[HN Gopher] Deep dive: the instability of op-amps
___________________________________________________________________
Deep dive: the instability of op-amps
Author : lapnect
Score : 80 points
Date : 2024-10-26 14:38 UTC (8 hours ago)
(HTM) web link (lcamtuf.substack.com)
(TXT) w3m dump (lcamtuf.substack.com)
| dragontamer wrote:
| It's almost criminal that all of these bode plots are missing
| their phase diagrams.
|
| Phase diagrams + OpAmp phase shift specs / phase margin are what
| you need to predict instability.
|
| -------
|
| EDIT: IMO it's also a lot easier to explain in the frequency
| domain. At 180-degree phase shift, all your negative feedback
| turns into positive feedback, causing instability. You need your
| amplifier to stay as far away from 180-degree phase shift as
| possible.
|
| I get that what the author was trying to get to with the 'Tape
| Delay OpAmp' example. But it should be double downed upon and the
| starting point of the discussion rather than something brought up
| later IMO.
| FunkyDuckling wrote:
| I agree.
|
| Phase Margin (How far away you are from 180 Phase Shift) is a
| critical parameter used whenever designing any kind of feedback
| loop and testing for stability.
|
| This is very to measure at the 0dB gain he pointed out, but
| lacked the phase diagram to show this shift.
| arijo wrote:
| This is only true for LTI (linear time-invariant systems).
|
| Nonlinear systems responses to a sine signal are in general
| not just a change in phase and amplitude.
|
| It works if the perturbation stays small and within a
| linearised version of the dynamics.
| dsv3099i wrote:
| I agree the overall math is easier in the frequency domain,
| especially because you don't know which frequencies are
| problematic so best to look at all of them, but I think the
| concept is best explained at first, in the time domain.
|
| Here's my attempt in a couple of sentences.
|
| It takes time for the signal to propagate from input to
| output in any real circuit. If that time is a substantial
| fraction of the period under consideration then the input of
| the amplifier, which includes the feedback signal, cannot
| effect the output before it has moved. And if the delay
| through the amplifier is just wrong relative to the signal
| period one can end up in a dog chasing its own tail situation
| and the output oscillates.
|
| The rest is just math. :)
|
| P.S. this explanation also explains why we use phase and not
| seconds to measure the delay of the circuit. Because
| everything is relative to the input signal period and if we
| use phase we get that for free. No extra divide.
| ChuckMcM wrote:
| I appreciate what you're saying. That said, not everyone learns
| the same way and for me, his explanations have always been
| clear and insightful. My theory is that different brain
| architectures ingest information in different ways (this is
| actually studied but not conclusively proven AFAICT) and that
| the language of exposition has a sort of 'impedance match' with
| brain architecture. So sometimes you can say something and the
| person hearing/reading it will just "get it" right away, and
| sometimes they will look at you like "that didn't help at all."
|
| That said, I agree that if you're used to thinking about things
| in the frequency domain it makes sense to explain it in those
| terms. Myself, as a young EE in college found thinking about
| things in the frequency domain to be useless, in part because I
| didn't understand the math, and in part because I didn't really
| understand sinusoidal waveforms. It wasn't until I started
| diving into SDRs and really unpacking the FFT and how it worked
| and why did I manage to connect a lot of dots that
| retroactively gave me a better insight into what my control
| systems professor was trying to teach me back in the day.
| buildsjets wrote:
| But the ringing, sustained oscillations, and excessive gain are
| sometimes desired characteristics of a particular op-amp
| implementation.
|
| https://youtu.be/SrS9EtfcANg?si=MwbtbuPWu85Tbjzq
| analog31 wrote:
| Sure, the Wien bridge oscillator. But then, oddly enough, even
| sustained oscillations need to be controlled in their
| amplitude, and the Wien Bridge has a secondary feedback loop
| for that purpose -- the temperature dependent resistance of a
| light bulb.
| ericwood wrote:
| Those op amp characteristics aren't really coming into play for
| that delay sound, it's just vanilla digital delay. Most of the
| "vibe" of the DD-3 is coming from the companding and filtering
| scheme it uses to work around the limitations of the digital
| pieces.
|
| You will hear the effects of this in many hard clipping
| distortion circuits, though, where the amplifier gain factor
| will far exceed the voltage rails and be pushed into undefined
| clipping territory. Behaviors in this range can be an important
| part of the sound, e.g. the Proco Rat and the infamous LM308 op
| amp with its slow slew rate. Some like the TL072 exhibit a
| really nasty phase inversion that results in a pretty horrific
| (usually undesired) distortion.
|
| It's a balancing act, though; search "op amp motorboating" in
| any DIY stompbox forum and you'll find thread after thread of
| people trying to keep op amp gain stages from oscillating. I
| know more than a few noisier artists who enjoy when designs can
| be tortured into doing that, though :)
| 082349872349872 wrote:
| to what degree does the Gartner hype cycle resemble the Gibbs
| phenomenon?
| marcosdumay wrote:
| To the degree that society is a complex system with feedback.
|
| But notice that the Gartner hype cycle is full of unjustifiable
| hidden assumptions (like the fact that the thing being hyped is
| useful at all) so it has no predictive power. It only happens
| that some times people act like that.
|
| Also, there's no guarantee that the society's response to a
| change will be stable.
| WJW wrote:
| One of the most "useful" insights I ever had is that
|
| 0. Any transfer function higher than 2nd order overshoots.
