I agree with what you said except this part.

gdjacobs wrote:

If you reduce the ESR, you reduce the amount of power which must be removed from the capacitor casing.

When you lower the ESR you raise Irip through the cap and that results in greater heating of the cap.
.
People forget there is a load in parallel with the cap(s).
.
The caps do not control the total amount of ripple ( in WATTS ) in the circuit.
That is controlled by the activity of the switchers, VR, MOSFET or whatever. I'll just say the IC for all since this is general.
The IC is in turn controlled by the DC at or near the load by some kind of feedback. (Usually controlled by DC voltage but could be DC current.)
Thus the DC needs of the load determine IC activity and the IC activity creates a corresponding amount of ripple - regardless of what the caps are doing.
The caps do not control the total amount of ripple *put in* - they only control what path it takes after it's created.
For ripple the ESR of the cap is (or caps are) one side a of a voltage divider with the load being the other side.
So, if you lower ESR you direct more ripple current through the cap and less ripple current goes through the load.
- With more ripple current passing through it the cap is heated more, not less.
- And the load (other side of the voltage divider) sees less ripple current and is heated less - which is the actual goal of all this.
.

To clarify for anyone watching, we're talking about filter applications here and not bulk caps.

Correct, ripple is not controlled by the cap, but the manner of it's dissipation is. It's a complicated concept that I didn't want to explain in totality. We have to consider ESR, reactance of the cap (at each harmonic frequency), and Thevenin impedance of the rest of the circuit. Ideally, we want to have a very low AC impedance to ground at the capacitor so ripple voltage will drop on the switching transistors, transformer, and traces before reaching the load.

In switching mode supplies, and assuming ideal load, SMPS output inductor ripple current is not determined by the filter capacitor, but voltage difference between voltage input and output, duty cycle, switching frequency and inductor value (and generally other operating parameters like avg/max load current).

The SMPS output inductor ripple current is then flowing into the filter capacitor so at the capacitor it is just seen as ripple voltage. Of course a real load like a CPU or memory pulling current transients, this is also ripple current from the cap to load which is seen as ripple voltage over the cap, and the cap must provide stable enough voltage. Cap does this by means of low enough ESR to prevent immediate drop from resistance and high enough capacitance so that during the current transient the voltage droops slowly enough to keep the load running until the SMPS can provide more current as requested.

Sure, smaller capacitor ESR means smaller ripple voltage for the same capacitance and same inductor ripple current.

But also more capacitance means smaller ripple voltage for same ripple current. Why? More capacitance takes more time to charge/discharge by same voltage amount at the same ripple current. So I do not agree on the part that where it was mentioned only lower ESR made something working, not more capacitance. They both help.

This is why bulk caps are generally chosen with quite high impedance. They're used to source current when the switching caps are not in conduction, so they must have significant charge storage. However, designing a low pass filter with a high impedance does move the -3 dB point higher as well, so massive caps are not desirable immediately preceding the DC output point.

I'm enjoying this thread .

Jepael wrote:

... ...

... But also more capacitance means smaller ripple voltage for same ripple current. Why? More capacitance takes more time to charge/discharge by same voltage amount at the same ripple current. So I do not agree on the part that where it was mentioned only lower ESR made something working, not more capacitance. They both help.

Very interesting Jepael, thanks for sharing.

To everybody: What do you think are the pros and cons of increasing bulk cap capacitance in a PSU (inrush current apart)? How (if) does it influence on output ripple?

As above, you will have superior ripple voltage characteristics and regulation before the application of the final shunt suppression caps on the DC output.

I think part of the problem with the cost of your procurements (for your friends' recap jobs) is your improper selection of capacitors. You always have to skim through lists of capacitors from different manufacturers for a given supply-house because oftentimes they will list a capacitor which they bought A LOT of, cheaper than another capacitor which may have been less expensive, save for the huge quantity-discount they received on the aforementioned cap.

As has been mentioned, you don't need to UP the voltage handling of the capacitors. In fact the only likely result you will have using this practice is that you will have a harder time finding the appropriate diameter/height capacitors, and they will probably cost more. It's better to find the appropriate voltage and use that. You'll often find 6.3V capacitors where there could have been a 3.3V or even a 1.5V--it all comes down to what the bean-counters could find cheapest (given whatever specifications they were trying to meet) when the board was manufactured.

