Yes. I would think less about capacitance in the PSU and maybe think more about getting decoupling capacitances as close as possible to the power output devices.
For speed (and the timing and phase accuracy that give the detailed soundstage image), you would need something like at least 10uF or 20uF per volt of rail voltage, with one inch or less of total connection and lead-spacing length, at each power output device.
Better yet, use multiple smaller caps that add up to that value, in parallel, with star-like connections that don't share conductors, to lower the total inductance (and ESR) below any of their individual inductances (or ESRs).
The total inductance and ESR would each be divided by the number of parallel caps. I would try to use at least three caps in parallel. (Note that the total connection length can be longer, with multiple caps, but I would still try to keep each cap's total connection plus lead-spacing length at one inch or less, if possible, or at least keep the total for all of them at one inch per cap or less.)
Then add more capacitors as close to each device as possible until you get a total of up to about 2200uF per max output amp, for each device. You don't have to go that high but it would improve the bass and mid response. If possible, make all capacitor connections like a star, where none of them share any common conductor. That will lower the total inductance (and ESR), very drastically. Using more smaller capacitors in parallel is better, especially if they can be placed to not share conductors until they get to the power and gnd decoupling points. But obviously there is limited space. So just do it as well as is practical.
Last edited:
Assuming 40000uF is needed, somewhere in between 40x 1000uF and 4x 10000uF. How about 8x 4700uF?
Interesting: So if the rails are held too tightly to a fixed voltage by too much reservoir capacitance, then maybe the decoupling caps that supply the fast transient currents can't respond as well, since the rail voltage would need to be able to drop just to make them send out the needed current! So then at least part of the current will have to try to come through the rail inductance, instead of from the decoupling caps. But for fast transients, it can't get there in time to provide an accurate transient response, because of the rail inductance. Also, the rail voltage gets disturbed by more than it would have (by the current trying to come through the rail inductance), possibly partially defeating the purpose of adding the reservoir caps in the first place.
(AS USUAL, we might know what kinds of things could be happening, but we have no idea of the relative magnitude or potential signficance of each one.)
So maybe the conclusion should be that if psu reservoir capacitance is raised to the overkill level, then the decoupling capacitances should also be raised to the overkill level, to increase the sensitivity of their current-output response to the smaller voltage variations caused by the too-large reservoir caps. <grin> (Or maybe the total decoupling capacitance should be kept equal to the total reservoir capacitance?)
This does point out the fact that more decoupling capscitance gives faster response and better dynamic accuracy, since when a power output device "opens wider" (lowers its resistance), and the rail voltage then dips a bit, more decoupling caps would produce more current, faster.
I wonder if there is a point where there could be so much decoupling capacitance (with low-enough inductance and ESR) that the whole system would become "too sensitive". Or maybe it would just keep getting better.
Last edited:
Since "there is not a perfect formula" and We go for experience... I would go for low E.S.R caps 105c 6.800X6 or 4.700X8 per rail. some people go crazy and go for 1.000.000uf per rail (LOL) thinking they will get "better" sound than a "normal" and moderated power supply.
for the transformer...if your amp gets 650watts at 4ohms We multiply 650x1.41 and we get 916VA so I would go for a 1K transformer...1.600VA I think is an "overkill" again!!!
Well, Krell amps use hefty power supplies...I guess 160.000uf per rail is "normal" with 3kVA transformers for them. For Krell there is not such as "overkill".
Another example is with Bryston.
Bryston 14BSST is 600watts at 8ohms using 44.000uf per rail.
Bryston 28BSST is 1,000watts at 8ohms using 80.000uf per rail.
so, as We can see here, Krell likes "overkill" power supplies compared to Bryston amps.But does that mean Krell amps sound better than Bryston amps because krell uses "overkill" power supplies?...I do not think so!!!There is not an exact capacitance for power supply. other way "all" 100 watts or so at 8 ohms amps will have the same capacitance in all power supplies.
Bryston 14BSST is 600watts at 8ohms using 44.000uf per rail.
Bryston 28BSST is 1,000watts at 8ohms using 80.000uf per rail.
so, as We can see here, Krell likes "overkill" power supplies compared to Bryston amps.But does that mean Krell amps sound better than Bryston amps because krell uses "overkill" power supplies?...I do not think so!!!There is not an exact capacitance for power supply. other way "all" 100 watts or so at 8 ohms amps will have the same capacitance in all power supplies.
