Research On Modulation Recognition Method Of Communication Signal Based On Incremental Learning

Automatic modulation classification（AMC）is a technology for modulation identification at the receiving end when the modulation class of communication signal is unknown.It can provide a basis for the demodulator to select the demodulation algorithm correctly and ensure that the receiver can obtain useful information.With the increasing complexity of electromagnetic influence in modern communication environment and the explosive growth of communication service data,modulation recognition system needs to learn new modulation categories in the process of data increment,adapt to the changes of communication environment and gradually obtain stronger modulation recognition ability.It has important research significance in modern communication scenarios to ensure that the modulation recognition system can learn new knowledge incrementally and continuously to adapt to the complex and changeable communication environment.Based on the signal modulation theory,this thesis researches the modulation classification of communication signal based on incremental learning to solve the above problems.The main research contents include the following aspects:（1）Firstly,based on the traditional modulation recognition methods and constellation diagram theory,a new signal representation method is introduced to obtains grid constellation diagram and color constellation diagram through special processing.This thesis could provide technical support for the following research.Compared with ordinary constellation diagrams,this method retains more complete characteristics of the signal,thereby improving the accuracy of modulation identification.Secondly,when the existing modulation recognition system encounters a new modulation category,it cannot incrementally learn the recognition ability of the new category,and people can only retrain the entire model.Above,by selecting balanced sample subsets and adding correction layers,the modulation recognition of communication signals based on class incremental learning is completed,guaranteeing that the modulation recognition system can incrementally learn new modulation classes under the premise of less computation and less storage pressure.（2）In practical applications,it often occurs that the modulation type of the signal remains unchanged but some characteristics of the signal change,resulting in a large drop in the modulation accuracy.To solve this problem,the signal-to-noise ratio is selected as a control variable,and a modulation recognition method based on sample increments is designed under the situation of rapid decline of signal-to-noise ratio（SNR）.So that the system can learn the characteristics of new samples incrementally and do not affect its memory of old samples as much as possible.Through this method,the recognition accuracy of signals with low signal-to-noise ratio can be improved.At the same time,it can also greatly reduce the training time and training cost,and improve the flexibility and scalability of the model.This thesis can also be extended to the scene where the recognition rate decreases due to the change of other characteristics of the signal,so that the system can adapt to various complex communication environments.