you have connected an amplifier as load in your model? (in this case is very simple)
To observe the behavior of a psu, you must connect a load of course.
This load can be fixed as resistive or interrupted as a pulse train.
depends on what you want to see.
a simple burst can be a signal eg.1 Khz that is cut off for 66ms and repeated (duration) for 33ms then call 33/66ms.
in the case of the psu, can also be used a mosfet which closes a resistance creating the load. the gate can be driven by a square wave.
-------
I asked you to see these signals in order to understand the behavior of your psu, when to supply an amplifier, if they fixed some problems that are not only the ripple.
Last edited:
Despite being quite confused some of the time and not being able to read all of the content, I've learned a few things.
A slightly undersize transformer dramatically increases the needed tank/reservoir capacitance. Thanks Tom!
Capacitance located more effectively is more effective capacitance. Thanks Nico and Terry!
P.S.
It seems that those gems can really reduce capacitor expense.
A slightly undersize transformer dramatically increases the needed tank/reservoir capacitance. Thanks Tom!
Capacitance located more effectively is more effective capacitance. Thanks Nico and Terry!
P.S.
It seems that those gems can really reduce capacitor expense.
DF,
That all sounds like it could be correct.
But still, in order to allow a particular current to flow from a rail to a load and induce an accurate voltage across the load that depends only on a known fixed gain and the input amplitude, every type of amplifier would have to look like exactly the same impedance between rail and load, for a particular gain and input amplitude, no? And for a particular rail voltage and load impedance and any particular commanded output amplitude across the load, only one impedance between the rail and the load would be correct, no?
If I am correct, then it seems like we should be able to find something that is independent of the amplifier details.
Or am I thinking completely wrongly about that? (It happens.)
I was thinkng that maybe the 3.3 Volt figure was mostly related to capacitor ESR, and possibly also other parasitics. I have also not yet varied the ESR, except with changes in capacitor value.
Tom
Last edited:
Yes, Daniel. I was amazed at how much difference it made. It seems obvious, NOW, of course. Also, make sure that there is enough headroom in both the Vrms and the VA ratings. Otherwise, the changes you hear when changing caps, especially when changing their values, might be mostly due to the transformer size rather than the caps themselves.
It complicates everything, and ESPECIALLY some of the "findings" of the subjectivists , and will continue to do so unless we can say something like "as long as the transformer has high-enough VA and Vrms ratings", and we can test everything with a way-oversized transformer, if it might matter.
Tom
If I understand you correctly, then no. You may be confusing DC resistance (V/I) with AC resistance (dV/dI). In fact, the concept of resistance is not very useful here. To get a particular current and particular voltage across two points can mean any combination between a voltage source (zero resistance) and a current source (infinite resistance).gootee said:
The 3.3V figure you have found probably has little to do with capacitor ESR. Vbe drops in the output stage account for much of it. Then there is voltage drop across driver resistors (e.g. the 22 ohm).
Tom & DF,
Isnt it great that all this detailed work has, once needless parasitics (by which I mean that the parasitics can be avoided at the design/layout/wiring stages by following the good engineering practices highlighted in this thread) have been eliminated (at little or no cost) we end up back at the fundamental - mayhap simplistic - approach given in many textbooks.
but the real magic of this thread, as Daniel alludes to, is the "topological dual" of the above statement: If one does not explicitly eliminate these parasitics then the simplistic approach doesnt really work.
Adding more caps doesnt really fix the ESL/ESR voltage ripple (and wiring/layout may make it worse) - eg doubling the capacitance wont halve the actual supply ripple, which leads directly to stupidly large amounts of capacitance.
