| Abstract: | Deep learning (DL)requires massive volume of data to train the network. Insufficient trainingdata will cause serious overfittingproblem and degrade the classification accuracy. In order to solve thisproblem, a method for automatic modulationclassification ( AMC) using AlexNet with data augmentation was proposed. Threedata augmentation methods isconsidered, i. e. , random erasing, CutMix, and rotation. Firstly, modulatedsignals are converted intoconstellation representations. And all constellation representations aredivided into training dataset and test dataset. Then,training dataset are augmented by three methods. Secondly, the optimal value ofexecution probability for randomerasing and CutMix are determined. Simulation results show that both of themperform optimally when executionprobability is 0.5. Thirdly, the performance of three data augmentation methodsare evaluated. Simulationresults demonstrate that all augmentation methods can improve theclassification accuracy. Rotation improves theclassification accuracy by 13.04% when signal noise ratio (SNR) is 2 dB. Amongthree methods, rotationoutperforms random erasing and CutMix when SNR is greater than - 6 dB. Finally,compared with other classificationalgorithms, random erasing, CutMix, and rotation used in this paper achievedthe performance significantlyimproved. It is worth mentioning that the classification accuracy can reach 90.5%with SNR at 10 dB. |
