Project:HUGnet Theory

From HUG Wiki

Polling

It takes 1-3 seconds for a polling event.

20 endpoints per minute.

Differential record filtering

What kinds of errors / anomalies

• Counter overflow
  • Rarely happens. Essentially covered by the next one.
• Counter reset
  • Easy to see, generally.
  • If the previous record is bigger than the next one, for the purpose of calculating the differential, set the previous record to 0.
    • This could created graph anomalies. It might be removed by user at run time or by trend analysis below?
• missing data
  • This will go away with averaging.
    • Averaging
• anomalies in the data
  • Take median of slope?
    • Make sure to take a odd number of samples.
  • Take median of the absolute?
    • Make sure to take a odd number of samples.
    • Take x samples before and x samples after + the current sample. Take the median of that and throw out this point if we are too far from the median.
    • What is an acceptable amount of deviation?
    • If we are the point furthest from the median in the sample, throw it out? Yes for a pulse counter.
      • Analog graph? -> How to not toss out peaks.

Bottom two require us to look at adjacent records or a trend.

How to deal with them

1. Save all generic pulse counters in Pulses per Minute (PPM). This will smooth out a lot of data anomalies.
2. Take short medians and long median.
   1. Keep the max and min of the short medians.
   2. If data > max + ((max - min)/2) throw it out.
   3. If data < min - ((max - min)/2) throw it out.
   4. This has to be done over a small time. Should probably be done at graph time.
3. Graph the median...
4. Graph the median and the raw data on the same graph.

This article or section is just a stub. If you want to help with this documentation please complete this article or section.

See Also