An adaptive time-frequency representation using a cone-shaped kernel

Cone-kernel time-frequency distributions (CKDs) preserve the outer hull of signal time support in the time-frequency plane, and appear promising in many applications. The authors present a technique for adaptively optimizing the performance of the cone-kernel distribution by varying the cone length in response to changing signal structure. The result is a time-varying spectral estimate that can track signal changes quickly in time, while still allowing the use of windows long enough to give good frequency resolution. The adaptive CKD generally provides much better performance than fixed-length CKDs, and requires only slightly more computation than a fixed-kernel distribution.<>