We are using the ADIS16480 in a project to track position and have observed noise on the x, y, and z acceleration sensors (and the delta x/y/z velocity entries in the ADIS-EVAL software). We have the ADIS16480AMLZ installed onto an EVAL-ADIS and are using the IMU Evaluation software to collect roughly 10 seconds worth of data, while we move the unit in a straight line for a total distance of 9 meters. We then took the data and ran it through our kinematics code and found that our position estimates varied from 8 meters to 11 meters. As for the specifics of the noise, when the device was left at rest and data was sampled for a few (~4) seconds, each sensor displayed noise. This noise was typically less than .01 m/s^2 (or it only existed in the thousandths place). A solution would then to only read through the hundredths place, I suppose, if you endorse that. Is there anything we can do to improve the accuracy?

While I am not in a position to review your kinematic formula, one of the key limitations in this type of computation is the orientation of the accelerometers, with respect to gravity. In dynamic situations, this computation typically relies on the gyroscopes. For example, if we assume that the gyroscopes are perfectly tuned and that their in-run bias stability determines the orientation accuracy over the 10 seconds, we can estimate the impact in the following way:

Average gyroscope error = 6 deg/hour x 1 hour/3600 seconds x 10 seconds x 1/2 = 0.0083 degrees

Bias on accelerometers = arcsin(0.0083 deg) = 0.00014g = 1.4mm/sec^2

Position error after 10 seconds = 1.4mm/sec^2 x (10ssec)^2 = 0.14 meters

If we look at the accelerometer in-run bias stability:

In-run bias = 0.1mg x ~1mm/sec^2

Position error after 10 seconds = 1 mm/sec^2 x (10sec)^2 = 0.1 meters

If we combine them, using a root sum square computation

Position error estimate = [0.1^2 + 0.14^2]^0.5 = 0.17meters

Since you are seeing an span of 3 meters, it would appear like better performance might be available.

I would expect that the orientation error, with respect to gravity and the in-run bias stability in the accelerometers would be the key limit. For starters, can you tell me what the EKF_CNFG register contents are?