Non-Stationary Signal Classification Using Joint Frequency Analysis

Abstract

Time-varying short-term spectral estimates have been successfully applied in many classification tasks. However, they are still insufficient for many non-stationary signals where time-varying information is useful. In this paper, we propose to improve the deficiencies of current short-term feature analysis by adding information to describe the time-varying behavior of the signals. Our proposed method which is motivated by the human auditory system can be applied to several non-stationary signal types. Real world communication signals were used for experimental verification. These experimental results, assessed with a conventional probabilistic classifier, showed significant improvement when the new features were added to short-term spectral estimates.

Modulation-Scale Analysis for Content Identification

“…This nonstationary information can represent various quantities (e.g., the symbol rate of a digital communications signal [19]). An example by a male speaker is illustrated in Fig.…”

“…If the time shift is relatively small compared to the integration time in (8), then the estimation is approximately invariant to the effect (18) Compared to CMS, cepstral features are the representation of time and time-lag; therefore, CMS can only remove a slowly varying convolutional noise, not time distortions. Equation (19) illustrates the effect of time-scale modification and time shift to CMS (19) For the purpose of signal classification, we define a modulation-scale feature matrix , where the modulation frequencies are estimated by wavelet filters with and representing the dyadic scale and discrete acoustic frequency, respectively. To initially demonstrate the advantages and generalization of the proposed features and channel compensation, subband normalized modulation-scale and cepstral energy features with mean subtraction of audio signal were compared using a 4-s clip of music.…”