A linear interpolatory algorithm for robust system identification with corrupted measurement data

Frequency domain identification of hammerstein models

“…Then, the obtained estimate satisfies as in the frequency range of interest. This algorithm is linear and see [3] for details. • If the whole frequency range is interested, it is well known [9] that there does not exist any linear algorithm which ensures the robustness in the presence of III TRUE VALUES AND THE ESTIMATES OF 'S TABLE IV TRUE VALUES AND THE ESTIMATES OF = ( ; ; ; ) small but nonvanishing errors in the point estimation of .…”

Robust identification from band-limited data

“…for all f 2 A( ): (1) This convergence requirement corresponds to a continuity property of the model f N with respect to the number of measurements and the noise level, as explained in Remark 1 below. To approach this problem, a two-stage algorithm has been found useful [8], [9], [11], [12].…”

“…In this connection, some work on band-limited identification has been published by Bai and Raman [1] in which they essentially approximate separately the real and imaginary parts of the transfer function by polynomials over the frequency interval , plugging 0018-9286/97$10.00 © 1997 IEEE in some arbitrary polynomial weight of sufficiently high degree to become the denominator off the approximant so as to end up with a stable and proper model. In doing so, they are not concerned about controlling the behavior of the set and, since their scheme is (real) linear, it is a routine matter to check, by the same arguments as in [12], that their sequence of estimates is unbounded outside for almost every noise in l 1 (i.e., for every noise sequence in a set of second category in the sense of Baire).…”