Animproved fitting algorithm forparametric macromodeling fromtabulated data
P. Triverio-t, S. Girivet-Talocia, Michel S. Nakhla · 2008
frequency response isgiven' This paper introduces anewscheme fortheidentification of . 1 multivariate behavioral macromodels fromtabulated frequencydomnain data. Themnethod produces closed-formi paramnetric ex- for different frequencies a -) andparamieter values A Al. pressions that reproduce withexcellernt accuracy theexternalTIhese datacan) be[Ie outcorne ofindependent measurernents or port: behavior ofthestructure, bothasfunctiorn offrequency full-wave simulations performed for fixed values A= A ofthe andoneormiore external parrneters. TIhte niumnerical robustness parameter within therange [A, A]. ofthemainalgorithm isdemonstrated ontwosignificant exam- Thefirst stepintheidentification ofaparametric model ples. E: JWHj, A)from (1)isthedefinition ofitsfunctional form. It turns outthat thechoice ofadirect pararneterization ofpoles anid residues isnotappropriate, duetothehighly non-smooth Model-based design isastandard practice inmanydifferent beIavior ofthie poles withrespect totheparameter A.A suitapplication areas. Frominitial conlcept toproduct finalization, ableformulation forparametric mriodels wasproposed inI1, various modeling steps take place, inorder tochoose thebestbasedonthefollowing general property oflumpedelectrical candidate forprototype construction, testing, anid tinial produc-networks [2]. tion. 'hisisparticularly trueirl thedesign ofelectrical interconnriect structuLres, wheresystem complexity calls forafullyTheoirem 1.LetAbethe value ofaresistor, inductor orcapacautomated design andoptimization workflow. itor inanarbitrary lumrped circuit. T7hen, anynetwork transfer Evenforsimple interconnect stiuctures, manyfree variables Jutnction, including Z,Y,andlSrepresentations. canbewritten mustbedetermined inorder tooptimize thesystem perfor- asaratio ofpolynomials maance. Examples canbegeometrical parameters suchasinE (Q + Q A)s'