Op-amp oscillation frequency goes up as C load is increased

[741]

I was trying to get some insight into instability using a TSV634 (880 kHz GBP, R2R i/o). The DS says it is stable up to 100pF, so I went ahead and overloaded it with 10nF, 20nF, 30nF...

I was unsurprised it oscillated, but I am still unsure why the frequency rises as I add capacitance.

My assumption was that a larger capacitance places the "-12db" slope at a lower frequency.

When configured a follower, then I see
10n: 60Hz
20n: 380 Hz
30n: 32 kHz //Sudden jump
40n 36 kHz
50n 40kHz
100n 41 kHz
200n 44 kHz

I had hoped to be able to work out the attenuation 'B' needed to just stop oscillation - but the rising F with larger C has unsettled me.

PS: I read somewheere that G(w).B(w) averages to 1 over a cycle when oscillations occur due to cut-off etc.

[danadak]

Two questions :

1) How were you measuring freq, with a counter or visual observation with scope ?

2) What do you have on the supply rails for bypassing ? Datasheet recomendations ?
And good low L ground path......


Regards, Dana.

[741]

Hi

(1) Using 'kHz' feature (up to 4MHz) on a DMM.

-->Do you think this might be the source of the surprising reasons? i.e. DMM picking up harmonics?
Seems like a distinct possibility.

(My old 'scope is with a friend 'on loan' at present...).



(2) 'On-adapter' 100n close to IC. (The IC adapter has on-board decoupling pads). Power is 4x'AA' + 78L05.

[danadak]

I would advise you look at it with a scope. DMMs have min levels to
trigger internal counters, could be that your osc amplitude changes
significantly with freq.

OpAmp slew rate enters into the osc result as well as GBW of the OpAmp.



Regards, Dana.

[T3sl4co1l]

Loop calculations don't necessarily apply, because the output stage can oscillate on its own, with the inputs tied back suitably*.

*Not that there's really a very suitable way to do this...  You can't simply ground the inputs, because then it amplifies its input offset voltage, probably leaving the output railed.  You add a very small voltage in series with one input to trim this out.  Then you can bias the output stage into the linear range, without having a closed feedback loop.  Except... you're still sensitive to parasitics: pin inductance (especially supplies) and pin-pin capacitance.  Even for a 880kHz amp.  So, "suitable" may be a bit strong, and it depends on the amp's internal design, but it may be possible to set up such a test.

Taking a page from RF design: you've studied one axis of the output load condition, capacitance of varying magnitude.  To complete the plot, you also need inductive load, and any resistance value inbetween.  This is rather daunting (a 2-D space of real numbers, good luck with that ), but the region of instability (where it oscillates) will generally be in a single range (namely, most capacitance values, at low ESR), and you can test a couple values at a time to walk along the boundary of that region, and thus plot its extents.  (Since you've found two oscillating frequencies, it might be that there are multiple overlapping regions, for different modes of oscillation.  Chaotic systems, go figure, right? )

The better datasheets have done this for you already, showing a plot of stability versus Cload + ESR.  Voltage regulators, too.   Indeed, this is a good criteria to grade such parts on, for purposes of voltage supply and line driving -- if they're too shy to tell you the stability range (or too lazy to measure it), beware!

Tim

[Zero999]

I was going to say that it will become stable again with a massive load capacitance, >10µF, but it probably won't, because it's a rail-to-rail op-amp with common source amplifiers on the output stage.

A series resistor can be added to drive capacitive loads, but that increases the output impedance. I wonder if an inductor, with a parallel damping resistor, in series with the output could stabilise it, without increasing the output impedance too much?

[741]

That the output stage could oscillate "on its own" is something I had not considered. I guess, as 'danadak' says, I have to 'scope it to see what is going on.

Re: "2-D space of real numbers", makes me think it'd be nice to have some generic, ultra high frequency response gyrator to explore this with.

[741]

Out of interest, how does "massive load capacitance" solve the issue for non r2r devices? Why does the frequency of oscillation not just fall?

I think you are right though - I have seen this done.

[Zero999]

Out of interest, how does "massive load capacitance" solve the issue for non r2r devices? Why does the frequency of oscillation not just fall?

I think you are right though - I have seen this done.

A capacitor causes oscillation because it forms an extra phase shift, with the resistance of the op-amp's output stage. The additional phase shift adds to the op-amp's internal phase shift, causing positive feedback and oscillation at some frequency.

A very large capacitor won't allow any oscillation, because it will dominate the op-amp's output resistance, at the frequency it will normally oscillate at and the signal will be attenuated to the point where positive feedback and oscillation can no occur. Imagine, trying to make a phase shift oscillator, except with one of the capacitors is huge, compared to the others. At some point the gain required for oscillation becomes very high. Beyond what an op-amp can provide.

This doesn't help with a rail to rail op-amp, which has common source/emitter amplifiers on the output, which may oscillate on their own. This is a same reason why low drop-out regulators oscillate, when driving high-Q capacitive loads and traditional linear regulators, with an emitter follower on the output don't.

[Audioguru]

An opamp has an internal capacitor to rolloff high frequencies at -6dB per octave (producing 90 degrees phase shift) so that the gain is less than one at a frequency where additional capacitance in the opamp causes 180 degrees of phase shift and oscillation when negative feedback is added.

Adding a capacitor to ground at the output causes an additional phase shift causing oscillation at a frequency where the gain is more than one.
Your multimeter might be measuring the frequency of clipping harmonics.

[David Hess]

Out of interest, how does "massive load capacitance" solve the issue for non r2r devices? Why does the frequency of oscillation not just fall?

With enough capacitance, the gain will fall below one before the phase margin reaches zero.  This is just another form of dominant pole compensation.  This is easier to do with an aluminum electrolytic or solid/wet tantalum capacitor because the ESR contributes a zero for some phase lead.  Most voltage regulators rely on this to one extent or another and the ESR issue is why they have to be specifically designed to use a big ceramic capacitor at their output.

Doing this yields a horribly low bandwidth but in some applications this is acceptable.

This can also be done with many comparators to turn them into poor operational amplifiers but usually a series RC pair is used on the output for better control of the frequency compensation.  High power operational amplifiers and audio amplifiers often have a series RC pair at their output to *prevent* oscillation of the output stage.