AD8066 buffer oscillates with inductor on input

Hi all,

Someone at work wanted a buffer to drive a coax cable with a signal from 
a high impedance device. I thought this would be a quick job so I built 
a buffer amplifier, using half of an AD8066 configured as a unity-gain 
buffer. I know that op-amps don't like capacitive loads, so I connected 
a 50 Ohm resistor between the cable and the buffer amplifier and also 
terminated the other end of the cable in 50 Ohms, so that the op-amp 
sees a purely resistive 100 Ohm load. When I tested it, the first thing 
I did was to short the input to ground with a clip lead, and it 
oscillated vigorously at about 50 MHz.

Here is a diagram and a photo of the construction:
https://www.flickr.com/photos/37778900@N06/36166979240/
https://www.flickr.com/photos/37778900@N06/36166979290/

Using a x10 passive probe on an oscilloscope I could see 2V p-p at the 
non-inverting input of the op-amp. Moving the shorting clip-lead, or 
connecting different inductances across the input of the amplifier would 
change the frequency. With low inductances it could be made to oscillate 
at up to 116MHz, or with 19 turns of hookup wire wound aroud a 60mm 
diameter, it oscillated down at 6.5MHz, sometimes still able to 
oscillate with 1kOhm in series with the inductor.

It seems like the input impedance of this buffer has a negative real 
part. If someone has a VNA handy, I'd be interested if you made any 
measurements of this chip configured as a buffer. Anyway, as a 
general-purpose lab amplifier, or as a FET active scope probe, this 
op-amp seems fairly undesirable, since one generally wants it to be 
stable regardless of what is being probed at the time.

The datasheet of the AD8066 gives no warning of this kind of 
misbehaviour. There is a linear tech app note that makes a brief mention 
of this sort of phenomenon for op-amps in general, under the heading 
"strange impedances":
http://cds.linear.com/docs/en/application-note/an148fa.pdf
Other than that I think this is not a commonly-known characteristic of 
op-amps. I vaguely recall encountering it in simulation, years ago with 
an on-chip op-amp that I designed, within part of a SOC.

As a very hand-wavy explanation, I suspect that there is a fair bit of 
capacitance between the two input pins, and that in combination with the 
inductor, this causes the differential input voltage of the op-amp to 
lag in phase behind the output voltage of the op-amp, leading to 
excessive loop phase shift at frequencies where the gain is still more 
than unity.

By trial and error with a variable resistor and some capacitors, I 
managed to mostly fix this buffer amplifier by adding some shunt 
capacitance to ground from the non-inverting input pin, and some 
resistance in series with the input connector. The 3.3pF shunt capacitor 
really helps to reduce the amount of resistance needed. It is not a very 
satisfactory fix because it increases the input capacitance, and also 
because I can't prove that there is not some other source impedance 
which, when applied to the input, would make it oscillate again.

If anyone knows a fast fet-input opamp or buffer that certainly won't 
oscillate with arbitrary passive impedances connected at its input, I'd 
be interested to know.

Chris

On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
<lugnut808@spam.yahoo.com> wrote:

