Comparison of Principal Components Analysis on Linear and Logarithmic Magnitude of Head-Related Transfer Functions

Principal components analysis (PCA) is widely used in compression of head-related transfer function (HRTF) database. In practice, PCA is often performed on linear or logarithmic magnitude of HRTFs. These two PCA models (Linear-PCA model and Log-PCA model) were compared in this paper. Cumulative Variance Percentage, Signal-to-Distortion Ratio and Spectral Distortion were used as criterions in comparison. Results show that Cumulative Variance Percentage is inadequate to evaluate the two models, while Signal-to-Distortion Ratio and Spectral Distortion may lead to contrary conclusions. Finally, monaural loudness spectra were calculated and the results show that the Linear-PCA model is superior at most of sound source positions.