I use 12 x 27,000 uf Panasonic 105C caps for each channel (324,000 uf) fed from seperate posivitve and negative RC snubbed TO 247 60 amp discrete hexfet bridges. Multicap 2uf film and foil bypasses then 1 mh chokes and then 4 x Blackgate 10,000 uf FK with 2,200 uf N/100v bypasses near the output stage. Thats for each channel in a Krell KSA 100 chassis. Front end is run from seperate PI filtered transformers and preregulated supplies - output is just 100 watts Class A per channel Borbely Milleniums all jfet/fet discrete direct coupled and servoed.
A 1/2 second AC line soft start is mandatory. Transformers run barely warm to the touch.
Runs parralled modified Martin Logan CLX (1/2ohm in the top end with a Raal ribbon) with Dynaudio subs - tomorrow i test 2 JL 113 Fathoms with it.
The Borbely is smooth and very dynamic despite just 100 watts, the Bryston 4B SST 2 was gritty , unresolving, even harsh on the same load despite more power.
A 1/2 second AC line soft start is mandatory. Transformers run barely warm to the touch.
Runs parralled modified Martin Logan CLX (1/2ohm in the top end with a Raal ribbon) with Dynaudio subs - tomorrow i test 2 JL 113 Fathoms with it.
The Borbely is smooth and very dynamic despite just 100 watts, the Bryston 4B SST 2 was gritty , unresolving, even harsh on the same load despite more power.
Last edited:
For those who joined the party late checkout
#49
#58
#60
#66
#70
#71
Then go adjust your power supply, sit back relax and enjoy the music
Lets not stray too much
Bigger isn't always best.
As you say, amplifier "tuning" or the sonic characteristics of an amp can be altered by its PSU design.
Again this is where objectivism and subjectivism might not agree.
A) The amp is better technically with the low ripple massive PSU.
B) The amp sounds better with a smaller "faster" PSU.
So which do you build ? A or B
Some good general useful info here... but it doesn't tell you how godd an amp will sound with a given PSU.
The Signal Transfer Company: Power Output
I don't understand the marked term. Mean a faster PSU a version with very low ESR capacitors?
Nicely put Hugh.
There's no hard and fast rules for all this. I find it's how a design comes together as a whole.
I don't think so. My experience is follow regarded power supplies for audio amplifier: The bigger the better. - completely irrelevant, whether it is a small amplifier (like Naim NAIT) or a bigger version (like Krell KSA250).
Unfortunately, this rule applies only to external (outdoor) power supplies, where is additional a big choke (low DC impedance air coil) in use between two sets of capacitors. By "built-in" resp. indoor power supplies (usually mostly the case) you are right of course.
I often observe this, even by very expensive units. In such cases one can always be sure that there are very major internal or external errors. Unfortunately often design errors. Currently there are not many good developers for audio amplifiers and many tried and tested recipes slip back into oblivion. Nevertheless - the power supply (and not the circuit topology) mainly determines the sonic results in the most cases.
Last edited:
I think so in general terms...
Yes , that is very important. Even on simple squarewave testing it's important to make sure the results are good down to millivolt output levels. If compensation is wrong you can have seemingly perfect results at say 1 or 10 volts output but ringing and signs of problems at 10 or 1 millivolt. Hugely important in low level signal circuitry whether discrete or opamp.
How wide is wide ? I prefer to think "adequate bandwidth".
I'm honestly not 100% sure on that and haven't done any critical listening comparisons. Many report favourable results though.
(I first encountered what I call "fast rectifiers" way back when I started out as a repair tech and wondered why certain diodes wouldn't work in TV deflection circuits or SMPS's) I can't help but feel that when you have added snubbers across diodes and across secondary windings that things equal out a lot. But I may be wrong
Absolutely. You can never have enough good local decoupling providing it doesn't inject into clean grounds.
Again that's another area where I lack experience of trying different values. 4, 6 or 10 uH.... I suspect it may be very system specific if you can reliably hear a difference. But I haven't tried
I often prefer a small series resistor to provide isolation perhaps with just a very low value coil as well. (The coil makes me feel better even if it's not always needed)
Those who have looked at pulse width modulation in class D and also switch mode power supplies will note that for a linear power supply with a recharge rate of 100 or 120 means that we do not need more capacitance to improve bass reponse beyond 5Hz as by this time if we were to play a 5 Hz tone the transformer would be able to sustain it without help from the capacitors.
Thus for an 8 Ohm load @ 5Hz (has nothing to do with wattage) 4000uF will suffice
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.