Noise smoothing

[Navid200]

Occasionally, a suggestion is made to add intelligence to xDrip to avoid false alarms.
I would like to talk about this and explain why it would be a mistake in my opinion. I hope to have a conversation to see if I may be mistaken and would be happy to have an open conversation.

I have put together a spreadsheet showing a typical set of numbers I could have:

I have definitely had sudden drops (or rises) of about 1.3mmol/L (23mg/dL).
At this point, if xDrip was to look at all the data, it could only extrapolate. But, without a crystal ball, there is no way to make a decision that the sudden drop is not real and be certain about it.

Next, I am showing two completely different possible outcomes.

I have had my glucose drop all of a sudden for real (hypo) and I also have had my glucose drop as sharply only for one reading (noise).

As xDrip developers, I can see two options:
1- Show the reading exactly as is. The outcome is occasional false alarms.
2- Use extrapolation and low-pass filtering to reduce sudden deviations. The outcome is occasional delayed alarms.

As an xDrip user whose life depends on it, I would not want to find out at 9:45 that I am at 3.5mmol/L (63 mg/dL) and xDrip decided not to tell me at 9:40 that I was at 4.7mmol/L (85mg/dL) and instead told me that I was 5.7mmol/L (103mg/dL) because of a DSP low-pass filter.
There is a big difference between 4.7 and 5.7.
I definitely hate a false alarm. But, I prefer it compared to a trip to the hospital.

[jonprogers]

@Navid200 thanks for such a clear example of the dilemma here.

My take on the situation is that the potential false alarms are important to me as I may have additional context that helps me understand what to do. For instance if the alarm woke me up and I'm lying on my sensor then I'll move, wait for more readings and then understand if it were a compression low. On the other hand if I had just undergone some unanticipated exercise then I would know that there is more chance of this being a real drop and act accordingly.

For me, xDrip is a tool that measures one indicator about my diabetes. It is then my job to take this with everything else that is going on to make treatment intervention decisions. A prompt warning that something weird may be happening - even if that is only sensor noise - is key to my effective management and I feel excessive smoothing would have a negative impact. The sooner I get this warning the better.

In addition to this I use rapidly increasing noise (both amplitude and frequency) as an indicator that I may be about to suffer a sensor failure. I find this warning sign very useful as it allows me to be prepared. More smoothing would lessen this advance indication.

Yes, false alarms are annoying, although I accept them as being better than the alternatives. I am therefore clearly in the option 1 camp above.

I look forward to the views of others, thanks for opening the conversation @Navid200...

BTW I currently use Dexcom G6