And although I personally feel subjective tests are anathema, I cannot help but conclude there really is something to it. Many (most?) amps have such poor (from an RF perspective) wiring and layout (IOW poor control of magnetic and to a lesser extent electric fields) that almost any change will affect them in some way.
which is exactly like EMI testing. EMI problems are seldom due to "one thing", they are invariably due to dozens if not hundreds of little antennas. which means that fixing any individual issue has little effect, and measured results seem to change as a function of time, temperature and sock colour. thats the true cause of male pattern baldness in engineers
And just like EMI fixes, the first approach should be to eliminate as many of the bad practices as possible before re-testing (EMI tests are expensive). And once one understands the basic principles (current flows in loops - minimise them, and Thou Shalt Control Thy Fields) you can find them by inspection rather than measurement (hard & $$) or, worse yet, analysis. unless one has full 3-D field solvers, which is how the big boys get stuff right first time, analytically accounting for a myriad of parasitics is useful for little more than writing papers.
Isnt it great that all this detailed work has, once needless parasitics (by which I mean that the parasitics can be avoided at the design/layout/wiring stages by following the good engineering practices highlighted in this thread) have been eliminated (at little or no cost) we end up back at the fundamental - mayhap simplistic - approach given in many textbooks.
but the real magic of this thread, as Daniel alludes to, is the "topological dual" of the above statement: If one does not explicitly eliminate these parasitics then the simplistic approach doesnt really work.
Adding more caps doesnt really fix the ESL/ESR voltage ripple (and wiring/layout may make it worse) - eg doubling the capacitance wont halve the actual supply ripple, which leads directly to stupidly large amounts of capacitance.
And although I personally feel subjective tests are anathema, I cannot help but conclude there really is something to it. Many (most?) amps have such poor (from an RF perspective) wiring and layout (IOW poor control of magnetic and to a lesser extent electric fields) that almost any change will affect them in some way.
which is exactly like EMI testing. EMI problems are seldom due to "one thing", they are invariably due to dozens if not hundreds of little antennas. which means that fixing any individual issue has little effect, and measured results seem to change as a function of time, temperature and sock colour. thats the true cause of male pattern baldness in engineers
And just like EMI fixes, the first approach should be to eliminate as many of the bad practices as possible before re-testing (EMI tests are expensive). And once one understands the basic principles (current flows in loops - minimise them, and Thou Shalt Control Thy Fields) you can find them by inspection rather than measurement (hard & $$) or, worse yet, analysis. unless one has full 3-D field solvers, which is how the big boys get stuff right first time, analytically accounting for a myriad of parasitics is useful for little more than writing papers.
Hey, DF, this particular type of case isn't that complicated.
Anyway, my post used the term impedance throughout, not resistance, except in the comment about ESR.
I was only talking about the complex impedance at any one instant.
What was being considered at the moment is a driven load that is a fixed ideal resistance, with a power rail voltage that is varying, but at any instant it is what it is and the input says that the load is supposed to have a particular voltage across it and so the amplifier, at any instant, "must" (try to) set its overall impedance (between rail and load resistor) so that Iload * Rload = the commanded output voltage across the pure resistance. I can worry about real speaker and crossover loads later, maybe, but the solution for this seems more-trivial. (nd our source is, we like to hope, very low impedance.)
Tom
This might be useful: The voltge across the amplifier, from the V+ rail to the high side of the load, when the load is 8 Ohms with 100 Watts RMS across it from that rail's share of a square wave, is about 16.85 Volts, average, with a ripple amplitude of about 2.1 Volts.
That is the real elephant in the room of this whole exercise, and it's more than the output stage if the input circuitry is fed from the same supplies. And provided a reasonably accurate sim of the amp circuitry is available then it is extremely easy to measure the PSRR, over the whole range of frequencies where it's potentially capable of being subjected to such "noise". I've already tested the PSRR of the output circuitry, and it's pretty good, about 80dB down over most of the spectrum. But, it starts to degrade at the higher end, just where you need it to be working at its best, for FB to function correctly, etc!
And then there's the full amp, by Bob Cordell, to be considered: its PSRR is much worse. And to really get the fur flying, at least one other amp circuit being discussed on diyAudio extensively has very poor PSRR, at least going by the sim ...
