Signal complexity estimation using time-frequency Short-term entropy
Sučić Viktor, Nicoletta Saulig, B. Boashash · 2015
This article has presented a method for quantifying the information on the local number of components present in a nonstationary signal using the R´enyi entropy of the signal TFD. The global R´enyi entropy can be used reliably only when the signal components have same time duration, equal amplitude and the reference signal is known in advance. However, the entropy estimation over a short time interval for components with different time and/or frequency supports can locally be compared to the local entropy of an arbitrarily generated analytic signal. Furthermore, the counting property of the short-term R´enyi entropy when extended to nonstationary signals offers important new insights on the signal structure, quantifying the local number of components present in the signal. To efficiently deal with the problem of estimating the number of components presenting different amplitudes, an iter- ative algorithm has been proposed. Simulations performed on synthetic signals in moderate noise and real data show that the obtained results can be used as reliable inputs in subsequent blind source separation procedures, as well as key information in dynamic memory allocation in multicomponent IF estimation applications.