| Abstrakt: |
A procedure is developed to identify frequency domain error bounds using data and deterministic prior information. This prior information consists of system priors, a bound on the system’s (generalized) pulse response, and noise priors, which can be time-domain, cross-covariance and frequency-domain bounds on the noise. The resulting error bounds are non-conservative, i.e. no smaller error bounds can be derived using the data and prior information given. The procedure is a parametric one, the class of systems under consideration is parametrized. For each frequency the actual worst-case transfer function consistent with data and prior information, is calculated with linear programming. Approximate knowledge about pole locations of the unknown system can be incorporated by the use of orthonormal basis functions. |
