Page 1 of 8
Yet More On Decoupling, Part 5 – When Harry Regulator Met Sally Op-Amp
Kendall Castor-Perry
This article was published on EDN: http://www.edn.com/design/powermanagement/4415318/Why-bypass-caps-make-a-difference---Part-5--Supply-impedanceand-op-amp-interaction
After the first four parts of “Yet more...” you are, I hope, curious about what happens
when everything is connected together. We’ve seen that some fearsome ringing can
appear on our simulated supply if we chose the values of the decoupling capacitance and
the ESR of the regulator’s output capacitor poorly. We’ve seen that it’s inevitable that
high-frequency variations of the power supply line will punch through to the amplifier’s
output pin, where they will at the very least add on top of what is supposed to be there
anyway.
So, without further ado, let’s get started on Part 5. The test fixture we’re using is the
same as shown in figure 3.10 of part 3. We’re running the amplifier at a closed-loop gain
of -1; the output of the amplifier is loaded so that later, when we drive it, we’ll take the
right amount of current from the rails (that’s the theory, anyway).
We can now plot the rejection transresistance of the system. It’s the output voltage of
the op-amp divided by the input test current injected into our composite power supply.
It’s basically the resultant voltage on the power supply multiplied by the power supply
rejection of the amplifier at the gain used, which as you’ll remember just looks like a line
that slopes down at 6 dB per falling-frequency octave from some frequency. So all the
real action is up at high frequencies, giving a rich tapestry of interaction between our
supply impedance resonances and the high frequency behaviour of the amplifier. Figures
5.1 and 5.2 show, on the red traces, the rejection transimpedance of the 200 MHz GBW
LT1723 and the 30 MHz GBW LT1630. The latter has so little high frequency rejection
that it faithfully follows the supply peaking above about 4 MHz. You can see that it has
aspirations to be more of a ‘precision’ amplifier, though, and its greater low frequency
gain pushes the low frequency power supply gain down, giving improved rejection
transimpedance below 1 kHz.
Page 2 of 8
3 0d B
V (p l u s )
V (m i n u s )
V (n 0 0 8 )
2 0d B
1 0d B
0 dB
-1 0 d B
-2 0 d B
-3 0 d B
-4 0 d B
-5 0 d B
-6 0 d B
-7 0 d B
-8 0 d B
-9 0 d B
-1 0 0 d B
1 00 H z
1KHz
1 0KH z
1 00 KH z
1M Hz
1 0M H z
1 00 M H z
Figure 5.1: rejection transimpedance of the LT1723 model. Basically, Vout versus
excitation current.
3 0d B
V (p l u s )
V (m i n u s )
V (n 0 0 8 )
2 0d B
1 0d B
0 dB
-1 0 d B
-2 0 d B
-3 0 d B
-4 0 d B
-5 0 d B
-6 0 d B
-7 0 d B
-8 0 d B
-9 0 d B
-1 0 0 d B
1 00 H z
1KHz
1 0KH z
1 00 KH z
1M Hz
1 0M H z
1 00 M H z
Figure 5.2: rejection transimpedance of the much lower bandwidth LT LT1630 model
(better below 1 kHz though).
Page 3 of 8
Let’s analyse the result in the time domain. Using the same, appropriately phased 10
mA, 100 kHz test currents as deployed in part 2, we see the expected signals on the
supply lines and a smaller signal on the output pin of the op-amp. Figures 5.3 and 5.4
show the results, with the worst-case 22 nF capacitor, for the high and low GBW
amplifiers; the advantages of a high GBW are quite obvious.
Now, by the way, in case you think these traces are plotted with 22 nF just for effect
because they give the most dramatic demonstration, remember back to the very first part,
and the voltage-dependency discussion. If you were remiss enough to use a 100 nF 16 V
Z5U capacitor for decoupling your 5 V rails, the actual value of that capacitor under bias
would be close to 22 nF. So this isn’t just whacky wrong-value simulation, it shows what
could really happen if you take your eye off the component ball!
V (p l u s ) -5
2 5m V
V (n 0 0 8 )
2 0m V
1 5m V
1 0m V
5m V
0m V
-5 m V
-1 0 m V
-1 5 m V
0 µs
5 µs
1 0µs
1 5µs
2 0µs
2 5µs
3 0µs
3 5µs
4 0µs
4 5µs
5 0µs
Figure 5.3: the positive supply and the output of the LT LT1723 model (lower, blue, with
22 nF decoupler).
Page 4 of 8
V (p l u s ) -5
2 5m V
V (n 0 0 8 )
2 0m V
1 5m V
1 0m V
5m V
0m V
-5 m V
-1 0 m V
-1 5 m V
0 µs
5 µs
1 0µs
1 5µs
2 0µs
2 5µs
3 0µs
3 5µs
4 0µs
4 5µs
5 0µs
Figure 5.4: the positive supply and the output of the LT LT1630 model (lower, blue, with
22 nF decoupler.
When we zoom into the output signal in the time domain (figures 5.5 and 5.6, with the
traces for 22 nF to 470 nF decoupling capacitors now spread apart for clarity) we indeed
see a high-pass filtered version of this ringing there on the output pin. The fundamental
frequency of our excitation doesn’t show up visibly. This points to the high frequency
behaviours of the decoupling network being most important when we consider the opamp output – no surprises there, say all you analog guys out there.
