Hi guys,

I bought an op-amp yesterday thinking it was unity gain stable, turns out that it's only stable at gains above 5 (the ADA4637):

http://www.analog.com/static/imported-files/data_sheets/ADA4627-1_4637-1.pdf

So I've spent today trying to stabilize this op-amp for gains below 5, with varied success. I'll explain what I think I understand about uncompensated op-amps first.

Definitions:

Open loop gain - Very high, the intrinsic gain of the op-amp

Closed loop gain - Vin*[1+(Rf/Rg)] for a non-inverting amp

Loop gain - open loop gain * feedback network gain (i.e. Rf/Rg)

At low frequencies, there's no roll-off and no phase shift at the output of the op-amp. As frequency increases, roll-off starts to occur due to a filter within the op-amp (first pole). This causes a 90 degree phase shift from the input to output (is this correct?) and 6dB/octave. As frequencies continue to rise, a second pole within the op-amp kicks in (for want of a better phrase). This creates a 180 degree phase shift between the input and output. If this second pole occurs at a frequency where the gain is less than 1 (i.e. it's being attenuated indefinitely) then there's no problem. However - if this pole is located at a frequency where the gain is greater than 1, you get oscillation (180 degree phase shift, positively amplified).

Now you can compensate the intrinsic instability by using it at a gain of 5 or more, the reason that works is excellently explained here:

https://www.youtube.com/watch?v=Db16d88ZziE

So with our chip, an inverting amplifier of gain 2 would be unstable (Rf = 2k, Rin = 1k say). If I wanted to make that chip stable and keep the low frequency gain below 5, how would I do that? Also, if I wanted to configure this amp as a unity gain non-inverting buffer, how could I possibly compensate that circuit? Does all of this have to do with the loop gain?

Sorry about the vague questions, I'll probably be more specific tomorrow after some sleep. Any help/teaching/guidance is greatly appreciated.

Dan

 

Before jumping in with any specific advice, what frequency range is your circuit actually supposed to respond to?

 

As high as possible - this is a part of some investigation equipment I'm designing, and we want to see as much high frequency information as we can. The GBWP of the ada4637 is 79MHz, and I only need a gain of 1, so I was hoping for a bandwidth of 79MHz if that makes sense. Just as a side, if I have two of these buffers in the signal path, one after the other, I'm guessing I wouldn't negatively affect the bandwidth, it'd just roll off quicker (6dB/oct) than if there were only one?

Cheers

 

By adding a compensation capacitor, you will reduce the bandwidth. If you want 79MHz bandwidth, you'll need a different opamp.

 

There's no easy way to compensate an op amp externally without significantly reducing its bandwidth. If you don't need a lot of amplitude, a dirty work-around is to run it with a noise-gain of 5 (non-inverting gain of 5 or an inverting gain of 4) and then attenuate the output by a factor of 5 or 4 with a resistive divider.

You could also attenuate the input by a factor of 5 or 4 respectively also, but that increases the op amp noise contribution by a factor of 5.

There are many op amps available with much higher unity gain bandwidth so you might consider buy one of those instead of trying to fight with this one.

 

Crutschow beat me to my suggestion...divide signal and run at minimum gain.

 

You need this:
http://www.analog.com/library/analogDialogue/cd/vol46n4.pdf#page=3

That was a decent stability series from Analog that I hadn't seen before... thanks for that.

 

As high as possible - this is a part of some investigation equipment I'm designing, and we want to see as much high frequency information as we can. The GBWP of the ada4637 is 79MHz, and I only need a gain of 1, so I was hoping for a bandwidth of 79MHz if that makes sense. Just as a side, if I have two of these buffers in the signal path, one after the other, I'm guessing I wouldn't negatively affect the bandwidth, it'd just roll off quicker (6dB/oct) than if there were only one?

Cheers

Hi,

Quick thought without looking up the data sheet...

If the true GBWP is 79e6 and the gain is 10, then the frequency is limited to 7.9e6 Hz. So with a voltage divider of any kind even without caps the bandwidth is already limited to around 8MHz which sounds much lower than what you need. With a voltage divider and gain of 2 it goes up to around 16MHz which still sounds too low for your app.

 

.........................
If the true GBWP is 79e6 and the gain is 10, then the frequency is limited to 7.9e6 Hz. So with a voltage divider of any kind even without caps the bandwidth is already limited to around 8MHz which sounds much lower than what you need. With a voltage divider and gain of 2 it goes up to around 16MHz which still sounds too low for your app.

That's true, I forgot about the gain-bandwidth limitation in this case. So, with a minimum gain of 5, the bandwidth will be about 15.8MHz. If you want more than that, you will need a different op amp.