Adaptive chirplet transform for the analysis of visual evoked potentials.

Description

Visual evoked potentials (VEPs) are electrical signals measured on the surface of the scalp in response to rapid and repetitive visual stimuli. These signals possess complex time-frequency structures and are difficult to characterize with conventional methods. In this work, we propose a new approach based on the adaptive chirplet transform (ACT) to represent a complete VEP response from the transient to the steady-state portion. Our implementation of the ACT involves both a windowed and a non-windowed approach. The non-windowed ACT employs a coarse-refinement algorithm to estimate multiple chirplets under low signal-to-noise ratio condition. The method decomposes VEPs into chirplet basis functions with four adjustable parameters (i.e., time-spread, chirp rate, time-center and frequency-center). We show how these parameters can be used to separate the transient from the steady-state portions of the response, arid that as few as three chirplets are required to represent a complete VEP signal. In the windowed ACT method, the signal is partitioned into equal-length non-overlapping segments before estimating one chirplet from each segment. The concept of the optimal window length with reference to the windowed ACT method is proposed and calculations are made in terms of the signal characteristics. Both the windowed and non-windowed methods reveal a similar pattern of the VEP response---that a short transient VEP precedes the steady-state VEP. It is shown, however, that the computational time of the windowed method is significantly lower. Finally, we demonstrate that the adaptive chirplet spectrogram (ACS) offers a clearer visualization of the time-frequency structure of the VEP signal than the conventional spectrogram, because the ACS avoids cross-term interference in the time-frequency plane. Possible applications of VEP chirplet analysis to on-line signal classification and the technical limitations of the ACT approach are also discussed.

Subjects

Links

• URL