|
| 1. Society is super complex, has a ton of internal feedback
| loops and would easily be higher than 2nd order.
|
| 2. So of course it overshoots all the time.
|
| I haven't been able to actually "use" it for anything but it
| does describe a lot of what we see every day in society.
| RossBencina wrote:
| > Any transfer function higher than 2nd order overshoots.
|
| That's false. Any LTI system higher than first order
| _might_ overshoot. But it 's easy to design high-order
| systems that don't overshoot. Consider for example a
| cascade of first-order sections. Related terms: Bessel
| filter, complex vs real poles, overdamped system.
| sobriquet9 wrote:
| The tape delay methaphor confused me. Tape recorders do not
| record DC or other frequencies much lower than, say, 20 Hz. So
| that circuit would run into one of the rails just as quickly as
| the previous cirquit without DC feedback.
| mrob wrote:
| It must be an FM encoded tape, which can record DC. There were
| upgraded versions of both VHS and Betamax with FM audio support
| (although I expect in practice the inputs were AC coupled).
| Buttons840 wrote:
| This is tangential, but I took an EE class at community college
| and the very first thing they did was start teaching op-amps. I
| don't remember ever getting any insight from working with them,
| only that we had to follow instructions and build one in the lab.
|
| When I see people asking questions about op-amps and doing "deep
| dives" into op-amps, I'm left wondering what's so deep about
| these things we do in week 2 of EE 101.
|
| I've forgotten almost everything from that class though, so maybe
| it was just a bad class? I switched majors and never took another
| EE class.
| fourier54 wrote:
| If your only interaction was "follow instructions and build one
| in the lab", doesn't that tell you exactly there is something
| deep you didn't understand at all?
| szundi wrote:
| These are pesky little things
| analog31 wrote:
| I'd say that nothing can be covered deeply in an introductory
| survey class. If it's being taught at the 101 level, the
| students don't yet have the math to scratch anything beneath
| the surface. And one of the points of op amps, if not the main
| point, is the correspondence between their mathematical
| representation, and their real world behavior.
|
| There are entire books about op amps and their uses. They're a
| cornerstone of analog design.
| Buttons840 wrote:
| Now that you mention it, I remember the point was the
| difference between the theoretical math and the actual
| behavior.
| WJW wrote:
| I can definitely see how a more advanced student can get a
| ton of valuable insights about the "abstraction leaks" of
| opamps, and I would even submit that opamps could be a
| particularly fertile place of such leaks (for our more
| software minded brethren: opamps are like databases: they
| are perfect until you start to push their boundaries and
| then you very quickly start to see just how imperfect they
| are), but I doubt you can teach a student enough about the
| theoretical math to get a useful intuition about the
| behavior of opamps in under 2 weeks.
| michaelt wrote:
| _> I 'm left wondering what's so deep about these things we do
| in week 2 of EE 101._
|
| Part core concept, part outdated nonsense that's taught due to
| tradition.
|
| If you've got components that can halve a voltage, then by
| putting that in a feedback loop you can double a voltage.
|
| And you can make an accurate amplifier even if some of your
| components - like the op-amp's gain - are inaccurate. So long
| as your voltage-halving components are accurate, your voltage
| doubler will be accurate whether your op-amp's gain is 10000x
| or 20000x
|
| You can chuck other components into the feedback loop too -
| want a higher current output from your voltage doubler? Have
| the op-amp control a high-power transistor.
|
| You know how a feedback loop can turn a voltage-halver into a
| voltage-doubler? It can invert other mathematical functions
| too. Put a capacitor into your op-amp circuit and you can
| integrator or differentiate. There are even op-amp circuits for
| summing inputs!
|
| You now understand feedback loops, precise gains, power output
| stages, integration and differentiation. You can now make a PID
| controller - a key concept in control theory! Just what you
| need to position control a robot's joints.
|
| Except making PID controllers out of op-amps is obsolete;
| they're all done in software these days.
| Metacelsus wrote:
| Yeah I was very confused by them in high school, I just
| internalized "op amps are weird" and then became a biologist.
| This article was a great explainer. Similar feedback loops
| happen in biology!
| masspro wrote:
| It is a headscratchingly bad idea to put op-amps in week 2 of
| course #1. I can't even remember for sure if we did them in the
| first course, but if so, it was at the very end, _after_ you
| already know how to do algebraic working-out of values in a
| circuit. From there, they give a couple algebraic rules to
| figure out what an op-amp circuit does. And a key point that is
| usually glossed over is: op-amps are basically useless when not
| in a feedback configuration, and some of the analysis rules are
| based on already assuming the op-amp is in a feedback
| configuration.
| neuralRiot wrote:
| Ahh the good times I remember designing and building some
| configurable audio crossovers with "memories" so they could be
| adjusted and stored. This was of course before DSPs were even a
| thing. We gave up because the only solutions were inserting
| mosfets on the opamp loop or using transconductance opamps. Both
| solutions were terrible in terms of audio quality, we decided for
| "cartridges" that were the whole x-over stage.
| Metacelsus wrote:
| Until reading this article I never really understood what an op-
| amp actually did (I remember coming across them as high schooler
| and being confused).
| davidhyde wrote:
| What an absolute goldmine of a website. So many varied topics.
| Just amazing.
| RossBencina wrote:
| If you want an actual deep dive watch James K. Roberge's OCW
| course and/or get the book "Operational Amplifiers Theory and
| Practice".
___________________________________________________________________
(page generated 2024-10-26 23:00 UTC)