Note also (hopefully PCBONEZ will chime in on this and help me have a better understanding) that, from what I've read in the past, upping the voltage can have a detrimental effect on the capacitor itself. The self-healing which was mentioned; I read that if you up the voltage significantly higher than it needs to be that the self-healing effect will not happen and can lead to capacitor... I don't recall, malfuction, failure? I don't think it would happen in your case because I believe you would have to increase the voltage handling to a ridiculous degree for this to happen. I'll wait for someone else to chime in on this--I'd like to have a bit better knowledge of this scenario.

Jepael wrote:

But also more capacitance means smaller ripple voltage for same ripple current. Why? More capacitance takes more time to charge/discharge by same voltage amount at the same ripple current. So I do not agree on the part that where it was mentioned only lower ESR made something working, not more capacitance. They both help.

No. Sorry. Wrong.
.
At approximately 100 kHz (the industry accepted typical SMPS ripple frequency and so what is used in datasheets) Xc = Xl .. It's not exact, but close enough to consider it so.
The equation for Z (total impedance) can be written thus:
Z = [(ESR)² + (Xc – Xl)²]½ [This equation is provided by several caps manufacturers.]
Since at 100kHz Xc = Xl it can be rewritten thus:
Z = [(ESR)² + (0)²]½
And simplified to this:
Z = [(ESR)²]½
And then this:
Z = ESR
(Recall this is at ~100 kHz)
.
You can verify what I just said in tech docs provided by multiple caps manufacturers.
.
And so it's very clear that the capacitance doesn't make a lick of difference regarding ripple at or near 100 kHz.
.
You made some other mistakes but I'm playing nursemaid to someone that had surgery today so I'll have to address them one at a time.
.

TELVM wrote:

To everybody: What do you think are the pros and cons of increasing bulk cap capacitance in a PSU (inrush current apart)? How (if) does it influence on output ripple?

As per my previous post it won't affect ripple at all. At least not to any measurable degree.
.
It will affect the PSUs timing which goes basically like this. (This is about DC voltage - not ripple.)
(ad libbing) The PSU voltages have to be at x-many percent of full voltage within y-many milliseconds or the protective circuit will shut the PSU back off.
You can read the exact wording (and numbers) in the ATX PSU specifications which aren't too hard to find on-line.
The exact values vary from one spec to another so it's impossible to be more specific.

Too much bulk capacitance -may- cause the PSU not to start (rather to not complete starting) because it takes the voltage too long to get to the x-many percent.
I don't think going up one standard value is enough to cause a problem though because the motherboard caps also affect how fast the voltage comes up so the OP caps in the PSU are only a small part of the total.
.
.
I must now go attend to my patient for a while.
.

Jepael wrote:

In switching mode supplies, and assuming ideal load, SMPS output inductor ripple current is not determined by the filter capacitor, but voltage difference between voltage input and output, duty cycle, switching frequency and inductor value (and generally other operating parameters like avg/max load current).

Glaring error in your circuit analysis and thought process here.
.
Yes the Irip through the switcher OP inductor depends on the ripple voltage drop across it - but you forgot WHY there is a ripple voltage drop across it.
The Irip through the filter cap drops the Vrip on the cap and that's where most of the voltage drop across the inductor comes from.
So the current through said inductor IS largely determined by the filter cap. - Specifically it's ESR.
.

Logistics wrote:

Note also (hopefully PCBONEZ will chime in on this and help me have a better understanding) that, from what I've read in the past, upping the voltage can have a detrimental effect on the capacitor itself. The self-healing which was mentioned; I read that if you up the voltage significantly higher than it needs to be that the self-healing effect will not happen and can lead to capacitor... I don't recall, malfuction, failure? I don't think it would happen in your case because I believe you would have to increase the voltage handling to a ridiculous degree for this to happen. I'll wait for someone else to chime in on this--I'd like to have a bit better knowledge of this scenario.

Howdy old BCN friend. - That discussion was so long ago I'm surprised anyone remembers.
.
Where that came from:
Tech/Training manuals I was using in the 80's (which were probably written in the 60's and 70's) used to stress not using a cap with a voltage rating more than 3 times the actual circuit voltage.
It was probably worded differently but that is basically what it said.
The reason given was that too little DC voltage (bias) would not provide/create enough ion flow through the electrolyte to rebuild the oxide layer. Not enough driving force.
- Remember, they call them electrolytic because a sort of electrolysis goes on inside, only in the case of these caps it's aluminum oxide that plates out from the solution.
I have not seen that mentioned in more recent documentation so presumably they have improved electrolyte chemistry to the point that much less bias is necessary to restore the oxide layer.
(Any TRUSTED more recent documentation. I did see something from a crap-cap company that said 10% of the rating for a minimum actual voltage but I don't trust crap-cap companies.)
I tend to follow the old rule anyway just to be safe. That is when I'm not just replacing existing caps.

