基于多任务学习的跳频调制方式识别与信噪比估计方法

打开文本图片集

中图分类号：TN911.72-34 文献标识码：A 文章编号：1004-373X（2026）01-0066-07

Frequency hopping modulation mode identification signal-to-noise ratio estimation method based on multi-task learning

，WANGHao1，²， （1.SchoolEnginering，，Oo94，China; 2.，，China）

Abstract：Animprovedmethodisproposedtoeliminatetheseesawefectinmulti-tasklearningFHsignalrecognition improve thegeneralizationability modelstrained with IQsignals.TheCGCmulti-task network framework combined with largeconvolution kernelstructuralreparameterization techniquesareadopted toenhancetheacuracyFHsignal modulationrecognitionSNRestimation.Thehardparametersharingisemployedforthemulti-tasknetworkarchitecture， the networkchannelsareclasified intoexpertchannelssharedchannels.Inaddition，the MobileBlock layers thatnclude largeconvolutionkernel structuralre-parameterizationresidual structuresareintroduced.IncomparisonwiththeMMOE structural model，theproposedmethodachieveshigherclassificationaccuracyforFHsignalmodulationrecognitionlower meansquarederror（MSE）forSNRestimation.Experimetalresultsdemonstratetheapplicationpotential theproposedmethod in modern military communication confrontations.In addition，the proposed method is robust for FH signal recognition parameter estimation.

Keywords：FHsignal；modulationrecognition;SNRestimation;multi-tasklearning;largekernelconvolution;structural reparameterization

0 引言

跳频信号具有低截获概率以及很强的抗干扰能力，因此在军事无线电通信中，跳频技术被广泛采用，既能提高通信的保密性与抗干扰能力，又能防止敌方的截获与干扰，极大地增强了作战能力。（剩余8206字）