Characterization of the quantization properties of similarity-related DSP structures by means of interval simulations

This paper presents a novel approach for the evaluation of the finite word-length performance of DSP structures. Realizations are simulated using interval arithmetic with support for feedback loops, thus allowing fast characterization of the overflow behavior and the quantization effects. When this technique is applied to unquantized and quantized filter realizations (linear case), previously obtained results, such as the improved performance of the normal form, are readily confirmed. When this technique is applied to nonlinear systems, it can be used to quantitatively measure the quantization effects and to provide guaranteed bounds for all signals, thus limiting the maximum amplitudes of possible limit cycles.