Design of Computer-Optimised Signals for Linear System Identification

Abstract

The design of computer-optimised signals is considered. Unlike pseudorandom signals which have fixed spectra, the objective now is to design a signal with a spectrum as close as possible to a specified spectrum. The optimisation algorithms come in different forms, depending on the class of signal, and particularly the number of signal levels. The first class of signals considered is the multisine sum of harmonics signals which can take any value between their minimum and maximum. In contrast, the second class dealt with comprises discrete interval signals, which are either binary or ternary. The third class considered is the multilevel multiharmonic signals which have a small number of signal levels, where this number is specified by the user. It is then shown that it is also possible to combine the advantages of pseudorandom and computer-optimised designs leading to a class of hybrid signals, which are generated as a combination of pseudorandom signals and computer-optimised ones. Finally, the concept of optimal input signals is briefly described where the power spectra are optimised based on initial models to satisfy an application-related objective.