I completely have the Filter Pt-by-Pt VI because it is not included in the base package (which some of my users use) and it is horribly inefficient.
I prefer to use the attached method. It is polymorphic and works for filtering single or multiple channels. You only need two single point buffers for each channel you are filtering and the size of the average does not effect performance. I believe that you could experience some jitter when using this method, but it has not been an issue for my applications. I don't think it can get any better than this, expecially for averages lasting thousands of points.
Note that whenever filtering data you need to make sure that a +/-Inf or NaN cannot get "caught" in your shift registers. In this example I chose to ignore those values and just pass the last valid data. You can use this to filter out invalid data points, just convert them to NaN before passing them to the filter. I like to use this method to filter out bad thermocouple values that are caused by broken wires, etc.
Thoughts? Comments?
