Research into Modulation Recognition Algorithm Based on Attention Mechanism
In contemporary communication technology,modulation identification occupies a crucial position.Especially in non-cooperative communication scenarios,its importance is particularly sig-nificant.Applying deep learning to modulation identification has become a research hotspot,and with the in-depth study of deep learning,it becomes difficult to improve the accuracy of modulation identification through optimizing the network performance.This paper chooses AlexNet network as the benchmark network,and proposes an improved squeeze and excition(SE)module,which takes both the absolute and relative importance of the channel into account,and then realizes the re-cali-bration of the channel importance through the excitation,improves the ability of capturing locally important information.Using this network to identify eleven signal modulation types,the accuracy is improved by about 4％compared to the network without the added module,and the overall iden-tification rate reaches about 86％,meanwhile the amount of calculation is greatly reduced.The ex-periments confirm the optimization effect of the improved channel attention module on the network performance.This method has research value and significance for the subsequent application of deep learning networks based on the attention mechanism in modulation recognition.
万方数据
维普
