Acceptable accelerator noise?

You don't define how you are acquiring the data but 4 mg rms is typical for 100Hz sampling with +/-2g range per the product spec PS-MPU-9150A.pdf.

I get 1.1 mg rms (X&Y) and 2.0 mg rms (Z) using /Datalogging/cube_python/eMPL-client.py which appears to sample accel data at 50Hz. I keyed the inva command with cube window in focus and keyed inva again 60 seconds later. The log file was recovered in Datalogging/cube_python/log/date_time.txt and plotted/analyzed it using Octave std() function. (ignoring the initial transient).

New Mystery: : When I turn on the SDK and then command "inva" via direct serial communication and capture the serial data directly I find the same noise level as with eMPL-client.py when the SDK is quiescent. When I rotate the device the python data tracks the local acceleration due to gravity, however, the serial capture does not change from its initial setting-- but it faithfully reproduces the same noise level ???

Anyone have an explanation?

The serial capture should change and have same output as the python data.

I would certainly agree with your statement, however I was never able to reliably command and acquire serial data directly. It turned out faster to learn python programming and modify eMPL-client.py to do the simple real time data plotting I needed using Pygame, rather than directly command and capture serial data using a higher level language such as Octave. (though some success with Matlab has been reported).

There may be some subtleties in the data framing or initialization of the SDK hardware that I couldn't deduce from the python script. The python seems very efficient and robust and leaves sufficient throughput for real time plotting. I only had to modify the data_display_viewer class and some logic in the main function so it looks like an update to eMPL-client that supports real time plotting, data logging and cube display and I would be willing to share it.

Now that I have programmable real-time access to all of the SDK data, converted to engineering units, in python I can likely build the application development code and displays that I need for the near term.