基于自适应空间通道收缩网络的自动调制识别算法

打开文本图片集

中图分类号：TN911.3-34 文献标识码：A 文章编号：1004-373X（2026）07-0012-07

引用格式：，等.基于自适应空间通道收缩网络的自动调制识别算法[J].现代电子技术，2026，49（7）：12-18

Automaticmodulation recognitionalgorithmbased onadaptive spacechannel shrinkagenetwork

GAOShaoyuan，GUOWenpu，KANGKai，SHIHao （CollgeofCombat Support，RocketForceUniversityofEngineering，Xi'an71Oo25，China）

Abstract：Inviewoftheinsuficientfeatureleaning inautomatic modulationrecognition（AMR）algorithmatlowsignal-tonoise ratio （SNR），thispaperproposesanAMRalgorithmbasedonadaptivespacechannelshrinkagenetwork.Thealgorithmismainly composedof adaptivespace channel shrinkage （ASCS）module，multi-scaleconvolution（MC）module，residual moduleand multi-headatention（MHA）module.TheASCSmoduleisusedtoextractfeaturesinthespatialandchanneldimensions，andthe improvedsoftthresholdfunctionisusedtoshrinkthefeaturestoreducenoisefeaturesandretainusefulfeatures，soasto improvethefeatureprocesingabiltyof thenetwork.Comprehensiveanalysisofexperimentalresultsshowthattheimprovedsoft thresholdfunctioncanprocess thefeatures beter.Theaveragerecognitionaccuracyof theproposedAMRalgorithmis62.94% onthe publicdatasetRML2016.10a，and 64.79%onRML2016.10b.Incomparisonwith theexisting AMRalgorithm，the proposedalgorithmcanachieve higheraccuracy.Inconclusion，itprovides anefective method forfeaturelearningatlow SNR.

Keywords： AMR; shrinkage network; threshold processing; channel feature; spatial feature; deep learning

0 引言

随着通信技术的快速发展，自动调制识别（AutomaticModulationRecognition，AMR）可以在无先验信息的情况下检测截获信号，并且为解调提供必要的信息，广泛应用于认知无线电领域1，对于信号的解调和电子对抗中的信号分析等任务都具有重要意义

传统的AMR研究可分为基于似然（Likelihood-based，LB）[3-4]和基于特征（Feature-based，FB）[5]的AMR两类。（剩余11316字）