EEVblog #1328 - uCurrent OP189 Measurements

[EEVblog]

Part 3 of designing a better uCurrent series.
Measuring the noise and consumption of the OPA189 compared to the MAX4239 using a dynamic signal analyser and an oscilloscope.

[Kleinstein]

The test with the split rail shown +-2.3 V. This is quite close to the lower 4.5 V limit for the OPA189 - noise may be a little higher there. It may be more the higher supply with the 2x3 V supply instead of the rail splitting circuit.

Ideally the noise of the OPA189 should be about equivalent to the noise of a 1.6 K resistor. It is kind of convenient to add noise sources as the corresponding resistance, as the take care of adding the powers and not the votlages. The  1 K resistor in the feedback network would be another small contribution to the noise. There is still a bit missing to the measured 10 nV/Sqrt(Hz) that corresponds to some 6 K.

For modding an existing µCurrent it would be enough to replace the OP at the input - the noise of the 2 nd OP would not matter that much.  Some filtering at the output would be a good idea, as not all scopes have BW limits much lower then 20 MHz.

[EEVblog]

The test with the split rail shown +-2.3 V. This is quite close to the lower 4.5 V limit for the OPA189 - noise may be a little higher there. It may be more the higher supply with the 2x3 V supply instead of the rail splitting circuit.

No difference at 6V.

[graybeard]

The chopping frequency I measured was ~300 KHz with a 150KHz subharmonic as well as many high order harmonics.

I would move your fixture and cabling away from the front of the CRT on your FFT.  I have had the magnetic fields from the sweep coils couple into circuits.   Even though they are in your shielded box, some of the magnetic fields can still make it in.

[BrianHG]

The chopping frequency I measured was ~300 KHz with a 150KHz subharmonic as well as many high order harmonics.

I would move your fixture and cabling away from the front of the CRT on your FFT.  I have had the magnetic fields from the sweep coils couple into circuits.   Even though they are in your shielded box, some of the magnetic fields can still make it in.

Oh yeah, that nasty 15KHz all out omnipresent signal which I also find in around 75% of the CDs I own as the mic and their preamps pick up that signal from the semi-modern CRT console on the studio audio mixers of the day.

It used to drive me nuts as my ears used to be good up to 25KHz when I was a teenager and no matter how small the signal, at times it had enough db for me to make out at high volume.

[graybeard]

I can't hear 15750 Hz any more

[iteratee]

How is the OPA WRT bandwidth vs supply voltage? Dave mentioned in the last video that the OPA189 is "wide supply". I have seen "high speed" opamps give banner specs for bandwidth that only really applies to the highest supply voltages that can get massively nerfed towards the low end.

[Kleinstein]

The supply current does not change much with the supply voltage, except a little fro 4.5 V to some 5 .5V.  So I would not expect much effect on the bandwidth and noise, except for the very low end, below some 5 V.

Because of noise one would often like a limited bandwidth (e.g. up to some 100-200 kHz, so that much of the chopper noise is left out).

If one would need high bandwidth of the whole amplifier, one would no longer use 2 stages x 10 in series, but a compound amplifier: one AZ OP and a normal higher speed OP inside the loop. The AZ OP would than do something like a gain of 3 or 5 and the other OP would do the rest.  This would also avoid possible interaction between the 2 chopper frequencies, may get away with less power and only needs 1 pair of precision resistors for the gain. So the compound amplifier is also attractive for a more normal version.

[iteratee]

I can't hear 15750 Hz any more

When I was little I could type "beep(23000)" into quickbasic (or whatever) and only I could hear it through the crappy PC speaker. 

30 years later, probably not. I'd test on this phone but I know the headphone output rolls off at 19 khz or so.

Nowadays I hear 20khz at all times whether it's there or not. 

[EEVblog]

The chopping frequency I measured was ~300 KHz with a 150KHz subharmonic as well as many high order harmonics.

I would move your fixture and cabling away from the front of the CRT on your FFT.  I have had the magnetic fields from the sweep coils couple into circuits.   Even though they are in your shielded box, some of the magnetic fields can still make it in.

Oh yeah, that nasty 15KHz all out omnipresent signal which I also find in around 75% of the CDs I own as the mic and their preamps pick up that signal from the semi-modern CRT console on the studio audio mixers of the day.

25KHz in this case.
I noticed this and did a quick video on it:

[SilverSolder]

[...]
Nowadays I hear 20khz at all times whether it's there or not. 

Think of all the electricity you're saving! 

[SilverSolder]

How much of the noise from the uCurrent actually comes from the rail splitter, rather than the op amps?  Is there anything to be gained by using two batteries instead of a rail splitter, from this perspective?

[EEVblog]

How much of the noise from the uCurrent actually comes from the rail splitter, rather than the op amps?  Is there anything to be gained by using two batteries instead of a rail splitter, from this perspective?

Watch the whole video 

[SilverSolder]

How much of the noise from the uCurrent actually comes from the rail splitter, rather than the op amps?  Is there anything to be gained by using two batteries instead of a rail splitter, from this perspective?

