Another op-amp stability question (LT1677)

[John Miles, KE5FX]

Tim W.'s question about the influence of resistance in the feedback path
of a voltage follower reminded me of a similar problem I ran into a while
back.

I've used LT1677s as an integrator in several low-noise PLLs in the
past and been satisfied with its performance. It's an old part and
a fairly expensive one, but it's quiet and well-behaved in loop filters,
with RRIO operation and guaranteed freedom from phase reversal. It's
slow (GBP=7 MHz) but not slow enough to be a problem in most RF PLLs.
You could ask for lower input bias current, I guess. But overall the
LT1677 is a good "jellybean" opamp for use between a low-voltage charge
pump and a high-voltage VCO.

It's also characterized explicitly for use at Av=+1; see page 13 of the
data sheet at http://www.linear.com/docs/2102 . At one point, I needed
a voltage follower for something I was breadboarding, and an LT1677 was
already lying around on the workbench. When I put it in the circuit,
I observed a nasty oscillation that didn't make any sense. It's easy to
reproduce the setup that misbehaves. Tie pin 2 to pin 6 as usual, put
a scope probe on pin 6, and feed a steady DC level or an LF tone to
pin 3:

http://www.ke5fx.com/1677.jpg

When the input (and output) level approaches 1.13V, I see this:

http://www.ke5fx.com/1677_osc_sine.png (when fed with sine wave)

http://www.ke5fx.com/1677_osc.png (when fed with DC from trimpot)

Things that don't have a significant effect on the symptom:

- Load resistance, either in series or parallel with the scope probe
- Load capacitance (lots of C will cause instability as expected)
- Source impedance (4.7K in this test bed, but doesn't matter)
- Supply voltage (might be a little less prone to occur at >10V)
- Resistance in feedback path (too much causes instability as expected,
and it might get a bit better with some values of resistance, but
I can always make it happen)
- The usual suspects like power supply impedance and bypassing, layout,
etc.

Things that do have an effect:

- The frequency of the input signal. If I increase the frequency from
the function generator beyond about 1 kHz, the oscillation goes
completely away (?!)

- Temperature. R134a spray lowers the amplitude of the oscillation,
while a heat gun makes it worse.

- Using a different opamp. No other parts rated for Avmin=+1 within easy
reach exhibit this behavior. Any given LT1677 from my parts drawer will.

So... what's up with this? Can anyone else repro the behavior? Maybe
I just got a bad batch of parts, or I'm doing something dumb. Hope so.

-- john, KE5FX

[pcdh...@gmail.com]

>So... what's up with this?  Can anyone else repro the behavior?  Maybe
>I just got a bad batch of parts, or I'm doing something dumb.  Hope so.

Where's the supply bypass? First guess is that 100 nF to ground will fix it. Keep the leads less than 1/4 inch long.

Not everything in the amp is as slow as 7 MHz--it looks like the two input stages (required for RRI operation) are fighting each other at the crossover point.

Cheers

Phil Hobbs

[John Larkin]

Did you use a 1x probe? That could make it oscillate.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

[piglet]

Datasheet indicates input stage NPN/PNP handover occurs at "approx 1.5V"
above negative rail so that figures with your observed 1.2 threshold.

piglet

---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus

[John Miles, KE5FX]

On Wednesday, March 29, 2017 at 5:18:47 AM UTC-7, John Larkin wrote:
> Did you use a 1x probe? That could make it oscillate.
>

As I said, it doesn't matter what the load is. 330 ohms to ground, a 10x
probe, a 1x probe, an active FET probe. Ditto supply bypassing.

-- john, KE5FX

[John Miles, KE5FX]

On Wednesday, March 29, 2017 at 4:50:57 AM UTC-7, pcdh...@gmail.com wrote:
> Where's the supply bypass? First guess is that 100 nF to ground will fix it.

Big glass-encapsulated axial ceramic in photo. I think it's 330 nF but it
might be 100. Makes no difference.

-- john, KE5FX

[tabb...@gmail.com]

I don't know either. I presume you've tried slotting in other opamps, were they ok?


NT

[John Larkin]

(If I don't make sense here, it's because I just had an endoscopy and
I'm still metabolizing IV drugs)

Looking at the data sheet figs on sheet 8, the middle figure is input
bias current vs cm voltage; if this is current out of DUT, isn't that
a negative input impedance?

RRI opamps usually have those tricky PNP-NPN diff pair swapping
circuits. I've always thought that the chip designers have to be very
careful about how the offset voltage and bias current transition has
to be handled as regards closed-loop behavior.

[Gerhard Hoffmann]

I have just built and tested it.
The setup is in

<
https://www.flickr.com/photos/137684711@N07/32912433303/in/dateposted-public/
>

I used a HP3325B as signal source, 500 Hz +- .
The built-in offset voltage is nice here.
There is a 49R9 termination resistor from Pin3 to gnd.

No problems whatsoever with Vcc from 3 to 15V.
But with Vcc < 2.4V hell breaks loose.
You can get any distortion or oscillation you can imagine
by just playing with the voltages.

