Small roundoff noise realization of fixed-point digital filters and controllers

A method for finding digital realizations of discrete-time transfer functions by fixed-point arithmetic is presented. The obtained realizations attain very small roundoff-noise intensity. This intensity is significantly smaller than the intensities that have been achieved so far. The method applies many 0-1 coefficients in the realization. It therefore requires a small number of multiplications and additions, and it is significantly faster than many other realization methods. A detailed comparison is made between this method and other existing methods for digital realization. Two different cases are investigated: a Butterworth filtering realization and the realization of an LQG (linear quadratic Gaussian) optimal controller.<
>