RFDynamic Behavioral ModelSuitable for GaN-HEMTDevices
M. Pirolal · 2006
This paper presents anewRFdynamic behavioral meaning thatM delaysamples foreachinput variable are modelbased onaneural network (NN)approach suitable foraccounted for,andtherefore, thatatotal ofteninputs is FETdevices inawiderange ofworking classes, andcapable to obtained. Thenumberofhidden neurons hasbeenchosen identify thedevice response, through thetraining procedure, for equal totheinput number, andtheoutput layer isa linear a widerangeofinputpowerlevels. Thepresented modelhas combination ofthenonlinear neurons.Thesystem output are beeneffectively applied toGaN-based devices at1GHz,workingthetimesamples oftheportcurrent IDS orIGS* InFig. 1 a inclass AandB. schematic picture oftheneural network isshown.Separated Indexterms-Dynamic behavioral model, Neural Network, neural networks havebeentrained forIGSandIDS, because GaNdevice. morecomplex networks aretrained hardly, especially withthe Levemberg-Marquardt algorithm. I. INTRODUCTION bh NNsusages arebecoming really attractive forRFblack-box VGs(n) device modeling duetotheir flexibility. Infact, they canbe directly trained on thedevice nonlinear measurements, without theneedneither oftheequivalent circuit topology, d