OpAmp to uniform amplification of signal from DC to ~100MHz

[gashtaan]

Hi,
I'd like to observe current consumption of various MCUs in sub-microseconds scale. The current consumption of MCU would be let say from 10uA to 30mA. My idea is to use 10R shunt resistor and amplify voltage drop with some high-bandwidth (~100MHz) operation amplifier with 10x gain to the level measurable with an oscilloscope. The problem is I want to see both fast changes in consumption on MCU cycle granularity and constant consumption during longer time periods too (i.e. I don't want to filter the DC component out). What I've seen there are OpAmps that performs great at high frequencies but they performs bad at low frequencies and vice-versa. Is there any OpAmp that would satisfy my needs or perhaps different solution?

I was going to use BGA2803 for it (I've seen it's being used in similar project), but I'm not sure if it's usable to amplify DC too. In the datasheet they say "General purpose low noise wideband amplifier for frequencies between DC and 2.2 GHz" - but there are DC blocking capacitors in application information schemes and in Fig.9 is obvious that the gain drops significantly for low frequencies.

[jonpaul]

use TEK crrent probe and amp, wide range capability, 50..100 MHz B

Alt: 4 term series R and wideband diff probe

See app notes eg Linear Tech, on diffeerential amps.

j

[gashtaan]

Thanks for advice. I think that differential amps are overcomplication for this and differential probes are too expensive. I want to put shunt on low-side, I hope there is an easier (and cheaper) way to do it with one or two ICs and some passives around. Am I too naive?

[David Hess]

Let me describe how I would make a high side measurement with your requirements.  There are two methods I would consider.

The first involves the AD8129 difference amplifier which is almost ideal and more than fast enough.  If its offset and drift are not good enough, then I might add a fast differential preamplifier made from a couple of current feedback operational amplifiers.  The AD8129 could also be useful for a low side measurement because its common mode rejection will remove common mode noise from a remote measurement.

Alternatively there is a 2 or 4 transistor circuit for measuring high side current.  An operational amplifier could be added to remove DC errors, but I am not sure how to apply it to a low side current measurement.

Aren't parts like the AD8007 fast and precise enough for your application?

[dobsonr741]

Are you doing security research, or just want to analyze/optimize for low power?

If the later, you won't see much sub microseconds detail in a real circuit. A good industry solution example is TI's EnergyTrace, https://software-dl.ti.com/ccs/esd/documents/xdsdebugprobes/emu_energytrace.html 
Their sampling is much lower, 250KHz: https://www.ti.com/lit/ug/spruie1a/spruie1a.pdf
The patent behind is https://patentimages.storage.googleapis.com/30/8a/a1/b9bbd09052c5e5/US20140253096A1.pdf

[gashtaan]

I would like to avoid high side measurement as I need to look on other signals from/to MCU on other channels, so same ground is convenient for me.

Looking on AD8xxx datasheets, I'm not sure how to use it in low side... I'll study it more, seems interesting.

Aren't parts like the AD8007 fast and precise enough for your application?

I don't need much precision in absolute numbers, I just need to see how current changes on the specific time interval. Now I'm considering different approach: use high-bandwidth opamp (like BGA2803) for fast current changes together with some basic opamp with low-pass filter to amplify lower frequencies. It hardly would be the same ratio between them, but it should be enough for my purpose.

[gashtaan]

Are you doing security research, or just want to analyze/optimize for low power?

It's the former, the security research. It would help me to capture precise timing of some actions inside MCU.

[jonpaul]

See the decades or R&D on cryptology use of CPU, IC power for security.

Paul KOCHER founeded Cryptography Research now part of RSA.

See "differential power analysis"

He had a floor full of latest TEK and Agilent top range equipment.

Jon

[David Hess]

I don't need much precision in absolute numbers, I just need to see how current changes on the specific time interval. Now I'm considering different approach: use high-bandwidth opamp (like BGA2803) for fast current changes together with some basic opamp with low-pass filter to amplify lower frequencies. It hardly would be the same ratio between them, but it should be enough for my purpose.