But the spec says 3V min, so that is ok.

regards, Gerhard

[John Miles, KE5FX]

On Wednesday, March 29, 2017 at 2:29:23 PM UTC-7, Gerhard Hoffmann wrote:
> There is a 49R9 termination resistor from Pin3 to gnd.
>

Thanks, that seems to be a good clue. I was mistaken when I said the
source Z "didn't matter" -- I must not have tried anything very low.
Can you take the 49.9 ohm termination out, and add a 10K series
resistance from the function generator instead? Set the 3325B to
50 Hz, 1V peak, 1V offset, and run the LT1677 from +5V. See if it
looks like this:

http://www.ke5fx.com/3325B_1Vpk_1Voff_10K_series_R_Vcc5.png

When I use a termination under a couple K, the instability goes away.

-- john, KE5FX

[John Miles, KE5FX]

On Wednesday, March 29, 2017 at 1:02:31 PM UTC-7, John Larkin wrote:
> (If I don't make sense here, it's because I just had an endoscopy and
> I'm still metabolizing IV drugs)

Hey, that's a better excuse than I usually come up with!

> Looking at the data sheet figs on sheet 8, the middle figure is input
> bias current vs cm voltage; if this is current out of DUT, isn't that
> a negative input impedance?

Not sure how to interpret it. If they measure Ib the way Analog Devices'
MT-038 note suggests, they're plotting the average of the currents at
both inputs. I imagine the effects of negative common-mode impedance
would be reduced by the relatively high CMRR.

> RRI opamps usually have those tricky PNP-NPN diff pair swapping
> circuits. I've always thought that the chip designers have to be very
> careful about how the offset voltage and bias current transition has
> to be handled as regards closed-loop behavior.

I'll bet the effect is related to that, in that the input bias current
varies slightly when different transistors take over. With a high source
resistance (as in my reply to Gerhard) the slight change in Ib might give
rise to positive feedback, this time in differential mode. Also might
explain why it's so temperature sensitive.

The OPA197 you mentioned the other day looks like a very good replacement
for the LT1677. I'll have to give one those a try.

-- john, KE5FX

[John Miles, KE5FX]

On Wednesday, March 29, 2017 at 5:14:52 PM UTC-7, John Miles, KE5FX wrote:
> http://www.ke5fx.com/3325B_1Vpk_1Voff_10K_series_R_Vcc5.png

By the way, don't use any so-called "high resolution" modes or other
averaging buttons on your DSO when you look for this effect. They clean
up the oscillation nicely.

-- john, KE5FX

[Gerhard Hoffmann]

Hi, John,

you have found a real bug with the LT1677 !

<
https://www.flickr.com/photos/137684711@N07/33598135802/in/album-72157662535945536/
>

cheers,

Gerhard, DK4XP

[John Miles, KE5FX]

On Friday, March 31, 2017 at 5:40:10 AM UTC-7, Gerhard Hoffmann wrote:
> Hi, John,
>
> you have found a real bug with the LT1677 !
>
> <
> https://www.flickr.com/photos/137684711@N07/33598135802/in/album-72157662535945536/

Thanks very much for checking, Gerhard. I was hoping I just got a bad
batch of LT1677s or some midnight-shift specials out of Shenzhen.

Maybe this is common/expected behavior when driving RRI opamps from
high-Z sources, but manufacturers need to document it if so.

-- john, KE5FX

[Phil Hobbs]

Nah, there are high-Z things in the datasheet's applications circuits.

That sort of mistake is pretty rare for LTC in my experience--they seem
to have learned from the LT1028 noise peak debacle. (Early LT1028
datasheets had noise plots that cut off at 20 kHz or so, neatly
disguising the big peak at ~300 kHz that disfigures what would otherwise
be almost the universal low noise, low-Z op amp.

(I much prefer the ADA4898 because it doesn't have the big peak and is
much less squirrelly generally. Not as fast, though, nor quite as quiet
outside of the peak.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

[tabb...@gmail.com]

It's a known issue with some opamps, but I've stayed quiet cos I can't remember where a datasheet (IIRC) talks about it. IIRC the advise was to keep source impedance under 10k, and the cause is stray capacitive coupling from output to input plus a faster input stage than output stage. But as I say, I can't remember where this was from so ICBW.


NT

[John Miles, KE5FX]

On Saturday, April 1, 2017 at 9:16:52 AM UTC-7, tabb...@gmail.com wrote:
> It's a known issue with some opamps, but I've stayed quiet cos I can't
> remember where a datasheet (IIRC) talks about it. IIRC the advise was
> to keep source impedance under 10k, and the cause is stray capacitive
> coupling from output to input plus a faster input stage than output
> stage. But as I say, I can't remember where this was from so ICBW.

