Parametrization and identification of bilinear systems

In this report, results on observable canonical forms for time-varying systems are extended to treat classes of bilinear systems whose structural properties guarantee that certain observability properties hold, regardless of the choice of input signal. The resulting canonical forms lead naturally to equivalent models which exhibit linearity with respect to (possibly unknown) system parameters, and which can therefore be used for online identification and adaptive observer design. This paper explicitly parametrizes selected classes of models, in both continuous and discrete time, and discusses the requirements for observer design and identification.