Watch the whole video 

I did!   Re-watching now, as I obviously missed something! 

[Edit] Senior moment...   I did see that section the first time, but somehow failed to conclude that the noise of the rail splitter was inferred to be very low even if not directly measured...   Going to get another cup of coffee, maybe that will help, but might need something stronger!

[Unixon]

Dave, did you have any issues with offsets/biases of U2 LMV321 in your uCurrent circuit?
According to the schematic U2 inputs are not impedance balanced. Isn't it a problem?

In one particular design I've used LMV321 to create a 2.5V reference (actually, to follow TL431) with unipolar 5V or 3.3V power supply and I got huge ~100-200mV negative output offset.
Without changing schematic, I solved this issue by replacing LMV321 with TSX561A (ST part), but this made me look at LMV321 and its relatives with a little apprehension.

[EEVblog]

Dave, did you have any issues with offsets/biases of U2 LMV321 in your uCurrent circuit?
According to the schematic U2 inputs are not impedance balanced. Isn't it a problem?

In one particular design I've used LMV321 to create a 2.5V reference (actually, to follow TL431) with unipolar 5V or 3.3V power supply and I got huge ~100-200mV negative output offset.
Without changing schematic, I solved this issue by replacing LMV321 with TSX561A (ST part), but this made me look at LMV321 and its relatives with a little apprehension.

Nope, no issues.
But any offset in the virtual ground reference won't matter, because it's the actual ground reference point.

[David Hess]

If one would need high bandwidth of the whole amplifier, one would no longer use 2 stages x 10 in series, but a compound amplifier: one AZ OP and a normal higher speed OP inside the loop. The AZ OP would than do something like a gain of 3 or 5 and the other OP would do the rest.  This would also avoid possible interaction between the 2 chopper frequencies, may get away with less power and only needs 1 pair of precision resistors for the gain. So the compound amplifier is also attractive for a more normal version.

Or make a compound amplifier with the chopper stabilized one correcting the low frequency noise and drift of the faster wideband amplifier or discrete differential pair.  I have done this many times for spectacular results.

[SilverSolder]

If one would need high bandwidth of the whole amplifier, one would no longer use 2 stages x 10 in series, but a compound amplifier: one AZ OP and a normal higher speed OP inside the loop. The AZ OP would than do something like a gain of 3 or 5 and the other OP would do the rest.  This would also avoid possible interaction between the 2 chopper frequencies, may get away with less power and only needs 1 pair of precision resistors for the gain. So the compound amplifier is also attractive for a more normal version.

Or make a compound amplifier with the chopper stabilized one correcting the low frequency noise and drift of the faster wideband amplifier or discrete differential pair.  I have done this many times for spectacular results.

How big is the problem of two chopping frequencies beating up on each other - could that lead to low frequency misbehaviour?

I like the compound amplifier idea.  This should make it possible to increase the bandwidth significantly?  OPA189 might do a super job in a "supporting role"?   I seem to remember seeing application notes with some fairly straightforward designs, but can't just recall where...

[Kleinstein]

The composite amplifier with the 2nd fast OP inside the loop is the simple solution, but the overall BW is limited to the GBW of the AZ OP (preferably a little lower to get extra phase reserve) and the noise would be from the AZ OP. For the compound amplifier I still have an old appl. note around that is attached.

An extra, fast low noise amplifier and the chopper OP only for stabilization can be faster and lower noise than the AZ OP for the higher frequencies. However the low frequency region may have additional noise and the bias current of the 2 amplifiers add. This was definitely the method of choice in the old days when the AZ OP were slow and relatively noisy. The circuit tends to be a little more complicated though. LT apply note 21 has a lot of such examples.

[David Hess]

How big is the problem of two chopping frequencies beating up on each other - could that lead to low frequency misbehaviour?

I have seen it happen with older parts which did not dither their chopping clock; the symptoms typically include low frequency tones or high offset voltage drift which may be indistinguishable from low frequency noise.  (1) Some of the early parts provide a way to synchronize the chopping clock to avoid problems and it is not only with other chopper stabilized amplifiers.  You might want to synchronize with a sampling analog-to-digital converter, switching power supply, or microprocessor clock.

(1) In the past, high noise prevented chopper stabilized amplifiers from replacing precision bipolar parts even in low frequency applications.  Maybe the OPA189 is different in medium impedance applications but I would sure want to empirically test it; the lack of a current noise graph and DC or 0.01 Hz noise specification makes me wonder if TI is hiding something.

I like the compound amplifier idea.  This should make it possible to increase the bandwidth significantly?  OPA189 might do a super job in a "supporting role"?   I seem to remember seeing application notes with some fairly straightforward designs, but can't just recall where...

Usually it was about noise instead of bandwidth.  Chopper stabilized amplifiers have flat flicker noise while linear parts have increasing flicker noise at lower frequencies and wide bandwidth parts can have astonishingly high flicker noise, but there is not much overlap between the requirements for low flicker noise and wide bandwidth except maybe in test instrumentation; I would not mind having an FFT signal analyzer with both but the common solution is just to use a separate low noise preamplifier.