Doesn't seem like a typical phase-margin gotcha, or one that involves external
coupling or feedback of any kind. There's no sensitivity to stray C at
all. There must be a thermal component to the phenomenon, because it's only
noticeable at very low frequencies and is quite sensitive to temperature. The
oscillation itself is around 700 - 800 kHz but I've never seen it appear on
signals above a few kHz. No electrical time constants are anywhere near that.

Given that people don't usually use pricy low-noise opamps with high-
resistance sources, I guess it's not too surprising that the issue isn't
well-known. That's no excuse for manufacturers not knowing about it and
warning accordingly, though.

-- john, KE5FX

[Winfield Hill]

John Miles, KE5FX wrote...

Yes, John. But I think its safe to say opamps going through
input-pair transitions at certain voltages are suspect to be
used at those voltages. Your experience illustrates that point.


--
Thanks,
- Win

[John Larkin]

On 1 Apr 2017 16:46:00 -0700, Winfield Hill <hi...@rowland.harvard.edu>
wrote:

But what's the use of a rail-to-rail opamp if, well, you can't use it
rail to rail? I mean, really!

Sounds like LTC scrooed up that one.

[tabb...@gmail.com]

On Saturday, 1 April 2017 23:03:18 UTC+1, John Miles, KE5FX wrote:

> On Saturday, April 1, 2017 at 9:16:52 AM UTC-7, tabby wrote:

> > It's a known issue with some opamps, but I've stayed quiet cos I can't
> > remember where a datasheet (IIRC) talks about it. IIRC the advise was
> > to keep source impedance under 10k, and the cause is stray capacitive
> > coupling from output to input plus a faster input stage than output
> > stage. But as I say, I can't remember where this was from so ICBW.
>
> Doesn't seem like a typical phase-margin gotcha,

what I described isn't. It's what happens when the input stage moves fast but the output stage doesn't.

> or one that involves external
> coupling or feedback of any kind.

You've tested for that?

> There's no sensitivity to stray C at
> all.

that too?

> There must be a thermal component to the phenomenon, because it's only
> noticeable at very low frequencies and is quite sensitive to temperature. The
> oscillation itself is around 700 - 800 kHz but I've never seen it appear on
> signals above a few kHz. No electrical time constants are anywhere near that.
>
> Given that people don't usually use pricy low-noise opamps with high-
> resistance sources, I guess it's not too surprising that the issue isn't
> well-known. That's no excuse for manufacturers not knowing about it and
> warning accordingly, though.

If my memory serves me correctly, they have.


NT

[John Miles, KE5FX]

On Saturday, April 1, 2017 at 8:59:59 PM UTC-7, tabb...@gmail.com wrote:

> You've tested for that?

Yep.

>that too?
Yep.

I routinely play with 10 GHz GBW parts. It's been a while since I stubbed my
toe on a 7-MHz one.

>If my memory serves me correctly, they have.

Not in the data sheet I linked to, they don't.

If a data sheet from Linear Technology tells me that a part can be used as a
unity-gain buffer, I'm inclined to believe it. I agree with the subtext
behind Win's comment, in that ignorance of fundamentals is no excuse, but I
don't agree that this is a "fundamental" topic. I believe that 99 out of 100
working EEs would have made the same ass-umption, so there needs to be an
explicit warning in the data sheet.

-- john, KE5FX

[tabb...@gmail.com]

On Sunday, 2 April 2017 05:41:32 UTC+1, John Miles, KE5FX wrote:

> On Saturday, April 1, 2017 at 8:59:59 PM UTC-7, tabby wrote:
>
> > You've tested for that?
> Yep.
>
> >that too?
> Yep.
>
> I routinely play with 10 GHz GBW parts. It's been a while since I stubbed my
> toe on a 7-MHz one.
>
> >If my memory serves me correctly, they have.
>
> Not in the data sheet I linked to, they don't.
>
> If a data sheet from Linear Technology tells me that a part can be used as a
> unity-gain buffer, I'm inclined to believe it. I agree with the subtext
> behind Win's comment, in that ignorance of fundamentals is no excuse, but I
> don't agree that this is a "fundamental" topic. I believe that 99 out of 100
> working EEs would have made the same ass-umption, so there needs to be an
> explicit warning in the data sheet.

Found it.
https://www.onsemi.com/pub/Collateral/LM339-D.PDF
page 6


NT

[John S]

There is an error in the schematic on that page. I could tell you where
it is, but then I'd have to kill you.

[John S]

On 3/29/2017 7:57 PM, John Miles, KE5FX wrote:
> On Wednesday, March 29, 2017 at 1:02:31 PM UTC-7, John Larkin wrote:
>> (If I don't make sense here, it's because I just had an endoscopy and
>> I'm still metabolizing IV drugs)
>
> Hey, that's a better excuse than I usually come up with!

If he can spell those high-power words correctly, he's just fine.

[k...@notreal.com]

A dot connecting the wire crossing the +15V line. How long did it
take you to fin it?

[John S]

A second or two. It was immediately obvious.

[John Larkin]

(If I don't make sense, it's because I just had an endocscopy and I'm
still a little drugged)