Lab Grade Instrumentation Amplifier - DIY or Buy A Chip?

[t1d]

My Background = Self-taught hobbyist with significant knowledge gaps, due to only studying what interests me. I have good skills... KiCad, ordering boards and soldering them up. I also have a really good hobbyist lab and can test most things.

I love to DIY my own EE bench tools. A Three-Op-Amp Instrument Amplifier design looks easy enough. However, it appears that there are all-in-one ICs that have better spec's than what I would end up with, if I built it, myself. Of course, there is the joy of learning and the satisfaction of building. But, the circuit is so straight forward that I think I have learned what there is to know. IMO, a chip price of say $20 would be worth it, for true lab-grade spec's. So, DIY, or Buy, and Why? If buy, what chip do you suggest? It could include more advanced features/techniques, of course. Thanks!!!

[Sensorcat]

If you enjoy to DIY the 3-OpAmp instrumentation amplifier, do it!

If you need it for another project to work, as a component of that project, it is wiser to find an integrated instrumentation amplifier. For help which one to choose, you must provide detail on your requirements.

[moffy]

A brief answer is that an all in one chip will tend to have a high CMRR due to matched/laser trimmed components which would require expensive matched resistor networks to replicate. As far as a recommendation, what is it intended for? Low noise, low offset, large voltage range, low bias current, high input impedance? etc. etc. The recommendation depends on the use cases.

[TimFox]

If you need to match resistors from a batch, a good DMM in ohms mode can be trusted for equality of a pair to its resolution, better than the calibration accuracy.  Also, it is easy to trim CMRR at DC by zeroing the offset with the two inputs connected to zero volts, then applying a reasonable voltage to both inputs (still connected together) and adjust a suitable trimpot in one resistor.

[t1d]

Excellent answers, everyone! Thank you.

To answer Sensorcat and moffy, my use is just a nice, general purpose instrument amplifier to have on the bench for general purposes. I do not have a specific need in mind. I have moved into the nicer grade of hobbyist instruments. A Siglent SDM3065X (DMM), SDS1204X-E (Scope), SPD 3303X (PSU) and Rigol 1062Z (FG), etc. So, I would like for you to recommend a capable IC... I am on a retirement budget, but the chip doesn't have to be dirt cheap. $25USD might be an idea of what I can manage. But, cheaper than that is fine to... <grin>

I appreciate your help and look forward to investigating your recommendations.

[moffy]

Without specs it is impossible to advise, at Digikey they range from $1 AU to over $500 AU here is a search selection:
https://www.digikey.com.au/en/products/filter/linear/amplifiers/instrumentation-op-amps-buffer-amps/687?s=N4IgjCBcpgbB0QGMoDMCGAbAzgUwDQgD2UA2iAMwCsAHAJw2wgC6hADgC5QgDKHATgEsAdgHMQAX0IBaAExRQKSAICuBYmRBUWUkNLoLkUVepKRyEZhN3bEbKGHb3IsqrqaJBAE27SwABgh2LkgQEEIOAE82XG50bBRrIA

[Kleinstein]

There is no universal best amplifier, the same amplifier can be good for one job and awful for another. It really depends on the signal source to look at, which amplifier is good. It is more about voltage noise and offset voltage for a low impedance source and more about current noise and bias for a high impedance cource.

For the more intermediate cases there are good off the shelf instrumentation amplifiers. For more unusual cases (especially need for "zero" drift or very low bias, but also a large voltage range) the separate build 3 OP-amp solution still makes sense.
For the chips it also depends on the supply range. INAs usually have a somewhat limited common voltage range - a point sometimes overlooked.

[Phil1977]

DIY or buy is from my experience heavily dependant on the frequency range you need.

If you have more or less static input voltages then DIY setups can quite easily be well calibrated to the required accuracy. You also get quick feedback during the calibration process if things are okay or not.

