Using Sensor RMS values to calculate deadband?

I'm using a MS5611 pressure sensor (datasheet) and it discusses total error band as well as Resolution RMS.

I'm fairly confident I don't understand what it means by Resolution RMS beyond RMS meaning Root-Mean-Squared as it lists the highest resolution giving an RMS value of 0.012 mbar, but also states that the error band at the highest resolution is +/- 0.5 mbar which seems a fair bit larger than 0.012 mbar.

I would like to create a deadband where I (theoretically) throw out bad reads caused by either random fluctuations or rounding errors using this RMS value but my web searches so far have not produced any useful information.

I know there are other options such as a weight filtered, averaging over a time span, or Kalman Filters, but I was specifically interested in figuring out more about RMS to either tailor make a filter or at least better understand my sensor.

Can anyone point me in the right direction for better understanding RMS in relation to a sensor or how I might use it to filter out jitter?


The error band applies to a single value while the Resolution RMS supposedly is the RMS of a number of samples, with 0.012mbar for 4096 samples. OSR=OverSampling Rate.

For that highest resolution you have to measure the pressure 4096 times and then calculate the RMS, or you use a circular buffer for 4096 samples and compute a moving RMS with every new sample. Or you use a smaller buffer and/or calculate a moving average or median or whatever you like.

Okay... hmm. The over sampling rate is done automatically by the sensor, I just tell it what rate I want it to use and it will do all the buffering and calculations for me. So in theory what that 0.012 is just a representation of the variance in the 4096 readings it took, and each of those 4096 readings has a resolution (accuracy?) of +/- 0.5 mbar.

So what would be a useful way to combine these two things together? If I set the resolution to 4096 then a plausible deadband could be 0.02 mbar (since the RMS of each value I get sent back is 0.012) which I could then smooth out further with a simple weighted formula of oldvalue 0.9 + newvalue0.1. Is that a reasonable application of the error and accuracy or am I still misunderstanding things?

The processing depends entirely on your needs in accuracy, stability, speed etc.

At this point I more interested if my understanding and potential application are well founded or if my assumption of a deadband of 0.02 based on an RMS of 0.012 is a misapplication of what is actually being represented by that figure.

My thinking is the sensor can only provide a resolution of 0.01 mbar. And since the RMS can swing one way or the other by 0.012 (<- that's the part I'm not sure is sound) any time my reading deviates by 0.02 it should be an actual change in pressure?

I guess one way to test this is a simple program that I left running for a couple hours that recorded the minimum and maximum values read by the sensor while it sat stationary.

This would also show how "stationary" your environment really is. What happens if e.g. you move around in the room...

Ah, I see that now. And would it surprise you to learn that after half an hour of being left alone the min and max values were right about 1 mbar from each other? Or in other words I had a range of +/- 0.5 mbar. I swear somedays my brain isn't turned on.

How can you be sure that the atmospheric pressure (temperature...) has not changed in that half hour?

Do you have a reliable barometer for testing your device?

This topic was automatically closed 120 days after the last reply. New replies are no longer allowed.