Research On Modulation Recognition Of Communication Signals Based On Deep Learning

Recognition of unknown signals in the electromagnetic spectrum is an important research direction in the field of signal processing and pattern recognition.In the case of non-cooperative communication such as electronic countermeasures,the receiver needs to identify the modulation type of the signal in order to obtain the information in the signal.However,there are many types of wireless modulation and different standards,which makes modulation recognition quite challenging.Two main traditional methods of modulation recognition: likelihood function method and feature recognition method.Because the latter has great advantages such as simple calculation and easy implementation,most of the research on modulation recognition is based on features for identification in the long-term phase.In recent years,with the development of machine learning and deep learning technology,more and more artificial intelligence algorithms have been applied to various fields,such as computer vision,language recognition,medical images,finance,etc.,and have achieved remarkable results.breakthrough.In the field of wireless communications,artificial intelligence algorithms have not been effectively utilized.The automatic recognition of wireless communication signal modulation by deep learning is an important step of software radio.Thanks to the computer computing power and the maturity of intelligent algorithms,this thesis mainly studies the modulation recognition of communication signals based on deep learning and is committed to building a more flexible,a more open radio environment,which will further promote the development of communication towards an intelligent road.Instead of the traditional modulation and recognition of complex signal preprocessing operations,this thesis uses a simple layer of end-to-end model to dozens of layers of complex network models to train and learn the sampled data to construct an intelligent wireless communication modulation recognition system.Simulation experiments show that when the 11 kinds of modulated signals such as analog amplitude modulation,phase frequency shift keying modulation,and quadrature amplitude modulation are identified,the improved depth residual model can be used to achieve the overall recognition rate more than 98.8% at a signal-to-noise ratio of 5 dB.During the research,it was found that the depth recognition network has a sharp decline in modulation recognition performance when predicting data in different scenarios.Therefore,this thesis combines the artificially designed communication modulation features with the strong fitting ability of deep neural networks,using more than 20 feature parameters,replacing the traditional decision tree and some other machine learning classification algorithms,applying deep neural network as the feature back-end classifier,the results show: The method based on depth learning to extract features can maintain better performance under additive Gaussian white noise channel,but the performance is drastically reduced when predicting samples under Rice channel;And the model of artificially extracting features to use deep network as backend classifier is able to maintain a relatively stable recognition accuracy under both channels.