If you have input signals with a higher bandwidth (100Hz or more) I´d strongly recommend to use a dedicated chip with matching specifications. It´s complex in the narrowest sense of the word to design amplification stages with matching frequency response. You definitely need simulations and very precise components for that.

If you want to have an idea about the maximal frequency that can be handled as quasi-static you can look into the datasheets of the respective opamps. Usually you find a "closed loop gain against frequency" plot. There you can look for the curve that´s matching your maximum gain and then see where the curve is flat. You need to be safely in the flat zone so that you can rely on a purely static calibration

[Conrad Hoffman]

Commercial ones can be expensive and I've sometimes had less than stellar results using them. I'd DIY your own and then try a commercial one if you have special needs that yours can't do. There are a couple different configurations you can try and you can certainly use trimmers to get good CMRR.

[t1d]

To each and all, thank you for the really great replies and information. You have pointed out a lot of type-of-use considerations that I did not know about. Wonderful! I like to learn...

Commercial ones can be expensive and I've sometimes had less than stellar results using them. I'd DIY your own and then try a commercial one if you have special needs that yours can't do. There are a couple different configurations you can try and you can certainly use trimmers to get good CMRR.

Conrad's suggestion seems to be the way to go, for me. I think I will try this one, just for fun: Any tips, tricks, or pitfalls to know of?

Thanks! Cheers!

[Kleinstein]

Unless one wants really high overall gain, the 2nd stage gain should be relatively low (like x2 or x5). The first stage gain improves the CMRR of the 2nd stage. So one wants much of the gain in the first stage and not in the 2nd.

The is a small trap for the young players: don't use an OP-amp that is not unitiy gain stable in the input stage. One usually has quite some gain for the differential part, but not for the common mode part. This is enough for the system to oscillate with OP-amps that are not unity gain stable (e.g. OP37 / LT1037, LF357).

I would avoid trimmers where no needed, like in the first stage.

[TimFox]

If you need to tweak the CMRR at high frequencies, you could add a small fixed capacitor across R1 and a trimmer capacitor across R2, or across R6 and R7+R9. 
However, I agree with Kleinstein's suggestion about high gain in stage 1 and lower gain in stage 2.

[Conrad Hoffman]

That's the classic circuit and works well. Capable of high gain, but follow the above recommendations. A hint- signal to noise is always established at the first stage of an amplifier. You can never improve it after that, only wreck it, so pay attention to that first stage!

[t1d]

More great information from everyone! Thank you! See my replies to quotes in the individual posts, below.

[t1d]

Unless one wants really high overall gain, the 2nd stage gain should be relatively low (like x2 or x5). The first stage gain improves the CMRR of the 2nd stage. So one wants much of the gain in the first stage and not in the 2nd.

The is a small trap for the young players: don't use an OP-amp that is not unitiy gain stable in the input stage. One usually has quite some gain for the differential part, but not for the common mode part. This is enough for the system to oscillate with OP-amps that are not unity gain stable (e.g. OP37 / LT1037, LF357).

I would avoid trimmers where no needed, like in the first stage.

Hi, Kleinstein! So great to have your expertise! I am still enjoying the e-load that you helped me design.

It is interesting advice about having the gain in the first stage, because that is somewhat non-intuitive. One would think that high upstream amplification might clip the downstream amplifier.

I do not understand why the design has such high gain. If we are looking at a signal source that needs 1000 times amplification, isn't it so far down in the noise floor as to be unmanageable (in terms of reading it, particularly with hobbyist instruments,) even with the built-in differentiation function?

If the second stage amplification would be better at x2-x5, might a third stage be useful to bring it up to say x10, or x100?

I would like to have some means to adjust the gain. A pot would introduce the problems that pots have - temperature instability, etc. Mechanically switching set gain brackets is simple and reliable, but may lack needed resolution. Digital pots seem to work around such issues and are doable, within my skill set. Well, what I could learn and employ... Thoughts?

