14
\$\begingroup\$

Since I've spent a good portion of my career trying to get opamp's rails as steady as possible at their intended voltage, I haven't really spent any time thinking about what would happen if the rails are moving away from a fixed value. Since I have only briefly studied the internal workings of op amps, I'm not so sure I could come up with a definite answer.

So, what does happen to the signal if the rails are moving? (lets just say there moving slowly, like less than 5Hz, maybe a 1V shift from time to time) Is it more than just clipping at different levels?

\$\endgroup\$
  • \$\begingroup\$ Have a look at opamp bootstrapping where the rails are modulated by the output signal to allow wider voltage swings \$\endgroup\$ – Colin Mar 7 '17 at 9:29
27
\$\begingroup\$

In theory, the OpAmp should perform well no matter what the supply is doing.

As we leave the theoretical model of an OpAmp (remember there aren't even supply pins on the basic symbol, just IN+, IN- and OUT), we have to consider more and more details brought in by the real circuit.

enter image description here

Many will of course be obvious to you, but trust me - we'll eventually get to an answer.

First, the output can never exceed the voltage supplied to the Amp.

Then, the performace gets worse when the output is trying to push or pull the voltage close to the rails. This will, of course, depend heavily on the design of the OpAmp - and Rail-to-Rail amps promise to give you all the available voltage at the output.

As long as we look at a DC-supplied OpAmp, any signal well within the specification of the maximum output swing will work, and you can supply the OpAmp with any positive and negative voltages allowed by the data sheet (with regard to each other and to ground, but note that the OpAmp has no way of knowing where ground actually is; supplying +3 V and -7 V is no problem at all - and your amp will try to remain working within this range of 10 V).

Internal current sources, differential stages and output drivers are designed such that the OpAmp cancels out any variations on the supply rails as quickly as it possibly can.

Only if the variations on the supply rails change quickly enough, you will start to notice an effect. Usually, this sets in somewhere between some 100 Hz to some 10 kHz.

And the best part: It's specified in the data sheet; look for PSRR (Power Supply Rejection Ratio).

The value is usually very high for DC to low frequencies (60...120 dB) and starts to degrade with what looks like a simple low-pass characteristic above a certain point. Note that we're talking about rejection, so it's actually a high-pass even though the slope goes down on the diagram:

enter image description here

Note that the text in the image says: ±15 V - so what is actually done to the OpAmp's supply pins?

As with any good data sheet specification, there's also a test circuit that tells you how it's measured:

enter image description here

This also explains why there are two lines in the diagram (-PSR and +PSR). The OpAmp's internal current sources, for example, are sometimes feeding their loads from the positive supply, sometimes into the negative supply, and the internal design is not absolutely symmetrical.

Take the good ol' 741 as an example:

enter image description here

Only the output stage at the very right is symmetrical, everything else is not. More advanced parts will still follow this basic principle to a certain degree.

In a nutshell: For DC and low frequencies, look at the DC specifications (rail-to-rail with what limitations for gain and distortion?). For higher frequencies, look at the PSRR. If you apply a step to the supply volatge, you have a mixture, because a step is composed of some high-frequency part besides the obvious jump from one DC level to another DC level, resulting in a disturbance at the output caused by whatever higher-frequency part of the step that can't be rejected by the OpAmp.

What I haven't covered here might be answered in Analog Devices' tutorial MT-043. This is also where I've taken the images from (except for the 741 circuit).

\$\endgroup\$
  • \$\begingroup\$ Excellent answer! To add a personal experience, I'm currently working on equipment where a power op-amp on an actuator driver was getting a 0.1Vrms ripple on the -45V supply. For most situations this would be no big deal, but we need position noise down to something like 5ppm. With the op-amp being less good at rejecting noise on the negative supply, this was something we needed to take seriously. \$\endgroup\$ – Graham Mar 7 '17 at 11:54
  • \$\begingroup\$ @Graham Looks like the math works out: 5 ppm is equal to 106 dB (if I haven't messed things up?!), so this may indeed be beyond the PSRR of your particular OpAmp, depending on what the 5 ppm refer to in your example (full-scale?), and considering that PSRR is often calculated "referred to input" (RTI), so any gain your OpAmp is configured in will multiply the noise caused by ripple on the supply rails. \$\endgroup\$ – zebonaut Mar 7 '17 at 12:23
  • \$\begingroup\$ This is awsome, yeah I know most of this stuff but I thought I'd ask the question for everyone else. Its also good to see how others view PSRR \$\endgroup\$ – Voltage Spike Mar 7 '17 at 16:59
4
\$\begingroup\$

Yes, there are AC effects. The op-amp datasheet should specify a Power Supply Rejection Ratio that gives you the maximum effect that a change in power supply will have on the output. It's a pretty high figure - even the ancient 741 has a typical figure in the 90dB range - but it can be significant if the change in output then produces further changes in power supply voltage and hence creates a feedback loop that could lead to oscillations.

Obviously, as you realise, this is in addition to any direct effects such as relying on rail-to-rail operation of inputs and outputs.

\$\endgroup\$
  • \$\begingroup\$ Yeah, I'm aware of PSRR, but what about slow changes? \$\endgroup\$ – Voltage Spike Mar 7 '17 at 7:09
  • 1
    \$\begingroup\$ Same answer, it's still AC! \$\endgroup\$ – Finbarr Mar 7 '17 at 7:11
1
\$\begingroup\$

There is an accepted answer, but I wanted to mention a specific example: audio power amps.

These are usually powered from unregulated rails. Expect several volts ripple at rectified AC mains frequency, often more depending on current demands. When the rectifying diodes are not conducting, which is most of the time, the supply voltage is decreasing according to output current divided by the value of the big supply capacitor.

Also, the rail voltage will vary depending on the signal's amplitude. When listening, the louder parts will draw more current, lowering rail voltage. Quiet parts will not. Thus rail voltage wiggles in the 0.1-2 Hz region in addition to the rectified mains frequency.

These amps are usually implemented as discrete opamps, which enables several tricks to increase PSRR. A discrete opamp has a GND terminal, so the internal nodes most sensitive to the power supply can be bypassed to ground by means of a cheap capacitor. The compensation capacitor is a major source of bad PSRR in opamps, since it has to be referenced to one of the supplies. In a discrete opamp, this can be mitigated.

The result is that you can get huge ripple on the rails without any issue. In fact, power amps with regulated rails are very exotic, only encountered in megabuck audiophile gear, and realistically, a waste of money.

So here's a real life example ;)

what does happen to the signal if the rails are moving? (lets just say there moving slowly, like less than 5Hz, maybe a 1V shift from time to time) Is it more than just clipping at different levels?

LF PSRR is huge, so nothing happens.

Opamps have low HF PSRR, and thus dislike bad decoupling which creates HF ringing on supplies, or other sources of HF noise like badly filtered switching regulators. LF supply voltage variation should not matter at all. Perhaps offset voltage could drift due to thermal effects, but this should be tiny.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.