Fixed-point-IIR-filter challenges

摘要

IIR FILTERS CAN MEET HIGHLY SELECTIVE MAGNIIUDE-FREQUENGY-RESPONSE SPECIFICATIONS WITH A LOW-ORDER APPROACH AS SUCH, AN IIR IS OFTEN THE TECHNOLOGY OF CHOICE IN REALIZING FREQUENCY-SELECTIVE TONE DETEC-TORS, NARROWBAND SPECTRAL FILTERS, NOISE REIECTERS, AND DIGITAL CONTROLLERS. THEIR DESIGN, HOWEVER ENTAILS SOME UNIQUE CHALLENGES. The historic role of IIR (infinite-impulse-reponse)-filter-design software is to translate a set of frequency-domain design specifications into a transfer function that is based on recognized IIR-filter models. You can use any of dozens of commercially available software packages to synthesize a transfer function from a set of user specifications. The SPT (signal-processing toolbox) and FDATool (filter-design and -analysis tool) in The Mathworks' (www.mathworks. com) Matlab, for example, contain many of the objects you need to synthesize an IIR-transfer function. These functions include Matlab's Butterworth, Chebyshev I, Chebyshev II, elliptic, Burg-AR (autoregressive), covariance-AR, and Yule-Walker-AR functions. You use the first four deterministic filter classes to synthesize a classic fixed-coefficient filter based on user-specified passband-critical frequencies and maximum attenuation and stopband-critical frequencies and minimum attenuation. The last three design methods synthesize AR, fixed-coefficient, feedback-only IIR filters in terms of measured or desired input/output spectral responses.