>I am currently working on a project where I am using sigma-delta ADCs and
>AFAIK, they are not cheap

A Sigma Delta converter is about as cheap as you can get for any given level
of accuracy. As the requisite accuracy increases cost goes up due to the
need for precision in fabrication and elimination of scond (& 3rd and 4th
....) order affects.

> but really high accuracy (although, I suppose
>that making 16 or 24 bit R2R ladder ADCs would be more expensive).

Yes, not mainly because of the number of components but because of the
required precision of some of them and the need to eliminate undesired
affects in more areas.

>IIRC,They work by having a capacitor attached to one input of a comparator.
The
>other input is the input to the ADC. If the voltage on the cap is lower
>than the main input voltage, a short pulse is fed to the cap thru a
>resistor to increase its voltage. If it is too high, the cap is allowed to
>discharge for a short bit through a resistor. The pulses fed to the cap are
>considered to be a fast bitstream,and their average over a thousand or so
>bits is computed, giving the average value of the input.


This is approximately correct. Usually the resistor which drives the
capacitor is ALWAYS either being driven high or low - not just in short
bursts. Decisions as to whether the capacitor is above the input level (so
that the capacitor should be being driven down) or below the input level
(when the capacitor should be being drivben "up") are made on a clocked
basis. If the transistion occurs between clock pulses the capacitor will
continue to be driven in the "wrong" direction for the remainder of the
clock cycle. This leads essentially to "pulses" similar to what you
described but they occur for the whole of the clock period and effectively
the capacitor is always being driven.

>So, the converter actually "samples" the input a thousand or more times per
>actual code output. Usually the output rate is around 100Hz and the 1-bit
>sampling rate is around 20-100kHz.

Very roughly the accuracy which can be obtained is proportional to the
number of samples. In practice it will be somewhat less. For  a 2000 sample
conversion you could "hope" to achieve approaching 1 in 2000 accuracy = 11
bits. In practice you wouldn't. I'm getting around 9 or 10 bits at 2000
samples. YMMV (actually YMWV). Accuracy varies depending on actual input
level  and various other somewhat arcane factors. Capacitor size is more
critical than may at first appear.

Obviously thedoubling in samples required every time accuracy increases by 1
bit makes high accuracy SD a slowish process.

Harris Semiconductors do a nice introduction in Sigma Delta in their AN9504.

The LTC2400 from Linear Technology is a 24 bit (! !! !!!!!!!!!) SD converter
at a reasonable price considering its accuracy.




regards



      Russell McMahon
_____________________________

>From other worlds - www.easttimor.com
                                www.sudan.com

What can one man* do?
Help the hungry at no cost to yourself!
at  http://www.thehungersite.com/

(* - or woman, child or internet enabled intelligent entity :-))