There are several ways to combine a slow precision operational amplifier with a high speed operational amplifier to get the best performance of both.

Given what you are doing, ultimate resolution will be limited by noise in the operational amplifiers and drift in the current shunt as its temperature changes do to self heating.  The advantage of a part like the AD8129 which I mentioned is that it operates like an instrumentation amplifier so it removes the effects lead resistance and common mode errors.  It translates a remote differential measurement to a single ended measurement at the point where you want to measure.

[dobsonr741]

In case have not seen the open sourced work for from New AE: https://www.newae.com/products/nae-cwlite-arm

[gashtaan]

In case have not seen the open sourced work for from New AE: https://www.newae.com/products/nae-cwlite-arm

I know this project, but I don't see it's doing something similar to what I'm asking here about.

[gashtaan]

There are several ways to combine a slow precision operational amplifier with a high speed operational amplifier to get the best performance of both.

Given what you are doing, ultimate resolution will be limited by noise in the operational amplifiers and drift in the current shunt as its temperature changes do to self heating.  The advantage of a part like the AD8129 which I mentioned is that it operates like an instrumentation amplifier so it removes the effects lead resistance and common mode errors.  It translates a remote differential measurement to a single ended measurement at the point where you want to measure.

What I care is the interference mixed into weak signal from current shunt, I bet no opamp is so bad that it'll make worse I just want to put opamp(s) and shunt close together to amplify the signal to the level that the interference will not be so significant and then measure it. Drift in the current shunt also don't bother me, as I said, I'm not interested in absolute numbers.

Anyway, AD8129 seems interesting, I'll study it more. I noticed that NewAE used it in their differential probe too... it's not very suitable to measure low-side shunt though.
http://media.newae.com/datasheets/NAE-DIFFPROBE_datasheet.pdf

[Calvin]

Hi,

for a Diode leakage testing device we designed in our company (range from 1µA to ~10mA) we used a ADA4523-1 chopper OPA in the configuration as I/V converter (inv input in).
We can truely measure down to 100nA and even below.
Post amplification with variable gains was added to boost the signal to a sufficient level for an ADC.
Bandwidth is of course not as high as the thread opener asks for, something about 10kHz-100kHz could be possible though.
But similarly to many scope probe circuits (like the P6042A current probe, et al), one could maybe add a a parallel HF-branch and join both outputs again?

regards
Calvin
 

[David Hess]

Anyway, AD8129 seems interesting, I'll study it more. I noticed that NewAE used it in their differential probe too... it's not very suitable to measure low-side shunt though.

The AD8129 also works fine for a low-side shunt, but the question then becomes why bother.  Its input noise of 4.5 nV/SqrtHz is about as good as most 100+ MHz parts, but its differential input has the advantage of removing common mode noise and errors.  This allows more than one ground connection for measurements without introducing common mode noise, and that by itself reduces noise of the measurement.

The alternative of using a non-inverting amplifier at the shunt works fine also, but requires the connection for the measurement to be directly at the shunt, and no other connections be made for lowest noise.  This can work fine if the ground connection to the shunt is used as a single point ground for *all* test connections.

Other than the AD8129 I am not sure what I would use.  Any fast current feedback amplifier will do the job.  I would look for one with high gain, low noise, and offset null capability but the later is not common.

[gashtaan]

Thanks for the valuable insights. Maybe it works fine, but at least that NewAE diff probe needs very specific positive and negative power supply in low-side configuration. To me it's very inconvenient, I would like to have only 5V and 3.3V power rails on the board. I thought it's limitation of diff amps, but maybe it's just the property of the probe. Am I understand it wrong?

Any fast current feedback amplifier will do the job.  I would look for one with high gain, low noise, and offset null capability but the later is not common.

