recommendations for reservoir capacitor type

[lordnoxx]

Hi all.
My 100W into 4 ohm class AB amplifier needs a power supply. I did a little calculation and the result is that I need 40000µF per rail (+Vdd=35V and -Vdd=35V) if I have to power a stereo setup. My calculation was the following:

Starting with a sine wave that produces 100W into 4 ohm gives me 7.07A peak current. Let's choose 10A here. Each power rail conducts for a half sine wave cycle. The latter results in a average current through the reservoir capacitor of 3.18A. So with the well known formula C=I/(dU*f) and a chosen ripple voltage of dU=2V I get C=3.18A/(2V*100Hz)=15900µF. I thought I better take 20000µF here . All the calculation is for a mono setup. But if I connect two amplifiers to that power supply I would need 40000µF per rail of reservoir capacitor size.

So finally my questions:
-Is the above calculation correct?

-How would you realize 40000µF? Two big 20000µF cans in parallel or better a few smaller ones in parallel (lets say 20 x 2000µF)?

-Which manufacturer would you recommend to choose for the capacitors?

Thank you.

[PlasticIsGood]

No. You've double counted the number of channels, and rounded up twice. Repeat, and then when you get the right answer, add a margin to that. That will make more sense.

Quality of manufacture varies within brands as well as between. I tend to favour Nichicon but that's largely familiarity, and the PWs look posh. You can find large variations in ESR and lifetime even between values in the same product range (eg Nichicon PW).

So, look for particularly well-specified caps of the particular value you decide on, choosing from a mainstream supplier who values a reputation for reliability, and who provides full data.

As for single or multiple caps, the same applies. Look at the data. Multiples should be of the same value I think.

[fas42]

You'll always be better off with multiples of small value units, no need to be concerned about the value sizes being the same. Put them as close as possible to where the voltage is actually used, try not to get hung up on making it look neat when you take the lid off, the length of leads becomes all important if you want the best bang for your buck ...

FRank

[PlasticIsGood]

If multiple caps are of the same type and value they may be cheaper, and interaction between the caps is minimised. If they are all connected in parallel, then they all need to go in the same position. This being impossible, there is nothing to be lost from a neat layout.

[fas42]

Originally Posted by PlasticIsGood

If multiple caps are of the same type and value they may be cheaper, and interaction between the caps is minimised. If they are all connected in parallel, then they all need to go in the same position. This being impossible, there is nothing to be lost from a neat layout.

If the caps are electrolytic, which they will be 99% of the time, they are not fussy about where they are, negative interaction between them is zero. I'm not sure where you get the idea that connected in parallel means being in the same position, that doesn't make electrical sense.

And, I would beg to differ: there's a lot to be lost from a neat layout, electrons don't care about things being visually "nice" ...

Frank

[lordnoxx]

Good morning everybody. Thank you for your replies so far.

Originally Posted by PlasticIsGood

No. You've double counted the number of channels, and rounded up twice.
.

Sorry, but I don't see that. One of my mono blocks can give 100W into 4 ohm.
-> 7.07A peak -> with a bit of margin 10A peak per supply rail.->average current per rail is then Ipeak/pi=10A/3.14=3.18A average current per supply rail. So if I connect 2 mono blocks to the same power supply and both give 100 W into 4 ohm leads to 2*3.18A=6.36A per supply rail as average current.->C=6.36A/(2V*100Hz))=31800µF. Thus I thought...hey what a strange value....I better take 40000µF per rail. And because I have two rails, +35V and -35V, both would need that 40000µF if I am going to connenct two mono blocks to end up with a stereo system.

[DF96]

If you want full bass at low frequencies then you need to design on the basis of peak current, not average current halved. This is because low bass will load one supply rail for a couple of mains half-cycles. So design for 7A load (x 2 for stereo).

2V ripple might be too tight a requirement, but that would depend on amplifier PSRR and how much spare voltage you have.

[PlasticIsGood]

Originally Posted by lordnoxx

Good morning everybody. Thank you for your replies so far.


Sorry, but I don't see that. One of my mono blocks can give 100W into 4 ohm.
-> 7.07A peak -> with a bit of margin 10A peak per supply rail.->average current per rail is then Ipeak/pi=10A/3.14=3.18A average current per supply rail. So if I connect 2 mono blocks to the same power supply and both give 100 W into 4 ohm leads to 2*3.18A=6.36A per supply rail as average current.->C=6.36A/(2V*100Hz))=31800µF. Thus I thought...hey what a strange value....I better take 40000µF per rail. And because I have two rails, +35V and -35V, both would need that 40000µF if I am going to connenct two mono blocks to end up with a stereo system.

The rounding up twice was from 7 to 10A, and then from 15,900 to 20,000uF. I'm mistaken about the double-counting: I was thinking rms rather than peak. Sorry

I find simulation useful for power supplies. Rectification makes maths awkward and I get lazy. Duncanamps PSUD is good, but SPICE allows testing with a particular amp circuit handling a range of signals.

I'm surprised your calculations don't include transformer regulation.

[PlasticIsGood]

Originally Posted by fas42

If the caps are electrolytic, which they will be 99% of the time, they are not fussy about where they are, negative interaction between them is zero. I'm not sure where you get the idea that connected in parallel means being in the same position, that doesn't make electrical sense.

And, I would beg to differ: there's a lot to be lost from a neat layout, electrons don't care about things being visually "nice" ...

Frank

In this context, the parallel caps would be serving a single purpose, in lieu of a single cap. To the extent that position is important, then their single purpose dictates a single ideal position. Since more than one cap cannot occupy the same physical position, there must be some compromise. Now, if you were to calculate the best arrangement for the cluster in 3D or 2D space, the result would be neat, but awkward to implement. So further compromise is usually necessary. In this context, the cost of neatness is very small compared to the gains in servicability, intelligibility, and aesthetic sensibility.

I know nothing of the secret desires of electrons, but they do seem to enjoy neat, orderly structures.

As for using various sizes of caps to make up a single bulk capacitance, the same concern for neatness applies. For low frequencies it is unlikely to make any difference, but higher up there is a risk of building a complicated mess of resonant circuits from all the various parasitics. Again, this may not matter in this case, but all the same it is much easier to characterise the performance of multiple caps in parallel if they are all of same nominal value and type. Superposition then applies, more or less, so combination is simple arithmetic.

Caps in parallel but serving different purposes are a different kettle of fish. I didn't intent to suggest that all caps in parallel must share the same position in the circuit. Actually though, if they need to be in different positions, then that must surely be to take into account circuit elements such as track or wire inductance, in which case the caps are not actually in parallel. So now I will say it: all caps in parallel should be of same nominal value and type and placed in a single neat array.

OTOH, some may find perverse pleasure in the juxtaposition of beautiful and orderly music coming from an ugly tangled mess. Not for me.