Power Supply Resevoir Size

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Daniel,
If you do use the Schottky diodes instead of a simple diode bridge would a snubber circuit take care of the rest of the filtering to keep the noise out of the circuit. I am talking about the resistor/capacitor snubbing, not just the capacitor across the diodes. I assume that you would want to run two separate bridge rectifiers one for each channel with the snubber circuits.

Steven
 
OnAudio #799 - feel free to try this; you will attempt to do so only once. regardless of how you choose to pick Zsec for termination, you'll end up dumping metric truckloads of power into your termination - anywhere from 100% to 2000% of rated VA.

an unused winding generates no H field - there is no loop within which current can flow. there is of course an E field, but its pretty well behaved with very low dV/dt = 2*pi*Fac*Vpk. this can be minimised by chopping the leads off, or just folding them back and forth a few times and securing with a cable tie (handy if you want to re-use the xfmr later).

Any other hairy mess (eg nastiness from ac line) that might be present on an unused winding will also appear on the other (presumably in use) windings.

Thanks for your contribution.

"Tonight" might have been a WEE bit optimistic.

I haven't had much time to work on anything in the last few days. I did have a nice set of distortion vs reservoir capacitance plots and a table of the actual minimum C_reservoir for SQUARE-wave signals, with various transformer and load combinations, similar to the stuff I posted for sine signals, but decided I needed to re-run everything, before posting them, which will take a few more hours.

Tonight is also good ;).
 
Daniel,
If you do use the Schottky diodes instead of a simple diode bridge would a snubber circuit take care of the rest of the filtering to keep the noise out of the circuit. I am talking about the resistor/capacitor snubbing, not just the capacitor across the diodes. I assume that you would want to run two separate bridge rectifiers one for each channel with the snubber circuits.

Steven

This power supply is deceptively simple. Try it first ;) If you no like then examine alternatives ;)

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


Also examine the power supply here: To reach such a design requires hours and hours of comparisons. This is vintage wine ;)

http://passdiy.com/pdf/BA2 r1.pdf
 
Daniel,
If you do use the Schottky diodes instead of a simple diode bridge would a snubber circuit take care of the rest of the filtering to keep the noise out of the circuit. I am talking about the resistor/capacitor snubbing, not just the capacitor across the diodes. I assume that you would want to run two separate bridge rectifiers one for each channel with the snubber circuits.
Steven
I don't know anything about running two bridge rectifiers from a single transformer. Perhaps it would complicate something?
P.S.
At post 840, I wasn't talking about bridge rectifiers at all.
P.P.S.
What I was actually talking about is, instead of regulators for v+ and v- private per left and right channels, schottky diode drops instead of regulators for accomplishing virtual dual mono more cheaply.
 
Last edited:
Hi Steven,
yes exactly, I was wondering if plainly move the capacitance where it's needed might be a simple solution to the issues given by parasitic inductance of the rails. I might be completely wrong but I understand the CRC as in Resistor-Capacitor (RC) Filters. In case of a power amp (like mine) with 2 caps/rail I was wondering if it might be the case to get rid one of one of them and move more capacitance (at least 1/2 capacitance for the +rail and 1/2 on the negative rail, with film bypass and a snubber) as close as possible to the output stage. This way we wouldn't have to add a R as the rails "offer" a resistive component "for free".

Hope it makes some sense,

Stefano

Exactly, yes!! You have seen the light!! (All on your own. I'm impressed.)

We definitely need local "decoupling" capacitance for each active device, usually between the power rail input and the load ground return point, usually impossibly-close to the device, if done well-enough.

The local capacitors act almost like small point-of-load power supplies. They are usually meant to supply the fast or large transient current demands, which can't accurately (in time) get through the rail inductance, and which would also induce a relatively-large voltage across the rail inductance. That large voltage is why most people want to "decouple" the load from the rest of the power distribution circuit.