Pretty please, can you give me some example of such amp?

[KE5FX]

Offset nulling isn't really a thing with CFB amps, at least as far as I understand it.  Their inputs are (intentionally) very different. 

[David Hess]

Offset nulling isn't really a thing with CFB amps, at least as far as I understand it.  Their inputs are (intentionally) very different.

Current feedback operational amplifier with offset null are not common, but they do or did exist, for example:

https://www.analog.com/en/products/lt1217.html
https://www.analog.com/en/products/lt1223.html
https://www.analog.com/en/products/lt1227.html

Thanks for the valuable insights. Maybe it works fine, but at least that NewAE diff probe needs very specific positive and negative power supply in low-side configuration. To me it's very inconvenient, I would like to have only 5V and 3.3V power rails on the board. I thought it's limitation of diff amps, but maybe it's just the property of the probe. Am I understand it wrong?

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

There are fast single supply operational amplifiers, but I think they are all voltage feedback.  They could be used, but current feedback parts have advantages at high speeds.

Any fast current feedback amplifier will do the job.  I would look for one with high gain, low noise, and offset null capability but the later is not common.

Pretty please, can you give me some example of such amp?

The only single supply part that I know of which can give enough gain at 100 MHz might be the LTC6268-10, but I have never used it so maybe someone else can comment.

Keep in mind that in order to be useful, the amplifier has to have enough gain-bandwidth product to provide useful gain at 100 MHz, so that means a GBP of many times 100 MHz, in this case 40 times.

[KE5FX]

Current feedback operational amplifier with offset null are not common, but they do or did exist, for example:

https://www.analog.com/en/products/lt1217.html
https://www.analog.com/en/products/lt1223.html
https://www.analog.com/en/products/lt1227.html

Interesting, never ran into those before.  Seems that the more-modern CFB parts don't even pretend to be any good at DC.

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

They can be used with a single supply like any other opamp, given a midpoint voltage.  This one ought to be pretty flat from AF up to 100 MHz or so, if you keep an eye on the slew rate:

Keep in mind that in order to be useful, the amplifier has to have enough gain-bandwidth product to provide useful gain at 100 MHz, so that means a GBP of many times 100 MHz, in this case 40 times.

GBP?  CFB amps don't got no steeeenkin' GBP... at least not in the traditional sense.

[David Hess]

Current feedback operational amplifier with offset null are not common, but they do or did exist, for example:

https://www.analog.com/en/products/lt1217.html
https://www.analog.com/en/products/lt1223.html
https://www.analog.com/en/products/lt1227.html

Interesting, never ran into those before.  Seems that the more-modern CFB parts don't even pretend to be any good at DC.

Their DC precision depends on matching between their PNP and NPN transistors so cannot be as good as the differential pairs in voltage feedback operational amplifier, but offset null still helps.

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

They can be used with a single supply like any other opamp, given a midpoint voltage.  This one ought to be pretty flat from AF up to 100 MHz or so, if you keep an eye on the slew rate:

Sure, but this application requires operation down to DC.  Level shifting could be used instead but would further compromise precision.

Keep in mind that in order to be useful, the amplifier has to have enough gain-bandwidth product to provide useful gain at 100 MHz, so that means a GBP of many times 100 MHz, in this case 40 times.

GBP?  CFB amps don't got no steeeenkin' GBP... at least not in the traditional sense.

GBP varies with the value of the feedback resistance so frequency compensation can be adjusted to match the gain.  Bandwidth still drops as gain increases, but not as quickly as with an amplifier that has fixed compensation.

Voltage feedback operational amplifiers can use external compensation to do the same thing, and some are available which are decompensated, or a fixed gain stage can be added within the feedback loop to do the same thing.

[KE5FX]

GBP varies with the value of the feedback resistance so frequency compensation can be adjusted to match the gain.  Bandwidth still drops as gain increases

Well, yeah, it had better, or things will get oscillatey at some point.

but not as quickly as with an amplifier that has fixed compensation.

