Abstract
There are two typical schemes for constructing the continuous-time adaptive identifier based on the equation error principle. One, which is called Scheme I in this paper, employs (n-1)-th order state variable filters for an n-th order plant model and the other, called Scheme II, uses n-th order filters. This paper discusses robustness properties of these schemes in the presence of parasitics and output noise. As a consequence, the following points are clarified: 1) Scheme II is more robust than Scheme I in the sense that the steady-state variation of the identification result is small. 2) Addition of the proportional term to the integral adaptive law is effective for improving robustness properties in the above both schemes. 3) It is profitable to adopt only the integral output as the identification result in place of the sum of the integral and proportional outputs when the integral plus proportional law is used. Furthermore, a new robust scheme with higher order state variable filters is also proposed here.