(You also need small-sized local "bypass" caps from power rail to load ground, for each device. But those are for HF stability, to short out (at HF) the hidden HF positive feedback loop through the power rail that almost all transistor amp circuits have. An extra couple of mm in connection length might make them almost useless, by the way.)

With reasonably-long power/gnd rail conductors from the PSU, the lower frequencies can mostly come directly from the PSU, so I don't think that the decoupling caps need to be as large as the power supply reservoir caps, although it couldn't hurt if they were. But the decoupling caps are absolutely essential for the higher-frequency response, which includes both the closed-loop "internal" response as well as fast edges etc in the signal. Depending on the amplifier's transient frequency response and max slew rate, they will probably need to be able to support frequencies up to 100kHz to 300 kHz, and for some equivalent full-scale rise-time, where frequency, f = 1 / ( π ∙ trise ) .

NOTE: You will have to calculate BOTH the required capacitance AND the inductance (impedance, actually) that can be tolerated in the connection length of the capacitance to the decoupling points. (In practice, you might have to just get as much capacitance as close as you can with the lowest total inductance and then calculate (or simulate) backwards to see how you did, and how much rail-voltage disturbance there will be, worst-case.)

I figured out how to do many of those calculations (and also came up with a way to make the PSU impedance, as seen by the active device, as low as you want). See Post 27, at:

http://www.diyaudio.com/forums/solid-state/216409-power-supply-resevoir-size-3.html#post3097232

and especially the links near the bottom of that post, which go to the actual calculations, et al.

NOTE that in many of those calculations, I was using the WRONG value for the self-inductance of a conductor. There are on-line calculators for when you need to be accurate but an accepted rule-of-thumb is 25 nH per inch, or 1 nH per mm. I think I was using 15 nH per inch, at some of those links. (25 nH per inch will make it even more difficult to implement it well, physically.)

Cheers,

Tom
 
Attached are a few results for the Cmin reservoir capacitance, with square waves.

Not enough data for comparisons etc, yet, but:

The 360 VA-per-secondary case for 4 Ohms and 150 Watts looks reasonable.

It looks like 240 VA per secondary is either working hard to get 150 Watts into 4 Ohms, or, the 360VA secondaries finds it to be very easy.

And even 480 VA per secondary looks very close to not being enough for 200 Watts into 4 Ohms, with only a 36 Volt transformer output.

Cheers,

Tom
 

Attachments

  • Schematic_square.jpg
    Schematic_square.jpg
    159.3 KB · Views: 198
  • meas_cmds.jpg
    meas_cmds.jpg
    196.4 KB · Views: 162
  • Data_Table4.jpg
    Data_Table4.jpg
    76.9 KB · Views: 399
  • 4Ohms_360VA_150W_5400uF_data.jpg
    4Ohms_360VA_150W_5400uF_data.jpg
    55.6 KB · Views: 155
  • 4Ohms_360VA_150W_5400uF_4.jpg
    4Ohms_360VA_150W_5400uF_4.jpg
    155.5 KB · Views: 86
  • 4Ohms_240VA_150W_10400uF_data.jpg
    4Ohms_240VA_150W_10400uF_data.jpg
    61.6 KB · Views: 70
  • 4Ohms_480VA_200W_50000uF_data.jpg
    4Ohms_480VA_200W_50000uF_data.jpg
    53.3 KB · Views: 64
  • Markers.jpg
    Markers.jpg
    120.8 KB · Views: 59
  • Markers2.jpg
    Markers2.jpg
    111.6 KB · Views: 60
Last edited:
Tom,
I know that it isn't a 1 to 1 ratio but could you approximately double the values for an 8ohm load and have reasonable values? I was surprised how much higher the Cmin value was when you increased the power output from 150 to about 200 watts at 4 ohms. It looks like a steep gain in capacitance value for a 30% increase in power output.
 
Tom,
I know that it isn't a 1 to 1 ratio but could you approximately double the values for an 8ohm load and have reasonable values? I was surprised how much higher the Cmin value was when you increased the power output from 150 to about 200 watts at 4 ohms. It looks like a steep gain in capacitance value for a 30% increase in power output.

