Spectral estimation of spontaneous otoacoustic emissions (SOAEs) depends on the methods used. The performance of different spectral estimation methods may depend on signal-to-noise-ratio (SNR) of SOAE signal and the measuring system. Accurate assessment of spectral characteristics of SOAE signals is needed for providing direct clues as to how the cochlea works. In this study, an average periodogram, a reduced variance estimate, and a model based high-order autoregressive (AR) estimate were applied to estimate the spectra of SOAEs. At higher SNR, the high-order AR spectrum demonstrated the highest frequency resolution owing to avoidance of the energy leakage problem associated with data windowing in the Fourier analysis. At lower SNR, compared to the average periodogram, the high-order AR spectrum showed an almost equivalent frequency resolution in the estimation of the SOAE spectrum, and showed a better performance in the estimation of the noise spectrum. The average periodogram had a reasonably good performance in the spectral estimation of SOAEs with the best algorithm efficiency. The reduced variance estimate of SOAEs gave a smoothed spectrum, which could be used to identify true SOAEs from the noise peaks associated with an inherently bigger variance in the spectral estimation using the FFT algorithm. When the high-order AR model was applied to estimate SOAEs, the analysis of minimum prediction error power revealed that a high-order all-pole process is involved in the generating of SOAEs. [References: 20]
展开▼