Frank
That may be why many good designs had a separate supply for the voltage gain stages, Electrocompaniet, Hafler XL, GAS Son, APT, to name a few.
The amp I mentioned above as designing that only had 3m3F had a regulated supply for the front end.
I really think that makes a big difference, more so than the bulk capacitance in the output stage (which is still suitably bypassed).
The amp I mentioned above as designing that only had 3m3F had a regulated supply for the front end.
I really think that makes a big difference, more so than the bulk capacitance in the output stage (which is still suitably bypassed).
Well, one great thing about Spice is it's easy to just measure things right off of the schematic, after a simulation has completed. (The numbers I gave after my name, in my last post, were the wrong ones, by the way. They were for an unrelated input waveform. See below for some relevant ones.)
First I ran a sim with a reservoir capacitance that was too low. Then I plotted the voltage across the amplifier terminals (V+ to the high side of the load) divided by the rail current, to give a plot of the amplifier's impedance. I also plotted many of the rest of the voltages and currents associated with the positive rail and the load.
When the positive part of a square wave comes up to 40 Volts, 5 amps start to flow through the 8 Ohm load resistor. As the reservoir cap discharges, the positive rail ramps downward, lowering the voltage across the amplifier by the same amount. The amplifier's impedance plot has exactly the same shape as the V+ rail voltage plot, because the amplifier has to ramp its impedance down with exactly the same profile, at the same time, in order to keep the load current the same.
If the reservoir capacitance is too small and the output waveform stays at high power for a long-enough time relative to the time between charging pulses, then the V+ rail can ramp downward until it's too close to the 40-Volt level of the square wave voltage across the load, which is the "bottom" reference voltage for the voltage across the amplifier leg that is between the V+ rail and the load resistor.
At some point, the amplifier either cannot lower its impedance any farther or cannot lower its voltage any farther, and the current in the load resistor cannot be maintained and it starts to fall, causing its voltage to also start downward. This typically only happens at the lowest rail voltage (a ripple voltage "trough"), which is also where the charging pulses occur, which is fortunate, because for marginal capacitances the signal waveform can recover quickly when the charging pulse occurs.
The main leg of the amplifier, between the V+ rail and the high side of the load, is the CE of an mjl21194C (Cordell's model) in series with a 0.22-Ohm resistor.
In the case I simulated, the impedance between the V+ rail and the high side of the load could not go below about 627 mOhms (i.e. about 407 mOhms for just the transistor) and the Vce voltage of the mjl21194C could not go below about 1.95 Volts (or 3.003 Volts if the 0.22 Ohm resistor voltage is also included).
So the 3.003 Volts across the amplifier looks like it is the voltage responsible for the difference voltage that I was using as a marker, before, to find the point when the gross distortion would stop as the C value was increased.
Thank you DF96 for the insight.
I guess that should have been obvious to me. The voltage across the amplifier can't go below some limit, if the amplifier is to continue to function.
This might provide a useful method for those who, for some reason, really do want or need to push their reservoir capacitance to the lowest limit that still avoids the gross distortion. The lowest voltage that can exist between the power rail and the high side of the load will determine how low the rail voltage can be allowed to fall, for a worst-case or maximum power signal.
I am not a BJT expert, by any means, but In the case of our simulation's amplifier topology, it looks like we could say that Vce must be greater than Vbe0 and the 0.22-Ohm resistor will have the maximum load current through it so its voltage can be calculated for a particular output spec. Other amplifier topologies will differ but it's a one-time exercise for each one.
The ripple voltage waveform's lowest minimum can be estimated using the standard equations, if 1/2 of the maximum peak load current value is assumed to be a constant DC load current (using 1/2 because the square wave is 50% duty cycle per rail). So, instead, the minimum value of the ripple voltage can be set to be just above the sum of the maximum peak load voltage PLUS the minimum sustainable voltage across the amplifier (between rail and load) and then an estimate for the minimum C can be calculated.
