Quantized Filter Analysis
Belle A. Shenoi · 2005
In this chapter, quantization of signal values to a finite number of binary bits is described. Binary number representation in fixed point and floating point form is explained. Signed magnitude form, one's complement and two' complement form of fixed point representation is also explained. Floating point number representation with bias of the exponent and different wordlength for the mantissa and the exponent is discussed. The FDA tool from MATHWORKS is described and is used in investigating many of its features in the design of a filter. This MATLAB tool is used to find a filter structure that realizes a given frequency response with the best wordlength and quantization format, after trying all structures, transfer functions and wordlengths, rounding, truncation, saturation, scaling modes for the filter coefficients. This chapter contains material which is not found in any other text book published so far.