Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 225

Moleti, 2002, Jedrzejczak et al., 2005]. OAE signal is a basic test of hearing impair-
ment in small children and a lot of attention has been devoted to clinical significance
and determination of latencies in neonates, e.g., [Tognola et al., 2005, Jedrzejczak
et al., 2007].
In the research concerning the mechanisms of OAE generation the relation be-
tween frequency and latency of components is one of the important tests of the mod-
els, for example, the scale-invariance hypothesis [Talmadge et al., 2000] leaded to
prediction of inverse relation between frequency and latency. However, this kind
of relation was not confirmed experimentally and exponential dependence with dif-
ferent values of exponents was reported [Sisto and Moleti, 2002, Tognola et al.,
1997, Jedrzejczak et al., 2004].
The non-stationary character of OAE and rising interest in frequency-latency
dependencies promoted the application of time-frequency methods to this signal.
Several methods have been devised to estimate the time-frequency distributions of
OAEs, including short-time Fourier transform [Hatzopoulos et al., 2000], minimum
variance spectral estimation [Zhang et al., 2008] methods based on the Wigner-Ville
transform [Cheng, 1995], or Choi-Williams transform [Ozdamar et al., 1997]. How-
ever, the last two methods are biased by the presence of the cross-terms (Sect. 2.4.2).
One of the first applications of wavelets to the analysis of OAE concerned time-
frequency decomposition of click evoked and synthesized emissions [Wit et al.,
1994]. Both continuous [Tognola et al., 1997] and discrete [Sisto and Moleti, 2002]
WT were used. For the purpose of improving pass/fail separation during transient
evoked otoacoustic emission (TEOAE) hearing screening, the method which com-
bined signal decomposition in scales by discrete WT, non-linear denoising, and
scale-dependent time windowing was used [Januauskas et al., 2001]. WT was also
applied for extraction of instantaneous frequencies of OAE [Delprat et al., 1992]
and for construction of multiscale detector of TEOAE [Marozas et al., 2006]. In the
above quoted contribution the detector performed adaptive splitting of the signal into
different frequency bands using either wavelet or wavelet packet decomposition. The
authors reported that the method performed significantly better than existing TEOAE
detectors based on wave reproducibility or the modified variance ratio.
OAEs are a superposition of components characterized by specific frequencies
and latencies. The dependence between their frequency and latency have been ex-
tensively studied in the context of verifying hypotheses concerning the mechanisms
of OAE generation. To this avail the latency has to be determined. The problem of
the identification of the OAE latencies was approached by discrete and continuous
wavelet transform. The method based on the combination of spectral analysis and
WT was proposed by Sisto and Moleti [Sisto and Moleti, 2002]. The limitation of
WT in this respect is octave-band resolution, which influences the accuracy, espe-
cially for high frequencies. In order to surmount this difficulty the authors proposed
a method relying on visual comparison between wavelet data and TEOAEs spectra,
followed by identification of the wavelet contribution to a given spectral line. How-
ever, as was pointed out in [Moleti and Sisto, 2003] wavelet transform has a tendency
to systematically underestimate the slope of the latency-frequency relation.
Substantial progress in the OAE field was achieved by the introduction to its anal-