That OPA189 is much better than past chopper stabilized parts I have used but total noise could still be improved by perhaps 4 times in a low impedance, 100s of ohms, application.

[Kleinstein]

There is a parallel thread about the low frequency noise of the chopper OPs. This includes some data on the OPA189.
https://www.eevblog.com/forum/metrology/low-frequency-noise-of-zero-drift-amplifiers/msg3094181/#msg3094181

So I don't think there is hidden flicker noise. The nasty parts is more like the input impedance in the higher frequency region (e.g. input capacitance, but may be more) can have an effect on the offset / input bias current. The new OPs often include some EMI filtering, but this seems to be not sufficient to fully suppress these effects. Worst case one may need RF gear like a VNA to build a precision circuit.

With the chopper OPs there is the tendency that the low noise parts (e.g. OPA189, ADA4522, MCP6V91, max44250) also have relative high bandwidth. I think this because low noise needs a high chopper frequency as the internal capacitors are limited in size. So they need higher speed anyway. With higher currents amplifiers also tend to get faster.

[Gandalf_Sr]

Dave,

I have a uCurrent Gold Rev 5 that looks like it is fitted with two MAX4239s (part marking 'ABAA').  A few quick questions for us existing owners:

1. Is the OPA189 going to be lower noise than the MAX4239?
2. Can we just replace the 4238/4239 with the OPA189 or are there other components that will need (or you recommend) changing too?
3. What's required as far as the power supply goes? My uCurrent Gold is fitted with a coin cell holder that would not allow 2xCR2016s as the side contact would touch both batteries' +ve terminal; replacing that holder with either a double coin cell holder or 3 x AAAs is fine but what is the specific max/min voltage spec?

Thanks in advance

[EEVblog]

Dave,

I have a uCurrent Gold Rev 5 that looks like it is fitted with two MAX4239s (part marking 'ABAA').  A few quick questions for us existing owners:

1. Is the OPA189 going to be lower noise than the MAX4239?
2. Can we just replace the 4238/4239 with the OPA189 or are there other components that will need (or you recommend) changing too?
3. What's required as far as the power supply goes? My uCurrent Gold is fitted with a coin cell holder that would not allow 2xCR2016s as the side contact would touch both batteries' +ve terminal; replacing that holder with either a double coin cell holder or 3 x AAAs is fine but what is the specific max/min voltage spec?

Thanks in advance

All answered in my latest video. I haven't done any further testing yet though.

[Kleinstein]

The OPA189 has 4.5 V min for the supply. The max4239 and LMV321 (or similar) have some 5.5 V as maximum for the supply. So there is only a relatively small window for the supply.

3 alkaline cells (e.g. AA) may fast run below 4.5 V and 4 x alkaline are usually more than 5.5 V when new, so it would need an LDO to get some 5 V.

Besides the supply limits the rest should be OK with changing the OP(s). The noise is relevant mainly at the input.
If all 3 OPs are exchanged (something like MCP6H01 instead of the LMV321) a voltage like 6-9 V could be OK.

[Gandalf_Sr]

Thanks guys

I went back and watched the whole video.  Therefore I think the answer is that the OPA189 is going to:
a. Be a drop in replacement except that it will be missing the enable pin but that is not going to affect anything
b. Lower the noise level and increase the bandwidth
c. Need a minimum Vbat of 4.5V so some way of upping this needs to be implemented
d. The LMV321 needs to be changed or removed and a split-rail power setup used.

Am I right?

Thanks for suggesting the MCP6H01 as an LMV321 replacement Kleinstein, I'll go check that out.  What are the pros and cons of just removing the LMV321 and going with a split rail with 2 x (2 x AAAs)?

[Kleinstein]

Using 2 x 2 AA is usually better than the virtual ground from the OP. It need less power and is usually lower impedance. The LMV321 allows to use just 1 coin cell or could work with 5 V from an LDO. So one does not absolutely need to remove the LMV321 if the OPA189 is used.
The MCP6H01 would allow using a 9 V block.

[Gandalf_Sr]

Using 2 x 2 AA is usually better than the virtual ground from the OP. It need less power and is usually lower impedance. The LMV321 allows to use just 1 coin cell or could work with 5 V from an LDO. So one does not absolutely need to remove the LMV321 if the OPA189 is used.
The MCP6H01 would allow using a 9 V block.

Thanks.  The MCP6H01 is good up to 16V (+/-8) but the TPS3809 voltage regulator used for the LED is rated at an absolute max of 7V so either we stay under 7V or that has to change too.

[EDIT] With 2 x (2xAAA) alkalines, the voltage would range from 6.4 down to 4.5 when the batteries were almost completely dead.  Although the MCP6H01 would work as a replacement for the LMV321, using a split battery supply means U2 and associated resistors could go away.

The TPS3809 could stay to give controlled battery level warning (LED goes out when battery is failing) but we would want the 4.55V version, part number TPS3809I50DBVR and we'd also want to change the value of the LED dropper resistor.

OK, I've ordered the parts from Digikey, I'll report back on the progress.