>Hi all,
>
>Someone at work wanted a buffer to drive a coax cable with a signal from 
>a high impedance device. I thought this would be a quick job so I built 
>a buffer amplifier, using half of an AD8066 configured as a unity-gain 
>buffer. I know that op-amps don't like capacitive loads, so I connected 
>a 50 Ohm resistor between the cable and the buffer amplifier and also 
>terminated the other end of the cable in 50 Ohms, so that the op-amp 
>sees a purely resistive 100 Ohm load. When I tested it, the first thing 
>I did was to short the input to ground with a clip lead, and it 
>oscillated vigorously at about 50 MHz.
>
>Here is a diagram and a photo of the construction:
>https://www.flickr.com/photos/37778900@N06/36166979240/
>https://www.flickr.com/photos/37778900@N06/36166979290/
>
>Using a x10 passive probe on an oscilloscope I could see 2V p-p at the 
>non-inverting input of the op-amp. Moving the shorting clip-lead, or 
>connecting different inductances across the input of the amplifier would 
>change the frequency. With low inductances it could be made to oscillate 
>at up to 116MHz, or with 19 turns of hookup wire wound aroud a 60mm 
>diameter, it oscillated down at 6.5MHz, sometimes still able to 
>oscillate with 1kOhm in series with the inductor.
>
>It seems like the input impedance of this buffer has a negative real 
>part. If someone has a VNA handy, I'd be interested if you made any 
>measurements of this chip configured as a buffer. Anyway, as a 
>general-purpose lab amplifier, or as a FET active scope probe, this 
>op-amp seems fairly undesirable, since one generally wants it to be 
>stable regardless of what is being probed at the time.
>
>The datasheet of the AD8066 gives no warning of this kind of 
>misbehaviour. There is a linear tech app note that makes a brief mention 
>of this sort of phenomenon for op-amps in general, under the heading 
>"strange impedances":
>http://cds.linear.com/docs/en/application-note/an148fa.pdf
>Other than that I think this is not a commonly-known characteristic of 
>op-amps. I vaguely recall encountering it in simulation, years ago with 
>an on-chip op-amp that I designed, within part of a SOC.
>
>As a very hand-wavy explanation, I suspect that there is a fair bit of 
>capacitance between the two input pins, and that in combination with the 
>inductor, this causes the differential input voltage of the op-amp to 
>lag in phase behind the output voltage of the op-amp, leading to 
>excessive loop phase shift at frequencies where the gain is still more 
>than unity.
>
>By trial and error with a variable resistor and some capacitors, I 
>managed to mostly fix this buffer amplifier by adding some shunt 
>capacitance to ground from the non-inverting input pin, and some 
>resistance in series with the input connector. The 3.3pF shunt capacitor 
>really helps to reduce the amount of resistance needed. It is not a very 
>satisfactory fix because it increases the input capacitance, and also 
>because I can't prove that there is not some other source impedance 
>which, when applied to the input, would make it oscillate again.
>
>If anyone knows a fast fet-input opamp or buffer that certainly won't 
>oscillate with arbitrary passive impedances connected at its input, I'd 
>be interested to know.
>
>Chris

The inductor is probably resonating with stray capacitances. A shunt
resistor might kill the oscillation. It would be interesting to know
the critical value.


-- 

John Larkin         Highland Technology, Inc
picosecond timing   precision measurement 

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

John Larkin wrote:

> On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
> <lugnut808@spam.yahoo.com> wrote:
> 
>>Hi all,
>>
>>Someone at work wanted a buffer to drive a coax cable with a signal from
>>a high impedance device.

A circuit I have used is a X2 amplifier with the 50 Ohm series termination 
resistor outside the feedback path.  This allows you to put whatever 
compensation cap across the feedback resistor to get the stability you want.

For driving large signals into 50 Ohms, my solution is a suitable current-
feedback op amp (such as LT1229) followed by an HA9P5002 buffer.  In this 
case, the feedback resistor value is selected to get the desired gain 
bandwidth.

Jon

On 15/08/2017 05:08, Jon Elson wrote:
> John Larkin wrote:
> 
>> On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
>> <lugnut808@spam.yahoo.com> wrote:
>>
>>> Hi all,
>>>
>>> Someone at work wanted a buffer to drive a coax cable with a signal from
>>> a high impedance device.
> 
> A circuit I have used is a X2 amplifier with the 50 Ohm series termination
> resistor outside the feedback path.  This allows you to put whatever
> compensation cap across the feedback resistor to get the stability you want.
I don't think that it would help much with this problem (other than that 
the loop gain would be 6dB lower).

> For driving large signals into 50 Ohms, my solution is a suitable current-
> feedback op amp (such as LT1229) followed by an HA9P5002 buffer.  In this
> case, the feedback resistor value is selected to get the desired gain
> bandwidth.
Ok thanks for the part numbers, that might be useful later, but the 
problem I was having was with the *input* of the op-amp oscillating, not 
with driving the 50 Ohm cable. I did really want the FET input as I had 
hoped for high input impedance (but I didn't bargain for negative 
resistance!).