The behavior is largely unrecognizable compared to a traditional voltage-feedback amp with a GBW product.  It really cannot be described as a GBW product.  When the available gain starts to decrease, it pretty much falls off a cliff.  Trying to compensate it is another good way to build an oscillator.

I have a lot of bench time with these puppies.

[gashtaan]

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

Keep in mind that in order to be useful, the amplifier has to have enough gain-bandwidth product to provide useful gain at 100 MHz, so that means a GBP of many times 100 MHz, in this case 40 times.

That's what I afraid of. The need for negative power supply is definitely something I want to avoid.

Thanks for the advice about gain-bandwidth product. Honestly, I don't know what does it mean I haven't seen such parameter in the datasheets...

It's probably dumb, but will something like this work?


Keep in mind that I don't need accuracy. I just need to see fast current changes on some "DC offset". When I observing current consumption of MCUs it mostly looks something like this:
https://circuitdigest.com/sites/default/files/inlineimages/u3/Power-Profiler-Kit-2-Data-Logger-View.png

[David Hess]

It's probably dumb, but will something like this work?

Keep in mind that I don't need accuracy. I just need to see fast current changes on some "DC offset". When I observing current consumption of MCUs it mostly looks something like this:
https://circuitdigest.com/sites/default/files/inlineimages/u3/Power-Profiler-Kit-2-Data-Logger-View.png

Yes, there are two path amplifier designs which would work but they are more complicated.  The gain (and phase) of the two paths need to be matched where the transition from one path to the other occurs or the frequency and phase response can get weird.

In your example a resistor needs to be added in series with the output of U7 with its resistance and value of C13 also controlling the transition.

[gashtaan]

Oh right, I naively added U7 buffer to avoid U5 feedback being affected by the output of U4, but it didn't ocurr to me that the same problem is still there

I thought I would experimentally set gain of the U5 to match the gain of U4 later, but that the signals would be at different phases didn't occur to me either.

[gashtaan]

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

I'm reading about that NewAE probe that uses AD8129 differential amplifier and it seems it can be used also with positive power supply, it just must be 2.25V higher than measured voltage. It even allow to measure DC bellow 20kHz (with lower gain, 3-5x instead of 10x). It must be used to high-side measuring in such configuration, but if I understand it right, it is possible to measure the output relative to GND (amplifier output will be somewhere between power-supply and GND?). It's starting to look very promising for me.

https://rtfm.newae.com/Tools/CW501%20Differential%20Probe/

[David Hess]

Current feedback operational amplifiers require positive and negative supplies because of how they are constructed with diamond buffer elements.  This is why they did not become available until complementary bipolar processes, fast NPNs and fast PNPs, became available.

I'm reading about that NewAE probe that uses AD8129 differential amplifier and it seems it can be used also with positive power supply, it just must be 2.25V higher than measured voltage. It even allow to measure DC bellow 20kHz (with lower gain, 3-5x instead of 10x). It must be used to high-side measuring in such configuration, but if I understand it right, it is possible to measure the output relative to GND (amplifier output will be somewhere between power-supply and GND?). It's starting to look very promising for me.

The datasheet is clear; the AD8129/AD8130 inputs, all 4 of  them, must be at least 1.5 volts away from each supply rail, and the output only reaches to within 1.1 volts of each supply rail.  The differential probe level shifts the input into range for the AD8129, and uses a virtual ground for the output which comes out at pin 4 of the connector.

A resistive level shift would defeat the purpose of using it in your circuit because it would increase noise, decrease sensitivity, and decrease bandwidth.  There *is* a way to use the AD8129/AD8130, but it involves a much more complicated circuit with a dual matched transistor to sense the current and do the level shift, like the first example I gave.

I always do these types of circuits with a separate bipolar dual supply so I can avoid the single supply issues which just add complication.