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.
Rather than buying a lot of pricey caps to a formula, try seperate PSUs with transformers for each channel and even seperate PSUs for the output stage and input+VAS stage of a given channel. I'd say dimishing returns kicks in more quickly for more caps than it does seperating the PSUs. A secondly, cheap amp boards with highly expensive caps just seems dumb.
 
That's just about equivalent to specifying a 180W into 8ohms, (53.7Vpk & +-60Vdc supply rails) amplifier and never using it @ >=100W into 8ohms, (40Vpk).
What a waste of resources and design effort.


Most "Rule of Thumb" guidelines will do far better than applying "-20V" to the supply rails!

If one were working with a 60W amplifier (+-36Vdc supply rails) then the "-20V" rule becomes silly.

The essential point is that the rails will sag under load due to the low frequency and limited duty cycle of the charging pulses. If you want full power output, then you need the unloaded rail to be much higher than the maximum output swing. How much higher depends on how long you want to sustain full power. I was really surprised by just how much the rails sag when I went and looked at it in simulation. Look at the simple simulation in the link that I posted earlier.

You bring up a good point that the unloaded rail voltage headroom above maximum output voltage shouldn't be expressed as a fixed value, perhaps expressing it as a percentage of the maximum output voltage - 120% to 140% makes it more generally applicable.
 
You may just have a good opportunity here Harrison.
Give each of these gentlemen one completed amplifier with exact same components then allow each of these gurus to paste on his KICK-*** power supply.

Now rotate to the left and each person gets his colleagues amp to listen and measure, then one final rotation - thereafter let them report their findings both objective and subjective..

The resultant is the best of the best rules of thumb compared objectively and will conclude this thread
 
That's just about equivalent to specifying a 180W into 8ohms, (53.7Vpk & +-60Vdc supply rails) amplifier and never using it @ >=100W into 8ohms, (40Vpk).
What a waste of resources and design effort.


Most "Rule of Thumb" guidelines will do far better than applying "-20V" to the supply rails!

If one were working with a 60W amplifier (+-36Vdc supply rails) then the "-20V" rule becomes silly.

with 10000uF per rail, at 60W into 8 ohms at 100Hz, 36V rails sag to the point that they limit the output in 50 msec. That's with ideal transformer, diode, and capacitor models, and an ideal output stage.

More capacitance means you can keep it up longer but not all night. :D Once you account for the multiple Vbe drops in the output stage and real components then to sustain 60W you need rails more like +/- 40V.
 
with 10000uF per rail, at 60W into 8 ohms at 100Hz, 36V rails sag to the point that they limit the output in 50 msec. That's with ideal transformer, diode, and capacitor models, and an ideal output stage.

More capacitance means you can keep it up longer but not all night. :D Once you account for the multiple Vbe drops in the output stage and real components then to sustain 60W you need rails more like +/- 40V.

While generally agreeing, it should be noted that ideal components might not necessarily always be a best-case scenario, in simulations. Sometimes, they exacerbate potential problems, or behave differently-enough that the results are invalid.

There are still real tradeoffs to be considered, between more capacitance and higher rail voltage, and between more and less capacitance. With enough capacitance, maybe you actually could keep it up forever. (But think about calling a doctor if it stays up for more than four hours straight.)

But one key to "keeping it up" might be the proper use of "decoupling capacitance" (otherwise known as "Viagramphetamine" :).

The main criterion, still, for me at least, is "accuracy of reproduction".

For example, I would insist on having an amp that would clip near its highest-possible power levels (which I typically wouldn't use, anyway), if that meant that it could be more accurate than would be possible with more reservoir capacitance.

Cheers,

Tom
 
You may just have a good opportunity here Harrison.
Give each of these gentlemen one completed amplifier with exact same components then allow each of these gurus to paste on his KICK-*** power supply.

Now rotate to the left and each person gets his colleagues amp to listen and measure, then one final rotation - thereafter let them report their findings both objective and subjective..

