1992 ACTmI 13 System Identification via Weighted Subspace Fitting

Ts paper preset anew m4ethd f*r the eiftien of lner systems paratvrise by stac space models. ne method relis ons tie concpt of s in which the goal is to find a particular inpw*/Output -tt model porietensed by the state matrices tat bet fits, in th last-squar sense, at dominant baspaeot of the J mesred ista. Central to this approach is the idw theta weighting w ay be applied to the obsered domisnat suopac in order to emphize certais direction wher thesatio is highest T has the advastage of making the agritm robust to system that are nearly unobscrvahle, or to those whose Astae ap" has not been sufficiestly ercitea Some empiriests are included to evidate thc algorithmad illhutte its advantages oecr previons techniques. Is addition to presting the theo and implemeotation of the -new method this pape alo illtatcs some isteresting coections betwee state -space data models sad those cecountere is Processing te signals recived Ib as array of1 sensors.