Stephano,
We are anticipating that your test will confirm the rule of thumb and that all DIY folks that followed this thread can stop guessing.
Hi Nico,
Years back I've done measurements on audio codec's in a dead room. We used a simple power amp with 2N3055 in the output.
The amp could deliver 70W. We found that the power supply modulation by current demands affecting the supply voltage was screwing up the performance.
By designing a power supply with CONSTANT output impedance over frequency, this performance could be optimized.
Still have s schematic someware if you want.
Years back I've done measurements on audio codec's in a dead room. We used a simple power amp with 2N3055 in the output.
The amp could deliver 70W. We found that the power supply modulation by current demands affecting the supply voltage was screwing up the performance.
By designing a power supply with CONSTANT output impedance over frequency, this performance could be optimized.
Still have s schematic someware if you want.
There's nothing ever really truly new under the sun ...
Every bit of input helps ...
Thanks,
Frank
I assume that the impedance would have to also be LOW-ENOUGH, not simply constant versus frequency regardless of magnitude. (Actually, if it was extremely low, then it shouldn't matter if it was constant vs frequency or not. But I guess that's just a mathematical curiosity.)
All,
I have played with AndrewT's transformer parameters. I have made per-unitized (scalable) spice transformer components with them. I have run simulations of at least two of the previous cases, with them, to compare to the original transformer model.
Results show that the new model does not let the rails sag nearly as much as the original model.
For a ± 28.28 V square wave into 8 Ohms, with a 44 VRMS 240VA 60 Hz secondary powering each rail, the absolute minimum reservoir capacitance only went from 770 uF (using old model) to 630 uF (using new model).
But for a ± 40 V square wave into 8 Ohms, with a 44 VRMS 240VA 60 Hz secondary powering each rail, the absolute minimum reservoir capacitance went from 6650 uF to 1950 uF. But with the new model the point at which the square wave is down to 0.01 % distortion is within 1000 uF of where it was with the old model.
All data is attached in an xls spreadsheet file.
Cheers,
Tom
I have played with AndrewT's transformer parameters. I have made per-unitized (scalable) spice transformer components with them. I have run simulations of at least two of the previous cases, with them, to compare to the original transformer model.
Results show that the new model does not let the rails sag nearly as much as the original model.
For a ± 28.28 V square wave into 8 Ohms, with a 44 VRMS 240VA 60 Hz secondary powering each rail, the absolute minimum reservoir capacitance only went from 770 uF (using old model) to 630 uF (using new model).
But for a ± 40 V square wave into 8 Ohms, with a 44 VRMS 240VA 60 Hz secondary powering each rail, the absolute minimum reservoir capacitance went from 6650 uF to 1950 uF. But with the new model the point at which the square wave is down to 0.01 % distortion is within 1000 uF of where it was with the old model.
All data is attached in an xls spreadsheet file.
Cheers,
Tom
Tarasque
I and the chaps following this thread would appreciate to having a look see what experience you can bring to the table. Thank you.
I would expect very little difference.
Why?
Because the capacitors are supplying the current. The current demand remains the same so the capacitance should substantially stay the same.
The difference the transformer makes can only be really be effective when the diodes are conducting and that is during the cap re-charge period, of the order of 10% duty cycle. During the other 90% the transformer is very effectively disconnected.
Why?
Because the capacitors are supplying the current. The current demand remains the same so the capacitance should substantially stay the same.
The difference the transformer makes can only be really be effective when the diodes are conducting and that is during the cap re-charge period, of the order of 10% duty cycle. During the other 90% the transformer is very effectively disconnected.
The size of the transformer, like the size of the capacitor, becomes important as you sail closer to the wind, where the difference between the off-load supply rail voltage and the minimum required rail voltage gets small. Elsewhere, a too small transformer will simply get too hot on continuous peak load but maybe not with real music. There will be small 'crossover' region where to some extent capacitor size and transformer size can compensate for each other - commercial designs, for cost reasons, may sit in this region but DIY need not.
PSUD2 simulations tell me transformer resistance plays a much smaller role than I expected. I am still thinking about why this is so. Part of it could be that resistance flattens and spreads the charging pulse, and this means less diode drop so the extra resistance is partially compensated for.
