Deep Learning Based Radio Signal Modulation Mode Identification Algorithm Research

Radio signal modulation mode identification is a technique to determine the received signal modulation mode by analyzing and processing the received signal without sufficient a priori knowledge.With the unceasing development of contemporary wireless communication technology,radio signal modulation mode identification becomes more and more difficult in the process of spectrum detection.In the case of increasingly complex electromagnetic environment and radio signal characteristics,improving the efficiency and accuracy of radio signal modulation mode identification is a hot research topic in the field of radio monitoring.In the recent years,with the boom of deep learning algorithms,researches set out to explore the application of deep learning algorithms in the filed of spectrum monitoring.A large number of experimental data and results indicate that the signal modulation mode identification algorithm combined with deep learning can not only make full use of massive data,but also eliminate the professional cost and labor cost of manually extracting features,and effectively improve the modulation mode identification accuracy.This thesis addresses deep learning based algorithms radio signal modulation method identification.Firstly,the background of signal modulation mode identification and its significance are briefly described,then the knowledge of traditional and the deep learning-based methods are introduced,and finally,the radio signal modulation mode identification based on Transformer network and based on lifelong learning approach is proposed to address the problems and shortcomings of existing deep learning based methods.1.Transformer-based radio signal modulation mode recognitionAlthough the deep learning method based on convolutional neural network achieves good recognition results,it does not grasp the global information of the signal is very well and has weak parallel computing capability.To address this problem,a deep learning method based on Transformer network is proposed,which uses a self-attentive mechanism to dynamically establish the importance of signal time feature sequences,effectively improving the relationship between sequences and sequences to extract more important features,and the self-attention mechanism in Transformer allows parallel computation at each time step to improve the training speed of the model.Experimental results show that the recognition results of the method on the RML2016.10 a dataset outperform existing methods.2.Lifelong learning-based approaches for radio signal modulation mode identificationThe analysis of existing deep learning-based algorithms for signal modulation mode recognition reveals that they all use fixed data sets for local or isolated tasks,which exists to make it difficult for models to produce more complex and autonomous intelligent behavior.Lifelong learning is a research direction to address such problems,and its goal is to extend the adaptive capability of the model so that it can learn knowledge about different tasks at different moments without forgetting the features of previous tasks.Based on this,a lifelong learning based radio signal modulation mode identification method is proposed.The method identification under different noises as different tasks,and tasks into account the problem of weaker signals with more noise,and the sequence of learned tasks is from signals with less noise to signals with more noise.Meanwhile,a convolutiona lmulti-layer perceptron based neural network is proposed,i.e.,a multi-layer perceptron module is added to the adopted VGG network.The experimental results show that the proposed network has better recognition results compared with other backbone networks;the recognition effect using the lifelong learning algorithm is significantly improved compared with the fine-tuning algorithm.