So it looks like the search for a rule of thumb for how much reservoir capacitance should be used might now shift back toward the plots of distortion versus capacitance, which now do have consistent and more-accurate measurements of the distortion. But we might now have to also evaluate the effects of different minimum amplifier voltages when the lower end of the capacitance range is considered.
It's getting late again. I hope this is intelligible.
I guess that should have been obvious to me. The voltage across the amplifier can't go below some limit, if the amplifier is to continue to function.
This might provide a useful method for those who, for some reason, really do want or need to push their reservoir capacitance to the lowest limit that still avoids the gross distortion. The lowest voltage that can exist between the power rail and the high side of the load will determine how low the rail voltage can be allowed to fall, for a worst-case or maximum power signal.
I am not a BJT expert, by any means, but In the case of our simulation's amplifier topology, it looks like we could say that Vce must be greater than Vbe0 and the 0.22-Ohm resistor will have the maximum load current through it so its voltage can be calculated for a particular output spec. Other amplifier topologies will differ but it's a one-time exercise for each one.
The ripple voltage waveform's lowest minimum can be estimated using the standard equations, if 1/2 of the maximum peak load current value is assumed to be a constant DC load current (using 1/2 because the square wave is 50% duty cycle per rail). So, instead, the minimum value of the ripple voltage can be set to be just above the sum of the maximum peak load voltage PLUS the minimum sustainable voltage across the amplifier (between rail and load) and then an estimate for the minimum C can be calculated.
So it looks like the search for a rule of thumb for how much reservoir capacitance should be used might now shift back toward the plots of distortion versus capacitance, which now do have consistent and more-accurate measurements of the distortion. But we might now have to also evaluate the effects of different minimum amplifier voltages when the lower end of the capacitance range is considered.
It's getting late again. I hope this is intelligible.
Last edited:
It might be interesting to check the difference between Panasonic FC (or Cornell LP) versus a standard cap like Cornell SK/SEK. I believe that the higher ESR of the standard cap might provide convenient ballast for charging, faster charging, of multi-cap arrays.
Far out oversize transformer incurs a slight "dull" problem, the same sort of effect that can be had if 1000uF caps are the smallest power caps on the amp board. This case is fixable by reducing the amp board's bypass caps. . . or for example, adding some small efficient 220u or smaller caps to the amp board (either 220u alone or 220u//1000u is similar and more "ear perk worthy" than 1000u alone). The situation is probably the opposite of a problem. I'd rather "wake up" an amp by putting some 220uF's (or smaller) on it rather than try to damp a shouty amp by putting Only large capacitance on it. Values given work for TDA7294 (will show dramatic results) and many discrete amplifiers, but aren't applicable to all amplifiers. Subjectivity can't always spot the right area to make adjustments; however, it can generate fascinating questions to explore.
So, you could consider it a question.
I just wanted to mention that far out oversize transformer can work a bit like too much of a good thing, or at least cause a redo of the amp board tuning. The problem is easily fixable and the fix (adding smaller, cleaner, clearer caps at the amp) is delightful, but the problem wouldn't be very nice as a mystery.
The difference between resistance and impedance is not important here. The issue, as I see it, is your 'model' of amplifier operation which requires it to modify "its overall impedance" to produce the correct voltage drop from supply rail to output port. This is not a helpful model to use, for most output architectures. It may work for some cases, but like any poor model it can lead you astray. Most output stages act like an imperfect voltage source (ultimately powered from the supply rail), while others act like an imperfect current source; I suppose a few might even act like a variable impedance although that would be less common.gootee said:
I can't even remember now what got us started on this point, but be careful of poor models!
In this case that was just something I needed to understand at the time. So as a conceptual model it seemed to work, there.
