Question asked by andersr on Nov 21, 2014
Latest reply on Nov 24, 2014 by andersr

We are using ADXL312 in a new design, and have measured the output of the accelerometer:

MinX    MaxX    AvgX    MinY    MaxY    AvgY    MinZ    MaxZ    AvgZ

-50.00  58.00   -5.68   -54.00  48.00   -4.25   228.00  410.00  309.97

Minimum, maximum and average (mean) are calculated from 1000-samples.

The sensor is laying solid on my desk, so i expect to have about 1G at Z, and 0G at X and Y.

The accelerometer is configuration:

3.3V supply

3200Hz data rate.

+-6G

Full resolution

Stream-mode

Watermark interrupt after 16 samples

As you can see, the output differences are very high, especially at the z-axis:

Min=228*2.9mG/LSB = 661mG

Max=410*2.9mG/LSB = 1189mG

avg=309.97*2.9mG/LSB = 899mG

I have already made some attempts to improve it, by changing decoupling of Vs, and to use a separate 3.3V supply for Vs. But it didn't help.

Have been looking into the data sheet which gives noise figures of 340uG/sqrt(Hz).  My interpretation of how to use that number is:

noise = (3200/2)^(1/2) * 340*10^(-6) = 17.6mG.

Even if i would add bit truncation and other, types of error sources would i never reach the figures i experience.

My question is: What should i expect with these settings, how much noise should it be there if the implementation of the sensor was perfect?