I'm sure Logistics already knows the rest. It's here for other readers.

- Ripple and time will thin the oxide layer. The oxide components go back into the electrolyte solution.
- DC voltage (the leakage current through the cap from it) restores (reforms) the oxide layer - this process is much faster than those that damage/thin the layer.
Once the oxide layer builds to a certain point (determined by the DC voltage/bias) the leakage will effectively slow to nothing and the layer will stop getting thicker.
(Thicker layer - little/no leakage current - nothing to drive (push) the ions back to the plate.)

Because of those processes if you use a 16v cap on 5v it will BECOME a 5v cap after a while because the thickness of the layer adjusts to 5v despite the caps voltage rating.
It does not work the other way. (Not usually.)
You can not turn a 5v cap (if they existed) into a 16v cap by applying 16v. The reason is there are not enough ions present in solution (the electrolyte) to build the layer that thick.
(In the case of caps with derated voltage specs it might work but it's not guaranteed.)
.

PCBONEZ wrote:

No. Sorry. Wrong. . At approximately 100 kHz (the industry accepted typical SMPS ripple frequency and so what is used in datashe […]
Show full quote

Jepael wrote:

But also more capacitance means smaller ripple voltage for same ripple current. Why? More capacitance takes more time to charge/discharge by same voltage amount at the same ripple current. So I do not agree on the part that where it was mentioned only lower ESR made something working, not more capacitance. They both help.

No. Sorry. Wrong.
.
At approximately 100 kHz (the industry accepted typical SMPS ripple frequency and so what is used in datasheets) Xc = Xl .. It's not exact, but close enough to consider it so.
The equation for Z (total impedance) can be written thus:
Z = [(ESR)² + (Xc – Xl)²]½ [This equation is provided by several caps manufacturers.]
Since at 100kHz Xc = Xl it can be rewritten thus:
Z = [(ESR)² + (0)²]½
And simplified to this:
Z = [(ESR)²]½
And then this:
Z = ESR
(Recall this is at ~100 kHz)
.
You can verify what I just said in tech docs provided by multiple caps manufacturers.
.
And so it's very clear that the capacitance doesn't make a lick of difference regarding ripple at or near 100 kHz.
.
You made some other mistakes but I'm playing nursemaid to someone that had surgery today so I'll have to address them one at a time.
.

Okay, makes sense. The coil, cap, and switching frequency are chosen to operate at resonance. Are consumer SMPS units generally utilizing single or two stage output filters?

gdjacobs wrote:

PCBONEZ wrote:

No. Sorry. Wrong. . At approximately 100 kHz (the industry accepted typical SMPS ripple frequency and so what is used in datashe […]
Show full quote

Jepael wrote:

But also more capacitance means smaller ripple voltage for same ripple current. Why? More capacitance takes more time to charge/discharge by same voltage amount at the same ripple current. So I do not agree on the part that where it was mentioned only lower ESR made something working, not more capacitance. They both help.

No. Sorry. Wrong.
.
At approximately 100 kHz (the industry accepted typical SMPS ripple frequency and so what is used in datasheets) Xc = Xl .. It's not exact, but close enough to consider it so.
The equation for Z (total impedance) can be written thus:
Z = [(ESR)² + (Xc – Xl)²]½ [This equation is provided by several caps manufacturers.]
Since at 100kHz Xc = Xl it can be rewritten thus:
Z = [(ESR)² + (0)²]½
And simplified to this:
Z = [(ESR)²]½
And then this:
Z = ESR
(Recall this is at ~100 kHz)
.
You can verify what I just said in tech docs provided by multiple caps manufacturers.
.
And so it's very clear that the capacitance doesn't make a lick of difference regarding ripple at or near 100 kHz.
.
You made some other mistakes but I'm playing nursemaid to someone that had surgery today so I'll have to address them one at a time.
.

Okay, makes sense. The coil, cap, and switching frequency are chosen to operate at resonance. Are consumer SMPS units generally utilizing single or two stage output filters?