PSUD2 simulations tell me transformer resistance plays a much smaller role than I expected. I am still thinking about why this is so. Part of it could be that resistance flattens and spreads the charging pulse, and this means less diode drop so the extra resistance is partially compensated for.
Me too.
However, Tom already showed that a transformer close to the verge of slightly to small makes a gigantic leap in required capacitance. That doesn't meet linear expectations.
My point:
I'd like to see a spreadsheet where one could enter capacitance and amplifier wattage and then the spreadsheet generates transformer VA as the answer.
Wouldn't that be great?
It is highly applicable because we are shopping for both transformers and capacitors.
Indeed, that is noteworthy. Thank you!AndrewT said:
Last edited:
Slight snag with that. Transformer VA measures how much heat is generated and how well the transformer gets rid of heat. It is only the former (secondary effective series resistance) which affects the calculations we are doing here. VA is used as a proxy for resistance, but what is actually needed is transformer regulation (e.g. 5% for fulll power) - from that Ohm's Law gives the series resistance. So the spreadsheet needs either Rs, or VA and regulation. (Typically, regulation varies somewhat with VA, getting better at higher VA.)danielwritesbac said:
Further snag: the relationship between Rs and PSU output voltage is non-linear and mathematically messy. However, if Rs is sufficiently small then it can be treated as zero or somewhat lost in the diode resistance. For larger Rs and sufficiently large capacitance I have found that a rough approximation is that DC voltage droop is proportional to (current x Rs) to the power 2/3.
8 ohm speaker, and amp wattage sized so that one or more transformers is too small, but not all of them too small.
OK... The 28VAC is RMS so that would give 39.6 V peak, which could give about 37.6 V DC, which would give a maximum peak output voltage across the load of about 34.6 Volts, for a maximum sinusoidal output power of 34.6²/8 = 149.6 Watts RMS.
I'll probably try 100 Watts output power and might adjust it up or down from there.
It would be interesting to sweep the output power, for each VA rating, and see where the minimum capacitance starts to rise disproportionately (if it does). But I wouldn't want to have to calibrate the output's level and DC offset for each step in the sweep. Maybe I can figure something out, for that.
OK, I erred in the max output power calculation, using peak voltage instead of RMS.
The 34.6V peak output voltage would be 34.6/sqrt(2) = 24.5 V RMS so the estimated theoretical maximum RMS output power should be about 75 Watts.
Last edited:
So, if we NOW look back and compare to the calculation method that was outlined in post 1246, i.e. at http://www.diyaudio.com/forums/power-supplies/216409-power-supply-resevoir-size-125.html#post3169077 , we can see that for the 50W and 100W cases, where simulation with the new-and-improved transformer model gave minimum C values of 630 uF and 1950 uF, the calculations predicted 705 uF and 1650 uF.
One is high by a factor of about 1.12 and one is low by a factor of about 0.85. But unless you use your system at its maximum output power rating, the predicted values should prevent gross clipping distortion.
I would probably want to multiply the predicted C value by 3x, to get the distortion metric down into the 0.01% region.
The equation derived in the post linked-to above is:
C_min ≥ i_load / (2 ∙ f ∙ (Vc_pk - Vout_pk - Vamp_min))
where we already know the desired max power and the nominal load resistance and the candidate transformer secondary output voltage, enabling us to calculate all of the terms in the equation above:
i_load_rms_max = √(Power_rms_max / R_load)
Vout_rms_max = √(Power_rms_max ∙ R_load)
Vout_pk = (√2)(√(Power_rms_max ∙ R_load))
Vc_pk = (Vrms_secondary x 1.414) - 1.4
Vamp_min = 3 or so (Min Vce plus worst-case voltage across the 0.22-Ohm R)
f = line frequency, e.g. 60 Hz or 50 Hz
post1338 supersedes.
There is something wrong with the sim if you are predicting 100W into 8r0 from a PSU that is connected to a 28+28Vac transformer.
That 100W into 8r0 typically needs a 35+35Vac transformer.
Here we go again.
There is something wrong with the sim if you are predicting 100W into 8r0 from a PSU that is connected to a 28+28Vac transformer.
That 100W into 8r0 typically needs a 35+35Vac transformer.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.