Data-driven And Model-driven Recognition Of Electrical Patterns In NR Terminals

With advantages such as low latency,high reliability and large bandwidth compared to 4G,5G networks are gradually replacing traditional 4G networks,and their terminals have already entered daily life on a large scale,showing a spurt of growth in scale.In recent years,smart homes,security and device authentication have also taken on an increasingly important role in daily life.The use of massive 5G terminals for these tasks will greatly reduce device usage and facilitate people’s lives,which is a new challenge and research direction.In this paper,we focus on the problems of RF signal reception,parameter identification,and parameter estimation of RF impairment in the 5G uplink channel,and achieve the extraction of 5G RF signal reception in the additive Gaussian white noise channel,the estimation of RF impairment parameters of different terminals,and the intelligent analysis of different terminal identification and authentication.Firstly,a mathematical modelling of the RF impairment of terminal hardware is proposed,which can be used as a fingerprint feature of the terminal due to the subtle variability between different hardware and used as a basis for identifying terminal devices and completing terminal authentication.Secondly,a correlation-based method is designed for the acquisition of the guided frequency signal.A simple and effective energy detection method is used to perform coarse synchronisation of the 5G pilot signal,followed by methods such as frequency domain correlation matching to obtain a specific format of the pilot signal and perform fine synchronisation.The combination of the two algorithms effectively reduces the complexity of the algorithm,improves the accuracy and convergence speed of synchronisation,and effectively reduces synchronisation errors.Finally,for model-driven terminal electrical pattern identification,Markov Chain Monte Carlo(MCMC)is used to complete the estimation of RF impairment parameters,including IQ imbalance,carrier frequency bias,DC bias and other parameters.A variety of machine learning algorithms are used to complete the classification task so as to perform the classification and identification of terminals,and the Hamiltonian Monte Carlo(HMC)method is used to optimise the parameter estimation process,which greatly improves the convergence speed.For data-driven terminal electrical pattern recognition,the signal is converted into a time-frequency map and the image classification is performed by using a vision transformer with integrated learning to improve the accuracy rate.