Research On Wireless Signal Recognition Technology Based On Deep Learning

Nowadays,with the birth of 5G and the Internet of Things,the access of mass devices and the emergence of more wide-area low-power service signals have become inevitable.In order to adapt to different business scenarios,more and more business signals have emerged,and the identification of signal service types has become a hot research topic.Whether in the military field or in daily life,there is a broad application prospect.Most of the research in the field of signal identification now focuses on the identification of modulation methods of wireless signals,and the traditional and commonly used methods are to extract expert characteristics of wireless signal IQ data such as high-order cyclic spectrum features,high-order cumulants,etc.,traditional expert features.Computational complexity and inability to uniformly identify all signals,multiple markings are required to identify a modulation scheme.Although in recent years,teams have combined deep learning and machine learning to improve signal feature extraction,making wireless signal feature extraction easier and more efficient,but most of them are still based on IQ data to identify the modulation of the signal.With the emergence of more and more business scenarios,different types of service signals emerge,various service signals adopt more complex multiple modulation methods,and even many new IoT signals are modulated in the traditional wireless signals.Obtained in the way.Therefore,only the modulation method of identifying the signal can no longer meet the demand.Different from the traditional signal recognition research,the main work of this paper is to use the power spectrum data of wireless signals combined with the deep learning algorithm model to identify the service type of wireless signals.The main results are as follows:1.A feature extraction method for extracting the fitting factor of wireless signal power spectrum waveform is proposed,and the neural network classifier model is used to identify the wireless signals of different services.The model is trained after 100,000 iterations.The recognition accuracy is above 97%,and the information about the recognition effect and model construction is given,and the difference from the statistical feature extraction is compared.2.The convolutional neural network model is constructed to realize the automatic extraction of the power spectrum waveform characteristics of wireless signals.The relevant information and experimental results of the model construction are given.The recognition accuracy of training models with more than 500 times is over 99%.And compared with traditional machine learning,the model generalized performance and migration performance were simulated and the results were obtained.When the data distortion is less than 80%,the recognition accuracy of the convolutional neural network classification model is still Can stay above 90%.3.An improved convolution model of spatial pyramid pooling layer is proposed,which realizes feature extraction and recognition of signal power spectrum data input to any size.The spatial pyramid layer structure information and the influence of the spatial pyramid pooling layer on the signal identification of the entire convolution model are given.Experiments show that the improved convolutional neural network can achieve more than 99%of data input for different sizes.Identify accuracy.4.A deconvolution model was constructed to visualize each layer of the convolutional neural network model.Deconvolution model construction information and visualization results are given.5.The following steps are first proposed to realize the template creation of the power spectrum waveform of the wireless signal by using the convolution self-encoder and the matching and identification by the template.The relevant information of the convolution self-encoder model structure and the visualization of the effect of template matching are given.Using the template as a feature vector for signal service identification accuracy can reach more than 99%.6.Using Python's Django framework to build a display system for wireless signal service identification.This paper combines the deep learning algorithm and the machine learning algorithm to extract the feature data of the wireless signal and realize the classification of the service type.Using supervised and unsupervised learning algorithms for feature extraction and classification,the feature extraction method based on fitting factor proposed in this paper is simpler than traditional signal feature extraction methods.Feature markers do not need to be different for different signal extraction.Features to mark.The final convolutional neural network classification model can achieve more than 99%accuracy of service identification for several experimental signals,and the superiority of the model is verified by a series of comparative experiments.Finally,an unsupervised learning method is introduced to realize the template for the feature of wireless signals,and the method of template matching is used to realize the identification of signal service types.