1989 vintage ADC replacement

[kubeek]

My dad's got a precision lab-grade scale, but recently it stopped working. Tracking the original manufacturer I got hold of the service manual for the scale, and figured out that the fault lies in the ADC module as it doesn't output any data and the pin is floating.

Now the ADC is a quite simple 15-bit multiple ramp converter module, but it is potted into the case quite well, and since it was made in the early 90s I doubt I can get a spare from anywhere, so the final choice is to replicate one.

As you can see from the DS5601 datasheet, the ADC simply does the conversion on its own, and then you can read out the data serially, the conversion takes 10ms so this should be easy to achieve.
The bigger question is what ADC should I choose? The main problem probably being linearity and low drift. I found this LTC2389-18 one at Linear which seems that should be far superior to the original, but I am a little worried that the very fast conversion time could introduce noise into the system.
Any thoughts?

 

[crutschow]

I believe the best fit for you would be a low conversion rate, high resolution sigma-delta A/D converter. They are typically used where high resolution is needed at a slow conversion rate such as scales. Because of their inherent high oversample rate it is relatively easy to use a simple RC input filter for anti-aliasing of any high frequency input noise.

 

[kubeek]

I came to conclusion that the scale uses 16x sampling and averaging to get it's required 19bits of precision.
So the question is, if I use a higher precision and lower noise ADC than the original, would I have to do dithering or something in order for the scale to get the correct result with this sampling?

 

[crutschow]

I came to conclusion that the scale uses 16x sampling and averaging to get it's required 19bits of precision.
So the question is, if I use a higher precision and lower noise ADC than the original, would I have to do dithering or something in order for the scale to get the correct result with this sampling?

I would think that, with the inherent oversampling of a high resolution Sigma Delta A/D converter, you wouldn't need to do any additional sampling or averaging.

 

[kubeek]

I didn't mean averaging or sampling, but see this - the scale asks sixteen times for a 15bit value. I got two choices, either use some 15+bit adc and truncate the result each time to 15 bits, or I can use a 19+bit adc but then I have to divide this result into 16 chunks such when the scale adds them up it gets the average actual value.

Now if I use for example this 18bit succesive approximation converter that I found and went with the option of sampling in each of the 16 times and truncate the result to 15 bits, wouldn't I actually be losing precision even though the scale adds the measurements and makes an artificial 19 bit number out of it?

 

[crutschow]

......................................
Now if I use for example this 18bit succesive approximation converter that I found and went with the option of sampling in each of the 16 times and truncate the result to 15 bits, wouldn't I actually be losing precision even though the scale adds the measurements and makes an artificial 19 bit number out of it?

Yes, you would be losing some resolution, but the result should be as good as obtained with the original A/D converter.