50nV Resolution

Thread Starter

joeyd999

Joined Jun 6, 2011
4,477
BTW, I wish I was proficient in writing Android apps. If so, I'd just transmit the data to my phone via bluetooth and let it do the analysis. It'd be much faster than the manual method I am using.
I've decided I am going to add Android app development to my repertoire of skills. That's how I'm going to spend the holidays. Merry, merry.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
4,477
Interesting project.

Why do you need resolution? Most people would want accuracy, or precision.
Accuracy and precision are already cooked in.

I need to be able to resolve a really small delta change within a large dynamic range.

Imagine a voltmeter with a 100V full-scale and 1 mV resolution.

I require that the least significant digit not change while a stable DC input is present, or to advance by 1mV (and only 1!) when an additional 1mV is added to the input signal.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
4,477
In essence, I am doing something like this*:



with a Bluetooth radio for less than $20.

*Edit: but shift the decimal point left twice.
 
Last edited:

NorthGuy

Joined Jun 28, 2014
611
Accuracy and precision are already cooked in.
Nope. Resolution is simply a number of digits. You can write any number of digits and get huge resolution.

Accuracy and precision are measures of how these digits correspond to reality.

Precision is a measure of expected difference between measurements made by the same device at a different time or under different conditions.

Accuracy is a measure of expected difference between measurements made by the device and actual values of the entities being measured. If you have a precise instrument you can calibrate it and thus make it accurate.

Imagine a voltmeter with a 100V full-scale and 1 mV resolution.

I require that the least significant digit not change while a stable DC input is present, or to advance by 1mV (and only 1!) when an additional 1mV is added to the input signal.
This is 0.001% precision. It is very unlikely your device achieves this. It may drift by 1mV because temperature changes slightly, a wind blows, or simply because time goes by. If it could be done for $20, people wouldn't pay thousands upon thousands for test equipment.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
4,477
This is 0.001% precision. It is very unlikely your device achieves this.
Well...tell it to my project. It's sitting on my desk at this moment doing exactly that.

Edit: Oh, and the accuracy is 0.0001%*. Just sayin'.

*Sorry! I meant 0.01% .... forgot to multiply by 100. It's the simple mistakes that'll kill ya.
 
Last edited:

Thread Starter

joeyd999

Joined Jun 6, 2011
4,477
It is very unlikely your device achieves this.
I would like to take this opportunity to point something out:

25 years ago, I gave up on working with other engineers because I got tired of them telling me that everything I wanted to do was impossible. It got to the point where, as an engineering manager with a staff of twelve, I was doing all the work because they said it couldn't be done. And I succeeded, each and every time, in meeting my own predetermined specifications -- but not without considerable push-back by my staff the whole time. If I had a nickel for every time someone told me it can't be done, I'd be a very rich man right now.

The same goes for software guys. Each and every time, "No, you can't do it that way." My response? "Get out of my way."

I've got patents for designs that do things no one else said could be done. And successful products all over the world that implement those patents.

I've said it before: I incorporate at least one impossibility into every one of my designs. This is why I am competitive at what I do.

Don't tell me I cannot do something. It only makes me want to prove you wrong.
 

cmartinez

Joined Jan 17, 2007
7,552
I'm not really a fan of tutorials -- I usually just jump to the reference documentation and work it out as I go along -- but I found this. It seems pretty good (and not dumbed down) -- if not a little bit dated.
Neither am I. What I do like to see is samples of code covering the most commonly used techniques, and then I take it from there. Thanks for the link, I'll take a look at it later tonight.
 

bogosort

Joined Sep 24, 2011
678
Nope. Resolution is simply a number of digits. You can write any number of digits and get huge resolution.

Accuracy and precision are measures of how these digits correspond to reality.
Yes, but you get precision from resolution by reducing noise.

It may drift by 1mV because temperature changes slightly, a wind blows, or simply because time goes by.
It's the designer's job to account for these effects with appropriate component selection and PCB routing. For example, in my 18-bit application, I used metal foil resistors for the resistors in the signal path, routed to minimize temperature gradients. With a temp-co of 1 ppm/C, I get almost 20-bits of precision from my resistors over typical operating conditions.

High-precision circuits are made by paying attention to these details in every aspect of the design.
 

NorthGuy

Joined Jun 28, 2014
611
Well...tell it to my project. It's sitting on my desk at this moment doing exactly that.
You tell her. :) Move her, breath on her, cycle the power, turn the lights in the room on and off, wait until tomorrow etc. See how the voltage changes.

*Sorry! I meant 0.01% .... forgot to multiply by 100.
That would be 10mV if your range is 100V.

It's the simple mistakes that'll kill ya.
Yep. Happens all the time.
 

NorthGuy

Joined Jun 28, 2014
611
Yes, but you get precision from resolution by reducing noise.
The precision exists in your front end, before the ADC. The sampling process further decreases the precision by introducing variable gain and offset errors. This is the base precision which you cannot increase. You cannot do better than this.

If your ADC has resolution which is not enough to represent your precision (e.g. you have 0.1% precision and 8-bit ADC) then you can increase resolution by averaging to catch up with your precision, but this decreases bandwidth and may not work at all. Such cases are rare. Usually people use ADCs with much higher resolution than is necessary for the actual precision/accuracy of their front ends.

Noise is a different beast. Some of it is normally removed by the front end - for example you must remove all frequencies below Nyquist before sampling takes place because you cannot remove them later. After sampling, a big question is what noise is. One man's noise is another man's signal. You would think that when measuring DC everything is a noise, but this is not entirely accurate. You want to get your measurements with some frequency, for example, once a second, or once an hour. You cannot remove any noise which has frequencies below your retrieval rate.
 

bogosort

Joined Sep 24, 2011
678
The precision exists in your front end, before the ADC. The sampling process further decreases the precision by introducing variable gain and offset errors. This is the base precision which you cannot increase. You cannot do better than this.

If your ADC has resolution which is not enough to represent your precision (e.g. you have 0.1% precision and 8-bit ADC) then you can increase resolution by averaging to catch up with your precision, but this decreases bandwidth and may not work at all. Such cases are rare. Usually people use ADCs with much higher resolution than is necessary for the actual precision/accuracy of their front ends.
Not sure what your point is. Incidentally, I wouldn't call the technique of increasing ADC resolution by oversampling and decimation rare by any stretch.

Noise is a different beast. Some of it is normally removed by the front end - for example you must remove all frequencies below Nyquist before sampling takes place because you cannot remove them later. After sampling, a big question is what noise is. One man's noise is another man's signal. You would think that when measuring DC everything is a noise, but this is not entirely accurate. You want to get your measurements with some frequency, for example, once a second, or once an hour. You cannot remove any noise which has frequencies below your retrieval rate.
I think we're talking about two different things. You're talking about the limitations of noise reduction through filtering, but I'm talking about low-noise design techniques, i.e., keeping the noise out of the signal in the first place. Your example of a 1 mV error due to a small temperature change is a good example. You can't filter that away, but you can certainly prevent it from happening with proper design.
 

BR-549

Joined Sep 22, 2013
4,938
I thought joey wanted the resolution to detect wee wee little changes. I must have mis-understood again. That's normal.
 
Top