Answer 1—This filter implements hysteresis (i.e., the output value won’t change until the input value is at least two counts different from the previous output value). This is a different type of nonlinearity than the previous filter; this one discriminates based on amplitude rather than duration in time.

Hysteresis is useful when a sequence of samples is toggling between two adjacent values (e.g., the output of an A/D converter that’s sitting right at the threshold between two output values) and you want to clean it up before displaying it on a user interface. One big advantage of this filter over the previous one is that large changes are reflected immediately in the output; there is no time delay.

You have to be careful with this sort of nonlinearity however. Although it completely eliminates low-level noise, it also completely eliminates any low-level signal information that actually might have been useful in some circumstances.

Contributor: David Tweed