More choices of filter design
angus-g opened this issue · 3 comments
The default choice of filter may not be suitable for all situations, or may be overkill in some cases. If we're taking the centre point of the filtered timeseries, do we need a high-order Butterworth filter? We may get better frequency separation, and still avoid ringing issues by simply using a step filter.
Related to #5.
The step filter mentioned above:
sharp_filter = signal.firwin(
N, cutoff, window="boxcar", pass_zero="highpass", fs=1/output_dt
)
f.inertial_filter = (sharp_filter, [1.0])Given that the refactor in #21 defers application of the filter to the filter object itself, we can support both time- and frequency-domain filters. In particular, the step filter is pretty trivial in frequency space, and is less likely to cause strange issues like we see with the time-domain step filter.
In addition to/instead of this, it would be nice to support lowpass as well as highpass filters.