TFR and ASTFT-based IFE

Post date: Jul 24, 2014 10:2:57 PM

Time-frequency representation (TFR) is a useful tool to study the characteristics of signals in many applications. 

Cohen's class of bilinear time-frequency distributions is a natural framework for analyzing non-stationary signal using TFR.

However, in practice, this form of TFR often suffers from the undesirable cross-terms. A TFR that is more common in practice is the short-time Fourier transform (STFT). [1]

IFE is instantaneous frequency estimation [2]. Simple (without tracking) STFT-based IFE returns the frequency index with the maximum power at each time frame. 

For non-stationary environment, this might cause noisy and bursty estimates. 

Hence, tracking, usually with an HMM, will help smooth it out over the time axis, providing more reliable estimates.

ASTFT-based IFE is IFE on ASTFT. The adaptive part in ASTFT stems from the fact that a different window can be used at each time frame.

There are many different adaptation rules that can be applied with the same ASTFT framework, most notable are adaptation rules with different window set, whose elements are different in type and length. 

In addition, the adaptation can also be different in the criterion it uses, most notable are the maximum correlation and maximum concentration rules. 

[1] Baraniuk, Richard G., and Douglas L. Jones. "A signal-dependent time-frequency representation: optimal kernel design." Signal Processing, IEEE Transactions on 41, no. 4 (1993): 1589-1602.

[2] Kwok, Henry K., and Douglas L. Jones. "Improved instantaneous frequency estimation using an adaptive short-time Fourier transform." Signal Processing, IEEE Transactions on 48, no. 10 (2000): 2964-2972.