Yeah, but the transformer VOLTAGE is then apparently way too LOW, for that (200 Watts), so it needs a whole lot more C just to keep the voltage and current ripple in the right zones, otherwise the signal gets bashed. (Or, maybe the crappy transformer model has something to do with it. I haven't done the calculations for that case.)
 
Last edited:
Tom,
I know that it isn't a 1 to 1 ratio but could you approximately double the values for an 8ohm load and have reasonable values? I was surprised how much higher the Cmin value was when you increased the power output from 150 to about 200 watts at 4 ohms. It looks like a steep gain in capacitance value for a 30% increase in power output.

For an 8 Ohm load instead of 4 Ohms, you would divide them by two, rather than double them.

See the results for sine signals, which also include some 8-Ohm cases, in Post 740, at:

http://www.diyaudio.com/forums/solid-state/216409-power-supply-resevoir-size-74.html#post3134179

Cheers,

Tom
 
So, DF, Terry, Fas, and other experts: Square-wave testing, with peak values equal to the peak sine values previously used, as in post 846, should be comparable to using a sine load current's or voltage's peak value instead of its RMS value, as the DC load current or voltage when calculating ripple and the needed reservoir capacitance using the standard approximate formulas, correct?

So it's like using double the power as a worst-case design margin, and throwing in huge fast transients, too.
 
Last edited:
So something like a 600VA transformer would handle the ripple problem with half the capacitor value then?

Not sure what we're comparing to. But increasing the VA "can", sometimes, enable lowering the minimum capacitance while still not bashing the signal. But that only works well if you're in a comfortable operating area for the transformer output voltage.

You want enough headroom, with the transformer output voltage. Then you can use reasonable and predictable capacitance values, and a reasonable and predictable VA rating.

If I had cranked up the input voltage for the 200 Watt case, the minimum required capacitance would have dropped quickly. At some point, I might try to parameterize the secondary voltage and fix one or two of the other parameters and see what other types of plots might be useful or enlightening.
 
So, DF, Terry, Fas, and other experts: Square-wave testing, with peak values equal to the peak sine values previously used, as in post 846, should be comparable to using a sine load current's or voltage's peak value instead of its RMS value, as the DC load current or voltage when calculating ripple and the needed reservoir capacitance using the standard approximate formulas, correct?

You could do that but then each half would conduct unrealistic amounts of current when averaged over time, requiring much more of the caps and PSU.

What I do with squarewave testing is limiting the gain untill the square wave average output power would equal that of a full power sine.

But the chance of running into full power square waves is slim in audio, unless you have an affection for playing loud chip tunes of computers of old ;)
 
298362d1346132498-power-supply-resevoir-size-data_table4.jpg

Do I read it correctly if I assume that 5400uF is enough for a 300 watt single rail amplifier if it also has a 360va transformer? Um, my understanding needs proofed a bit prior to making a graph with the information.

It also seems interesting to compare with some very low power amplifiers. As for line graphs, it seems to need 3 graphs to get the small, correct, large transformer selections represented (in addition to capacitance per watts). I'd do it with an all small transformer graph, an all correct correct transformer graph and an all large transformer graph on a 3d (3 color) type line graph for getting the transformer information visible. If we get enough datapoints, we can fill in the gaps, but the low power data points are currently missing.
 
But the chance of running into full power square waves is slim in audio, unless you have an affection for playing loud chip tunes of computers of old ;)
Or popular music: One Direction - What Makes You Beautiful - YouTube Wait for the chorus. They brick walled it. I think that coloration indicates first cousin of a square wave. Since it was a hit, many others will copy--and there is your frequently occurring square wave. Duck and cover. :)
P.S.
But I appreciate this audio track because it is challenging to decompress in live time during replay. So, more learning for me.
 
