Research On Seven Kinds Of Wireless Signal Simulation And Recognition Algorithm

As the most eye-catching fifth-generation mobile communication technology in the wireless communication technology goes to commercial use,the future wireless communication network will move towards the era of interconnection of all things.This means that the number of devices connected to the wireless communication network will increase exponentially,and the types of wireless communication standards used will also be various,and the aerial radio signals will become more and more complex.The efficient identification of wireless signals has become a hot spot in radio monitoring research.The research content of this article is mainly divided into the following three aspects:First: At present,researchers on wireless communication signal identification only conduct research on several types of signals such as GSM,UMTS,and LTE.The signal types involved are few,and the type of fading channel considered is single.To this end,this article takes common wireless signals in recent years as the research object,and uses Matlab to construct a large-scale wireless standard signal data set containing 7 categories and 22 types,including frequency domain data sets and time domain data sets.At the same time,the data set constructed in this paper uses three channel models,namely AWGN channel,Rayleigh fading channel model and Rice fading channel model,and considers a wide range of signal-to-noise ratio from-10 d B to 20 d B to study different The influence of the channel on wireless signal recognition.The simulation signal and data set generation code of this article has been openly shared on Git Hub for subsequent researchers to study the recognition algorithm.Second: For frequency domain data sets,the recognition effects of different models are studied,including image recognition-based models,machine learning-based sequence recognition models,and deep learning-based sequence recognition models.In order to further improve the accuracy of the recognition algorithm,this paper combines the residual network and the Inception network to build an improved CNN model: Res Net,Inception and Inception-add.The experimental results show that the highest average accuracy rates of the image recognition model,the sequence recognition model based on machine learning,and the sequence recognition model based on deep learning under the fading channel are 78.76%,77.07% and 84.02%,respectively.In the Rice fading channel,the average accuracy of the improved Inception-add model is 3.56% higher than that of the original CNN model.Third: For the time domain data set,the recognition effect of the above model is studied.The experimental results show that the highest average accuracy rates of the image recognition model,the sequence recognition model based on machine learning,and the sequence recognition model based on deep learning under the fading channel are73.51%,31.89%,and 90.52%,respectively.In the Rice fading channel,the average accuracy of the improved Inception-add model is 7.12% higher than that of the original CNN model.