Pressure Plug measurements
The JeeLabs Pressure Plug uses the Bosch Sensortec BMP085 sensor. The data sheet claims +/- 2.5 hPa accuracy under normal conditions, and 0.03 hPa noise (standard deviation of 10 readings) in the "ultra high resolution mode" which internally takes 8 samples and has a total conversion time of 25.5 msec (max).
By comparing the readings of two Pressure Plugs in the same environment we can get an idea about repeatability, if not absolute accuracy. Two plugs were installed on the same JeeNode, in a buffered, insulated box so the ambient temperature changes very slowly (typically < 0.002 degrees C per minute). Readings were recorded over a period of many hours using code based on the Plugs library BMP085 demo. I ran the sensors continuously and recorded the average and standard deviation of each group of 100 samples on each plug. A set of 100 temperature + pressure readings plus computation of calibrated values takes about 5 seconds. For each reading I also recorded current temperature with 0.0002 C resolution, using a separate LM35 temperature sensor going into the Analog Plug MCP3424 in 18-bit mode.
Both BMP085 units give me a standard deviation of pressure values of around 0.045 hPa, although this is not a stationary value. It goes as low as 0.03 hPa, and occasionally above 0.1 hPa for unknown reasons (not related to ambient pressure change). The difference in averaged reading between the two sensors is about 0.4 hPa but varies from 0.0 to 0.6 hPa, with a typical ambient pressure of 1010 hPa and ambient temperature of 22.5 C +/- 0.5 C. Looking at the pressure difference (B-A) plot in the PDF below, you can see how the readings differ between the two sensors, in this plot averaging about 0.35 hPa. Ideally they should always be the same value, so this plot gives some idea of the noise level of the sensors. In this case the difference does not appear strongly correlated with pressure, temperature, or rate of change of either. Note: graph labelled (A-B) is in fact (B-A). See also the code file, below. The noise in the pressure reading is not simple "white" noise, it has a significant 1/f component (there is more noise at lower frequencies). That means averaging (n) readings together gives you less than SQRT(n) improvement in noise.
The absolute pressure readings from both the BMP085 sensors are about 3 hPa below the airport weather station a few miles away. My elevation is 78' (23.8 m), so I expect to read 2.8 hPa lower (11 C, dew point 6 C). The measurements agree to within the resolution of the airport pressure report of 1 hPa.
To eliminate the variables of atmospheric pressure, I did another measurement with two sensors in a sealed container (1.5 L glass canning jar). Once sealed, no air goes in or out of the container, so from the ideal gas law PV=nRT (with n,R,V all fixed) the pressure should be exactly proportional to temperature. The temperature wanders around slowly as it is in an insulated box, in a room that varies by a few degrees C. This plot covers 34 hours worth of data, one data point every 5 seconds.
In this case, you can see that Sensor A does a better job of giving us a straight line on the graph of Pressure vs Temperature, while Sensor B wanders more. Looking at the data from this experiment in a different way, in a plot of pressure and temperature values vs. time, we see that Sensor B increases its offset from Sensor A between hour 15 and 20. These plots gives you a feeling for how accurately these sensors can be calibrated, due to presence of short-term noise and long-term drift.