On 15/08/2017 03:31, John Larkin wrote:
> On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
> <lugnut808@spam.yahoo.com> wrote:
> 
>> Hi all,
>>
>> Someone at work wanted a buffer to drive a coax cable with a signal from
>> a high impedance device. I thought this would be a quick job so I built
>> a buffer amplifier, using half of an AD8066 configured as a unity-gain
>> buffer. I know that op-amps don't like capacitive loads, so I connected
>> a 50 Ohm resistor between the cable and the buffer amplifier and also
>> terminated the other end of the cable in 50 Ohms, so that the op-amp
>> sees a purely resistive 100 Ohm load. When I tested it, the first thing
>> I did was to short the input to ground with a clip lead, and it
>> oscillated vigorously at about 50 MHz.
>>
>> Here is a diagram and a photo of the construction:
>> https://www.flickr.com/photos/37778900@N06/36166979240/
>> https://www.flickr.com/photos/37778900@N06/36166979290/
>>
>> Using a x10 passive probe on an oscilloscope I could see 2V p-p at the
>> non-inverting input of the op-amp. Moving the shorting clip-lead, or
>> connecting different inductances across the input of the amplifier would
>> change the frequency. With low inductances it could be made to oscillate
>> at up to 116MHz, or with 19 turns of hookup wire wound aroud a 60mm
>> diameter, it oscillated down at 6.5MHz, sometimes still able to
>> oscillate with 1kOhm in series with the inductor.
>>
>> It seems like the input impedance of this buffer has a negative real
>> part. If someone has a VNA handy, I'd be interested if you made any
>> measurements of this chip configured as a buffer. Anyway, as a
>> general-purpose lab amplifier, or as a FET active scope probe, this
>> op-amp seems fairly undesirable, since one generally wants it to be
>> stable regardless of what is being probed at the time.
>>
>> The datasheet of the AD8066 gives no warning of this kind of
>> misbehaviour. There is a linear tech app note that makes a brief mention
>> of this sort of phenomenon for op-amps in general, under the heading
>> "strange impedances":
>> http://cds.linear.com/docs/en/application-note/an148fa.pdf
>> Other than that I think this is not a commonly-known characteristic of
>> op-amps. I vaguely recall encountering it in simulation, years ago with
>> an on-chip op-amp that I designed, within part of a SOC.
>>
>> As a very hand-wavy explanation, I suspect that there is a fair bit of
>> capacitance between the two input pins, and that in combination with the
>> inductor, this causes the differential input voltage of the op-amp to
>> lag in phase behind the output voltage of the op-amp, leading to
>> excessive loop phase shift at frequencies where the gain is still more
>> than unity.
>>
>> By trial and error with a variable resistor and some capacitors, I
>> managed to mostly fix this buffer amplifier by adding some shunt
>> capacitance to ground from the non-inverting input pin, and some
>> resistance in series with the input connector. The 3.3pF shunt capacitor
>> really helps to reduce the amount of resistance needed. It is not a very
>> satisfactory fix because it increases the input capacitance, and also
>> because I can't prove that there is not some other source impedance
>> which, when applied to the input, would make it oscillate again.
>>
>> If anyone knows a fast fet-input opamp or buffer that certainly won't
>> oscillate with arbitrary passive impedances connected at its input, I'd
>> be interested to know.
>>
>> Chris
> 
> The inductor is probably resonating with stray capacitances. A shunt
> resistor might kill the oscillation. It would be interesting to know
> the critical value.

The thing is that I wanted to be able to use it as a general purpose lab 
amplifier like a FET active scope probe. You don't have to worry about 
putting shunt resistors on your circuit board to de-Q the traces when 
you use a normal FET active scope probe, that would be a pain. What I 
did instead was to put a series resistor and shunt cap right at the 
op-amp, as shown in red on my schematic. That can at least be done once 
and for all inside the amplifier/probe, and doesn't involve messing with 
each device that you want to probe with it.

