High precision inclinometer

Hi all,
I've got a digital inclinometer with +-0.1° accuracy. I'd like to get +-0.01° accuracy (an order of magnitude higher). I searched for commercial products, but couldn't find any.
So I looked here, and saw ADXL335 mentioned, as well as some other AD chips. Searching through AD's product selector I came across ADIS16203. I'm a bit confused about the accuracy of these chips...
Looking at ADIS16203 datasheet, I read "Linear output, 0.025° resolution" which would be ok, but on the next page I see:

Relative accuracy +-0.6°
Sensitivity 0.025°/LSB
Accuracy temp coeff +-0.0167°/°C
Output noise: no averaging 1.0°rms, max averaging 0.1°rms

My question is: supposing I put this chip into a tilt measurement instrument, could I say it would measure its tilt with a precision of +-0.025° ?

I've read on the AD site that by combining two ore more chips and averaging their readings one could improve accuracy because the noise curve is gaussian. I think this would be very challenging though, as it would require a very high precision while assembling the pcbs and other mechanical parts...

I've read some threads here about angle and position measuring. I'm starting to grasp the meaning of it all, but I'd like to have a hint about where to go for more information. Even a "can't be done" response would be helpful

PS: I'm not sure if this goes into "sensors" or "project guidance" section...

0.025° resolution" ...
with a precision of +-0.025° ?

You answered your own question, aka resolution vs. precision.

2 degrees at both extremes, my experience.

I've read on the AD site that by combining two ore more chips and averaging their readings one could improve accuracy because the noise curve is gaussian. I think this would be very challenging though, as it would require a very high precision while assembling the pcbs and other mechanical parts...

When you compensate the error in SW it is not so difficult, just be sure they are in a fixed position with respect to each other.
Then you put them in position 0°, read out the sensors and note their response.
repeat that for a number of angles e.g. every 5 degrees

Now you can average the delta's for the different sensors and correct for these offsets in your sketch.

Example:
sensor1: 359.9, 10.0, 19.95, 30.0 ...
sensor2: 0.1, 10.2, 20.15, 30.10 ...
sensor3: 0.2, 10.25, 20.20, 30.20 ...

Assume the angles are exact 0,10,20,30,....
the average error can be calculated e.g.
sensor 1: -0.05
sensor 2: +0.14
sensor 3: +0.21

code:
float angle = (sensor1+0.05 + sensor2 -0.14 + sensor3-0.21) / 3; // optimized (sensor1 + sensor2 + sensor3 - 0.30) / 3;

Got the idea?

Yes, I think I got it, rob. Thank you very much.

IIRC the AD faq page said with n chips one can get improve by sqrt(n). Have to re-read it carefully to better understand what this means... How many adxl335 boards would I need to match one ADIS16203 ?

dhenry:

0.025° resolution" ...
with a precision of +-0.025° ?

You answered your own question, aka resolution vs. precision.

Going to read Accuracy and precision - Wikipedia

dhenry:
2 degrees at both extremes, my experience.

I don't understand what you mean...

I've done some more research... The ADIS16203 is quite expensive, especially if one wants to get the evaluation board O_o

I found a commercial inclinometer that declares a precision of 0.5mm over 1 m, which if my math is correct means atan(0,5mm/1000mm) = 0,0286 degrees.

The price is quite high here, too, but this would be ready to use

(edit: the technical details page reports a precision of +-0.1° on measured angles 0-90°... )