The Xl in that equation is from the ESL (Equivalent Series Inductance) of the capacitor itself - not the inductor elsewhere in the filter.
IOW as used above the equation is just for the cap, not the whole filter.

If by single stage you mean a basic L-C and two stage means a Pi (C-L-C) then both ways are common.
Higher quality models get the Pi.
.

Note to everyone
This has gotten steered this towards being all about PSUs.
I'm not opposed to talking about PSUs but as I said earlier my intentions were to be more general in nature.
The basic concepts also apply to VRMs and POL VRs and I don't want to exclude those from the discussion.

[edit/add]
POL VR = Point of Load Voltage Regulator.
It's mostly those single transistor/MOSFET regulators you'll see scattered about motherboards, video cards and what not.
I guess technically a CPU VRM is also a POL as they are located next to the CPU but the component count sort of makes them a different animal.
.
Also VRMs and POL VRs are technically SMPS, just very simple ones.
.

I think we're okay as long as we stick to filter design. That's all germane to capacitor selection.

Some filters in the literature feature a series inductor between the switch converter and the Pi filter. The article in question is specifically regarding PSU construction by IBM for their iron.

Definitely some useful information here. Not that I had to replace a capacitor yet but, when that day comes, I know where to come and look for some info 😀

gdjacobs wrote:

I think we're okay as long as we stick to filter design. That's all germane to capacitor selection.

Some filters in the literature feature a series inductor between the switch converter and the Pi filter. The article in question is specifically regarding PSU construction by IBM for their iron.

Their iron?
.
The equation I used earlier as applied was for the equivalent circuit of a capacitor. - Just the cap.
Inductors external to the cap as in a filter aren't part of it.
Yes, they do use the same equation for filters but that isn't what that was.
.
PC PSUs do have inductors between the switchers and the OP filters.
Most have toroidal transformers after the diodes (which are after the switchers and their transformer) and before the OP filters.

A bit of googling will pull up dozens of ATX PSU schematics. (Many are very old designs though.)
The designs vary widely (particularly after the toroidal transformers) so saying something is 'always' true at the OP filter stage can be problematic.
I try not to do that because I don't want some whiz-kid finding a random counter example of something and arguing about it for a week.
.

jesolo wrote:

Definitely some useful information here. Not that I had to replace a capacitor yet but, when that day comes, I know where to come and look for some info 😀

Good thing to know and skill to have if you are into retro computers.
Almost guaranteed you will need to replace caps in/on something eventually.
.

PCBONEZ wrote:

Their iron?

S390 / zSeries

PCBONEZ wrote:

. The equation I used earlier as applied was for the equivalent circuit of a capacitor. - Just the cap. Inductors external to th […]
Show full quote

.
The equation I used earlier as applied was for the equivalent circuit of a capacitor. - Just the cap.
Inductors external to the cap as in a filter aren't part of it.
Yes, they do use the same equation for filters but that isn't what that was.
.

It's the same for tuned LC band stop filters, but that wasn't what they were using. More like a second order RLC filter, but split into two sections.

PCBONEZ wrote:

PC PSUs do have inductors between the switchers and the OP filters. Most have toroidal transformers after the diodes (which are […]
Show full quote

PC PSUs do have inductors between the switchers and the OP filters.
Most have toroidal transformers after the diodes (which are after the switchers and their transformer) and before the OP filters.

A bit of googling will pull up dozens of ATX PSU schematics. (Many are very old designs though.)
The designs vary widely (particularly after the toroidal transformers) so saying something is 'always' true at the OP filter stage can be problematic.
I try not to do that because I don't want some whiz-kid finding a random counter example of something and arguing about it for a week.
.

Definitely lots of stuff out there, but many of the intricacies and subtleties don't seem to be publicly available. To the best of my knowledge, consumer level PSU and power conversion is a low margin business, so maintaining confidentiality and forcing competitors to reverse engineer is understandable. That's why this is a great opportunity to learn.

gdjacobs wrote:

but many of the intricacies and subtleties don't seem to be publicly available.

Agreed.
There are parts of a PC PSUs operation that I still don't fully understand because I could never find good info.
I can get through a block diagram from end to end but fine details inside some of the blocks - not so much.
At some point I lost interest in learning more because I'd rather work on motherboards anyway.
If a PSU needs more than just caps I'm more inclined to toss it and get another than try to fix it.
I know capacitors because of my interest in motherboards but almost all of that applies to PSUs as well.
.