16-bits ADC anyone?

In article <5Zk9i.16082$C96.15101@newssvr23.news.prodigy.net>, 
antispam_bogus@hotmail.com says...
> 
> 
> Paul Keinanen wrote:
> 
> 
> >>I am working on a project where I need some 16 bits ADC to retrieve
> >>information from a sensor. I also need a small microcontroller such as
> >>a PIC, AVR or 8051, and I got surprising quotes for the ADC: Around $5
> >>(qty 1000), which is 5 times more expensive than the controller!
> > 
> > 
> > The problem with microcontrollers with on-chip ADC/DACs is that you
> > might not get the nominal 98 dB SNR  due to the noise from the
> > controller. 
> 
> Besides, the high performance ADCs and the microcontrollers are the two 
> different technologies. The MCUs with the good ADC/DACs usually contain 
> two separate dies in one package. For that reason they are more 
> expensive then the equvalent micro + equvalent ADC/DAC.
> 
> > 
> > Is DC accuracy (drifts) important in your application ?
> > 
> > Is this ADC part of a control loop, in which case it would be
> > preferable that the ADC is monotonous.
> 
> I am wondering of what could be a sensor which requires the ADC with the 
> true 16-bit accuracy. For the sensor application, that sounds 
> unreasonable to me. Especially considering that the rest of application 
> is handled by a small micro. Apparently there is a problem with the 
> concept.
> 
> 
There are a lot of oceanographic variables that need something near 16-
bit resolution.  When I was working with sensors to measure the optical
properties of seawater,  our minimum standard was one part in 10,000 
sensitivity and noise levels.  


Mark Borgerson

On Tue, 05 Jun 2007 16:56:20 -0500, Vladimir Vassilevsky wrote:

> I am wondering of what could be a sensor which requires the ADC with the 
> true 16-bit accuracy. For the sensor application, that sounds 
> unreasonable to me. Especially considering that the rest of application 
> is handled by a small micro. Apparently there is a problem with the 
> concept.

I worked with a 16-bit cirrus logic sigma-delta adc in a design with a
non-linear gas sensor and 16-bits was _just_ enough. If needed, I could
have gone up to the 23/24 bit device in the same footprint, for more
resolution. Absolute accuracy wasn't the main issue (I could have got away
with 12/13 bits for that), resolution was. I used it with an 8-bit micro.

The really good thing about the A/D, was that its programmable internal
amplifiers had a gain stability against temperature of only a few
ppm/degree celcius, which meant that I could get away with a cheapish
reference, of about 5ppm (iirc).

Can't remember the exact cirrus part number now, but would look at those
again as the price was pretty good too.

Regards,

Paul.

"Paul Taylor" <paul_ng_pls_rem@tiscali.co.uk> wrote in message 
news:pan.2007.06.08.20.29.01.108147@tiscali.co.uk...
> On Tue, 05 Jun 2007 16:56:20 -0500, Vladimir Vassilevsky wrote:
>
>> I am wondering of what could be a sensor which requires the ADC with 
>> the
>> true 16-bit accuracy. For the sensor application, that sounds
>> unreasonable to me. Especially considering that the rest of 
>> application
>> is handled by a small micro. Apparently there is a problem with the
>> concept.
>
> I worked with a 16-bit cirrus logic sigma-delta adc in a design with a
> non-linear gas sensor and 16-bits was _just_ enough. If needed, I 
> could
> have gone up to the 23/24 bit device in the same footprint, for more
> resolution. Absolute accuracy wasn't the main issue (I could have got 
> away
> with 12/13 bits for that), resolution was. I used it with an 8-bit 
> micro.

My experience was similar (except with discrete integrating ADCs [1]). 
Application was temperature control; we needed as much resolution as we 
could get, with low noise and total monotonicity (mainly for the 
derivative term), while absolute accuracy was somewhat less important - 
around 0.25% of full-scale, IIRC.

[1] By "discrete", I mean that our cost and resolution constraints 
didn't allow for buying in an ADC (this was some years ago). We used a 
couple of methods: one was the classic dual-slope integrator (where we 
measured time), the other was a voltage-to-frequency design (where we 
counted pulses).

> The really good thing about the A/D, was that its programmable 
> internal
> amplifiers had a gain stability against temperature of only a few
> ppm/degree celcius, which meant that I could get away with a cheapish
> reference, of about 5ppm (iirc).

We used the classic ratiometric approach: we used a 4-channel input mux, 
and measured the input signal, a reference voltage, ground, and either 
the cold junction for thermocouples, or the third wire for RTDs. That 
allowed us to characterise everything except linearity.

Bottom line was that we were achieving something like 20- to 24-bit 
resolution with components (in addition to the micro) costing a few tens 
of pence. The tradeoff, of course, was speed - these were not flash 
converters ;).

Steve
http://www.fivetrees.com 

Steve at fivetrees expressed precisely :

> We used the classic ratiometric approach: we used a 4-channel input mux, and 
> measured the input signal, a reference voltage, ground, and either the cold 
> junction for thermocouples, or the third wire for RTDs. That allowed us to 
> characterise everything except linearity.

Can you give me any link about this approach?

Thanks!

"djordj" <djordj@despammed.com> wrote in message 
news:mn.5ac67d7613d4d06f.73215@despammed.com...
> Steve at fivetrees expressed precisely :
>
>> We used the classic ratiometric approach: we used a 4-channel input 
>> mux, and measured the input signal, a reference voltage, ground, and 
>> either the cold junction for thermocouples, or the third wire for 
>> RTDs. That allowed us to characterise everything except linearity.
>
> Can you give me any link about this approach?

Which bit: ratiometric measurement? If so, Google is your friend:
  http://www.campbellsci.ca/Download/VoltAccy.pdf

Amongst others.

Steve
http://www.fivetrees.com