Page 5 of 8
1 0m V
v (n 0 0 8 )@ 1
v (n 0 0 8 )@ 2 - 0 . 0 0 5
v (n 0 0 8 )@ 3 - 0 . 0 1
v (n 0 0 8 )@ 4 - 0 . 0 1 5
2 5µs
3 5µs
v (n 0 0 8 )@ 5 - 0 . 0 2
5m V
0m V
-5 m V
-1 0 m V
-1 5 m V
-2 0 m V
-2 5 m V
-3 0 m V
0 µs
5 µs
1 0µs
1 5µs
2 0µs
3 0µs
4 0µs
4 5µs
5 0µs
Figure 5.5: LT LT1723 model output ringing changing as the decoupler is stepped from
22 nF to 470 nF.
1 0m V
V (o u t )@ 1
V (o u t )@ 2 - 0 . 0 0 5
V (o u t )@ 3 - 0 . 0 1
V (o u t )@ 4 - 0 . 0 1 5
V (o u t )@ 5 - 0 . 0 2
5m V
0m V
-5 m V
-1 0 m V
-1 5 m V
-2 0 m V
-2 5 m V
-3 0 m V
0 µs
5 µs
1 0µs
1 5µs
2 0µs
2 5µs
3 0µs
3 5µs
4 0µs
4 5µs
Figure 5.6: same as figure 5.3 but with the LT1630, showing how low GBW makes it
worse.
5 0µs
Page 6 of 8
Since we saw that the supply ringing was much reduced by using a tantalum capacitor on
the regulator, we shouldn’t be surprised if the output voltage of the op-amp is also much
cleaner if we try that. Figure 5.7 shows what happens to the output of our lowerbandwidth LT1630 if we just change over to our standard tantalum output caps on the
regulators. The high frequency thrashing is almost eliminated, and what’s more, there’s
not a lot of difference in the peak-to-peak value of what’s left when we change the
decoupling capacitor value.
The output of the LT1723 on the same scale now shows so little detail that we need to
zoom much further in. Figure 5.8 is starting to look interesting! Here we at last start to
see the detail changes to the dynamics of the amplifier that might occur with changes in
decoupling capacitor value. Who was that guy who said they couldn’t make any
difference to the performance?
1 0m V
V (o u t )@ 1
V (o u t )@ 2 - 0 . 0 0 5
V (o u t )@ 3 - 0 . 0 1
V (o u t )@ 4 - 0 . 0 1 5
V (o u t )@ 5 - 0 . 0 2
5m V
0m V
-5 m V
-1 0 m V
-1 5 m V
-2 0 m V
-2 5 m V
-3 0 m V
0 µs
5 µs
1 0µs
1 5µs
2 0µs
2 5µs
3 0µs
3 5µs
4 0µs
4 5µs
Figure 5.7: LT LT1630 model’s output, just changing the regulator capacitors to
tantalum.
5 0µs
Page 7 of 8
V (n 0 0 8 )
2 .0 m V
1 .6 m V
1 .2 m V
0 .8 m V
0 .4 m V
0 .0 m V
-0 . 4 m V
-0 . 8 m V
-1 . 2 m V
-1 . 6 m V
-2 . 0 m V
1 9.80 µ s
1 9.85 µ s
1 9.90 µ s
1 9.95 µ s
2 0.00 µ s
2 0.05 µ s
2 0.10 µ s
2 0.15 µ s
2 0.20 µ s
Figure 5.8: zooming in on the LT1723 output with tantalum output caps on the
regulators.
We are almost there. At the moment we are still applying an external excitation stimulus
to the supply rails. What would be really interesting is to see how the amplifier might
interfere with its own performance, by making demands on the supply rails which turn
into accuracy-robbing output effects.
To do this we’ll feed a signal directly into the amplifier. And sadly the disappointment
with the results of the manufacturers’ models is not over yet. To be continued...
Take-aways from this part:


The choice of decoupling and regulator output capacitors will affect the
accuracy of your op-amp circuits, and you can model these effects.
Transient signals on the supply pins of an op-amp will find their way to the
output, in amounts that might be critical in a high accuracy system
2013 Postscript
This was the part where we started to see things happening at the output of the amplifier
(OK, the amplifier’s simulation model) when we did something to the supply rails.
Strangely enough, I got very little pushback on this, not even to the specially-invented
phrase rejection transimpedance.
The incredible voltage coefficient of “high K” ceramic capacitors came as a surprise to
one or two people. To this day I see people using 0.1 uF Z5U capacitors on a 5 V supply
Page 8 of 8
where the IC manufacturer has recommended a capacitor value of 0.1 uF – not a
capacitor marking of 0.1 uF!
The lossiness of chemical capacitors really starts to show its value here. If you can’t use
a chemical (i.e. electrolytic) capacitor, make sure you build in some low-value resistors in
series with some of those bypass capacitors. It goes against the grain to add impedance
when you feel that lower is better, but the benefits are unmistakable.
Apart from that, not much happened in this part. It was the lull before the storm…