulated signals are deeply studied through deep learning methods.In the aspect of unknown signal recognition,this paper proposes an unknown signal recognition method based on the combination of multi stream ConvNeXt network and Mahalanobis distance Distance Metric(MDM)decision,which includes signal feature extraction and distance metric decision.Firstly,the method of parallel processing with multiple convolutional cores is used to improve the ConvNeXt network,which more specifically improves the network’s ability to extract features from modulated signals,making the distinction between known and unknown signal features more significant.Secondly,in terms of distance measurement decision,the idea of outlier detection and incremental learning is introduced to make two distance measurement decisions for the extracted features.The first distance measurement decision detects unknown signals,the second distance measurement decision further subdivides the detected unknown signals,and can automatically update the parameters of the model according to the increasing unknown signals,so that the model has the ability to self evolve and achieve more effective recognition of unknown signals.In terms of parameter estimation,this article combines convolutional neural networks with Transformer networks to propose a parameter estimation model based on multi-stream ConvNeXt Transformer for estimating signal parameters.This model first uses a multi-stream ConvNeXt network to extract the local spatial features of the signal,then uses a Transformer network to extract the long-distance global features of the signal,and finally outputs the final parameter estimation value through the fully connected layer.The experimental results show that compared with other parameter estimation models,the parameter estimation model proposed in this paper has smaller estimation errors and can effectively complete the task of accurately estimating signal parameters.