The resultant is the best of the best rules of thumb compared objectively and will conclude this thread

hahaha Nico, you think these instructions would successfully detroll the DIY forum? :up:
 
Last edited:
Its proportionally cheaper to up the transformer VA than capacitance. Online you'll find people doing classic class A amps with capacitance in the 100,000s UF with a single transformer. If mono blocks are better due to the seperation of PSU and signals, then to me thats the direction to go in, to which rules of thumb and technical analysis can then be applied.
 
Finally, I am ready to start some actual comparison testing. I now have the error plotting working well. It gives the difference, at each point in time, between the output voltage and what the output voltage should be.

With that now in place, subtle differences in reproduction accuracy should be easily seen. I plan to first compare a small but broad set of sample reservoir and decoupling capacitance configurations. Then we can "go from there".

To properly-implement the error plotting, I had to try to remove the other sources of error, i.e. those that were always present due to the amplifier itself. So I replaced the amplifier output stage with an idealized amplifier made from spice Behavioral Current Sources (B-sources). I was able to use the u() "unit step" function to provide a "polarity" factor that is 1 when the input voltage is on one side of zero, and is zero otherwise. That way, I could make perfect push-pull current sources, driven by equations.

The error level, with 3x 10000 uF reservoir capacitance (and minimal decoupling) is now typically in the low millivolts, at up to +/- 60 Volts output.

I also corrected the problem with the transformer output labels, which probably meant they were wired oddly. I have only tested that with this latest circuit, so far, which unfortunately also has other changes: parasitics for two inches of conductor between the rectifiers and caps, and between the "center tap" and caps' ground, and also has 10 uF of decoupling capacitance to make the B-sources behave better. So it's difficult to say if the differences are due to the transformer correction or due to the other changes. I will go back and try the corrected transformers with the original circuit, eventually.

But note that with the current model, the rail voltage dips even more (about 33% more) than it did before.

Below is the latst schematic, sample plots (new and old simulations) for comparison, and, a zip file with everything needed to run the same simulations, except for the WAV file and the LT-Spice software itself (free download from linear.com !).

P.S. With this simulation, I think that the LT-Spice "Alternate Solver" runs much faster, and better.

Cheers,

Tom
 

Attachments

  • PSU_Schem_30kuF_freq-dep_45C_6inch-rails_Ideal-Amp.jpg
    PSU_Schem_30kuF_freq-dep_45C_6inch-rails_Ideal-Amp.jpg
    65.1 KB · Views: 246
  • gootee3.zip
    13 KB · Views: 40
  • PSU_30kuF_freq-dep_45C_6inch-rails_Ideal-Amp.jpg
    PSU_30kuF_freq-dep_45C_6inch-rails_Ideal-Amp.jpg
    162.5 KB · Views: 234
  • PSU_30kuF_freq-dep_45C_6inch-rails.jpg
    PSU_30kuF_freq-dep_45C_6inch-rails.jpg
    142.9 KB · Views: 231
Last edited:
You may just have a good opportunity here Harrison.
Give each of these gentlemen one completed amplifier with exact same components then allow each of these gurus to paste on his KICK-*** power supply.

Now rotate to the left and each person gets his colleagues amp to listen and measure, then one final rotation - thereafter let them report their findings both objective and subjective..

The resultant is the best of the best rules of thumb compared objectively and will conclude this thread

With more funding , this is possible. as is since the BOM is up and standardized, they may be able to assemble accurately.
 
Rather than buying a lot of pricey caps to a formula, try seperate PSUs with transformers for each channel and even seperate PSUs for the output stage and input+VAS stage of a given channel. I'd say dimishing returns kicks in more quickly for more caps than it does seperating the PSUs. A secondly, cheap amp boards with highly expensive caps just seems dumb.

Its proportionally cheaper to up the transformer VA than capacitance. Online you'll find people doing classic class A amps with capacitance in the 100,000s UF with a single transformer. If mono blocks are better due to the seperation of PSU and signals, then to me thats the direction to go in, to which rules of thumb and technical analysis can then be applied.

