Title :
Error analysis of algorithms in linear prediction
Author :
Wang, Yunbiao ; Krishna, Hari
Author_Institution :
Dept. of Electr. & Comput. Eng., Syracuse Univ., NY, USA
Abstract :
Errors caused by the finite wordlength of the computer in several popular algorithms in linear prediction are studied. Both the classical algorithms and the split cases of the classical algorithms are superior to other algorithms since the errors of computation at each stage propagate more slowly than in the Levinson and Schur cases. The results of the Levinson and Schur algorithms are almost the same. For the split cases, the errors accumulate much faster than in the classical situations since the three-term recurrence is used
Keywords :
error analysis; filtering and prediction theory; signal processing; Levinson algorithms; Schur algorithms; computation error analysis; finite wordlength; linear prediction algorithms; signal processing; split algorithms; three-term recurrence; Algorithm design and analysis; Computer errors; Error analysis; Filters; Large Hadron Collider; Lattices; Prediction algorithms; Quantization; Reflection; Stability analysis;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on
Conference_Location :
Toronto, Ont.
Print_ISBN :
0-7803-0003-3
DOI :
10.1109/ICASSP.1991.150752