comparing two voltage down to microvolt level

[nour]

I want to be able to compare two voltage down to the microvolt level say for example 1.000,010 and 1.000,000
When I tried to do this, it doesn't work and it take the comparator a certain amount of lower and upper range to be able to flip its output
Like for example, if the ref is 1.000,000 I am expecting it to flip the output when it is greater than the ref like 1.000,001, but that doesn't happen it will flip after maybe 1.001,000 or so (don't remember the exact figure)

I have done that a while ago and got those results and I want to buy some comparator for an upcoming project that I will require this level of comparison.

What is the thing that I am missing?

[Zero999]

What comparator IC are you using?

This is most likely caused by noise, the voltage offset and drift and the fact it will most likely be operating in the linear region.

Read the data sheet, paying close attention to the open loop gain, voltage offset and drift specifications. Noise is probably important but isn't always specified on comparator data sheets.

I don't even know if this is practical. You need low offset with a drift in the 100s of nV range, very low noise and open loop gain of 140dB.

[bktemp]

Comparators are similar to opamps: Both have a finite gain.
If you want to compare very low voltages there are better ways, like using an precision opamp for preamplifying the signal before feeding it into a comparator.
http://www.linear.com/solutions/1585

[tszaboo]

A 3458A in 1000 NPLC can do this with the two inputs.
How many pointless topics are you going to open?

[nour]

A 3458A in 1000 NPLC can do this with the two inputs.
How many pointless topics are you going to open?

Untill cows come home 

[TimFox]

You need to specify the time allowed to make the comparison.  The shorter the time, the higher the required bandwidth, and the more noise.  Also, at very long times you will be fighting 1/f (flicker) noise.  There are old-school analog voltmeters that do this, but they are slow.  The suggestion above with a digital voltmeter has a long averaging time.

[nour]

What comparator IC are you using?

This is most likely caused by noise, the voltage offset and drift and the fact it will most likely be operating in the linear region.

Read the data sheet, paying close attention to the open loop gain, voltage offset and drift specifications. Noise is probably important but isn't always specified on comparator data sheets.

I don't even know if this is practical. You need low offset with a drift in the 100s of nV range, very low noise and open loop gain of 140dB.

At that time I was using 393 as I remember

[Zero999]

What comparator IC are you using?

This is most likely caused by noise, the voltage offset and drift and the fact it will most likely be operating in the linear region.

Read the data sheet, paying close attention to the open loop gain, voltage offset and drift specifications. Noise is probably important but isn't always specified on comparator data sheets.

I don't even know if this is practical. You need low offset with a drift in the 100s of nV range, very low noise and open loop gain of 140dB.

At that time I was using 393 as I remember

Then it's no surprise it won't work. Its open loop gain is only 200k The offset is 4mV and drift probably far too high, even if you trim it. Noise isn't specified but is also probably too high.

http://www.ti.com/lit/ds/symlink/lm393-n.pdf

I think bktemp hit the nail on the head: use an op-amp to increase the signal to a usable level before the comparator stage.

[The Soulman]

look at the null amplifier here:

http://conradhoffman.com/mini_metro_lab.html

[bktemp]

Circuit from MCP6V31 datasheet:


There are probably better opamps, but it gives an idea how it can be done.
The precision opamp amplifies the difference between both voltages before it feeds that signal into the comparator.
If both inputs need to be high impedance, you can build the precision amplifier in a fully differential configuration.

[nour]

Circuit from MCP6V31 datasheet:


There are probably better opamps, but it gives an idea how it can be done.
The precision opamp amplifies the difference between both voltages before it feeds that signal into the comparator.
If both inputs need to be high impedance, you can build the precision amplifier in a fully differential configuration.

When I thought about it before writing this thread, I thought I would do it like this but I wasn't sure enough about the idea and I thought maybe there is one integrated chip that can do this for a reasonable price.

[David Hess]

You will need to add low noise gain before the comparator.  In practice this means either adding a precision gain stage or using a precision operational amplifier as a comparator.  A comparator by itself will have have enough open loop gain to work at such low voltage differences.  Most operational amplifiers will not either.

[schmitt trigger]

It goes without saying that board layout is very critical. Dedicated ground planes, guard rings, excellent supply decoupling, perhaps even a shield.

And configure the preamp such that it also includes low pass filtering. The corner frequency depends on the required bandwidth.

Even so, you may still have random flips. If you are willing to have some speed vs. accuracy trade-offs, you could actually validate that the signal has flipped for at least -let's say- 10 msecs before assuming that the value change has occurred.