[Gandalf_Sr]

Well, I was about to perform this upgrade today when I realized that, if moving to a split rail supply using 2+2 AAA batteries, I'm going to have to figure out a way to switch both the +3V and the -3V rails or there will be a half-powered circuit drawing current all the time.  Either that or maybe I should find a replacement for the LMV321 that can handle 4 x AAAs like Dave suggests.

[Gandalf_Sr]

I figured it all out.  I did away with the short functionality which allowed me to switch the +3V and the -3V lines at the same time.  Here's what I did:
1. Used a 4 x AAA battery holder and added an extra wire (green) that's the center tap of the 4 batteries.
2. Cut away the ridges inside the plastic case using a chisel modelling knife and Dremel - the battery box just fits inside.
3. Removed U2, R6,R7,R10,C2
4. Replaced U1 and U4 with OP189IDBVRCT
5. Replaced U3 with TPS3809I50DBVR (4.55) volt version and changed R4 to 909 Ohms
6. Removed battery holder and cut tracks on PCB to remove shorting connection (see picture)
7. Soldered wires onto PCB (see picture)

On testing, I was able to set up a 0.5 uA current through the uCurrent and see 500 mV on the output (1mV/nA range) with my calibrated 34461A.  It looks like it was all successful.

[apoorv3in]

Modified the uCurrent based on V3 with OPA189 and LM321MF
running now with 9v Battery

[EEVblog]

I figured it all out.  I did away with the short functionality which allowed me to switch the +3V and the -3V lines at the same time.  Here's what I did:
1. Used a 4 x AAA battery holder and added an extra wire (green) that's the center tap of the 4 batteries.
2. Cut away the ridges inside the plastic case using a chisel modelling knife and Dremel - the battery box just fits inside.
3. Removed U2, R6,R7,R10,C2
4. Replaced U1 and U4 with OP189IDBVRCT
5. Replaced U3 with TPS3809I50DBVR (4.55) volt version and changed R4 to 909 Ohms
6. Removed battery holder and cut tracks on PCB to remove shorting connection (see picture)
7. Soldered wires onto PCB (see picture)

On testing, I was able to set up a 0.5 uA current through the uCurrent and see 500 mV on the output (1mV/nA range) with my calibrated 34461A.  It looks like it was all successful.

Have you done a full frequency sweep?

[Gandalf_Sr]

No, not yet Dave. Any tips on how to set that up, say with a Rigol MSO5074?

[SilverSolder]

No, not yet Dave. Any tips on how to set that up, say with a Rigol MSO5074?

[Andreas]

Hmm,

shure that the OPA189 is the right candidate for a (high impedant) current source input.
My measurements give a relative high (average) input bias current.
with a 10K source resistance I get a 23 uV offset voltage or 2.3 nA effective input bias current.

see also here:
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2819014/#msg2819014

with best regards

Andreas

[sourcecharge]

Part 3 of designing a better uCurrent series.
Measuring the noise and consumption of the OPA189 compared to the MAX4239 using a dynamic signal analyser and an oscilloscope.

So after watching this video, it leads me to think that the OP189 is better, but the following post has me scratching my head:

Hmm,

shure that the OPA189 is the right candidate for a (high impedant) current source input.
My measurements give a relative high (average) input bias current.
with a 10K source resistance I get a 23 uV offset voltage or 2.3 nA effective input bias current.

see also here:
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2819014/#msg2819014

with best regards

Andreas

Does this effect uA and mA measurements?

sorry to bring this topic up so late, but I've been waiting for those that were testing it to post their results.

I think I want to change this to the OP189 if uA and mA measurements are unaffected.

[Kleinstein]

The input bias current effectively adds to the measured current, a little like additional offset. One would see the sum of the voltage offset and the bias current. 23 µV (2.3 nA) looks like quite high. The typical current should be lower.
The bias current can vary between units and can also vary with the input impedance (in the higher frequency range).

In most cases on can tolerate a small offset and just subtract it. Some extra 2 nA should not be unnoticed in the mA range and hardly visible in the µAs.

One does not have to change both of the AZ OPs. Only the OP at the input is critical for the noise.

[sourcecharge]

The input bias current effectively adds to the measured current, a little like additional offset. One would see the sum of the voltage offset and the bias current. 23 µV (2.3 nA) looks like quite high. The typical current should be lower.
The bias current can vary between units and can also vary with the input impedance (in the higher frequency range).

In most cases on can tolerate a small offset and just subtract it. Some extra 2 nA should not be unnoticed in the mA range and hardly visible in the µAs.

One does not have to change both of the AZ OPs. Only the OP at the input is critical for the noise.

Ya, it seems that would make sence about the voltage offset, but why didn't dave see that voltage offest in his testing?

The decrease in noise is great improvement for it, but I think I want to change both ops because of their frequency range and the higher frequency of the chopper.

