Power Supply Decoupling

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
What is the best three capacitor combination for power supply decoupling on amp PCBs? I know to use three caps (one large, one medium, one small), but I get conflicting results as to what kind the medium and small ones should be. This is what I can gather so far:

Large: electrolytic (10uF)
Medium: film or poly (but what kind of film or poly? 1uF)
Small: ceramic or mica (100pF)

What have you all found that works best (and is easy on the wallet)?
 
Hi,
my view is use the biggest polyfilm you can afford and keep this and the ceramic right at the last connection point before leaving the board.

Remember to add mains caps (X & Y rated) and switch suppression.

You might also want to experiment with a snubber on the output and on board the PCB.
 
Paralelling dissimilar capacitors directly does not produce a lower impedance PSU, since each capacitor resonates with the rest producing peaks in the impedance verses frequency characteristic of the PSU that may reach 10 ohms or even 100 ohms and may be an octave wide.

A single medium sized (modern) electrolytic capacitor (around 100uF) will do a better job and produce a nice flat and resistive low PSU impedance up to at least 1Mhz. On the other hand, PCB track inductance of the typical dumb amplifier layout (rails in the sides and ground in the center of a wide PCB) is far more inductive than any electrolytic capacitor, so the only way to reduce PSU impedance above 1Mhz is to learn to lay out PCBs properly (most audio designs are terrible).
 
Eva said:
Paralelling dissimilar capacitors directly does not produce a lower impedance PSU, since each capacitor resonates with the rest producing peaks in the impedance verses frequency characteristic of the PSU that may reach 10 ohms or even 100 ohms and may be an octave wide.

A single medium sized (modern) electrolytic capacitor (around 100uF) will do a better job and produce a nice flat and resistive low PSU impedance up to at least 1Mhz. On the other hand, PCB track inductance of the typical dumb amplifier layout (rails in the sides and ground in the center of a wide PCB) is far more inductive than any electrolytic capacitor, so the only way to reduce PSU impedance above 1Mhz is to learn to lay out PCBs properly (most audio designs are terrible).

FWIW, Analog Devices recommends 100nF ceramic in parallel with 1 tantalum for low-noise measurement applications.

The real world problem with all the parasitics is illustrated below:
An externally hosted image should be here but it was not working when we last tested it.


here's one layout which they recommend.
An externally hosted image should be here but it was not working when we last tested it.


The entire article is referenced here:
http://www.analog.com/library/analogDialogue/archives/39-09/layout.html
 
This article is useles for our purpose, as it ignores completely mutual capacitor interaction and does not even model capacitor inductance in the drawings!!

An externally hosted image should be here but it was not working when we last tested it.


First, this graph shows that a 2.2uF tantalum has very good high frequency performance, as it does not show much higher impedance than the ceramics at higher frequencies (it may be an artifact of the measurement setup, though).

Also, a resonance peak will be formed when paralelling the 2.2uF and the 100nF, it will be located where both impedance graphs cross, at 10Mhz, and will probably produce a peak of 1 ohm or more, thus ruining the small advantage that the 100nF could have provided.

In other words, the PSU impedance will be actually *lower* when the 100nF paralell capacitor is not used. Note that the 2.2uF tantalum provides by far the lowest impedance in the 1Mhz to 10Mhz range, that is critical for audio amplifier stability since it's where most circuits transition from feedback to no-feedback, while the rest of the capacitors show much higher impedance here.
 
Eva said:
This article is useles for our purpose, as it ignores completely mutual capacitor interaction and does not even model capacitor inductance in the drawings!!

An externally hosted image should be here but it was not working when we last tested it.


First, this graph shows that a 2.2uF tantalum has very good high frequency performance, as it does not show much higher impedance than the ceramics at higher frequencies (it may be an artifact of the measurement setup, though).

Also, a resonance peak will be formed when paralelling the 2.2uF and the 100nF, it will be located where both impedance graphs cross, at 10Mhz, and will probably produce a peak of 1 ohm or more, thus ruining the small advantage that the 100nF could have provided.

In other words, the PSU impedance will be actually *lower* when the 100nF paralell capacitor is not used. Note that the 2.2uF tantalum provides by far the lowest impedance in the 1Mhz to 10Mhz range, that is critical for audio amplifier stability since it's where most circuits transition from feedback to no-feedback, while the rest of the capacitors show much higher impedance here.
What makes you think that 2.2uF graph is for tantalum? Looks like MLCC surface mount ceramic cap to me.

2.2uF tantalum with 10mOhms ESR? No Have, Sir.

What comes to paralleing lets say 10uF tantalum+ 100nF ceramic, there is usually NO resonance as tantals have enough high ESR to damp any problems like that.

Paralleing 100nF ceramic with 10uF foil or ceramic would be a bad idea just like Eva said.
 
Indeed you will find a lot of people recommending paralell capacitors, but they will do it without analysing capacitor interaction (most of them without ever knowing these effects).

I have just taken the following capture from one of my circuits. It's low power logic and it's made with a 7555 CMOS timer and some 4000 CMOS gates, all them together can only draw approx. 50mA transients at best from the PSU (so that's nothing in comparison with what a DSP can do to a resonant PSU).

