» Download Article

Authors
Mats Ekman; Mei Hong; Torsten Soderstrom

Identifier
10.3182/20060329-3-AU-2901.00022

Index Terms
parameter estimation,least-squares method,noisy input-output systems,separable nonlinear least squares

Abstract
It is well-known that the least-squares identification method generally gives biased parameter estimates when the observed input-output data are corrupted with noise. Previously, an extended version of compensated least-squares (ECLS), based on an overdetermined linear system of equations, was proposed as a method for handling problems where the input and output data are corrupted by white noise. This paper considers the problem where the noise is colored and, thus, extends previous results of the ECLS method. By considering the ECLS problem as a separable nonlinear LS problem, it is shown that the parameters, associated with the noise terms, can be obtained from solving a variable projection minimization problem. The accuracy of the parameter estimates is investigated, and it is also shown that the estimates, under some general assumptions, are consistent.

References

[1] Åström, K. J. and B. Wittenmark (1990). Computer-Controlled
     Systems. Prentice-Hall International
    Editions, New Jersey.
[2] Beghelli, S., R. P. Guidorzi and U. Soverini (1990).
    The Frisch scheme in dynamic system identification.
     Automatica 26, 171-176.
[3] Cramer, H. (1946). Mathematical Methods of Statistics
    . Princeton University Press, Princeton.
[4] Ekman, M. (2005). Identification of linear systems
    with errors in variables using separable nonlinear
    least squares. In: 16th World Congress of IFAC.
    Prague, Czech Republic.
[5] Fernando, K. V. and H. Nicholson (1985). Identification
     of linear systems with input and output
    noise: the Koopmans-Levin method. Proc. IEE,
    Part D 132, 30-36.
[6] Golub, G. H. and V. Pereyra (1973). The differentiation
     of pseudo-inverses and nonlinear
    least squares problems whose variables separate.
    SIAM Journal of Numerical Analysis 10(2), 413-
    432.
[7] Hannan, E. J. (1971). Non-linear time series regression.
     J. Appl. Prob. 8, 767-780.
[8] Levin, M. J. (1964). Estimation of a system pulse
    transfer function in the presence of noise. IEEE
    Trans. on Automatic Control AC-9, 229-235.
[9] Mahata, K. and T. Söderström (2002). Identification
    of dynamic errors-in-variables model using prefiltered
     data. In: 15th World Congress of IFAC.
    Barcelona, Spain.
[10] Söderström, T. and K. Mahata (2002). On instrumental
     variable and total least squares approaches for
     identification of noisy systems. Int. J. Control
     75(6), 381-389.
[11] Söderström, T. (1974). Convergence properties of
     the generalized least squares identification algorithm.
      Automatica 10, 617-626.
[12] Söderström, T. (2002). Discrete-Time Stochastic Systems
     . 2nd edition, Springer-Verlag, London, UK.
[13] Söderström, T. (2006). On computing the Cramer-Rao
     bound and the covariance matrix for PEM estimates
      in linear state space models. In: Proc. 14th
     IFAC Symp. on System Identification, Newcastle,
     Australia, March 29-31.
[14] Söderström, T. and P. Stoica (1989). System Identification
     . Prentice Hall, Hemel Hempstead, UK.
[15] Söderström, T., U. Soverini
     and K. Mahata (2002). Perspectives on errors-invariables
      estimation for dynamic systems. Signal
     Processing 82(8), 1139-1154.
[16] Stoica, P. and T. Söderström (1982). Bias correction
      in least-squares identification. Int. J. Control
     35(3), 449-457.
[17] Zheng, W. X. (1999). On least-squares identification
     of stochastic linear systems with noisy input-output
      data. International Journal of Adaptive
     Control and Signal Processing 13, 131-143.
[18] Zheng, W. X. and C. B. Feng (1989). Unbiased parameter
      estimation of linear systems in presence of
     input and output noise. International Journal of
     Adaptive Control and Signal Processing 3, 231-
     251.