High on my to do list was the intention to measure the current noise for OpAmps with Input bias-current compensation or IBC as in the AD797 and the LT1028 to name a few.
Not a very sexy item, but nevertheless an issue where many different opinions are being uttered, reason enough to dig a bit deeper.
IBC is a very effective way of reducing the input bias current of a bipolar amp where a current of opposite direction and the same magnitudeis injected to both inputs.
See Image below where the input currents of Q1 and Q2 are cancelled by the currents from Q3 and Q4.
In an ideal situation, input bias current from the outside would become zero, but in practice there will always be a small difference between both, leaving only a fraction to be sinked or sourced through +In and –In.
However, injecting current also adds additional current noise from Q3 and Q4 to the already existing current noise of Q1 and Q2.
When this current noise being produced by Q3 and Q4 would be 100% correlated and the CMRR of the amp very high, contribution of IBC on input current noise would be cancelled if using matched input resistors.
Only the uncorrelated input current noise of Q1 and Q2 would then contribute to the total noise production.
And that's where my search started, on one hand because a fully correlated IBC current noise seems impossible because of the use of three different transistors Q3, Q4 and Q6 involved in producing this current, see fig 1 and on the other hand because figures from manufacturers are not conclusive.
That’s why I have given Q3 and Q4 two different current noises, In3 being 100% correlated and In4 and In5 being completely uncorrelated.
My assumption is that noise from Q3, Q4 and Q6 is mainly caused by the base spread resistor Rb1, Rb2 and Rb3.
When that being the case, and all three resistors having equal values, then Q3 and Q4 will add their own noise from Rb2 and Rb3 to Q6’s noise from Rb1, all being of equal magnitude.
That would mean that ca. half the noise currents from Q3 and Q4, being generated byRb2 and Rb3, would be uncorrelated and the other correlated half would come from Q6’s Rb1 ………….…….[1]
Let’s see in how far this postulate agrees with the outcome of this test.
Not a very sexy item, but nevertheless an issue where many different opinions are being uttered, reason enough to dig a bit deeper.
IBC is a very effective way of reducing the input bias current of a bipolar amp where a current of opposite direction and the same magnitudeis injected to both inputs.
See Image below where the input currents of Q1 and Q2 are cancelled by the currents from Q3 and Q4.
In an ideal situation, input bias current from the outside would become zero, but in practice there will always be a small difference between both, leaving only a fraction to be sinked or sourced through +In and –In.
However, injecting current also adds additional current noise from Q3 and Q4 to the already existing current noise of Q1 and Q2.
When this current noise being produced by Q3 and Q4 would be 100% correlated and the CMRR of the amp very high, contribution of IBC on input current noise would be cancelled if using matched input resistors.
Only the uncorrelated input current noise of Q1 and Q2 would then contribute to the total noise production.
And that's where my search started, on one hand because a fully correlated IBC current noise seems impossible because of the use of three different transistors Q3, Q4 and Q6 involved in producing this current, see fig 1 and on the other hand because figures from manufacturers are not conclusive.
That’s why I have given Q3 and Q4 two different current noises, In3 being 100% correlated and In4 and In5 being completely uncorrelated.
My assumption is that noise from Q3, Q4 and Q6 is mainly caused by the base spread resistor Rb1, Rb2 and Rb3.
When that being the case, and all three resistors having equal values, then Q3 and Q4 will add their own noise from Rb2 and Rb3 to Q6’s noise from Rb1, all being of equal magnitude.
That would mean that ca. half the noise currents from Q3 and Q4, being generated byRb2 and Rb3, would be uncorrelated and the other correlated half would come from Q6’s Rb1 ………….…….[1]
Let’s see in how far this postulate agrees with the outcome of this test.
Attachments
I took the LT1028 specs, a low noise bipolar where the specs mention In=1pA/rtHz when using matched source resistances.
A total noise graph from LT is shown below for resistance values from 1Ohm to 10kOhm.
With a specified input voltage noise of En=0.85nV/rtHz@1KHz, a current noise In=1pA/rtHz@1KHz and a 10KOhm resistor producing 12.9nV/rtHz, total noise should be:
Sqrt(2*(12.9e-9)^2+2*(In*10K)^2+En^2) = 23.1nV/rtHz.
However the added graph shows a total noise of ca 35nV/rtHz for matched 10K resistors.
So there is something quite far off and this can only be the noise current In, in this case the dominant contributor to the noise.
The AD797 having the same specs is just as far off with 1pA/rtHz.
I won’t go further into details, but the unmatched resistor graph is also wrong.
Therefore I tested the AD797 with resp two 14K resistors and with one single 14K resistor at resp. the minus and the plus input.
With the two matched resistors R3 and R4 voltage noise by taking the FFT was measured at 47.6nV/rtHz and with only one resistor noise 40.1nV/rtHz, in this case even 1.5dB less.
It made no difference whether the single resistor was used in front of +in or–In.
A total noise graph from LT is shown below for resistance values from 1Ohm to 10kOhm.
