3-OP amp instrument amplifier "non-linearity" issue

[PeterF]

Hi!

I am working on a design with a strain gage amplification.
(see attached image of schematic).

I have tested it using LT6005 quad OP-amp, which is working quite fine, except for a strange nonlinearity in the middle of the range.
(see attached image)


I have an alternative OP amp part which I could use, which I also did the measurements with (see attached image).

The LT6085 is not showing the non-linear behavior in the middle area of the range.



I performed the measurements by stepping of the DAC which I have for offset adjustment, and measuring the input and output in many steps, adding them in Excel and plotting the curves I've attached here.


To the question:


Can anyone help me figure out what the cause of this nonlinearity is?

Why is the LT6085 not having this issue?


(I am planning on perhaps changing to a totally different part in the future, but right now I want to understand what the cause is of my issue here.)


Thanks in advance 

[tszaboo]

What are the power supplies feeding the opamp? Do you have a dual power supply? Are you violating the output swing?

[danadak]

There is a phenomenon on RRIO input CMOS OpAmps regarding input stage crossover
distortion. However the 065 is a bipolar inout, so not sure whats going on, whether
its input or output stage crossover distortion.

On a more important point given you have a G = 1000 are you using high precision
feedback resistor array to insure you are getting the CMR you are expecting, as well
as matching of drifts, offsets ? You might want to consider an integrated IA as they are
laser trimmed at production to eliminate most errors. Analog Devices excellent
supplier of those.

Attached an eye opener analysis of CMR vs feedback ratio and Aol OpAmp to OpAmp impact
on CMR.


Regards, Dana.

[PeterF]

What are the power supplies feeding the opamp? Do you have a dual power supply? Are you violating the output swing?

LT6005 PDF:
http://cds.linear.com/docs/en/datasheet/600345fd.pdf

LTC6085 PDF:
http://cds.linear.com/docs/en/datasheet/60845fa.pdf

I am using a single 2.5 V supply, which to my knowledge should be fine on both (even more fine on the LT6005).

[PeterF]

There is a phenomenon on RRIO input CMOS OpAmps regarding input stage crossover
distortion. However the 065 is a bipolar inout, so not sure whats going on, whether
its input or output stage crossover distortion.

On a more important point given you have a G = 1000 are you using high precision
feedback resistor array to insure you are getting the CMR you are expecting, as well
as matching of drifts, offsets ? You might want to consider an integrated IA as they are
laser trimmed at production to eliminate most errors. Analog Devices excellent
supplier of those.

Attached an eye opener analysis of CMR vs feedback ratio and Aol OpAmp to OpAmp impact
on CMR.


Regards, Dana.

Thanks Dana,

I am at this point using 0.1% resistors. I am having soome issue with 50 Hz noise on the output, but I can live with that.
I may try to improve this in the future, but right now I am getting the performance I want (well except for the noninearity).

The reason for not using an integrated solution was for me the price point.
I may however have to look some more into perhaps using an integrated solution in the future. But I was quite happy with this performance before I saw this behavior   

In the future, I am planning on finding a cheaper alternative. Since offsets are handled by my DAC, I am hoping to achieve similar performance to the current solution at a lower price point.
If only I can get my head around this issue.

[tszaboo]

Datasheet page 3:
Input Voltage Range Guaranteed by CMRR l 0.3 V+ – 0.3V V
So I guess you understand, that in case you feed in 2mV, the opamp will not work correctly, because it is outside its input voltage range. I guess the second stage will have the same issue for low voltages.
Dont forget that it is enough to overload just one amplifier to make the other three behave badly.
I dont btw that building INAs is a good idea. For example, an INA331 is cheaper than the LT6008. Or an INA333, if you want the traditional arrangement. And you need the precision resistors, and stuff, which adds to your BOM. I just dont see the point.

[danadak]

Maybe I am missing something but I thought the CM Range was supply rated -

Attached partial spec.

On another matter you might contact FAE at LTC to see if there is some crossover distortion
at close to rail behavior in this OpAmp. He will check with factory.

Regards, Dana.