NOTE: You will have to calculate BOTH the required capacitance AND the inductance (impedance, actually) that can be tolerated in the connection length of the capacitance to the decoupling points. (In practice, you might have to just get as much capacitance as close as you can with the lowest total inductance and then calculate (or simulate) backwards to see how you did, and how much rail-voltage disturbance there will be, worst-case.)

Hi Tom,
first of all you I didn't but apply the logic you pointed out, so no honor at all...

Anyway, I shared the idea with an ex University fellow of mine and he warned me against some possible oscillation of the amp saying he couldn't say anything more precise (without being able to measure amp behavior). Now, it's a long time since my last Bode diagram, H(S) poles and zeroes distribution, and so on, and I can't recall/imagine how just moving half of PSU caps near the output device (in my case 2SC2922 and 2SA1216) would add anything to zeros and poles of the transfer function, so impairing stability... (while adding new caps such as bypass and that one in the snubber do affects H(S))

Would some of you be so kind to show me (point to links, or whatever) how/why this might happen and what is possible to do aside from using a signal generator feeding square wave trains and an oscilloscope? :confused:

Thus said I need anyway to buy the latter... ;)

Thank you,

Stefano
 
Last edited:
Hi Tom,
first of all you I didn't but apply the logic you pointed out, so no honor at all...

Anyway, I shared the idea with an ex University fellow of mine and he warned me against some possible oscillation of the amp saying he couldn't say anything more precise (without being able to measure amp behavior). Now, it's a long time since my last Bode diagram, H(S) poles and zeroes distribution, and so on, and I can't recall/imagine how just moving half of PSU caps near the output device (in my case 2SC2922 and 2SA1216) would add anything to zeros and poles of the transfer function, so impairing stability... (while adding new caps such as bypass and that one in the snubber do affects H(S))

Would some of you be so kind to show me (point to links, or whatever) how/why this might happen and what is possible to do aside from using a signal generator feeding square wave trains and an oscilloscope? :confused:

Thus said I need anyway to buy the latter... ;)

Thank you,

Stefano

My assumption has to be that he misunderstood what was meant. He must have assumed that you were putting the caps in the small-signal path or in the feedback loop, which would mean that he completely misunderstood.
 
Do I read it correctly if I assume that 5400uF is enough for a 300 watt single rail amplifier if it also has a 360va transformer? Um, my understanding needs proofed a bit prior to making a graph with the information.

It also seems interesting to compare with some very low power amplifiers. As for line graphs, it seems to need 3 graphs to get the small, correct, large transformer selections represented (in addition to capacitance per watts). I'd do it with an all small transformer graph, an all correct correct transformer graph and an all large transformer graph on a 3d (3 color) type line graph for getting the transformer information visible. If we get enough datapoints, we can fill in the gaps, but the low power data points are currently missing.

It says 150 Watts PER RAIL (i.e. per voltage rail; one positive and one negative, for a total of 300 Watts if BOTH rails are operating).

The RMS Power (dissipated in the load) of the square wave is done as if the negative portion were folded around the axis to be positive. With only one rail, the negative parts would be missing so the power would be half of the total given (half of 300 W).
 
Tom @ #851:

Yeah, I think that is precisely what is required. And here's a justification for it using an entirely different approach: With a rectifier conduction interval on the order of 1-2ms, any LF output (eg 20Hz) is very long by comparison, and its also quite long compared to the cap conduction interval (6~8ms). So the rectifier/filter needs to be designed to handle the peak power, not the average power. And the peak power is twice the average (assuming sinusoidal output), so Voila, there's your factor of two.

And thats why the transformer leakage inductance is so important. lovely work BTW, showing quite clearly how excess leakage requires stupidly large amounts of capacitance to reduce rail droop.

Given that very few builders will have custom-made magnetics, it probably makes sense to develop a design procedure based on xfmr parameters. If you normalise ("Per-Unitise" as power systems engineers like to say) the figures you have, you'll pretty much end up with a set of design nomographs (by varying the output power you have achieved the same effective results as varying the transformer parameters). Add in a little explanatory text (eg how to measure & normalise transformer parameters) and its an awesome design tool. And of course ROTs are as simple as picking points on the nomograph that give the "best" (whatever that means) bang for ones buck.
 