But I do often like to think of both power output transistors and power output stages as current valves between the PSU rail and the speaker, which have their resistance/impedance controlled by the small signal input (either voltage or current), to vary the current they allow to flow from the PSU.
It might be interesting to check the difference between Panasonic FC (or Cornell LP) versus a standard cap like Cornell SK/SEK. I believe that the higher ESR of the standard cap might provide convenient ballast for charging, faster charging, of multi-cap arrays.
Far out oversize transformer incurs a slight "dull" problem, the same sort of effect that can be had if 1000uF caps are the smallest power caps on the amp board. This case is fixable by reducing the amp board's bypass caps. . . or for example, adding some small efficient 220u or smaller caps to the amp board (either 220u alone or 220u//1000u is similar and more "ear perk worthy" than 1000u alone). The situation is probably the opposite of a problem. I'd rather "wake up" an amp by putting some 220uF's (or smaller) on it rather than try to damp a shouty amp by putting Only large capacitance on it. Values given work for TDA7294 (will show dramatic results) and many discrete amplifiers, but aren't applicable to all amplifiers. Subjectivity can't always spot the right area to make adjustments; however, it can generate fascinating questions to explore.
So, you could consider it a question.
I just wanted to mention that far out oversize transformer can work a bit like too much of a good thing, or at least cause a redo of the amp board tuning. The problem is easily fixable and the fix (adding smaller, cleaner, clearer caps at the amp) is delightful, but the problem wouldn't be very nice as a mystery.
Thanks for the information.
I'm not sure if higher ESR could be better, for reservoir caps. Higher ESR would cause a larger voltage disturbance for a given charging or discharging current. And higher ESR would cause greater heating of the cap for any given charge or discharge current, lowering the cap's ripple-current rating. But if you find (or have found) that there is some benefit in having a higher ESR, then it's possible that it could outweigh those costs. And the overall ESR could always be lowered by using enough smaller caps in parallel, if that wouldn't undo the benefit you found.
High ESR will damp HF resonances, so may be particularly useful if 'bypassing' is being done. It may also broaden and soften the charging pulses, although transformer resistance is likely to dominate here. High ESR might also increase PSU-related IM, which some might interpret as a 'richer' sound.
Tom, where higher ESR will work is if there are further caps downstream, with much lower ESR, as you implied. Because, what you then have is equivalent to a CRC circuit during the discharge phase, which is well known to improve final voltage stability - the intermediate R is supplied by the high ESR of the first cap(s).
Frank
Frank
Thanks!
It doesn't undo the benefit--at the power supply board, I was comparing standard (higher esr) caps in parallel to high efficiency (low ESR) caps in parallel. The standard caps in parallel, like 5 or 6 of 2,200u mallory sk/sek (any "standard" cap) per rail seem to work better at the power board. Lovely big bass. Yes, I do heatsink bridge rectifiers.
Some extra work to do:
~33n polyester dip cap (quite lossy--those are RC's) at the output of the power board.
and then on the other end of the power interlink cable. . .
One of 3.3u 250v (range 2u to 4.7u) (probably an RC works better) at the point where power interlink cable hooks up to the amplifier board--desired effect is more relaxed mids and amp idle temp cut in half.
Oh, good explanation. Thank you!
Yes, I always favor Low ESR caps on the amplifier board. It is a modest clarity enhancement, but inexpensive and easily done.
One can use a step bigger Low ESR caps and not harm the clarity of the amp (a step bigger size of Low ESR amp board power cap = less forwards without being less clear, so have your cake and eat it too). Apparently amp board power caps are equally important to signal caps?
Last edited:
That "less forward without being less clear" is a good way of describing a lowering of treble distortion, which is exactly what's it all about, IMO. And so, yes, amp board caps are very, very significant in getting good sound - if you think of the amp as just being a valve adjusting the feed from the power supply, as Tom also looks at it, then the importance of clean voltage rails at the precise point where they're controlled by the amp becomes clear ...
Frank
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.