|
[Part 1 looks at the overall supply impedance seen by a component when the decoupling caps, the voltage regulator and the board traces are taken into account. Part 2 looks at how a real-world supply impedance - comprising decoupling capacitors, regulator output impedance, and PCB traces - responds to a small test current step.]
Previously on "Yet More..." we looked at the impedance of a typical regulator and decoupling capacitor combination, and showed what happened when you 'shock' it with some dynamic current consumption. Now we're going to look at whether we should be concerned about the wobbling supply voltage that analog components will experience when connected. We'll focus on a staple ingredient of the analog designer's toolbox: the op amp.
"Bah, Humbug!", I hear from somewhere in the audience. "Modern precision op amps are so fantastic that their enormous power supply rejection will extinguish any possible effect that varying supplies could have on a circuit! My dear boy, decoupling capacitors are only there to quell the evil spirits of oscillation, they don't actually affect the performance of the circuit! Simulate the power supply rail? Haven't you got anything better to do with that new-fangled slide rule of yours?"
Well, we'll see. By now, most analog engineers are (or should be) wise to the myth that 'op amps have almost infinite gain and everything is sorted out by the negative feedback loop, whose properties completely dominate the closed-loop performance'. However, despite huge advertised 'open loop gains', it is the destiny of any op amp's gain curve to trend downwards at roughly 6dB per octave until at some frequency (the unity-gain frequency, numerically equal to gain-bandwidth product in simple cases, though the terms don't mean the same thing), there's none left; it's just unity, or 0dB.
As frequency rises, falling loop gain means a falling amount of feedback to correct errors inside the loop. What's more, some amplifiers don't even have very high low-frequency open-loop gain either. This is particularly true either of very fast amplifiers or very cheap amplifiers.
OK, so we may not have that much loop gain, but why does it matter " surely a competent designer can suppress the sensitivity to supply rail variations just through architecture choices. Not so! Here's a blunt assertion you may not have seen before: it is not possible to build a conventional op amp (a pair of differential inputs, two power pins, one single-ended output pin) that is insensitive to voltage variations on one or other of its power pins. The device converts the voltage difference between the signal inputs into a single-ended output " but that output has to be referenced to something other than ground, because the op amp doesn't have a ground pin! Depending on the design, the output voltage will be referenced either to one or other supply pin, or to a voltage somewhere along a potential divider between those two pins (think of it as produced by the ratio of output conductances of the current paths in the main gain stage to the two supply pins).
So, you can't build a precision op amp that behaves as if it had its own private voltage regulators inside to prevent the output voltage from moving when the supplies do. That kind of device would require a ground connection to reference those private regulators to, which our standard pinout does not give us. And indeed this is borne out by the PSRR plots in amplifier data sheets, which usually show curves with the same general shape as the open-loop gain. Figure 3.1 shows the open loop gain of LT's LT1723 and figure 3.2 shows the PSRR of the amplifier when connected as a unity-gain buffer. The curves for rejection from the +ve and -ve supplies deviate at low frequencies but are very similar above 1MHz.
Figure 3.1: the open-loop gain response of the LT1723 to signals applied at its input pins.
Figure 3.2: the closed-loop gain response of the LT1723 to signals applied at its power pins.
The question of common-mode performance effects on the input stage is a complicating factor and I want to skirt around that for this article. Of course the input stage is attached to the power supplies, and the apparent input voltage at the device can be distorted by the input stage's reaction to supply variations. But, unlike the power supply rejection problem, this one is amenable to design, layout and processing fixes. Also, the common-mode rejection capabilities of current-feedback amplifiers often fall short of good voltage-mode op amps. For clarity of purpose, I'll use voltage-feedback op amps in an inverting amplifier configuration as my circuit under test. This means that the input stage only has to reject the (hopefully small) supply variations, and not any applied input signal as well.
|
