M. Cehan-Racovita (Romania)
Identification, variable parameters, noise filtering.
Based on a novel filtering principle, an identification method suitable to time-depending linear systems is developed. Any conventional stochastic treatment will be given up. One takes into consideration the very low probability of certain events (occurring during process decomposition); as well as the idea that any stochastic process over a finite time interval may be considered a deterministic function. In contrast to Fourier’s series a polynomial-exponential decomposition of the process is accomplished. The produced parametric spectrum gives rise to a sharp characterisation of the noise and test signal. Digital filters based on functional properties of process components are built up. These ones tuned on two spectral parameters (instead of one as usual) ensure a high selectivity. The computing algorithm embodies linear modules only, so that a cut down of time processing is achieved. On the other hand, the procedure involves a short incipient segment of the output, thereby a parameter estimation unaltered by the system drift will be performed.
Important Links:
Go Back
