EEVblog #572 - Cascading Opamps For Increased Bandwidth

[EEVblog]

Fundamentals Friday.
Dave explains Gain Bandwith Product and how it's possible to increase your system bandwidth by cascading opamps in series. Also, a discussion on the associated noise issues.
A breadboard example shows how variable GBWP can be, and how it can relate to distortion.

[tree]

I'm rather interested in the derivation of that formula in the upper right hand corner.

[Rerouter]

A more interesting point to me was that the phase shift seemed to radically increase when you lowered supply voltage,

(phase margin and feedback loops is one i have really wanted to see for a while for an upcoming video)

[PeterL]

Nice video, I really like this fundamental friday series.

You mention  that it can be hard to find the right combination of resistors to get a precise gain. I made an excel sheet for this long time ago. Just put in the ratio you need, and the tolence you permit, and all possible combination turn green.
Also the android app 'ElectroDriod' is very handy for this.

[mikeselectricstuff]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

[electronics man]

Yay more fundermentals Fridays thanks dave

[EEVblog]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

At these volumes and item count it makes no difference.
Series can have some advantage in terms of single sided SMD layout, and I found they were in stock at the time so just went with that.

[andtfoot]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

At these volumes and item count it makes no difference.
Series can have some advantage in terms of single sided SMD layout, and I found they were in stock at the time so just went with that.

Do you mind expanding on that? Having recently started looking at designing SMD boards, I'm curious...

[Rerouter]

you can jump 2 traces instead of just one, equally if you rotate one 180 degrees your trace follows the same line it entered on, (or so i assume based on how i have been using them)

[mikeselectricstuff]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

At these volumes and item count it makes no difference.
Series can have some advantage in terms of single sided SMD layout, and I found they were in stock at the time so just went with that.

...but you don't know what size the next batch might be.

[Melkor]

You said that most OpAmps have a constant GBWP. However for that particular OpAmp you tested you got values between about 2.4 and 1.4
Is there any way to determine from the datasheet if an OpAmp has a constant or variable GBWP ?

[Wytnucls]

I'm rather interested in the derivation of that formula in the upper right hand corner.

Have a look at these notes for cascading inverting opamps:
http://opencourseware.kfupm.edu.sa/colleges/ces/ee/ee303/files%5C3-Lesson_Notes_Lecture_12.pdf

The logic should be the same for non-inverting opamps, to work-out the right formula.

[EEVblog]

...but you don't know what size the next batch might be.

But I know what it's not going to be!
The point is these precision resistors don't plummet in price when you jump up a volume notch. You only save 3 cents jumping from 1000qty to 100,000qty for example. It becomes a non-issue unless you are after every last reel space on a machine.

[EEVblog]

you can jump 2 traces instead of just one, equally if you rotate one 180 degrees your trace follows the same line it entered on, (or so i assume based on how i have been using them)

Yes and yes. Two in series gives you the option to jump two other traces, and in different locations, and/or returning to the direction you started.
That can mean the difference between making a design realisable on a double sided front panel board (like in the uCurrent) or not.
Of course on regular double sided board this might not matter at all, but when you are doing a front panel you don't want any traces on the other layer.

[robrenz]

Great video Dave, I learned a lot

[daqq]

Thanks for the video Dave, useful.

For a 9:1 divider in one application we used this "IC": http://www.farnell.com/datasheets/572234.pdf

It's pretty precise, but pricey. There are many other ratios available from various makers.

[Rufus]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

If you don't mind E24 200R and 1k8 or 300R and 2k7 does it with 2 resistors.

[free_electron]

GBW is not to be confused with unity gain bandwidth !

GBW is a meaningless number. it is mainly determined by how high the internal stabilising capacitor is between the longtail and the output driver. This capacitor is responsible for the gain collapse over frequency. A far more interesting number is the slew rate of an opamp.


Unity gain bandwidth tells you up to what frequency an opamp can 'follow' the incoming signal when set in unity gain.
unity gain frequency is only specified for 1 input level : 0dB. Send a larger signal in there and UGB will go down. Send a smaller signal in and UGB may go up. You can't just arbitrarily send a level into the opamp.