Originally Posted by gootee
So, DF, Terry, Fas, and other experts: Square-wave testing, with peak values equal to the peak sine values previously used, as in post 846, should be comparable to using a sine load current's or voltage's peak value instead of its RMS value, as the DC load current or voltage when calculating ripple and the needed reservoir capacitance using the standard approximate formulas, correct?

You could do that but then each half would conduct unrealistic amounts of current when averaged over time, requiring much more of the caps and PSU.

What I do with squarewave testing is limiting the gain untill the square wave average output power would equal that of a full power sine.

But the chance of running into full power square waves is slim in audio, unless you have an affection for playing loud chip tunes of computers of old ;)

Actually, I already DID do that. I was merely asking about the relative equivalence to the calculations for sines.

Using the same average output power for the square wave as for the sine wave was discussed and tried already. It results in nothing interesting being learned, since the voltage peaks are then too low to challenge the power supply.

The idea behind using the full-peak square waves as the test signal was to find something close to the worst-possible case that the reservoir caps and power supply would ever have to be able to handle, such that if the worst case could be handled without the possibility of any distortion that was due to the power supply "choking on it", then maybe one could "truly" be CERTAIN that any and all music that did not exceed that peak voltage could also be handled without any PSU-induced problem.

Also, conveniently, the square waveform covers half of the total time during which a sine peak could have occurred, requiring testing of only two phase angle cases, i.e. 0 and 180 degrees, whereas with a sine as the test signal we would need to test at many, many phase angles, to try to ensure that the worst-case alignment of the test peak with the PSU's charging pulses and ripple voltage waveform's troughs had actually gotten tested.

A single sine seems to obviously not be the worst case for testing the PSU's robustness. All we would have to do to make it worse would be to play more than one sine at a time, so that there were more peaks in each unit of time. If we had enough tones playing at once, with different relative phase angles, then the peak level would be completely filled with sine peaks and would look like the top of a square wave.

Actually, it could be much worse than that, and far worse than even the square wave, whenever there were multiple tones and their peaks happened to line up with each other in time and all overlap, i.e. all be in phase, even if just for an instant. Then their peak values would be SUMMED, and the resultant peak voltage could try to be MUCH higher than we anticipated, very possibly far above the transformer output voltage. But I guess we are allowed to assume that the volume control will be set at an appropriate level, to avoid such problems.

At any rate, any particular frequency of square wave is only about HALF as bad as it could very-easily get. All we'd need, to make it immediately almost twice (2X) as difficult for the reservoir caps would be a square wave of half the frequency, which would draw from the caps for twice as long at a time.

We should also expect that a fixed maximum-peak-value DC signal voltage could be applied. But I don't think we'd be worrying about the PSU distorting its sound, if that happened.

So very-low-frequency square waves with peaks at the expected peak sine voltage seem like a pretty good test. The lowest frequency of square waves that can be handled gracefully might even be a good additional general "figure of merit" to use, for a PSU.

For a particular transformer voltage and V-A rating, it looks like we should be able to literally "assume", with high confidence, that we should never need to use more than the Cmin value that worked during the peak-level square-wave testing.

I will also note that this type of test seems like it might be relatively easy and practical to implement on our test benches, with a square wave small-signal source and an oscilloscope: Just add capacitance until the amplifier's output square wave looks good-enough. Then it should be "guaranteed bullet-proof" for real music signals.

Of course we still need to figure out and agree on what frequency of square wave is "low-enough", for this type of test, and what is an acceptable maximum amount of distortion of the square wave.

