Abstract
This thesis describes a novel tree search, autoregressive- model-based interpolation method for recovering lost packets of DPCM encoded speech. This method is described as an evolution of an adaptive AR model-based interpolation method presented by Janssen, Velhuis, and Vries. The tree search method is contrasted to a minimum mean squared error interpolator described by Jayant and Christensen that is extended, herein, to 6 coefficients. Finally, through simulations with real speech, the new tree search interpolator is compared to an embedded DPCM system with a probability greater than zero of losing bits in the predictor loop. It is shown that when this probability is large enough an interpolative recovery method performs better than EDPCM. Finally, it is shown that the tree interpolator performs better than the Jayant interpolator in an SNR sense as well as providing flexibilities that the Jayant interpolator does not have.
Erhart, Wesley Ronald (1993). Odd-even sample packet recovery using an AR model based tree search interpolation method. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1993 -THESIS -E68.