Integrating MEMS accelerometer output 1

[TimCodd]

Hi, I'd like to explore the use of a Micro-Electro-Mechanical System (MEMS), i.e. micromachined, accelerometer.

I'm interested in hearing about experience using one of these for "low g" applications.

To experiment, I want to build a circuit using one of the 3-axis accelerometers from SGS Thompson (ST). Each output would be coupled to two integrator op amp stages. These outputs would be converted to digital (12bit?) and then read by a uP. I know I can use analog or digital mux's to
minimize part count, but for now I'm just experimenting.

I have a lot of questions I'd like to get answers. Here are a few:

1) Realistically, if I integrate the output (0-2.5v=neg g's, 2.5-5.0v=+g's) and integrate to get rate, and then integrate again to get position how close can I track the movement of a slow moving object over time given some of the temperature caused errors?
2) What is a good choice for an op amp part from Linear Technology, ST, or Intersil to use in integration of these output?
3) What are some techniques for compensating for temperature errors in the accelerometer, integrator stages, etc?

I'm intrigued by this technology and want to learn more.

Thanks

 

[IRstuff]

You should find a good text on navigation systems.

As for integrations, the big problem is DRIFT, particularly for slow movers

TTFN

 

[mgopalan]

I would watch this one carefully...

Integration is a smoothing process in the frequency domain so high frequency effects that you might see in the accelerometer data will not hurt you too much.The low frequency data adds a lot to the integration process...So good signal levels and linear frequency response helps..

So the high frequency is ok, the low frequency is hard but not impossible...why is this problem hard? The 0Hz Stuff.

If you are not careful, you will get killed by DC offsets...MEMS devices are notoriously bias unstable and hysteretic...If you are integrating analog then amplifier offsets and millivolt level inaccuracies in the accel will integrate over time to give you gross errors in velocity and displacement..Evan at 1V/G, 1mV is only 1mG which is at the bleeding adge of these devices today.

You problem is one of Inertial guidance. Think about why we have GPS at all if we could integrate accelerometers and gyros accurately..

Now, the good news....If you want to monitor slow moving objects and you want low to average accuracies, you can do this. If you want true accuracy over time...then you will have to calibrate very very carefully..

Temperature calibration of accelerometers use three sets of coefficients...

Bias
Scale
Orthogonality

Bias corrects for the drift in the 0G reading for temperature.
Scale corrects fot the sensitivity (V/G) change with temperature
Orthogonality corrects for misalignment between your sense axis and the sensor's sense axes over temperature.

Of these three, the first two will require second order or third order curve fits...At low temperature changes, you might get away a linear

The third rarely requires temperature calibration at all. But I do a linear just to be safe. Heck PC Boards bend over temperature!!..

The best commercially available Low G accelerometers in the world for inertial guidance are the Servo Quartz Flexure Accels...They provide us AT BEST with about 0.1 to 1mG total accuracy after corrections..even with a 0.1% error in the data....it does not take too long to integrate yourself into a wall....try doing some rough math assuming some ballpark values and see what kind of positional accuracy you get over time.

Also, keep you noise levels very low...integrating noise is always messy.
MG