With a specified input voltage noise of En=0.85nV/rtHz@1KHz, a current noise In=1pA/rtHz@1KHz and a 10KOhm resistor producing 12.9nV/rtHz, total noise should be:
Sqrt(2*(12.9e-9)^2+2*(In*10K)^2+En^2) = 23.1nV/rtHz.
However the added graph shows a total noise of ca 35nV/rtHz for matched 10K resistors.
So there is something quite far off and this can only be the noise current In, in this case the dominant contributor to the noise.
The AD797 having the same specs is just as far off with 1pA/rtHz.
I won’t go further into details, but the unmatched resistor graph is also wrong.
Therefore I tested the AD797 with resp two 14K resistors and with one single 14K resistor at resp. the minus and the plus input.
With the two matched resistors R3 and R4 voltage noise by taking the FFT was measured at 47.6nV/rtHz and with only one resistor noise 40.1nV/rtHz, in this case even 1.5dB less.
It made no difference whether the single resistor was used in front of +in or–In.
Attachments
A simple calculation from the above measurements learns that with matched resistors In=2,14pA/rtHz and In=2,65pA/rtHz with a single resistor.
Suggestions that current noise with unmatched resistors may become 3 times as large, from 1pA/rtHz to 3pA/rtHz, has been proved here as absolutely wrong.
My feeling is that the specified 1pA/rtHz by AD and LT may be valid for the bare Amps current noise without IBC and that the assumption was made that the IBC carries 100% correlated current noise.
Assuming this “guess” being correct, then the uncorrelated part of the IBC current noise for In4 and In5 in the first posting will be the just measured 2.14pA/rtHz minus the specified 1pA/rtHz resulting in:
sqrt(2.14^2-1^2) = 1.89pA/rtHz
There is also a correlated IBC noise part, named In3 in # 1, being the difference in noise with two matched resistors versus only one resistor inserted,resulting in:
sqrt(2.65^2-2.14^2) = 1.56pA/rtHz
In that case the IBCcurrent noise being close to 50/50 correlated /uncorrelated (1.56 vs 1.89) , which seems much better than the 100% correlated assumption and very close to postulate [1] in the first posting.
Anyhow, in the overview below one can see the Equivalent Input Noise or EIN for different values of the matched and unmatched input resistors making use of the measured noise figures in this test, where the unmatched situation always gives a better overall result, quite contrary to what is often thought.
Hans
Suggestions that current noise with unmatched resistors may become 3 times as large, from 1pA/rtHz to 3pA/rtHz, has been proved here as absolutely wrong.
My feeling is that the specified 1pA/rtHz by AD and LT may be valid for the bare Amps current noise without IBC and that the assumption was made that the IBC carries 100% correlated current noise.
Assuming this “guess” being correct, then the uncorrelated part of the IBC current noise for In4 and In5 in the first posting will be the just measured 2.14pA/rtHz minus the specified 1pA/rtHz resulting in:
sqrt(2.14^2-1^2) = 1.89pA/rtHz
There is also a correlated IBC noise part, named In3 in # 1, being the difference in noise with two matched resistors versus only one resistor inserted,resulting in:
sqrt(2.65^2-2.14^2) = 1.56pA/rtHz
In that case the IBCcurrent noise being close to 50/50 correlated /uncorrelated (1.56 vs 1.89) , which seems much better than the 100% correlated assumption and very close to postulate [1] in the first posting.
Anyhow, in the overview below one can see the Equivalent Input Noise or EIN for different values of the matched and unmatched input resistors making use of the measured noise figures in this test, where the unmatched situation always gives a better overall result, quite contrary to what is often thought.
Hans
Attachments
Last edited:
Fascinating
You have ignored the current sink I2 in your noise model, leaving only Q6 as the source of correlated noise.
Your circuit diagram shows positive feedback on the AD797 BTW
When you take out one of the resistors you'd expect the noise to reduce, and everything being equal by 3dB. Your 1.5dB reduction is actually 1.5dB more than expected? I may be confused here of course.
I calculate the expected matched 14k noise as 29.2nV/rtHz, you measure 47.6nV/rtHz, meaning there is 37.6nV/rtHz unaccounted for (assuming the device is matching the typical performance figures).
With one 14k resistor I expect 20.7nV/rtHz (ignoring IBC), you measure 40.1, leaving 34.3nV/rtHz unaccounted for.
This begs the question: is there some other source of noise involved (it could be dominant, if my figures are right).
Did you try removing both resistors?
You have ignored the current sink I2 in your noise model, leaving only Q6 as the source of correlated noise.
Your circuit diagram shows positive feedback on the AD797 BTW
When you take out one of the resistors you'd expect the noise to reduce, and everything being equal by 3dB. Your 1.5dB reduction is actually 1.5dB more than expected? I may be confused here of course.
I calculate the expected matched 14k noise as 29.2nV/rtHz, you measure 47.6nV/rtHz, meaning there is 37.6nV/rtHz unaccounted for (assuming the device is matching the typical performance figures).
With one 14k resistor I expect 20.7nV/rtHz (ignoring IBC), you measure 40.1, leaving 34.3nV/rtHz unaccounted for.
This begs the question: is there some other source of noise involved (it could be dominant, if my figures are right).
