Distortion Product Otoacoustic Emissions (DPOAE) offer great potential for hearing diagnosis, but are complicated by the interaction of components generated by different mechanisms. Separation of these components from DPOAE measurements may allow conclusions to be drawn about the functionality of these separate mechanisms of a cochlear. However, the signal processing methods for performing this separation are imperfect. Existing methods are based on time windowing of DPOAE generated from frequency sweep stimuli.
This paper presents a method in which the entire spectra of both distortion (D) and reflection (R) components are simultaneously estimated. This approach has several advantages. Firstly, the method removes the need for a compromise between frequency precision and signal to noise ratio. Secondly, the method can be made to include models of the stimulus signals, so that the stimulus does not strongly interfere with the estimation process. Thirdly, the method can be arranged to make efficient use of data that has been corrupted by measurement artefacts. Fourthly, the method can be easily adapted to track DPOAEs that are changing in response to chemical or acoustic treatments.
The basic modelling assumptions made are that the sum of R and D measurements can be represented as the sum of convolutions with the stimulus signal, the frequency representation of the D component is more smooth than the R component, and that a reasonable estimate of the noise level in the signal is available. These assumptions are combined into a linear convex problem. In this paper we compare the proposed approach with three other methods. While it is not superior to the earlier methods at every frequency, it does offer some improvement, particularly with regards reducing the contamination of D by R.