I figured it all out.  I did away with the short functionality which allowed me to switch the +3V and the -3V lines at the same time.  Here's what I did:
1. Used a 4 x AAA battery holder and added an extra wire (green) that's the center tap of the 4 batteries.
2. Cut away the ridges inside the plastic case using a chisel modelling knife and Dremel - the battery box just fits inside.
3. Removed U2, R6,R7,R10,C2
4. Replaced U1 and U4 with OP189IDBVRCT
5. Replaced U3 with TPS3809I50DBVR (4.55) volt version and changed R4 to 909 Ohms
6. Removed battery holder and cut tracks on PCB to remove shorting connection (see picture)
7. Soldered wires onto PCB (see picture)

On testing, I was able to set up a 0.5 uA current through the uCurrent and see 500 mV on the output (1mV/nA range) with my calibrated 34461A.  It looks like it was all successful.

I've also got a question about this design change.

First, I think I'm going to use dual PS's at +/-10V instead of batteries because I'm using the PS's in my FG isolator anyway, so that would eliminate U2,R6,R7,R10,C2,U3,R4, the battery holder, and the LED.  I'm thinking because the PS's would show the input voltages and therefore I don't really need a low voltage indicator. 

What I don't understand is why you attached the negative to the switch and then cut the shorting traces to the switch.  Doesn't this take the function of the switch away from Off/On/On(Shorted), to Off/On/On(Not Shorted)?

Can't the negative voltage wire simply go the negative battery terminal pad and leave the switch and traces alone?

Just in case, I included the schmatic of the uCurrent


Edited:

I figured out why you did that, because you are using batteries and not PS's that can be turned off.

I'm guessing that the ops will still drain the negative supply side of the batteries because their is no virtual ground.

I wasn't thinking about that because I had PS's in mind with their own on/off switch.

[Kleinstein]

With more changes to the circuit one could consider to build an independent circuit (e.g. with adapter and raster board or dead bug style) instead of starting from a µCurrent board and replacing much of the parts. A modified circuit with a compound stage instead of 2 stages in series could get away without the second Az OP and use a more conventional one.

Keep in mind the µCurrent uses low tolerance resistors for the gain and the shunts to get the gain right without any adjustment.

The OPA189 is not that much faster than the MAX4239. There is however the different copper frequency which may be a factor when using it with a scope. One may need to change the second OP, as the Max4239 has a limited maximum supply. It still works with 5 V total.

[Gandalf_Sr]

Sourcecharge
Yes, you figured it out.  Because I moved from a single supply (where the center voltage is created by the electronics) to a V- 0 V+ battery setup, I changed to switch both sides of the battery supply.  You're right that I lose the shorted position but that's easy to replicate, just turn it on and short the inputs.

If you were not using batteries then you could keep the shorted position and not modify the tracks around the switch.

[SilverSolder]

Just for fun, I tested the uCurrent background noise level using a venerable old spectrum analyzer, as an exercise in first principles.



The results:




On the 3571A we see a uCurrent noise floor of -93dBV, if we ignore the chopper noise.

We must add 2.5dB to that due to logarithmic averaging, taking us to -90.5dBV

Then we convert dBV to RMS, so  -90.5dBV -> 30uV RMS

Now we have to adjust RBW since the noise power bandwidth of the filters are about 12% wider than the 3dB bandwidth.  So 100Hz RBW --> 120Hz RBW for noise.

Finally we divide the RMS signal by rt-hz, i.e. RMS/sqrt(AdjRBW).
Presto,   30uV/10.95 = 2.75uV/rt-hz



This number is lower than Dave's measurement of 3.5uV/rt-hz, but still on the same planet...   there could be a hole in my methodology, this analyzer is not exactly user friendly!  Comparing it to a car, it is like driving a 1965 Mustang, with manual steering and transmission, without power brakes...

[sourcecharge]

With more changes to the circuit one could consider to build an independent circuit (e.g. with adapter and raster board or dead bug style) instead of starting from a µCurrent board and replacing much of the parts. A modified circuit with a compound stage instead of 2 stages in series could get away without the second Az OP and use a more conventional one.

Keep in mind the µCurrent uses low tolerance resistors for the gain and the shunts to get the gain right without any adjustment.

The OPA189 is not that much faster than the MAX4239. There is however the different copper frequency which may be a factor when using it with a scope. One may need to change the second OP, as the Max4239 has a limited maximum supply. It still works with 5 V total.

So here is where I am.
I did the mods that I described plus some, but there is a couple of problems.
First, the virtual ground of the circuit is actually required for the op amps to be biased correctly.  You can do this with the balenced comparitor's output to virtual ground or you can use two 100nF caps from +V and V to virtual ground.
Second, the unconnceted voltage readout on 3 of my mastech meters for both of my ucurrents (2 of them) are about 0.4mV and 0.12mV in mA and uA settings but the nA setting seems to be way off which I'm guessing is the problem that was previously described.
They are both on the same + and - V PS with the same vitual ground.
I double check the resistor networks and found them in spec.
I've also replaced the mA 0.01 ohm shunt resistors in both of them so they are new too just in case I overloaded the first ones.
I replaced the 270 ohm resistor on the output of the last opamp and put a 0 ohm jumper in it's place, while scratching off the masking near the output terminal and cutting the trace and jumping that cut with the 270 ohm resistor.
I've spiced this and it shouldn't change anything really, but it always nagged me on the idea that the resistor network is using 0.05% resistors on the last op amp resistor network but it has a whoping 270 ohms in series with the 9k ohms.  0.05% of 9k is 4.5 ohms, and 270 ohms included into it just doesn't make any logical sence.
The amout of current (voltage readout) under load is offset by the same amount that they are reading under unconnected conditions.

