In a digital filter realized with fixed-point arithmetic, there is a peculiar phenomenon known as limit-cycle oscillation, which is due to roundoff errors. For floating-point arithmetic, it has been conjectured that its amplitude is negligibly small, if it does exist. This paper shows that limit-cycle oscillations can exist in floating-point digital filters and that their amplitude can be large. Also, conditions for the existence of limit-cycle oscillations are derived.
No takes yet. Share an insight, caveat, or question.
T. Kaneko (1973) studied this question.
Synapse has enriched 5 related papers. Consider them for comparative context: