Subsampling data: terminology

Question

I'm writing software and the data reporting frequency is slower than the acquisition cycle, so effectively sub-sampling is being done. Different data types require different methods of sub-sampling which I'm tempted to call a Gating method where oversampled data in one time domain goes through a Gate to another time domain. Windowing also comes to mind.

I hesitate though as although these terms imply an interval of focus/ interest, they might not necessarily imply continuity between intervals; back to back intervals where no input is ignored.

Such methods might be:

• Average/Median since last report (or low pass filter)
• Min/Max since last report
• Sticky flags (a bit field who's bits are cleared when read)

Is there a better term than Gate or Window?

Note 1: This is borderline a programming question or code review, but I think still more for signal processing.

Note 2: Although sub-sampling causes signal artifacts such as aliasing. This is not part of the question.

Answer 1

This is more likely a question of what the data reporting application is trying to achieve, rather than a DSP question.

Yes, there is a different term for Gate or Window. In fact, if you "subsample" and low pass filter, as per your first intuition, then you are effectively doing decimation. For more information please see: http://en.wikipedia.org/wiki/Decimation_%28signal_processing%29

The rest of your suggestions would be equivalent to subsampling followed by non-linear filtering such as median, min, max which would have the corresponding effect on the reported data.

For example, a sliding window median filter would "absorb" any spikes in the reported data and the min and max filters would tend to report the highest or lowest values. For more information please see: http://en.wikipedia.org/wiki/Nonlinear_filter and http://en.wikipedia.org/wiki/Median_filter

So, it is really a matter of what does that GUI element try to convey to its user.

Hope this helps.

Supplementary Information: Yes decimation does imply a stable sampling frequency but most DSP operations can be defined over irregular domains as well.

Irrespectively of that however, I think that we are indirectly looking at the real problem here which is one of visualisation.

The real problem, or at least my understanding of it so far, is that you are trying to convey a comprehensive view of a window of the history of a signal at a rate lower than its actual rate of change. Furthermore, this signal is not used in any further computation. If that is mostly accurate and assuming that your signal is coming from a stochastic process then no single value will be able to represent this history reliably.

Therefore, I would suggest two alternatives:

• To use a single boxplot that would represent your last signal's frame history in a comprehensive way including its min, max, median and quartile information. For more information please see: http://en.wikipedia.org/wiki/Box_plot

• To convey the signal's history to your user via a Sparkline which could furthermore be annotated with min,max,median lines. Please note that Sparklines can also be created via the use of special fonts, which makes them really easy to quickly insert to a GUI. For more information please see: http://en.wikipedia.org/wiki/Sparkline and https://www.fontfont.com/how-to-use-ff-chartwell or http://nsfmc.github.io/chartjunk/

Hope this helps