As supply decoupling I have used a 22uF 35V standard electrolytic (high ESR so less likely to ring) and a 100nF multilayer ceramic paralelled quite closely. The measurement is done directly across the legs of the 100nF ceramic with a common mode filter inserted in the oscilloscope probe to remove common mode noise.

See how happily the two paralelled capacitors are singing their 2Mhz song after each logic transient :D:D:D:D:D

An externally hosted image should be here but it was not working when we last tested it.


BTW: The ringing disappears if either capacitor is removed.

All that discussion is just a matter of lazyness, it's very easy to recommed stacking dissimilar capacitors just because you have read about it in some paper or forum, but whithout having ever investigated the resulting effects on supply impedance.
 
In the previous capture, the capacitors were placed approx 5cm away in the PCB and the total inductance figure was PCB inductance plus capacitor inductance. PCB inductance must not ve overlooked when paralelling

This is what happens when the legs of both capacitors are soldered together just near the bodies, the inductance reduction is just enough to tame the oscillation down to a single bump after the transient that triggers it.
An externally hosted image should be here but it was not working when we last tested it.



This is what happens when the ceramic is removed leaving the electrolytic alone... I wonder if the ceramic was doing some useful job...

An externally hosted image should be here but it was not working when we last tested it.



This is what happens when the electrolytic is removed and the ceramic is left alone. It resonates like crazy with the 30cm of twisted pair wire that go from the PCB to my bench PSU and with the 1000uF output capacitor of the PSU!!!
An externally hosted image should be here but it was not working when we last tested it.



And this is what happens when no supply bypass at all is employed. As it can be seen, the 22uF electrolytic alone was doing almost all the useful work. Note that here I had to change the scale from the 5mV/div of all previous captures to 100mV/div!!
An externally hosted image should be here but it was not working when we last tested it.



BTW: Avoid 100nF ceramics except when they are soldered directly across the legs of a two orders of magnitude bigger high-ESR electrolytic.
 
Eva said:
In the previous capture, the capacitors were placed approx 5cm away in the PCB and the total inductance figure was PCB inductance plus capacitor inductance. PCB inductance must not ve overlooked when paralelling
.

5cm apart... It makes sense now, 5cm distance on PCB with unknown loop area could make around that 100nH what I was questimating.

Note that Analog's example layout uses groundplane underside of board and power traces upside, resulting very low PCB stray inductances. Plus caps are very close, maybe 5mm apart from each other.
 
This is a 10uf tantalum electrolytic alone, a big winner in my opinion:
An externally hosted image should be here but it was not working when we last tested it.



This is the 10uF tantalum paralelled with a 100nF multilayer ceramic just where the legs come out of the packages. There is not much difference:
An externally hosted image should be here but it was not working when we last tested it.



This is a 2200uF 25V low-ESR capacitor (0.045 ohms) from Panasonic FC series (hey, some of my components have known make and model):
An externally hosted image should be here but it was not working when we last tested it.



And this is the same 2200uF 25V low-ESR capacitor with the 100nF ceramic paralelled just where the legs exit the package. It sings a 5.5Mhz song while not improving much anything:
An externally hosted image should be here but it was not working when we last tested it.
 
This is the 2200uF low-ESR capacitor paralelled with a 1uF 63V MKT from Toshiba (not much difference, despite the price of the MKT and its fancy yellow square case). Note the slight 1.2Mhz ringing lurking around (it will sing the full melody as soon as you place the capacitors a fer centimeters away):
An externally hosted image should be here but it was not working when we last tested it.



And finally, this is the Audiophile Special: The 2200uF low-ESR paralelled with the 1uF 63V MKT and the 100nF ceramic. Don't laugh, this is high-end 4Mhz ringing that can be only appreciated by the most select ears out there...
An externally hosted image should be here but it was not working when we last tested it.


Note that the transition from global-feedback to no-global-feedback happens in the 4Mhz range for most power amplifiers and op-amps, and low PSU impedance is critical here in order to obtain the required phase margin for good stability.
 
Oh, I forgot that audiophiles doesn't like ceramic capacitors...

This is what happens when a 10nF MKT film is paralelled with the 2200uF 25V low-ESR. Nice, eh?:
An externally hosted image should be here but it was not working when we last tested it.



And this is the result when a 100nF 250V MKT film is added to the mixture (it has ringing in all flavours):
An externally hosted image should be here but it was not working when we last tested it.
 
Good, Eva!

It was great fun! :D

But now could we get serious, and do something like this on the chip what you are measuring?

http://www.diyhifi.org/forums/viewtopic.php?p=8676#8676

in place of that ceramic 100nF in the pic, there should be [beside each other] a 10nF 0805 smd // 100nF 0805 smd // and finally a 10uF 10v X5R 1210 pana, or mur, or kemet, on top of the previous two?
But really any kind of high density multilayer at hand would do in this position, from 5 - 22 uF. Nothing else.

Ciao, George

Ps.: I could [and maybe will] measure it myself, but would be nice to see a direct comparison in your setup. ;)
 
If there's no signal with high enough frequency component that can excitate the resonant circuit into oscillation and possibility by the circuit to maintain oscillation there's no problem to use several capacitors in parallel, that's partly why opamps may survive with 2 decoupling capacitors.

Cheers Michael
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.