ABSTRACT

This chapter deals with the effects of finite word-length in 2-D digital filters. It introduces the quantization errors by the use of finite word-lengths for the representation of coefficients and signals are examined and methods for their computation and characterization are considered. The chapter provides a method for the minimization of output round-off noise in 1-D state-space structures due to Mullis and Roberts based on a state-space noise model proposed by Hwang is extended to the case of 2-D digital filters. It also provides two types of parasitic oscillations, namely, quantization and overflow limit cycles, are examined and conditions for their elimination are delineated. Coefficient-quantization errors are due to the representation of the filter coefficients by a finite number of digits. Quantization of numbers can be carried out by rounding or truncation, and the magnitude of the error introduced tends to depend on the type of arithmetic and on the number representation.