On 2017-08-15 15:32, Chris Jones wrote:
> On 15/08/2017 03:31, John Larkin wrote:
>> On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
>> <lugnut808@spam.yahoo.com> wrote:
>>
>>> Hi all,
>>>
>>> Someone at work wanted a buffer to drive a coax cable with a signal from
>>> a high impedance device. I thought this would be a quick job so I built
>>> a buffer amplifier, using half of an AD8066 configured as a unity-gain
>>> buffer.  [...]
>>>
>>> Chris
>>
>> The inductor is probably resonating with stray capacitances. A shunt
>> resistor might kill the oscillation. It would be interesting to know
>> the critical value.
>
> The thing is that I wanted to be able to use it as a general purpose lab
> amplifier like a FET active scope probe. You don't have to worry about
> putting shunt resistors on your circuit board to de-Q the traces when
> you use a normal FET active scope probe, that would be a pain. What I
> did instead was to put a series resistor and shunt cap right at the
> op-amp, as shown in red on my schematic. That can at least be done once
> and for all inside the amplifier/probe, and doesn't involve messing with
> each device that you want to probe with it.

Did you notice the resistors in the feedback of the G=+1 and
G=+2 circuits in the datasheet? Put 25 ohms in series with
both inputs of the unused section too.

I think you can remove the shunt C at the positive input,
but keep the series resistor. You'll need a shunt resistor
too, say a Mohm or so, if you want the amplifier to behave
with its positive input not connected to anything else. FET
bias current is small, but not zero.

The picture doesn't seem to match the schematics. Take
the feedback straight from the op-amp output, not from
the far end of the output series resistor.

Don't try to insert buffer amplifiers in the loop. They'll
make you miserable.

Jeroen Belleman

On Monday, August 14, 2017 at 9:14:44 AM UTC-4, Chris Jones wrote:
> Hi all,
> 
> Someone at work wanted a buffer to drive a coax cable with a signal from 
> a high impedance device. I thought this would be a quick job so I built 
> a buffer amplifier, using half of an AD8066 configured as a unity-gain 
> buffer. I know that op-amps don't like capacitive loads, so I connected 
> a 50 Ohm resistor between the cable and the buffer amplifier and also 
> terminated the other end of the cable in 50 Ohms, so that the op-amp 
> sees a purely resistive 100 Ohm load. When I tested it, the first thing 
> I did was to short the input to ground with a clip lead, and it 
> oscillated vigorously at about 50 MHz.
> 
> Here is a diagram and a photo of the construction:
> https://www.flickr.com/photos/37778900@N06/36166979240/
> https://www.flickr.com/photos/37778900@N06/36166979290/
> 
> Using a x10 passive probe on an oscilloscope I could see 2V p-p at the 
> non-inverting input of the op-amp. Moving the shorting clip-lead, or 
> connecting different inductances across the input of the amplifier would 
> change the frequency. With low inductances it could be made to oscillate 
> at up to 116MHz, or with 19 turns of hookup wire wound aroud a 60mm 
> diameter, it oscillated down at 6.5MHz, sometimes still able to 
> oscillate with 1kOhm in series with the inductor.
> 
> It seems like the input impedance of this buffer has a negative real 
> part. If someone has a VNA handy, I'd be interested if you made any 
> measurements of this chip configured as a buffer. Anyway, as a 
> general-purpose lab amplifier, or as a FET active scope probe, this 
> op-amp seems fairly undesirable, since one generally wants it to be 
> stable regardless of what is being probed at the time.
> 
> The datasheet of the AD8066 gives no warning of this kind of 
> misbehaviour. There is a linear tech app note that makes a brief mention 
> of this sort of phenomenon for op-amps in general, under the heading 
> "strange impedances":
> http://cds.linear.com/docs/en/application-note/an148fa.pdf
> Other than that I think this is not a commonly-known characteristic of 
> op-amps. I vaguely recall encountering it in simulation, years ago with 
> an on-chip op-amp that I designed, within part of a SOC.
I saw something similar in the past... I think (like you) I added some
series resistance.  

George H. 
> 
> As a very hand-wavy explanation, I suspect that there is a fair bit of 
> capacitance between the two input pins, and that in combination with the 
> inductor, this causes the differential input voltage of the op-amp to 
> lag in phase behind the output voltage of the op-amp, leading to 
> excessive loop phase shift at frequencies where the gain is still more 
> than unity.
> 
> By trial and error with a variable resistor and some capacitors, I 
> managed to mostly fix this buffer amplifier by adding some shunt 
> capacitance to ground from the non-inverting input pin, and some 
> resistance in series with the input connector. The 3.3pF shunt capacitor 
> really helps to reduce the amount of resistance needed. It is not a very 
> satisfactory fix because it increases the input capacitance, and also 
> because I can't prove that there is not some other source impedance 
> which, when applied to the input, would make it oscillate again.
> 
> If anyone knows a fast fet-input opamp or buffer that certainly won't 
> oscillate with arbitrary passive impedances connected at its input, I'd 
> be interested to know.
> 
> Chris

