A function calculus for identification and system analysis

Techniques are presented for use in FFT network analyzers to improve estimates of long lags in the impulse response. A key step involves short-term correlation of time-shifted copies of the input-output data. We introduce new Banach signal spaces for the analysis of such data. As a preliminary result, new transfer function estimation algorithms are given with bounds on the effect of undermodelling, noise and lack of empirical data in the H/sup /spl infin// norm. Comparisons with optimality are given using metric-complexity concepts. The algorithms and error calculations are implemented in MATLAB to realize a software network analyzer which includes error bars.