6x oversampling compared to the highest frequency of interest is fairly minimal for time domain work. The rule of thumb I've encountered must often is to aim for at least 10x with 20x being preferable.
I've been known to run through 3 stages of software filtering:
1. A narrow median filter (left rank=right rank=1 sample) to eliminate "spiky" noise data. I run this first to suppress such outliers on equipment that needs it.
2. A regular IIR or FIR filter, most commonly Butterworth for its flat and predictable passband gain. This suppresses higher frequencies that are captured but not of interest. (Note: it's important to do median filtering *before* IIR or FIR filtering or else the spikes just get partially attenuated and spread out.)
3. Resampling to the desired data logging rate, often using averaging-based decimation. Averaging helps suppress random noise, and even moreso with larger oversampling factors like 25x.
-Kevin P
ALERT! LabVIEW's subscription-only policy came to an end (finally!). Unfortunately, pricing favors the captured and committed over new adopters -- so tread carefully.
