Hi Adam,
=20
> If you continue to use your current design I believe you'll obtain the =
 accuracy you need=20
> except for the temperature compensation issues. The simple fact is that=
 unless you calibrate=20
> onsite at the current temperature, you may be off by as much as 0.8 deg=
rees just due to=20
> temperature variations. Using the 213 moderates it somewhat since the m=
easurement is=20
> relative - you could say that a little compensation is built into the d=
uty cycle modulation.
=20
Thanks for pointing that out. I found a design note on Analog's site on h=
ow to embed temperature information in a PWM output using a TMP36. The on=
ly problem is that I can't order samples of it or order it from Farnell i=
n a SOT23 package, so I'll use a TMP37 instead. It has 20mV/=B0C instead =
of the 10mV/=B0C the TMP36 has. Now if I only could figure out how T2 is =
related to the voltage over Rset :).
=20
> In my case I used 4.7uF caps and a 1.2Mohm resistor on the 213. This gi=
ves about a one=20
> second response time to changes, and counting the duty cycle at around =
250kHz (ie, the=20
> counter increments 250k times per second to find the size of the pulse)=
 gave me the 0.1=20
> degree resolution I needed. Since this device is expected to be re-zero=
ed before each use I=20
> didn't need to worry about temperature compensation.
=20
Just to see if I understand, T2 is 104kHz with Rset 1.2Mohm, right? How e=
xactly do you calculate the resolution? Since I've got to know the exact =
angle the sensor has in relationship to the earth and because I can't be =
sure it will be on a flat surface when in use, I can't rezero it before e=
ach use.
=20
> Then it gets interesting. Unfortunately this part of the exercise is le=
ft up to the reader. You=20
> can simply use those numbers directly, but the duty cycle is formed fro=
m a triangle wave,=20
> which complicates matters when you need high accuracy. Further, doing a=
n inverse tangent=20
> function (ATAN2) is not trivial in an 8 bit microcontroller. In my case=
 I used a calculation, but=20
> a table with interpolation would likely be easier for the lower end mic=
ros, and if done well will=20
> give you the needed accuracy.
=20
How exactly does it complicate things when needing high accuracy? I can u=
se pretty much any =B5C I want (well, as far as I'm able to program it of=
 course), because size isn't really a problem for the =B5C board. So I mi=
ght use a PIC18 so that I can use a high order Taylor approximation (well=
, that or a big lookup table).
=20
> Check out the documents on analog's site. They have some good algorithm=
s.
Yes, indeed, I've found some interesting stuff.
=20
> I assumed that long wire length given the environment my device is goin=
g into would cause=20
> too many problems with noise to get the needed accuracy. Further I was =
using 3.3v which=20
> gives a poor signal to noise ratio on the line. I never tried it though=
, and it may work out just=20
> fine depending on the environment, cable, etc. I put a small microcontr=
oller next to it which=20
> did all the timing, and sent a simple checksummed data and clock signal=
 to the display=20
> where the real calculations are done.
=20
Oh, then I think I'll probably be fine, because there won't be a lot (if =
any) noise and I'll also be using 5V.
=20
> Good luck!

Thanks, Anthony


Website: http://members.lycos.nl/anthonyvh

	=09
---------------------------------
Do you Yahoo!?
 Yahoo! Mail - now with 250MB free storage. Learn more.
--=20
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist