Researc On Communication Signal Recognition And Position Based On Deep Learning

In recent years,with the rapid development of radio-electronic technology,wireless network technology gradually covers a variety of applications,but also leads to the difficulty of wireless radiation source equipment authentication increased.There is an urgent need for a kind of devices identification with good real-time performance and high accuracy in Internet of Things(IoT).First,this identification method is the authentication of device identity.The existing signal recognition methods are mostly based on cryptography algorithms and protocols,which are difficult to deploy on the IoT terminals because of their high complexity.The second is the identification of device position information.Currently,the mainstream methods are easy to be interfered by various factors in the environment.The fingerprint-based method has the advantages of strong robustness and high precision,which is the current research hotspot.In order to realize the two key authentication tasks of recognition and position coherently,a deep-learning based communication signal recognition and position method is proposed,which mainly includes the following two key research contents:(1)communication signal recognition method based on deep learning;(2)unknown communication signal position method based on deep learning.1.A hierarchical predictive neural network based communication signal recognition method is proposed.The method will have better robustness to ambient noise by using the normalized characteristic constellation map as the input of the network.A power amplifier based on the memory polynomial mode is created.The power amplifier with the greatest influence is modeled by using the memory polynomial model,and the overcomplicated Volterra series cross terms are deleted,which can better balance the direct relationship between accuracy and complexity.In view of the advantages and disadvantages of both CNN network and Transformer network in feature extraction,the advantages of the two neural networks are complemented by series to build a feature extraction network.After the feature extraction network extracts the feature,considering that the conventional deep learning methods only use the characteristics of the last layer of the network for classification,which cannot reflect the problem that different layers of the network have different sensitivities to different types of data.This thesis extracts the output of different network layers for communication signal recognition,and then synthesizes the output results of each layer by voting algorithm as the final prediction.The method in this thesis is 2.5dB ahead of the traditional CNN network to achieve 85%recognition accuracy.2.A two-path prediction based communication signal position method is proposed.This thesis design experimental scene and collects the actual RF fingerprint data for communication signal position,and the fingerprint database is established.During this period,the collected data is filtered to reduce the influence of accidental errors on the experiment,and the possible interference in the process of data acquisition is studied and eliminated.Aiming at the problem that CNN network and LSTM network have different emphases when extracting features of different dimensions,the two neural networks are connected to complement each other’s advantages,and the feature extraction network is constructed.Since the data collection area in this thesis is a square area,considering the symmetry of the coordinate,the two dimensional coordinate is divided into two-path prediction,and the complexity of classification task is reduced by two-path prediction.Considering the continuity of position information,the output of Softmax function is optimized.The final experimental results show that the positioning accuracy is improved by 10%compared with the traditional method.