On 16/08/2017 00:56, Jeroen Belleman wrote:
> On 2017-08-15 15:32, Chris Jones wrote:
>> On 15/08/2017 03:31, John Larkin wrote:
>>> On Mon, 14 Aug 2017 23:14:35 +1000, Chris Jones
>>> <lugnut808@spam.yahoo.com> wrote:
>>>
>>>> Hi all,
>>>>
>>>> Someone at work wanted a buffer to drive a coax cable with a signal 
>>>> from
>>>> a high impedance device. I thought this would be a quick job so I built
>>>> a buffer amplifier, using half of an AD8066 configured as a unity-gain
>>>> buffer.  [...]
>>>>
>>>> Chris
>>>
>>> The inductor is probably resonating with stray capacitances. A shunt
>>> resistor might kill the oscillation. It would be interesting to know
>>> the critical value.
>>
>> The thing is that I wanted to be able to use it as a general purpose lab
>> amplifier like a FET active scope probe. You don't have to worry about
>> putting shunt resistors on your circuit board to de-Q the traces when
>> you use a normal FET active scope probe, that would be a pain. What I
>> did instead was to put a series resistor and shunt cap right at the
>> op-amp, as shown in red on my schematic. That can at least be done once
>> and for all inside the amplifier/probe, and doesn't involve messing with
>> each device that you want to probe with it.
> 
> Did you notice the resistors in the feedback of the G=+1 and
> G=+2 circuits in the datasheet? 
Yes I saw that in Fig. 42. I thought it was for the reason in the 
paragraph: "For the best settling times and the best distortion, the 
impedances at the AD8065/AD8066 input terminals should be matched. This
minimizes nonlinear common-mode capacitive effects that can degrade ac 
performance.". They don't bother with the resistor in Fig.55 or Fig.57. 
The value of 25 Ohms is probably too low to stop it oscillating anyway, 
I needed around 400 Ohms in some cases, without the shunt capacitor 
anyway, and with some inductive impedances that was still not enough.

>Put 25 ohms in series with
> both inputs of the unused section too.
That one didn't misbehave (as the wire from the non-inverting input to 
the ground plane is too short to allow it to oscillate) so I don't see 
the need. If the wire were 50mm or longer then I would expect it to 
oscillate though.

> I think you can remove the shunt C at the positive input,
> but keep the series resistor.
Yes, I tried that first. There are some inductanctive source impedances 
for which it will oscillate even with 1k series resistance. I didn't try 
higher series resistors, at some point the resistance starts to affect 
noise and bandwidth. With the added shunt C, much less series resistance 
is needed and it prevents oscillation more reliably.

> You'll need a shunt resistor
> too, say a Mohm or so, if you want the amplifier to behave
> with its positive input not connected to anything else. FET
> bias current is small, but not zero.
Yes, but in the intended application 1MOhm would disturb the measurement 
so instead I warned the user to make sure the input is connected. If I 
had more time, I might add a resistive divider to the output of the 
op-amp and connect a 100M resistor from the non-inverting input to a tap 
at about 0.9 of the output voltage. I could also add a clamp as part of 
the resistive divider which would allow me to raise the positive supply 
voltage with less risk of damaging the instrument that receives the 
output signal. That would improve the positive input voltage range of 
the op-amp.

> The picture doesn't seem to match the schematics.
Sorry for the low-resolution picture, but it does match the schematic. 
The only change is that I have replaced the trimpot with a fixed 
resistor in the final construction.

> Take
> the feedback straight from the op-amp output, not from
> the far end of the output series resistor.
Yes, I did that.

> Don't try to insert buffer amplifiers in the loop. They'll
> make you miserable.
Generally, yes I agree, much extra testing would then be required, and 
trouble is likely unless the buffer is much faster that the op-amp whose 
loop it has been inserted into, which would not be easy with this fast 
op-amp.