Research On 5G New Multi-carrier Signals Recognition Algorithms

In non-cooperative signal processing systems,such as cognitive radio and communication reconnaissance,signal identification is an intermediate operation between detection and demodulation,and it also provides a basis for the realization of subsequent demodulation and other processing links.With the complexities of the signal environments and transmission systems,the tasks of signal identification have also become more complicated,which including signal modulation identification,code pattern identification,and multi-carrier signal identification.It should be noted that the orthogonal frequency division multiplexing(OFDM)modulation used in the fourth generation(4G)communication systems has shortcomings due to its high out-of-band radiation,loss of spectrum efficiency,and strict synchronization of receivers.It is difficult to meet the technical requirements of the fifth generation(5G)communication systems.Therefore,several new multicarrier access schems methods have been proposed in the 5G communication system,such as filteredorthogonal frequency division multiplexing(FOFDM)with sub-band filtering,filter bank multi carrier(FBMC)and universal filtered multi-carrier(UFMC).Therefore,for 5G-oriented noncooperative signal processing systems,the identification of new multi-carrier signals has become an indispensable technical link,and the research of related identification algorithms has also become a new topic that needs to be resolved in this field.In order to solve the flaws of the existing convolutional neural network(CNN)-based multi-carrier signal recognition algorithm,such as limited recognizable signal types,strong dependence on training samples,and poor performance under low signal-to-noise ratio conditions,several effective recognition algorithms for multi-carrier signals are proposed in this thesis.They utilize extreme value distribution theory,wavelet transform and neural network models by combination of the traditional statistical theory and emerging machine learning methods.They can effectively identify the multicarrier signals commonly used in 5G,such as OFDM,FOFDM,FBMC and UFMC.The main contributions of this thesis can be briefly concluded as follows:(1)In order to reduce the algorithm's dependence on training samples and to improve the realtime performance of the algorithm,a 5G multi-carrier signal recognition algorithm based on block maximum(BM)and wavelet transform feature is proposed.The algorithm can be divided into two steps: inter-class recognition and intra-class recognition.Firstly,the BM distribution fitting test based on the time-domain modulus square sequence of the multi-carrier signal is utlized to distinguish OFDM-type and non-OFDM-type signals.Then,the Haar wavelet transform of the spectrum of the two classes of signals,and the peak characteristics and the line spectrum existence characteristics of the front-segment spectrum are extracted as features to distinguish the signals within the classes respectively.Simulation results show that when the signal-to-noise ratio is greater than 4d B,the average recognition accuracy of the algorithm is up to about 83%,which expands the recognisable multi-carrier signal scopes of the existing CNN-based identify algorithm,and has a better real-time processing performance without training samples.(2)In order to reduce the dependence of the algorithm on training samples and enhance the recognition performance of the algorithm under lower signal-to-noise ratios,a 5G multi-carrier signal recognition algorithm based on peaks over threshold(POT)extreme value and wavelet transform characteristics is proposed.The algorithm is also divided into two steps: first,the POT distribution fitting test based on the squared modulus of the multi-carrier signal in time-domain is used to identify the OFDM and non-OFDM type signals;then,the POT extreme value characteristics and the line spectrum existence characteristics in the middle section of the Haar wavelet transform of the two classes of signals are extracted respectively to realize the recognition of the signals within the corresponding classes.Simulation results show that when the signal-to-noise ratio is greater than 4d B,the average recognition accuracy of the algorithm can reach more than 92%.In addition,compared with the algorithm based on BM and Haar transformation,the algorithm has a better performance under the lower signal-to-noise ratio,but the computational loader increases about two times.(3)In order to further increase the recognition performance of the algorithm under lower signalto-noise ratios,three types of multi-carrier recognition algorithms based on wavelet transform and neural network are studied respectively.First,the characteristic differences among the wavelet transformations of spectrum modulus of the four types of multi-carrier signals are analyzed.Then,using the characteristic as the input features,and selecting the BP,Alex Net and VGGNet neural networks as the classifiers,three neural network-based multi-carrier signal type recognition algorithms are designed respectively.Simulation results show that when the signal-to-noise ratio is greater than 4d B,the average recognition accuracies of the proposed algorithms can reach more than92%.In addition,compared with the two multi-carrier recognition algorithms based on extreme value theory,the proposed neural network-based algorithms have better performances under the lower signal-to-noise ratios,but computational costs are increase as well.