Anyways, if anyone has any ideas with whats going on, let me know....

[Kleinstein]

The 270 Ohms at the output of the last OP are not directly on series to the 9 K,  but the output is taken from behind the resistor.
The idea behind the 270 ohms resistor is to avoid capacitive load to the OPs output, that may lead to oscillation. AFAIR the resistor was at a slightly odd place, not where it ideally should be. There may be slight differences in circuit versions.

The virtual ground driver had some problems with oscillation, depending on the exact OP type used. Here capacitive loading is part of the problem. So it is either the OP or passive with resistors and capacitors. Combining the OP with capacitors is tricky.

Some of the AZ OPs react to the capacitance or more accurately the impedance at some 10-500 MHz for the input nodes. The modern ones include some EMI filtering, but it may still need more external filtering to avoid odd effects from the higher frequency bands, like capacitive loading or cable length. For the higher current ranges the shunt resistors essentially short the input, but not the 10 K for the nA range. 
Filtering maybe something like 100 pF+100 Ohms in series in parallel to the 10 K shunt.  A ferrite bead at the input may also help.

[sourcecharge]

so the rev5 boards have the 270 ohm resistor in series with the 9k ohm resistor and it also goes to the output terminal.

I simply couldn't leave it alone, although all spice sims show no difference in any output.

What's really got me confused is the 0.4mV to 0.5mV and the 0.1mV to 0.2mV output of the ucurrents when they are unconnected.  These seem to be affecting the mA and uA measurements, although the mA and uA measurements are within 0.5%.

So, I guess the question is, can this be improved or is there something wrong with both of my ucurrents?  Come to think about it, they had the same problem with the max4239 at about 0.2mV each.

I measured 0mV on the input of the 1st opamp for both, and then 0.05mV on the input of the 2nd opamp for the 0.4mV ucurrent, and 0.01mV input on the input of the 2nd opamp for the 0.1mV ucurrent.  But when I measure current I'm within 0.5% even though the offsets are included in both the mA and the uA measurement ranges.

I'm only using matched caps to make the virtual ground, and I'm wondering if an opamp virtual ground would be better.

Edit, I think I just figured out where the measurement was comming from.

If the output of the 1st opamp has a bias voltage of somewhere between 0uV  and 5 uV, then the 10x amplification makes it 0 to 50uV?  Then the next op amp, amplifies it to 0.4mV and 0.1mV for the unconnected measurement?

Can anyone verify this logic? I'm not sure.

[Kleinstein]

The max4239 can have an offset of some 1 µV, maybe a little more with unequal impedance at the inputs. This amplified 100 times and this some 100 µV = 0.1 mV range offset at the output is well possible. It may be a little more if the input offset is higher.
In the nA range there is also bias current from the OP.

The OPA189 has more bias current and may react more sensitive to unequal impedance.

The 270 Ohms from the OP to the output and the Feedback from behind the 270 Ohms, has little effect. It essentially adds the 270 Ohms to the open loop output impedance. So the OP will compensate for it. The resistor may help a little with EMI, but does not help to prevent oscillation from capacitive loading. So be careful with a capacitive load (some DMMs).

[SilverSolder]

The max4239 can have an offset of some 1 µV, maybe a little more with unequal impedance at the inputs. This amplified 100 times and this some 100 µV = 0.1 mV range offset at the output is well possible.[...]

Is the compound amplifier talked about earlier in the thread less sensitive to the offset issue -  i.e. does a compound amplifier have only the offset of the input amplifier, whereas the offset of the second op-amp inside the loop disappears due to feedback?  - if so, it might be a good reason to try the compound amplifier idea for this application?

[sourcecharge]

The max4239 can have an offset of some 1 µV, maybe a little more with unequal impedance at the inputs. This amplified 100 times and this some 100 µV = 0.1 mV range offset at the output is well possible.[...]

Is the compound amplifier talked about earlier in the thread less sensitive to the offset issue -  i.e. does a compound amplifier have only the offset of the input amplifier, whereas the offset of the second op-amp inside the loop disappears due to feedback?  - if so, it might be a good reason to try the compound amplifier idea for this application?

If the only "significant" amount of offset is from the 1st opamp, can this be nulled by using some other opamp with manual pots which can be set by the user while "zeroing" the unconnected measurement?

[SilverSolder]

The max4239 can have an offset of some 1 µV, maybe a little more with unequal impedance at the inputs. This amplified 100 times and this some 100 µV = 0.1 mV range offset at the output is well possible.[...]

Is the compound amplifier talked about earlier in the thread less sensitive to the offset issue -  i.e. does a compound amplifier have only the offset of the input amplifier, whereas the offset of the second op-amp inside the loop disappears due to feedback?  - if so, it might be a good reason to try the compound amplifier idea for this application?