In my most-recent (square wave) testing, I more-or-less tried to keep any individual instant's peak deviation (i.e. the max error) of the output voltage (versus the theoretical ideal output voltage) to less than 0.1% of the expected ideal peak output voltage, while also rejecting any PSU-induced "gouging" of the output voltage waveform that further-deviated by more than 1 mV (while still also keeping the maximum error within 0.1% of the ideal peak output voltage, including the 1 mV or less).

The only "problem" I still perceive, throughout the above discussion, is the uncertainty of the degree of "overkill" in the Cmin values obtained with the square-wave method described. (i.e. "A Cmin that's too large is not the real Cmin.")

Nico's ORIGINAL GOAL for this thread was related to "how much reservoir capacitance SHOULD be used" (and the possibility of developing a "rule of thumb" for that).

So I thought that to just BEGIN the process of finding that answer, maybe I should start by finding a way to determine the absolute-minimum C that would work, i.e. without allowing "obvious" PSU-induced problems to occur.

The idea was to try to be able to determine at least one BOUNDARY in the range of candidate C values, and hope that maybe we could narow it down further, from there, somehow, later.

So, at this point, even without having complete data, yet, for the many possible configurations of transformer, load, and output power specs, we think that we can at least determine, for a given viable configuration, a minimum C value that will prevent the possibility that the power supply, itself, could mangle the music signal due to insufficient capacitance. And we think that we will also be able to determine what are "good enough" combinations of transformer output voltage, transformer VA rating, reservoir capacitance, load resistance, and output power, but with the "good enough" being ONLY in terms of the PSU being guaranteed to not be so badly designed that it might betray its only purpose and possibly be unable to even support the faithful reproduction of the music signal, at up to the rated maximum output power.

In some ways, the progress so far seems embarrassingly minor, given the time and effort that have been applied, so far. And we don't even have nearly enough data produced, yet, to be able to actually use it as I just described.

BUT, we MIGHT have almost enough data produced to be able to check whether or not the quick-and-easy standard "shortcut" formulas can always (or even just usually, or often) predict comparable worst-case C values, which would be an extremely-nice result that would save a lot of simulator time, and simplify things a lot.

(So hey, DF96 (or anyone): "Got equations?" How do their predicted C requirements line up against the simulation results? Or... what fixed DC load voltage or current gives the same C value, for one [or each] of the cases?)

My point was going to be that while being able to find a lower-bound for C_reservoir seems necessary, and important, it is rather coarse, in the overall scheme of things, here. I am wondering what should be the next step, if we eventually want to be able to have a good and repeatable method for determining what reservoir capacitance "should" actually be used?

I guess it's obvious that having a way to find an "upper bound" for C would be nice. If we knew how to measure the "goodness" of the amplifier, we could then know how to raise C until the measurement's improvement rate versus the C's rate of increase could no longer justify raising C any more, or something like that, for whatever criteria we wanted to use (assuming that such a diminishing-returns "convergence" phenomenon would indeed occur).

In reality, such an effort could be quite complex and even just developing the methods for doing it, either with simulations or physical testing, might be difficult to achieve with a comfortable (and believable) confidence level.

BUT, it helps if we remember that the goal would be to develop a reasonable "rule of thumb", early in that process, we hope, rather than developing the full process and then churning out an exhaustive array of test configurations to guide us, instead.

Or is a lower bound all that is needed?

On the other hand, we DO ALREADY have the simulation setup that can calculate the sine signal THD or square wave signal distortion as a function of C_reservoir. And I think people already do know what levels of distortion are low-enough that no further improvement could be audible. So that might suffice for judging the steady-state response. Maybe "all" we need is some way(s) to measure the transient-response's relative fidelity, and probably a way to relate that to audibility. Perhaps the already-existing data on the lowest-discernable time-of-arival differences for sounds reaching ears could be used to declare an initial minimum required transient-response accuracy (i.e. the maximum acceptable timing deviation). And non-linearity of phase errors versus frequency would show up as ringing on a square wave. C_decoupling, near each active device, would also have to be considered.

Sorry to have belabored all of that for so long. It's been a rough few weeks so I'm probably about "done", for tonight.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.