MS5611 "blips"

[hpmaxim]

I have a weird, and unfortunately very open-ended question, but I'm not sure how else to put it or even where to start.

I have an Opensource project which I am trying to get to work but it is being problematic.  It is an embedded Linux machine running on a "CubieBoard2" (similar to a Raspberry Pi).  The Linux box has a "sensor board" on it which has a number of I2C sensors, including two MS5611s.

There is a daemon that is run on boot-up of the Linux box which periodically reads the sensor data and sends it out a TCP Port (and records it to the SD card).  In particular, it polls both MS5611s at approximately 50 ms intervals.  If I look at the data, I see "blips" in it.  A blip is typically 1-3 samples of + .3 to.5 mbar, followed by several samples of -.1 to -.5 mbar (about 10 total).  The blips appear to occur between 92 and 94 samples apart, pretty consistently.  And the blips on both MS5611s are PERFECTLY synchronized in time, and pretty much in magnitude.  As far as I can tell, if one MS5611 blips, the other does too.  The program kicks off the two conversions, and then does a usleep (12500); before reading both out.  Timing in the program is entirely controlled by usleep, which of course isn't exact.  I suppose it's possible that some of the usleeps aren't waiting long enough, but the MS5611 documentation says that a read on an incomplete conversion will give 0 and that's not what happens.  Temperature is sampled (but not recorded) every 9 or 10 pressure samples during the period between the pressure samples.  Is it possible that a temperature conversion is getting messed up?  Maybe... But it seems unlikely this would result in the pressure output consistently jumping up and then down.

Now, it this point, I'd think... Okay, maybe there is noise on the power supply.  Seems unlikely since it's two separate power supplies, but, maybe... Unfortunately, it gets weirder...

If I kill the process, and re-start it, the blips either completely or mostly go away.  I recorded data off a restarted process for 20 minutes and never saw a blip.  I recorded data off a restarted process and saw one blip in 2 minutes (obviously it normally blips every 4-5 seconds). 

The sensors are both sampling atmospheric pressure outside, and they are not moving.

I'll also add that the sensor board is powered by a 5V supply and there are multiple 3.3V linear regulators on it.  Each MS5611 has its own regulator.  So the 5V supply is shared, but the 3.3V is not.

Schematic/layout/BOM for the board can be found here: https://www.openvario.org/lib/exe/fetch.php?media=projects:series_00:electrical_tests:pressure-sensor.pdf
Software can be found here: https://github.com/Openvario/sensord

The bad news is, it's in a self-contained box and would be difficult to probe the hardware.