There are other problems to be faced : the feedback divider will have stray capacitances due to the PCB material , soldermaks ,greasy fingers etc .. This may have an impact on the frequency/gain response. If you are working in the khz range you may not notice this , go to higher frequencies and it becomes problematic. That is why there are current feedback opamps for high frequencies. These do not care about stray capacitances on the pcb.

[tree]

I'm rather interested in the derivation of that formula in the upper right hand corner.

Have a look at these notes for cascading inverting opamps:
http://opencourseware.kfupm.edu.sa/colleges/ces/ee/ee303/files%5C3-Lesson_Notes_Lecture_12.pdf

The logic should be the same for non-inverting opamps, to work-out the right formula.

Thanks. I worked through it and found out how the formula is derived.

[tree]

Why would you want to increase the bandwidth in an application such as this (uCurrent)? I though this would measure ~DC current consumption, especially if you're measuring it with a handheld meter (which don't have 100s of kHz of bandwidth).

[Marco]

A far more interesting number is the slew rate of an opamp.

Take the first opamp in Dave's cascade ... why would slew rate be interesting for it at all?

I can see how it's important to know it can add an additional limit on the bandwidth, but I don't see why it would necessarily be more interesting.

[robrenz]

Why would you want to increase the bandwidth in an application such as this (uCurrent)? I though this would measure ~DC current consumption, especially if you're measuring it with a handheld meter (which don't have 100s of kHz of bandwidth).

I signed up for a µcurrent just to use as a current input to a scope. I don't need it for measuring DC current. In this case the bandwidth increase makes it much more useful.

[tree]

Why would you want to increase the bandwidth in an application such as this (uCurrent)? I though this would measure ~DC current consumption, especially if you're measuring it with a handheld meter (which don't have 100s of kHz of bandwidth).

I signed up for a µcurrent just to use as a current input to a scope. I don't need it for measuring DC current. In this case the bandwidth increase makes it much more useful.

I thought about that, but then wouldn't its [the measurement setup] accuracy be limited by the scope's accuracy? I suppose when you need to make a rough measurement, then it would be fine, but when you're trying to make a measurement with .05% accuracy then I don't think it would work well.

[robrenz]

Why would you want to increase the bandwidth in an application such as this (uCurrent)? I though this would measure ~DC current consumption, especially if you're measuring it with a handheld meter (which don't have 100s of kHz of bandwidth).

I signed up for a µcurrent just to use as a current input to a scope. I don't need it for measuring DC current. In this case the bandwidth increase makes it much more useful.

I thought about that, but then wouldn't its [the measurement setup] accuracy be limited by the scope's accuracy? I suppose when you need to make a rough measurement, then it would be fine, but when you're trying to make a measurement with .05% accuracy then I don't think it would work well.

The µcurrent only guarantees a .05% conversion accuracy from current to voltage. The voltage measuring accuracy contribution to the max error has to be considered whether scope or DMM.  But it is a very nice pre amplified shunt for a scope.

[rfbroadband]

Nice video. To gain more insight on these kind of topics one could also use a simulator like LTSpice/PSpice, especially  when you want to look at noise and distortion. A comment on non-linearity: If one can see non linearity of CW signals in the time domain, it is usually too late. It is more accurate to characterize non linear effects in the frequency domain  (-> simulator or a scope with decent FFT capability, or even an SA).
I like the series.

[Dr. Frank]

Dave,
That's been your best Fundamental Friday:
Storytelling is very straightforward, every aspect very well explained.

A minute critique:
You did not mention, that by cascading those OpAmps, your initial/superior goal of high precision DC gain will deteriorate.
One stage only will give 2 x 0,05% DC max. error, 2 stages will double that worst case value by 2 because you will have 4 instead of 2 uncertain resistors.
The probable error might follow a SQRT law, therefore 2 stages will increase uncertainty by a factor of 1,41 at least.

Anyhow, a fundamental error analysis, plus data from your series production / testing would make up a nice next-next-next fundamental Friday, I suppose.

Frank

[Echo20-127]

Dave,
That's been your best Fundamental Friday:
Storytelling is very straightforward, every aspect very well explained.

A minute critique:
You did not mention, that by cascading those OpAmps, your initial/superior goal of high precision DC gain will deteriorate.
One stage only will give 2 x 0,05% DC max. error, 2 stages will double that worst case value by 2 because you will have 4 instead of 2 uncertain resistors.
The probable error might follow a SQRT law, therefore 2 stages will decrease uncertainty by a factor of 1,41 at least.

