Research On Adaptive Code And Modulation (AMC) Based On Machine Learning

In order to meet the further requirements of system throughput,transmission rate and transmission reliability in 5G digital communication,Adaptive Modulation And Code(AMC)has been applied in various fields.AMC technology can timely adjust the modulation and coding scheme of wireless link transmission according to the change of communication environment,so as to ensure the quality of communication transmission.At the same time,with the modern society ushered in the era of big data and artificial intelligence,research on the combination of communication and AI(Artificial Intelligence)is gradually becoming a hotspot.This paper studies the AMC based on machine learning,in which the automatic modulation recognition technology and the adaptive modulation coding technology are studied in detail.Automatic modulation recognition based on machine learning can save the signaling overhead of AMC to a certain extent,and AMC based on machine learning can improve the throughput and stability.Firstly,this paper studies the automatic modulation recognition technology based on machine learning for 5G modulation signal,which can reduce signaling overhead for AMC.A new feature combination is proposed for QPSK and its advantages are verified by simulation.Due to the lack of current research and poor performance of high-order QAM modulation recognition,we put forward a modified KNN(k-Nearest Neighbor)based on weighted HOCs feature combination.This algorithm overcomes the defects of existing algorithms by reducing the dimension of high-order cumulant features and combining features with weights.Finally,the simulation results compare the performance of several better algorithms(including constellation clustering and SVM(Support Vector Machine))and the proposed improved KNN algorithm.It is proved that our algorithm is better.Then we studied the AMC based on machine learning.In 5G scenario,the coding and modulation,physical resource calculation,partial bandwidth transmission and TBS selection are studied,and the mapping table of SNR(Signal-Noise Ratio)and CQI(Channel Quality Indicator)in AWGN(Additive White Gaussian Noise)channel is obtained by simulation.In this paper,the CQI index of SNR estimation in the classical AMC scheme is replaced by machine learning algorithm.Three existing technologies are studied and RF-AMC scheme is proposed.Besides,the schemes of KNN-AMC,ANN-AMC,SVM-AMC and RF-AMC are modeled based on 5G modulation and coding parameter,and key parameters of the model are decided through experiment.In addition,the throughput and time complexity of the four schemes are discussed and analyzed,and the feasibility of RF-AMC scheme is verified.The advantages and disadvantages of the four algorithms are introduced.Finally,the adaptive scenarios are analyzed to make the best choice according to the characteristics of different data scenes.