If the only "significant" amount of offset is from the 1st opamp, can this be nulled by using some other opamp with manual pots which can be set by the user while "zeroing" the unconnected measurement?

The easiest is to zero the voltmeter that the uCurrent is driving (or add a small offset if you're using a scope).

So an offset isn't a big deal in most normal use cases.

But obviously the smaller the offset of the uCurrent itself, the better!

[sourcecharge]

The max4239 can have an offset of some 1 µV, maybe a little more with unequal impedance at the inputs. This amplified 100 times and this some 100 µV = 0.1 mV range offset at the output is well possible.[...]

Is the compound amplifier talked about earlier in the thread less sensitive to the offset issue -  i.e. does a compound amplifier have only the offset of the input amplifier, whereas the offset of the second op-amp inside the loop disappears due to feedback?  - if so, it might be a good reason to try the compound amplifier idea for this application?

If the only "significant" amount of offset is from the 1st opamp, can this be nulled by using some other opamp with manual pots which can be set by the user while "zeroing" the unconnected measurement?

The easiest is to zero the voltmeter that the uCurrent is driving (or add a small offset if you're using a scope).

So an offset isn't a big deal in most normal use cases.

But obviously the smaller the offset of the uCurrent itself, the better!

ya you're probably right, but it was just a thought.

[Kleinstein]

The relevant offset if from the 1st OP only.  It is somewhat tricky to use the trim pins at one OP to compensate more than the OPs own offset.  At least for BJT based OPs the offset trim also effects the drift. In the ideal picture the offset is proportional to the absolute temperature. For a very small offset, like a few µV at the input this may still be acceptable. It would not be zero drift, but could still be low drift. However modern OPs rarely have the  trim pins and something like an OP27 needs quite a lot of supply current and more than 3 V.

The compound amplifier idea does not help much with the offset. It helps with needing only 1 AZ OP and only 1 pair of precision resistors. This can help reducing the overall power needed (a non AZ OP may be lower supply current at the same speed) and speed as the overall BW could reach something like 1/2 the GBW of the AZ OP.  However this would be quite some change in the circuit - so nothing for the existing board, more like some a thing for a new version with the OPA189. Because of noise, a high BW version may be better with the AZ OP to stabilize a low noise amplifier.

The OPA189 will be tricky for small currents, as it may need added filtering to avoid an effect from external capacitance at the input. This would interfere with a high speed. Because of this a OPA189 version would likely be for the 2 higher current ranges only, maybe with an extra option as a µV amplifier (e.g. for thermocouples or an external higher current shunt). If needed, the very low currents (e.g. < 10 µA) would be better done with a TIA like configuration anyway, with a different OP.

[SilverSolder]

The relevant offset if from the 1st OP only.  It is somewhat tricky to use the trim pins at one OP to compensate more than the OPs own offset.  At least for BJT based OPs the offset trim also effects the drift. In the ideal picture the offset is proportional to the absolute temperature. For a very small offset, like a few µV at the input this may still be acceptable. It would not be zero drift, but could still be low drift. However modern OPs rarely have the  trim pins and something like an OP27 needs quite a lot of supply current and more than 3 V.

The compound amplifier idea does not help much with the offset. It helps with needing only 1 AZ OP and only 1 pair of precision resistors. This can help reducing the overall power needed (a non AZ OP may be lower supply current at the same speed) and speed as the overall BW could reach something like 1/2 the GBW of the AZ OP.  However this would be quite some change in the circuit - so nothing for the existing board, more like some a thing for a new version with the OPA189. Because of noise, a high BW version may be better with the AZ OP to stabilize a low noise amplifier.

The OPA189 will be tricky for small currents, as it may need added filtering to avoid an effect from external capacitance at the input. This would interfere with a high speed. Because of this a OPA189 version would likely be for the 2 higher current ranges only, maybe with an extra option as a µV amplifier (e.g. for thermocouples or an external higher current shunt). If needed, the very low currents (e.g. < 10 µA) would be better done with a TIA like configuration anyway, with a different OP.

As seen in the spectrum plot, the noise from the auto-zero amps is "significant" (not a problem for most use cases, but definitely visible).  So, having a "normal" op amp disciplined by an auto-zero servo sounds like a good idea to explore, just to get rid of the noise.   Furthermore, if we now don't have to worry about the offset at all any longer, perhaps there are a lot more options for which op amps to use.

The downside with any servo type arrangement is that now, a DC current will "fade out" over time, as the servo corrects it away.  So perhaps this fact alone makes that idea still born?

Funny how much thinking goes in to analog electronics, even when the problem you are trying to solve is well understood!

[Kleinstein]

The AZ OP to correct an normal amplifier would not be a normal DC servo to bring the DC output to zero, essentially simulation AC coupling, but a DC servo to force the input difference to zero, so correcting the offset. There is no principle problem with this - it just needs extra parts and in most implementations there will be a little more low frequency noise than just the AZ op. This is from extra resistors used for isolation to keep out the chopper spikes from the fast amplifier.
The AZ amplifier plus some resistors from the filter (some 10-100K) would set the noise at low frequencies like < 10-100 Hz and the fast amplifier would set the higher frequency noise.  Compared to just an AZ amplifier one usually has a little more noise at low frequencies, but less noise at higher frequencies (e.g. > 100 Hz).
Just an AZ OP may also need some filtering, though this can often be lower resistance.

