piclist 2000\05\04\203718a >
Thread: [EE] 24-bit A/D. Are We in the Twilite Zone Here?
www.piclist.com/techref/io/atod.htm?key=a%2Fd
BY : Brandon, Tom email (remove spam text)

In regards to adding more noise, remember, you're not actually adding any
accuracy (repeatability), in fact you're decreasing the accuracy of the
will have more variation. What you are doing is adding precision. The
averaged result is now closer to the 'true' value.

However, I think this would only apply with analog noise. You need
fractional noise or it falls apart. Adding 1LSB of digital noise would just
give you 1\3rd correct values, 1\3rd up 1 LSB and 1\3rd down 1LSB (assuming
true Gaussian noise). Average that out and it just returns the correct
value.

Yes, averaging the values would reduce the noise. If it's natural noise then
it's Gaussian, thus it should average out to 0 over multiple samples. But
the digital noise is irrelevant. Assuming the digital noise is also Gaussian
it too will be averaged out, but it will not increase accuracy or precision
(but as suggested it may speed the averaging).

In terms of how you 'add' the random noise, you don't. Ever designed a
(mixed signal) circuit with 0 noise? You don't add analog noise, you just
use the existing noise (assuming it's Gaussian, which almost all naturally
occuring noise is). You could digitally produce the noise and then subtract
it out for even better accuracy but this is rarely used as it isn't
justified. After spending 6months trying to eliminate the #\$%\$ing analog
noise, it's nice to be able to put it to good use.

In terms of digitally producing Gaussian noise, you just apply the Central
Limits Theroem. It states that:
A sum of random numbers becomes normally distributed as more and more of the
random numbers are added together. The Central Limit Theorem does not
require the individual random numbers be from any particular distribution,
or even that the random numbers be from the same distribution.

This is why Gaussian noise is so common. Whenever multiple random processes
(any distribution including pseudorandom computer generated numbers)
interact, the overall output will show Gaussian distribution.

Tom.

-----Original Messages-----

>From Andy:
<SNIP>
It worked much better than w/o the noise correction.

Andy

>From Scott:
<SNIP>
I suspect if you go back and repeat your experiments with the "digital
random
number" stuff removed that you'll get the same results - perhaps even
improved
results.
<SNIP>

Scott

>From KŸbek Tony:

Hi,
I'm intrigued by this discussion, and I'm probably
in over my head here :-)
Anyway regarding bit jittering, increasing accuracy etc.
If I understand correctly if one adds a random bit to the
reading then one possibly ( likely ? ) could increase the accuracy ?
Is this vaild for, let's say one has 6 bit's of noice ?
Would one then add several random bit's ?
Ore are this only valid for the LSB ?

Further how does one accomplish this 'random bit' ?
to be truly random I guess it would be very hard,
I've read the last month disscussion regarding
an random byte generation, but for a single bit
there must be an easier way.

BTW Just 'playing' with the 24 bit AD7730 ;-) DS

/Tony

Tony KŸbek, Flintab AB
ÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓ
E-mail: tony.kubekflintab.com
ÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓÓ

<F08121B734E0D311804300105A7890F771FE@psynt.psy.unsw.EDU.AU>