基于卷积神经网络的调制样式识别

DOI10．19392/j．cnki．1671-7341．201910211050000。。1。。。、、1112。1。st()=∫xa()ωt－ada1ωxst()。2。st()=x*ω()t()=∑#a=－#xa()ωt－a()2ω。IK3。Sij()=I*K()ij()=∑m∑nImn()Ki－mj－n()3pooling。。。4。ulj=fβljdownxl－1j()+blj()4uljljxl－1iβblj。downxl－1jn×nSUM、MAX。2。。2。2CNN2Dropout3。ＲeluSoftMax。32TimothyJO＇SheaＲML2016．10a_dict．pkl4。50%50%150Adam。tensorflowNVDIACudaGPU。3。3ＲML2016．10a_dict．pkl45。1．D．201614-15．2．．201537091212-1217．3N．SrivastavaG．HintonA．KrizhevskyI．SutskeverandＲ．Salakhutdinov．DropoutasimplewaytopreventneuralnetworksfromoverfittingJ．JournalofMachineLearningＲesearchvol．15no．1pp．1929-19-582014．4TJ．O＇Sheahops//radioml．com/datasets/radioml-2016-04-dataset/．5Azzous．E．ENandi．A．K．AutomaticidentificationofdigitalmodulationsJ．SignalPocessing．199547155-69．。14220194