Blind Modulation Recognition Technology Of Aliasing Signal Based On Deep Learning In Cognitive NOMA

In order to cope with the rapid growth of communication services and users' demand for wireless spectrum resources effectively,the combination of cognitive radio(CR)and nonorthogonal multiple access(NOMA)technology can provide a new solution.Among them,successive interference cancellation(SIC)is one of the key steps for CR-NOMA to achieve higher spectral efficiency.As a priori information,the modulation type of each user in the system needs to be transmitted to the user from the base station through high-level signaling to realize SIC.However,due to the application requirements of the cognitive scene and the limited signaling cost of the system,users usually cannot obtain the modulation information of NOMA signal,resulting in the failure of SIC.To solve the above problems,this paper proposes the modulation recognition technology route based on blind signal processing in CR-NOMA systems,introduces the deep learning method into the blind modulation recognition,and focuses on the research on the challenges such as aliasing signal interference and the dynamic change of transmission power ratio in the blind modulation recognition of CR-NOMA.Specific work is as follows.Aiming at the problem that multiple aliasing signals in CR-NOMA have strong interference to each other,a blind modulation recognition algorithm of aliasing signals based on recurrent neural network(RNN)was designed by extracting and selecting the high-order cumulants of aliasing signals.On the basis of channel estimation and equalization,the paper adopts the high-order cumulants of different orders of NOMA signals as alternative features.In order to further reduce the amount of data and complexity,this paper uses genetic algorithm to implement feature selection according to the Euclidean distance between different modulation types,and selects several features with high importance as the input of the RNN for classifier training.Simulation results show that the algorithm can achieve good recognition performance with low complexity.On the basis of the above research,a modulation blind recognition enhancement algorithm based on transfer learning(TL)is designed to enhance the adaptability of the algorithm to application scenarios,focusing on the challenge of dynamic change of CR-NOMA transmission power ratio.Considering the similarity of recognition tasks before and after power change,we can effectively use the existing knowledge in the original network to achieve a higher correct recognition rate by retaining part of the structure of the original neural network and combining with a small amount of new sample data.In addition,in order to further improve the performance of transfer learning,data augmentation is proposed for a small number of samples in the new environment to ensure the full training of the classifier.The simulation results show that the algorithm can improve the adaptability of the algorithm effectively.