Maybe the formula will end up telling us it's better to buy just a few cheap caps. ;-)

Your opinions may be correct. We can't say, yet. But here in Nico's thread we are trying to actually determine what tends to be better and what does not, and possibly even uncover a "formula" or two for later, so we don't have to do all of this work again every time, or just rely on opinions and hearsay. But suggestions for what to try could be useful. Anyway, it's been a slow process so please be patient.

Cheers,

Tom
 
Last edited:
But one key to "keeping it up" might be the proper use of "decoupling capacitance"

The main criterion, still, for me at least, is "accuracy of reproduction".

If you mean "decoupling capacitance" to mean additional reservoir capacitance placed local to the output transistors, it will help with accuracy of reproduction due to minimizing the effects of reservoir ESR and the inductance of the supply leads. But the rails will still sag, a little slower due to the added capacitance, but to the same level as without for the same output power.
 
I am still not understanding what is happening inside that transformer model.
How can a 25Vac transformer give 73Vdc, inless the regulation is enormous? We know that transformers used for power amplifier PSUs have a regulation between 3% and 10%, that limits the off load voltage.
Because, as Tom stated, the sim has been fudged, the transformer is actually working at well above its design parameters, eg, the primary voltage. This has only a marginal effect on getting reasonable results in the sim, but if you want to play completely fair simply add another, same spec'd, transformer in each leg and wire the secondaries in series. Then it's all ridgey didge, and doesn't cost any moolah, either ... ;)

Frank
 
Welcome back, Andrew! We really missed you!

I don't know what makes transformers have regulation. But the model is probably way too simple to account for that.

But the way a 25-volt-output transformer gives 71 Volts out is for it to get about 342 Volts p-p as input. It's just a simple matter of turns ratio. Yes, it was not meant for this job. Yes, I used it only because it was convenient and I thought it would be better than an ideal voltage source and an ideal transformer model.

Or are you wondering how I am getting 71 Volts when I only put in 321 Volts (instead of 342)?

If someone looks at the transformer model schematic (in post 400 of this thread, which is at http://www.diyaudio.com/forums/solid-state/216409-power-supply-resevoir-size-40.html#post3119281 ) and tells me what component values to change to make it more suitable for higher current and voltage than the one the measurements came from, I can change it with just a few mouse and keyboard clicks.

Better yet, if anyone has a variac and a suitable transformer to measure, it only takes a few minutes.

Cheers,

Tom
 
Last edited:
Like this:

AltTrans.gif

Trouble is, now have double the resistance in the secondary windings feeding the rectifier, so the sag is even worse at full load, down to around 53V. Which is why you need a transformer with sufficient VA rating, giving better regulation, more cost! Or, use switching mode supplies, etc.

Frank
 
Last edited:
To properly-implement the error plotting, I had to try to remove the other sources of error, i.e. those that were always present due to the amplifier itself. So I replaced the amplifier output stage with an idealized amplifier made from spice Behavioral Current Sources (B-sources). I was able to use the u() "unit step" function to provide a "polarity" factor that is 1 when the input voltage is on one side of zero, and is zero otherwise. That way, I could make perfect push-pull current sources, driven by equations.
Just a cautionary thought, Tom ... my belief is that real circuitry should be used to load the PS, otherwise one's thinking can possibly be led down an unnecessarily torturous detour. Ultimately, what one comes up with in terms of results has to relate to real, on the bench, bits of circuitry, because that's what the whole exercise is about ...

Cheers,
Frank
 
Frank & Tom:

If you remove the entire amplifier & modulated load from the sim, and replace with a resistor, you will see the droop caused by the transformer "regulation" - which will be mostly due to the leakage inductance and the narrow conduction angle. I'd also look carefully at the rectifier model as well - its probably fine, but there are, for example, 10BQ100 spice models with 100R slope resistance (I found them the hard way)
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.