AVR-Chat

>>> dave_mucha@yahoo.com 03/11/2004 1:48:24 pm >>>
>I figured the 4.096v was to generate an exact integer for a 12 bit 
>ADC.

Yep. It'll do the same for any number of bits actually.

>Guess that means a 0-5 requires a voltge divider to get to the 4.096 
>value at 5 volts.

Yes, although you are probably better off using a 5V reference if you
want 0-5V range. Otherwise if you use 4.096V you have to use say 0.1%
precision resistors for the divider. Extra components, extra cost,
greater error.
Regardless of whether you use 4.096 with a divider, or 5V without a
divider, you will still have an oddball value per bit. i.e for a 5V
input and 12bit ADC you will have 1.22mV per bit. Can't escape that
fact.

>I assume that 4.096 volts is exactly 12 bits also and not some 80% of

>the range.  If I am not mistaken, the whole idea of the voltage ref 
>is that with that specifc voltage, the input will be exactly 12.000 
>bits so even if it were a 2 volt ref, 2 volts in on an ADC pin would 
>be exactly 12 bits.

Correct.
Use a 5V reference you get a 0-5V range, use a 2.5V ref you get 0-2.5V
range etc. All of them will still give you 12 bits of resolution.
Although the lower the voltage reference gets, the greater impact you'll
have from noise. i.e. a 2.5V ref gives 610uV per bit. So if you have 1mV
of noise in your system, that is almost 2 least significant bits worth
of noise.

Dave :)

At 02:16 PM 11/3/04 +1100, you wrote:

> >>> dave_mucha@yahoo.com 03/11/2004 1:48:24 pm >>>
> >I figured the 4.096v was to generate an exact integer for a 12 bit
> >ADC.
>
>Yep. It'll do the same for any number of bits actually.
>
> >Guess that means a 0-5 requires a voltge divider to get to the 4.096
> >value at 5 volts.
>
>Yes, although you are probably better off using a 5V reference if you
>want 0-5V range. Otherwise if you use 4.096V you have to use say 0.1%
>precision resistors for the divider. Extra components, extra cost,
>greater error.

2 resistors at 0.1%, a 0.1% reference, the ATmega8535 has a 10bit converter 
so that's a best case accuracy of a little under 0.1% (the data sheet 
suggests it's actually probably closer to 0.3% depending on what kind of 
accuracy you are looking for).  That doesn't leave much of an error budget 
for sensor errors layout or other considerations.

I am assuming that it's unlikely for the source signal level range to match 
the A/D input range and that it will either have to be increased or reduced 
(and will thus likely have at least 2 resistors involved in the input).

Some of that can be calibrated out but that will mean some sort of 
precision source (better certainly than 0.1%) and additional care in 
testing/calibration.

If I was presented with a request for any particular accuracy my first 
question would be to ask what the accuracy of the source transducer 
was.  To get 0.5% using the ATmega I would at first blush estimate the 
source would have to 0.1% accurate or better.  A lot of sources don't 
approach that, a fair number have trouble even being 8 bits accurate.


Robert

" 'Freedom' has no meaning of itself.  There are always restrictions,
be they legal, genetic, or physical.  If you don't believe me, try to
chew a radio signal. "

                         Kelvin Throop, III

--- In AVR-Chat@yahoogroups.com, Robert Adsett <subscriptions@a...> 
wrote:
> At 02:16 PM 11/3/04 +1100, you wrote:
> 
> > >>> dave_mucha@y... 03/11/2004 1:48:24 pm >>>
> > >I figured the 4.096v was to generate an exact integer for a 12 
bit
> > >ADC.
> >
> >Yep. It'll do the same for any number of bits actually.
> >
> > >Guess that means a 0-5 requires a voltge divider to get to the 
4.096
> > >value at 5 volts.
> >
> >Yes, although you are probably better off using a 5V reference if 
you
> >want 0-5V range. Otherwise if you use 4.096V you have to use say 
0.1%
> >precision resistors for the divider. Extra components, extra cost,
> >greater error.
> 
> 2 resistors at 0.1%, a 0.1% reference, the ATmega8535 has a 10bit 
converter 
> so that's a best case accuracy of a little under 0.1% (the data 
sheet 
> suggests it's actually probably closer to 0.3% depending on what 
kind of 
> accuracy you are looking for).  That doesn't leave much of an error 
budget 
> for sensor errors layout or other considerations.
> 
> I am assuming that it's unlikely for the source signal level range 
to match 
> the A/D input range and that it will either have to be increased or 
reduced 
> (and will thus likely have at least 2 resistors involved in the 
input).
> 
> Some of that can be calibrated out but that will mean some sort of 
> precision source (better certainly than 0.1%) and additional care 
in 
> testing/calibration.
> 
> If I was presented with a request for any particular accuracy my 
first 
> question would be to ask what the accuracy of the source transducer 
> was.  To get 0.5% using the ATmega I would at first blush estimate 
the 
> source would have to 0.1% accurate or better.  A lot of sources 
don't 
> approach that, a fair number have trouble even being 8 bits 
accurate.
> 
> 
> Robert

The transducer that will be measured will have a 0-5Vdc output that 
is really about 100mV to 5v and has an accurace of better than 0.001% 
if you correct for the termperature and voltage.

In point of fact, some people use only the first 3% for the entire 
range into a 12 bit ADC.

I know of one guy using a 24bit ADC.

Our application will have the input into two channels.  One for the 
full 0-5V and the other with some range of about 15% of the full 
range.

But our first try with the TLC2543 showed that without doing 
something about the board voltge we were way off.

Dave