[danadak]

Just a thought but seems you should back into this by doing an
end to end error budget.

To start :

1) Temp range you want to do this over
2) Speed of measurement needed
3) Absolute accuracy desired
4) Resolution of measurement (I gather 1 ppm)
5) CM permissible

There are delsig ADCs that exceed 30 bits one could consider. But then you would
need a UP or PLD to do the acquisition and comparison. Your 1 part in a million
implies 21 bits, not including effects of T & V & PSRR &....

You can get a 20 bit delsig + ARM core in one part, a PSOC. That would allow you
to do averaging as well to improve S/N. Or use Corellated Double Sampling, see -


http://www.cypress.com/documentation/application-notes/an66444-psoc-3-and-psoc-5lp-correlated-double-sampling-reduce-offset



Regards, Dana.

[nour]

Just a thought but seems you should back into this by doing an
end to end error budget.

To start :

1) Temp range you want to do this over
2) Speed of measurement needed
3) Absolute accuracy desired
4) Resolution of measurement (I gather 1 ppm)
5) CM permissible

There are delsig ADCs that exceed 30 bits one could consider. But then you would
need a UP or PLD to do the acquisition and comparison. Your 1 part in a million
implies 21 bits, not including effects of T & V & PSRR &....

You can get a 20 bit delsig + ARM core in one part, a PSOC. That would allow you
to do averaging as well to improve S/N. Or use Corellated Double Sampling, see -


http://www.cypress.com/documentation/application-notes/an66444-psoc-3-and-psoc-5lp-correlated-double-sampling-reduce-offset



Regards, Dana.

1) Temp range you want to do this over => 20~28 (45 considerable) C
2) Speed of measurement needed => 100~100,000 /s
3) Absolute accuracy desired => -+ 1 ppm
4) Resolution of measurement (I gather 1 ppm)
5) CM permissible => not sure!

I started this because I have read about integrated ADCs to be more specific the dual slope integrating ADC and I wanted to try to experiment with doing it from discrete components, I am not sure how silly this idea is or how easy or how applicable it would be but this is how I learn stuff! and I want to try it very badly!

Thanks!

[BrianHG]

I've seen that many super precision voltmeters, above 5 digits, may use chopper-stabilized op-amps in their acquisition circuits.  To add gain to your comparator, and keep that 0.0000005v differential offset voltage or even better, have you considered a chopper stabilized op-amp, whether it be custom built, or per-packaged IC, or, is this overkill for you?

I know Analog devices and TI have at least one model each.  You might not even need a comparator here.  You wont get high bandwidth though.

[BrianHG]

It would be funny to build a chopper stabilized op-amp with chopper stabilized opamps like the TLC2652AC, that must give you an insane amount of precision, though very low frequency response and pricey since most old DIY chopper op-amps used multiple normal instrumentation op-amps in their design.  Let's now talk about pico-volts... 

[nour]

It would be funny to build a chopper stabilized op-amp with chopper stabilized opamps.

that was good 

I am not yet ready to do this, I will investigate more about it

[David Hess]

Chopper and chopper stabilized amplifiers expect to be run in a closed loop and will saturate if used in a comparator application.  There is an alternative although I do not remember ever hearing of it being used.  Chop the differential input signal like you would do with a chopper amplifier, amplify it, and demodulate it.  Now the output represents the comparison.  Add clamping from the output to the input stage to prevent saturation.

It would probably be better to chopper stabilize a fast low noise clamped input amplifier and use that to drive the comparator.  Clamping is needed to prevent the input amplifier from saturating which will cause the chopper to saturate.  But some precision amplifiers have a low enough offset drift that chopper stabilization would not be needed.

[ConKbot]

Is there something wrong with using an instrumentation amplifier that I'm missing here? Use one with whatever gain/bandwidth you need gain the uV level difference up to a mV level difference, offset it with whatever voltage plays nice with whatever cheap comparator you use that doesn't have horrible offset, all the trimming and matching is done already, with an error budget specified in the datasheet.

[danadak]

One issue with high G IA is GBW since application wants up to 100K SPS.

Examine this as an issue.


Regards, Dana.

[David Hess]

Is there something wrong with using an instrumentation amplifier that I'm missing here?

An instrumentation amplifier could be used but it would have no advantage over a differential amplifier and most will perform worse.

[Marco]

Is there something wrong with using an instrumentation amplifier that I'm missing here?

They don't necessarily have great offset voltage drift.

The LTC1052 has a clamp feature to prevent saturation when used as a comparator. Since it's a zero-drift/autozero/chopper opamp that would be a single IC solution.