The AZ OP to correct the error of a normal amplifier is nothing new and standard technique.

[SilverSolder]

[...]
The AZ OP to correct the error of a normal amplifier is nothing new and standard technique.

I found this example of...  something! -  in the LTC2057 data sheet.  Is that the kind of technique you are referring to?

[David Hess]

I found this example of...  something! -  in the LTC2057 data sheet.  Is that the kind of technique you are referring to?

Yes, that is the idea.  The integration constant controls the breakpoint frequency between the two amplifiers.  If you know the sensitivity of the offset null terminal, then you can calculate the optimum breakpoint frequency.

Many years ago I extended it to differential operation using a pair of LT1028s and an LTC1151 dual.  (1) First I measured the sensitivity of the LT1028's offset null and then calculated the optimum frequency breakpoint, which ended up being right on.  Noise performance was somewhere between incredibly good and impossibly good.

Linear Technology also published several examples of this with low noise JFET differential pairs which would be suitable if you also want low input bias current for higher source impedance.

(1) I used the dual LTC1151 because separate chopper stabilized amplifiers might have resulted in intermodulation between the clocks.  Modern automatic zero and chopper stabilized operational amplifiers lack any ability to control or synchronize their clock which suggests that this problem has been solved.

[SilverSolder]

[...] Noise performance was somewhere between incredibly good and impossibly good. [...]

So the chopper noise is completely suppressed in this design?

[David Hess]

[...] Noise performance was somewhere between incredibly good and impossibly good. [...]

So the chopper noise is completely suppressed in this design?

Yes, but the operating conditions for the chopper stabilized amplifier are ideal with very low AC impedance.  It is more like the broadband noise of the chopper is completely suppressed, or the flicker noise, including drift, of the main amplifier is completely suppressed.

I absolutely saw an improvement over either amplifier alone, and as I adjusted the frequency breakpoint between the amplifiers, I could directly measure an increase in 0.1 to 10 Hz noise, which is how I calibrated it.

[SilverSolder]

Any gotchas?   Bias current perhaps?

[Kleinstein]

Both OP still contribute to the noise: the chopper (+ a little from the filter) at low frequency and the other OP at higher frequency.

However the current noise of both OPs add. So the current noise will be the sum of both, possibly with some filtering for the chopper OP at really high frequencies. The bias current also adds - so one may want more like a JFET based OP.

The chopper still sees the input impedance direct. So ideally there should be some additional filtering at the positive input of the chopper (e.g. some 1 K to the input and some 100 pF to round or/end the neg side input). How the filter should look like can depend on the OP. I am still not sure about the choice here.

A slight difficulty is that many modern OPs don't have offset trim pins anymore. So the choice of OPs is not that large.

[sourcecharge]

Both OP still contribute to the noise: the chopper (+ a little from the filter) at low frequency and the other OP at higher frequency.

However the current noise of both OPs add. So the current noise will be the sum of both, possibly with some filtering for the chopper OP at really high frequencies. The bias current also adds - so one may want more like a JFET based OP.

The chopper still sees the input impedance direct. So ideally there should be some additional filtering at the positive input of the chopper (e.g. some 1 K to the input and some 100 pF to round or/end the neg side input). How the filter should look like can depend on the OP. I am still not sure about the choice here.

A slight difficulty is that many modern OPs don't have offset trim pins anymore. So the choice of OPs is not that large.

https://www.eevblog.com/forum/projects/modern-opamps-with-null-offset/

Seems you guys were trying to figure this out before.

Here are the ops that were listed in that thread:
OP277
LT1636
OPA192
OPA2192
Are any of these able to be used?

[SilverSolder]

OP277 - Rare, expensive
LT1636 - Slow
OPA192 - Does not actually have pins for offset adjustment
OPA2192 - Same as OPA192 (dual version) - no pins for offset


The LT1037 / LT1007 is still available... not cheap, but not stupid expensive either.

Maybe it is possible to design a composite amp that doesn't rely on offset adjustment pins?

[Kleinstein]

The simple circuit uses the offset trim pins, so the only OP from the list would be an OPA277 (not sure if there was an OP277.  With this type I am no so sure it would make much sense, as it is not that fast and low noise to start with.

It would be more like something like OP27, OP37, LT1037, AD8675 or LT1028 as relatively fast low noise BJT based ones or a low noise JFET based one. However many of the modern low noise ones don't have the trim pins. The AD8610 and ADA4627/ADA4637 are a few candidates that still have trim pins.
There are also some plans around to use a discrete JFET based amplifier with an AZ OP for zero stabilization.

[David Hess]

Maybe it is possible to design a composite amp that doesn't rely on offset adjustment pins?

In the inverting configuration, the non-inverting input can accept the correction signal.