DocumentCode :
2337334
Title :
Synthesis of minimum roundoff noise structures for extended state-space digital filter implementations
Author :
Bomar, Bruce W. ; Smith, L. Montgomery
Author_Institution :
Dept. of Electr. & Comput. Eng., Tennessee Univ., Space Inst., Tullahoma, TN, USA
fYear :
1990
fDate :
11-13 Mar 1990
Firstpage :
531
Lastpage :
535
Abstract :
An algorithm for minimizing roundoff noise in extended state-space (e-state) realizations of recursive digital filters, where the order of the e-state equation is 2, is developed. It is shown that previous efforts to minimize roundoff noise in e-state structures have not provided a global minimum. The algorithm presented applies an unconstrained transformation matrix to an arbitrary starting state-space structure to produce an intermediate structure. A second matrix transforms the intermediate structure to e-state form. A conjugate-gradient optimization scheme is used to determine the coefficients of the first matrix that minimize the roundoff noise gain of the e-state structure produced by the second transformation. A numerical example illustrates that orders-of-magnitude improvement over previous results can be achieved with this approach
Keywords :
digital filters; filtering and prediction theory; matrix algebra; roundoff errors; signal processing; state-space methods; conjugate-gradient optimization scheme; e-state realizations; extended state-space digital filter implementations; minimum roundoff noise structures; noise structure synthesis; recursive digital filters; second-order equation; unconstrained transformation matrix; Computational complexity; Difference equations; Digital filters; IIR filters; Noise reduction; Nonlinear filters; Signal processing algorithms; State-space methods; Transfer functions; Transforms;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
System Theory, 1990., Twenty-Second Southeastern Symposium on
Conference_Location :
Cookeville, TN
ISSN :
0094-2898
Print_ISBN :
0-8186-2038-2
Type :
conf
DOI :
10.1109/SSST.1990.138203
Filename :
138203
Link To Document :
بازگشت