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

A linear, robustly convergent interpolatory algorithm for system identification in the presence of bounded noise is presented. The algorithm converges in the actual, but unknown, system in the frequency domain in the noise-free case and maintains the robust convergence result in the face of bounded noise. This robustness property distinguishes the algorithm from existing linear schemes. A key idea of this robust linear algorithm is that the approximation is separated into real and imaginary parts and into Fejer (Hermite) interpolation. Because of the Fejer interpolation the data points are required at the Chebyshev points