Thanks! Next...

[t1d]

If you need to tweak the CMRR at high frequencies, you could add a small fixed capacitor across R1 and a trimmer capacitor across R2, or across R6 and R7+R9. 
However, I agree with Kleinstein's suggestion about high gain in stage 1 and lower gain in stage 2.

Great, Tim! It is easy enough to add auxiliary footprints to the board and populate these components, at a later date, if needed/desired.

[t1d]

As I am just learning about this particular circuit, I could use help with the specifics of what components to add, or change, to improve the circuit with the groups suggestions. Calculations (for Tim's caps,) what to place where, etc...

Here is an interesting video. But, I have not finished watching the series.

[t1d]

What about breadboarding a proof of concept with TL081s and a LM741?

What about using an 8-pin 10K resistor array, for better matching and thermal coupling. Three of the resistors for R1,2&6 and the remaining five in parallel for R3/2K1-1%? And a dual array for R4&5/100R-1%?

[Kleinstein]

A resistor array makes sense. It is only the 2nd stage to effect the CMRR. The first stage resistors are for gain stability only.
Quite often the 2nd stage is with a gain of 1 and thus 4 equal resistors (e.g. 4 x 10 K to 50 K depending on the supply / voltage range). An array of 7 or 8 may be an idea for the gain of 2 or 3 case (e.g. CM range more important than CMRR).

[t1d]

EDIT: While not considering the circuit correctly, I posted several comments that were just plain wrong. I have removed those, to try to prevent further confusion. I will have to try again when my health has improved.

[Terry Bites]

A regular differential amplifier design, assuming an ideal opamp and discrete resistors, CMRR≈ 20* log((1+Av)/4t)
Where t is the per unit resistor tolerance. That means that the best you could get with 0.1% resistors is about 54dB. That still assumes that the amplifier has infinite CMRR which a real world opamp does not. Using an array reduces drift but only provides a very modest improvement in CMRR. CMRR≈ 20* log((1+Av)/3t)
See www.eetimes.com/the-effects-of-resistor-matching-on-common-mode-rejection.

Discrete INAs are best avoided because of these limitations. The 3 opamp design is pretty much a just a text book thing these days. Still highy instructive though.
Compare the price of precision parts, tweaky bits and a precision opamp with the price of a decent INA eg an INA849. No contest!

Consider the practicalities such as power supplies, gain control and input protection/ coupling.

Here are a couple more CMRR tweaks that can be applied to the 3 amp design.

[t1d]

Thank you, Terry. Your conclusions that a chip would have better spec's and economies is quite true. However, there seemed to be a conscience that it would be difficult to select a particular chip, without a well defined stated use.

I have briefly scanned the 849 DS. I do not generally work with high frequencies. So, maybe this chip will work for me as a general bench amplifier. It can't hurt to give it a try.

I wonder if there might be a demo board PCB in the documents. I am sure that the board layout is critical.

A battery power supply seems advantageous... What are the needed considerations for selecting a single, or split, power supply design?

Thanks for your help and participation!

[Kleinstein]

The simple CMRR estimate applies to a simple 1 stage difference amplifier. With the 3 OP-amp design the CMRR can get better by the gain of the initial stage. The input stage amplifies the differential signal, but not the common moder signal. With high gain even the separate build version can reach high CMRR.

Which INA fits really depends. The INA849 is an example with very low voltage noise, but also very high current noise. So it may be good, but could also fail with high impedance source resistance.

Especially with battery supply one has to watch the input voltage limitations. INAs are usually not rail to rail input, but may need quite some headroom. How much depends on the type and gain. Especially with the very low noise types like INA849 a high supply can cause quite some power consuption.

[Terry Bites]

Quite so. There are loads of types, none of which are ideal. Horses for courses.
I'm guessing that a bench amp would largely be used as a low gain differential to single ended converter.
T1d needs to define general purpose of course.