Anyhow, a fundamental error analysis, plus data from your series production / testing would make up a nice next-next-next fundamental Friday, I suppose.

Frank

This gain accuracy might actually be a problem, but I expect Dave to know what he is doing with this...


On the bandwith problem, I guess this is a rather clever solution. I'm not very experienced in this domain but I would expect the improvement in the bandwidth domain to come at the detriment of an increased phase delay as the signal propagates through one more OpAmp. (tell me if I'm wrong  )
It would be an improvement when you are trying to read the amplitude of some short transient current, but the delay would be something to keep in mind.

[casper.bang]

Great video... entirely too few fridays in Dave's calendar. 

[qno]

I normally use slew rate.
The GBW is nice but says nothing about the amplitude of the output.
Before the amplitude of the opamps drops to -3 dB the output sinus waveform has been
changed to a triangle due to the slew rate limit.

See wiki

https://en.wikipedia.org/wiki/Slew_rate

[electrocat]

Hi
A method I used to get a 10:1 divider is a 8 resistor network.

 ----R---R---R---R---R---R-
    !     !     !
    --R---R--
          !   

2R in parallel is the 1R in the divider
2R in parallel is series with 4 R's is the 9R

Use a 1% network with 8 individual resistors
Generally the match between individual resisters is much better then 1%, for a radiometric divider the exact value does not matter
The resistors are all from the same mix of resistive material therefore they temperature stability tracks very well.
I have used the method over the years with great success, if space is not a problem
Generally get better then 0.1% ratios with high temperature stability, only have to by 1 part plus not very expensive.

Regards


 

[BravoV]

A method I used to get a 10:1 divider is a 8 resistor network.

 ----R---R---R---R---R---R-
    !     !     !
    --R---R--
          !   

2R in parallel is the 1R in the divider
2R in parallel is series with 4 R's is the 9R

Interesting & noted down, thanks. 

[German_EE]

Sorry but I'm missing something here, I cannot see why the bandwidth of an op amp will change depending on what gain you set it at. All that is in the feedback path is two resistors so what causes a device that works at over 2 MHz at unity gain to only work at a few KHz?

Peter

[Bored@Work]

Sorry but I'm missing something here, I cannot see why the bandwidth of an op amp will change depending on what gain you set it at. All that is in the feedback path is two resistors so what causes a device that works at over 2 MHz at unity gain to only work at a few KHz?

Peter

In simplified terms:

To get a stable opamp it is internally frequency compensated with what is called a dominant pole. The result (again simplified) is that an opamp behaves like a first order low-pass filter with -20 dB/decade frequency response. I.e. the low-pass behavior limits amplification at higher frequencies.

Actually, as an EE you should know that from your introduction course to opamps.

[German_EE]

EE, Electrical Engineer, normally I work on bigger stuff such as 10Kw motor drivers but I do the smaller stuff like this for fun.

Peter

[vigsgb]

I think you should create a video on bandwidth as it applies to electronics as I don't even know why you needed more bandwidth in your micro current..  was it for response time of the input?  Make tests quicker.   When I search online and YouTube I find so many variants but nothing a newbie like myself can understand.

[mexakin]

You need more bandwitdh in opamp circuits to see fast changing frequencies on your test signals,
for example try to measure a rectangle signal with a too small bandwitdh, you will see a sinus of it, so you must always know what to expect from your test and measurment equipments, otherwise there is no need to measure at all

[The Chump]

For 9K, surely 2x18K in parallel is better than 2K2+6K8 as it reduces your BOM.

For a two resistor ratio of 9:1, what could be better than 1k8 and 200R?

[vigsgb]

You need more bandwitdh in opamp circuits to see fast changing frequencies on your test signals,
for example try to measure a rectangle signal with a too small bandwitdh, you will see a sinus of it, so you must always know what to expect from your test and measurment equipments, otherwise there is no need to measure at all

Thanks!  That answers my question.   Tomorrow I will be playing with some op amps.   Need to get a scope soon so I can see what the wave form looks like and such.   

[VanitarNordic]

in the video, Dave mentions that it is better that we apply the whole gain in one opamp for a couple of reasons, what are the reasons behind this?