A measurement method based on an improved version of the chirplet transform for instantaneous frequency estimation

Abstract

A measurement method for instantaneous frequency estimation is presented in the paper. The method is based on the use of the chirplet transform, a linear time-frequency representation (TFR) allowing additional modifications of each cell on the time-frequency plane with respect to other TFRs. In particular, a modified version of this transform is here proposed; a bending effect can further be imposed on the cells. Thanks both to this feature and a suitable measurement procedure, properly set up by the authors, the method grants both a satisfactory accuracy in reconstructing the instantaneous frequency trajectory of monocomponent signals and an good resolving capability in the analysis of multicomponent signals whose instantaneous frequency trajectories are strongly nonlinear and very close to one another. Details concerning the theory underlying the chirplet transform and the proposed modified version are first given. Then, the measurement procedure for optimizing the values of the parameters of the transform according to the local features of the analyzed signal is described in detail. Finally, the results of several experimental tests conducted both on monocomponent and multicomponent signals are presented; advantages over other solutions are also highlighted.