does anyone know how to tell if an opamp is unity gain stable? My guess is to check out the open loop phase margin and make sure it's over... I dunno ... about 30 degrees... is this it? Don't most opamps have decent open loo p phase margin? Should I be saying open loop phase margin or does just say ing phase margin imply that you are talking about the open loop phase margi n of the opamp?
check out the open loop phase margin and make sure it's over... I dunno... about
30 degrees... is this it? Don't most opamps have decent open loop phase margin? Should I be saying open loop phase margin or does just saying phase margin imply that you are talking about the open loop phase margin of the opamp?
You need to check the phase margin with the OpAmp loaded with any expected capacitance. I like 45° or more, but it's all personal choice... how much overshoot can you tolerate?
(Some OpAmps are NOT unity-gain stable... check the data sheet.) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
is to check out the open loop phase margin and make sure it's over... I dun no... about 30 degrees... is this it? Don't most opamps have decent open l oop phase margin? Should I be saying open loop phase margin or does just s aying phase margin imply that you are talking about the open loop phase mar gin of the opamp?
"open loop phase margin" is an oxymoron, the very term phase margin is part and parcel with feedback, how you could possibly have feedback in a open l oop???
to check out the open loop phase margin and make sure it's over... I dunno... about 30 degrees... is this it? Don't most opamps have decent open loop phase margin? Should I be saying open loop phase margin or does just saying phase margin imply that you are talking about the open loop phase margin of the opamp?
parcel with feedback, how you could possibly have feedback in a open loop???
Nope. Phase margin is defined at the point where the _open_loop_gain_ passes thru 0dB.
However, it is usually simulated with the loop closed to ensure proper operating point conditions.
See...
formatting link
for analysis and a simulation tool based on Dr. R.D. Middlebrook's laboratory method. ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
Then why aren't you calling it "open loop phase margin?".
You're disagreeing with something that Bloggs didn't say. He didn't say that the phase margin isn't where the open loop gain passes through 0dB. He did say that the whole concept of phase margin (or gain margin, for that matter) only means something in the context of a closed loop -- which statement I agree with, wholeheartedly.
--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
Strictly speaking, the amplifier circuit will be unity gain stable if the open-loop gain has any phase margin at all. With 1 degree of phase margin you will still have an amplifier that is stable in the strict sense, but that is (a) going to show a lot of peaking and ringing, and (b) is going to be a hair's breadth away from bursting into song.
Look at what Jim said about the operation of the op-amp in question in the circuit for which it is designed: nearly all op-amps suffer from additional phase shift if they're driving a capacitive load, or if they have significant capacitance on the negative feedback. Thus you can have an op-amp that is stable at some gain in a circuit with purely resistive loads and sources, but which is unstable in _your_ circuit.
Jim quoted 45 degrees of margin as being OK. That implies 3dB of peaking in the frequency domain (I can't remember what percentage of overshoot in the time domain). You may need more phase margin if you care about the amplifier being flat all the way out. An amplifier that has 30 degrees of phase margin is going to be peaky, exhibit overshoot, and may have problems with manufacturing variations, temperature variations, and large- signal behavior (because phase shift often gets worse when the output is really banging around).
Note that gain margin matters, too.
--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
But the phase margin isn't equal to any phase you measure on the closed-loop system--especially if it's negative. It's inherently an open-loop quantity, which is why nobody says "open loop phase margin"--it would be redundant.
check out the open loop phase margin and make sure it's over... I dunno... about
30 degrees... is this it? Don't most opamps have decent open loop phase margin? Should I be saying open loop phase margin or does just saying phase margin imply that you are talking about the open loop phase margin of the opamp?
Unity-gain unstable is rare enough that data sheets always point it out. But sure, check the graphs for phase shift at the zero-gain frequency, and consider loading too.
--
John Larkin Highland Technology, Inc
jlarkin at highlandtechnology dot com
If you're not closing a loop around an element, it's phase shift does not affect stability. So "phase margin" doesn't mean anything (unless you have some other critical phase shift to worry about).
And while I freely admit that this _is_ picking nits, I measure the open- loop response of things in closed loop all the time. Googling on "frequency response measurement", "control system analyzer" or "transfer function measurement" should get some relevant hits.
Or read this (it probably needs updating):
formatting link
--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
Traditional op-amps with one dominant pole followed by parasitic poles at m uch higher frequencies can be adequately analyzed by gain/phase margin. Lat ely there has been increasing use of conditionally-stable op-amps (mostly o n-chip) that achieve incredible gain-bandwidth products. These op-amps alwa ys have at least one frequency where there is more than 0 db gain around th e loop with 0 degrees of phase shift at that frequency, and yet they do not oscillate when the loop is closed. Such op-amps have many low-frequency po les. The time-domain response is often pretty ugly, but usually you don't c are.
Clearly you are not at all familiar with the method.
No, you do not. You need to build a spare summing junction into the system at the input of the portion of the system you wish to measure, but once you do that you just inject a sine wave into the system at the summing junction, and measure the gain and phase shift from the summing junction output to the output of whatever part of the system you want to measure (which, if it is the entire open-loop response, is the system's input to the summing junction).
--
Tim Wescott
Control system and signal processing consulting
I'm unfamiliar with a whole lot of things. That's one of the main reasons that I like SED--I learn stuff.
So lay this out for me a bit further. Say you have a system with a phase margin of 2 degrees. You construct a point in the closed-loop system where you actually measure 2 degrees directly, without needing to calculate anything? How does that work exactly?
And this method works if the phase margin is -2 degrees?
It doesn't work on unstable systems. On the two-degree phase margin system you'd have to do a lot of hand-coaching (and hold your breath hoping that you don't hit a nonlinearity that takes the phase margin away).
But if you can keep the system stable, and keep the amplitudes high enough to adequately measure things away from the loop closure point, while keeping them low enough to not knock into the rails at the loop closure point with its 38dB of peaking (or whatever 2 degrees works out to), then it'll pretty much work.
Of course, in the normal course of events you'd notice the long ringing, and you'd adjust the loop to something more stable before you took your measurements.
You can, however, use it for an unstable _portion_ of the system, as long as the _overall_ loop is stable. Basically, as long as you have a reasonable facsimile of a sine wave running around the loop you have set up the conditions to do a single-frequency measurement of gain and phase shift. Do that over and over again for multiple frequencies (or use an instrument that does it for you) and you can get a Bode plot of the bit you're interested in.
Search on "control systems analyzer", or read the article on my web site
-- it should be clear how it all works.
--
Tim Wescott
Control system and signal processing consulting
Thanks. The point we were arguing about is whether, in order to obtain that 2 degree number, you measure it directly, or have to compute it, i.e. transform it back to the open-loop situation mathematically.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.