[wraper]

The reason for not using an integrated solution was for me the price point.

Then you are doing this wrong, because it sucks at the performance and the price too.

[tszaboo]

Maybe I am missing something but I thought the CM Range was supply rated -

Attached partial spec.

On another matter you might contact FAE at LTC to see if there is some crossover distortion
at close to rail behavior in this OpAmp. He will check with factory.

Regards, Dana.

Sorry, it is page 5...
But you are right, a different table has different values.
(LT6003C/I, LT6004C/I, LT6005C/I) apparently has different input voltage range than the (LT6003H, LT6004H, LT6005H).
That looks like there are some shenanigans going on in the datasheet. 

[PeterF]

Another reason for this strategy is that I want to offset the input from software, which is what the DAC is for.
(To zero out the big offset in the wheatstone bridge between different samples).

For that I added a buffer, and I liked the idea of just having a single 4-OPamp part to allow all of this functionality.


Therefore, I can handle offsets, which was my point of going down to a cheaper 4-OP part in the future.


The CM voltage is 1.25 V on both inputs, with a few mV change when measuring.

[tszaboo]

Which you can absolutely do with a regular INA. Just scale the DAC output and route it to the REF pin. Must be low impedance drive (saying just in case).

Therefore, I can handle offsets, which was my point of going down to a cheaper 4-OP part in the future.

It is not cheaper.

[danadak]

As an aside if you are using a UP Cypress PSOC has a 20 bit A/D on board,
a .1% onboard reference, and differential input. Additionally .2mV offset,
and a 8 bit Current (or V) DAC that can be used to do offset correction.

Or you could do correlated double sampling

http://www.cypress.com/file/119666/download

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


Regards, Dana.

[wraper]

In one of mine designs I interface MCU with a load cell for about EUR 1 total BOM cost @100 pcs,  less than EUR 1.5 @1 pcs. I use AD8293G160 which works perfectly for this, offset is almost non existent too, nothing like the crap you get. Don't forget that there is offset and offset temperature drift too, latter will be just horrible for your circuit.

[danadak]

@wraper, that is a good looking part, dirt cheap I might add. Only issue,
minor, I see is fixed Gs of 80 and 160, that must have been driven by a
specific high volume customer they had/have. Or a design with one of their
A/Ds that has a specifc common mode range to maximize dynamic range.


Thanks for that post, i will use that part now myself.


Regards, Dana.

[wraper]

@wraper, that is a good looking part, dirt cheap I might add. Only issue,
minor, I see is fixed Gs of 80 and 160, that must have been driven by a
specific high volume customer they had/have. Or a design with one of their
A/Ds that has a specifc common mode range to maximize dynamic range.


Thanks for that post, i will use that part now myself.


Regards, Dana.

I was in search search for such dirt cheap part with a good specs for a few days in a row, brute forcing major distributor catalogs . When found it was like blessing, exactly what I desired. The idea behind my previous post was - spend a good time for investigating suitable parts at reasonable price. Don't make ridiculous schematic decisions when saving the costs.

[Ammar]

C16 and C17 would have to be very well matched. What type of capacitors are you using there and what are their tolerances?

[StillTrying]

The CM voltage is 1.25 V on both inputs

Which is where the kink in the plot is.

0-2.5V on DAC gives -2.5 to +2.5mV offset adjustment

As long as the input doesn't move from 1.25V.

[splin]

The CM voltage is 1.25 V on both inputs

Which is where the kink in the plot is.

Errm where? Perhaps I'm missing something obvious but could you explain what you mean? Do you literally mean the kink(s) in the OP's transfer plots?

PeterF. It seems that no-one has answered your questions.

First point is that the DAC offset range is +/-1.25mV not the +/-2.5mV shown.

More importantly is that you appear to have too much gain: 2010 X 2mV input swing (as shown in your plots) is over 4V, but you only have a 2.5V supply. However your plots actually show gains of only 1900 which is odd given that your cct shows 0.1% resistors. I would check your cct actually matches the diagram.