Did you try removing both resistors?
Last edited:
Thank you for your reaction.Fascinating
You have ignored the current sink I2 in your noise model, leaving only Q6 as the source of correlated noise.
Your circuit diagram shows positive feedback on the AD797 BTW
When you take out one of the resistors you'd expect the noise to reduce, and everything being equal by 3dB. Your 1.5dB reduction is actually 1.5dB more than expected? I may be confused here of course.
I calculate the expected matched 14k noise as 29.2nV/rtHz, you measure 47.6nV/rtHz, meaning there is 37.6nV/rtHz unaccounted for (assuming the device is matching the typical performance figures).
With one 14k resistor I expect 20.7nV/rtHz (ignoring IBC), you measure 40.1, leaving 34.3nV/rtHz unaccounted for.
This begs the question: is there some other source of noise involved (it could be dominant, if my figures are right).
Did you try removing both resistors?
You are right about switching the AD797 inputs in the image, the intention however may be clear 🤗
Have you looked at the calculation at the end of the third posting, there you can see how all figures where calculated and that no single current noise was ignored.
And indeeed did I measure without the 14k resistors to find the amps voltage noise. But with the 14k resistors inserted, voltage noise plays no role with these OpAmps as can be seen in the calculations.
For any further clarification , I will be happy to answer.
Hans
The main issue of my postings as the title said was to find the difference in matched versus unmatched current noise for IBC amps.
The input current of 1pA/rtHz that LT specifies was just used as a starting point, but the outcome from measurements was that unmatched resistors produce less noise as matched resistors and that was the whole purpose.
Hans
With unmatched I followed the way that LT used, being one resistor having a certain value and the other one being zero.
Last edited:
Why do you make that assumption? I think a competent designer should have the good sense to use transistors with a reasonably low base resistance, such that collector shot noise dominates.
In the case of the LT1028, the schematic in the datasheet clearly shows that Q6 runs on a much smaller current than Q3 and Q4. That is, it is an amplifying current mirror of which the noise will be dominated by the input transistor. The model NPN that produces the model base current (I2 in your schematic) produces as much base shot noise as Q6's collector shot noise. As these noise contributions are common to both outputs, the noise coming out of Q1 and Q2 should be strongly (but not completely) correlated.
Last edited:
We had this discussion before. There is no reason to assume that the noise current specified with equal source resistances is uncorrelated between the positive and negative inputs. In fact the equations in the datasheet of the LT1028 show explicitly that it is the differential noise component that's being specified, so:
sqrt(2*(12.9e-9)^2+(In*2*10K)^2+En^2) ~= 27.084 nV/rtHz
which clearly doesn't match reality either. It starts matching for In ~= 1.5 pA/sqrt(Hz).
I just wanted to point out that the datasheet was true to your initial measurements even if they are wrong or not...So your calculated noise is not 3 times larger just 1.325 times larger in the worst case scenario and 1.07 times lager in your best scenario....which for me sounds perfect.
I'm not such a smart guy as you, just happened to remember the true value mentioned in the datasheet.
Last edited:
That was just an assumption, but only the measurements counted.
And as they showed, I wasn’t that far off, was I ?
Hans
Sorry but to my opinion you are making the mistake by treating In1 and In2 as correlated noise.
I see no reason why there should be any correlation between In1 and In2 apart from the fact that they have equal magnitude.
If they where fully correlated, the amps CMRR would cancel their noise with matched resistances which is definitely not the case, true ?
Hans
Well guys ...i think you're arguing on some components that are way closer to the datasheet than more modern ones form the competition and I think that it's no wonder why in most advanced low noise medical circuits i only find three manufacturers constantly: Analog Devices, Linear Technology and Maxim Integrated.
Last summer i built a cassette tape head preamp which aimed at very low noise and opa1642 which in the datasheet had at least half the noise of OPA2132, proved the other way around.I tried them in two different topology and they performed the same in both of them.So opa2132 was the one to have half the noise of opa1642 and that was a real shock as i made the same measurements on 2 separate op amps of each series.
I'd be interested more in OPA1632 real world measurements.
Last summer i built a cassette tape head preamp which aimed at very low noise and opa1642 which in the datasheet had at least half the noise of OPA2132, proved the other way around.I tried them in two different topology and they performed the same in both of them.So opa2132 was the one to have half the noise of opa1642 and that was a real shock as i made the same measurements on 2 separate op amps of each series.
I'd be interested more in OPA1632 real world measurements.
See this post from last year:
https://www.diyaudio.com/forums/ana...ge-riaa-accuracy-computer-19.html#post5394157
I know, that’s where you mentioned the same.
Just think of the CMRR that I referred to and you will understand that there is no correlation between In1 and In2.
Hans
At least AD datasheet are very clear in their statements i think...
The last photo shows some resistor values for the input impedances when measuring the noise for lt1028 too...
Attachments
Last edited:
You're not getting the point. The point is that that 1 pA/sqrt(Hz) from the datasheet is not the spectral density of In1 and In2 but of their differential component (In1 - In2)/2. That reduces the discrepancy by a factor of sqrt(2).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.