The excess gain might explain why your first plot shows the output to be saturated until the input reaches 0.7mV but not the non-linearity mid scale and beyond. Another oddity is the LT6005's 74mV saturation voltage - that suggests that the output, TP104 is connected to 2.5V. If it were only connected to a high impedance voltmeter the o/p should get much closer to ground. However that still wouldn't explain the non-linearity.

Simulating your cct in LTspice shows that it should work fine providing you reduce the gain by half and get the offsets right so I don't know what is going off in your case. Even keeping the gain at 2010 it stays linear until the o/p saturates. It doesn't appear to be a common mode voltage problem as all the CM voltages stay within 200mV of mid supply. The datasheet does show a huge input offset shift when CM gets within 1.1V of the positive supply, of between 150uV to 220uV depending on package. With your gain of 2010X that would be significant but the simulation doesn't show Vcm getting above 1.4V. It might be worth measuring though.

My guess is that your circuit doesn't actually match the diagram or perhaps the LT6005 is damaged. Please let us know how you get on.

[danadak]

A couple of tools that make IA design a lot easier -

http://www.analog.com/designtools/en/diamond/

http://www.analog.com/en/design-center/interactive-design-tools/adi-diffampcalc.html


Regards, Dana.

[StillTrying]

The CM voltage is 1.25 V on both inputs

Which is where the kink in the plot is.

Errm where? Perhaps I'm missing something obvious but could you explain what you mean? Do you literally mean the kink(s) in the OP's transfer plots?

I was miss-reading mV for V. But still I would expect output kinks using lt6005 at very high gains.

As I understand it the last graph on page 9 of the data sheet shows a strange change in the input offset voltage as the output passes through the mid point, =0V on the data sheet with +-8V rails probably =1.25V here! - as soon as there's any load on the output.

Page 13 says:
the input offset voltage shift over the entire common mode range (CMRR) is typically 160uV, maintaining the precision characteristics of the amplifier.

What is doesn't say is that the typically 160uV shift occurs suddenly 0.9V below V+, which is only 1.6V with V+ of 2.5V. 6th graph on page 8.

...And a GBW of 2KHz.

[splin]

The CM voltage is 1.25 V on both inputs

Which is where the kink in the plot is.

Errm where? Perhaps I'm missing something obvious but could you explain what you mean? Do you literally mean the kink(s) in the OP's transfer plots?

I was miss-reading mV for V. But still I would expect output kinks using lt6005 at very high gains.

As I understand it the last graph on page 9 of the data sheet shows a strange change in the input offset voltage as the output passes through the mid point, =0V on the data sheet with +-8V rails probably =1.25V here! - as soon as there's any load on the output.

Page 13 says:
the input offset voltage shift over the entire common mode range (CMRR) is typically 160mV, maintaining the precision characteristics of the amplifier.

What is doesn't say is that the typically 160mV shift occurs suddenly 0.9V below V+, which is only 1.6V with V+ of 2.5V. 6th graph on page 8.

...And a GBW of 2KHz.below

Page 13 does say that the input switches from a pnp to an npn stage at inputs above .9V from the positive supply which as you say will be 1.6V. However in the OP's circuit none of the inputs should see voltages above 1.4V, given that the OP says that the common mode voltage on the inputs is 1.25V, so this can't be the explanation of the non-linearity. It any case I would expect it to manifest itself as a change in offset but the slope thereafter shouldn't be affected.

[danadak]

Sure makes on think that because this is a very low V opamp and what compromises
bias networks to get the performance.

In R-R input CMOS OpAmps there is a problem with crossover between N and P MOS
transistors. Some vendors do not discuss this in datasheet, others do. But the problem
exists in ALL switched input stages. Some make the transition better than others.

I guess 40 years ago after class AB output stage distortion was conquered its time to give
us a new distortion to deal with.


Regards, Dana.

[StillTrying]

Whatever it is I can simulate something that looks very similar to me by adding a sudden 80 uV offset in series with the -ve input, 1 opamp, gain 940, +ve input 1.25V +-1mV, load 10K.

The .asc is the above circuit with a fix, gain is 2000, BW is 3 Hz.

I don't think this LT6005 is useful for anything meself!