Hi!
I'm trying to understand how to design a class B power amp, and I am a little confused with the method to define the peak of power dissipated by the output transistors.
All the mathematical methods I have seen are based on sinewaves.
But it look to me that with an assymetrical heavilly clipped signal, like you could have from a fuzz, things could be really worse. With a big phase shift, you could have all the voltage (Vcc - Vee) accross a transistor in the same it would pass full curent. So Pd(peak) could be I(peak) * (Vcc - Vee). Is this really pessimistic?
Here, https://sound-au.com/soa.htm, the result look the same, but the method look strange to me :
"Having discounted the idea of any 'rules-of-thumb', I'm going to give you one anyway . Let's assume that you want to deliver 100W into 8 ohms, so you need a power supply with ±42V rails (I'm going to ignore losses here). The amp must also be able to drive nominal 4 ohm loads, so expect the minimum impedance to be 3 ohms. Worst case (resistive load) dissipation is therefore ...
I = V / 2 / R = 21 / 3 = 7 Amps
P = V / 2 * I = 21 * 8 = 168 Watts (peak)
This accounts for the resistive part of the load, and as we saw above, the reactive part of the load causes dissipation to double. Just like second breakdown, we aren't interested in the average dissipation - this influences the size of heatsink needed, but not the transistor's safe area. Therefore, Ppeak will be ...
Ppeak = P * 2 = 168 * 2 = 336 Watts"
How could the transistor pass full current with the output being at Vcc/2 with a resistive load?
Edit : this is only half current.
What would you assumed to be the whorse impedance and phase shift from a 8ohms guitar speaker please?
I am right assuming at the lower impedance, the load would look only resistive, so there won't be phase shift, so the the worse cases couldn't come hand in hand? I have not found anything about this yet.
Thanks a lot for your help!
Please apologise my poor English.
Check this thread over at MEF.
Link: https://music-electronics-forum.com/forum/fun-with-computers/36422-testing-zip-files?t=35493
If you are not a member there, I think you will have to join to see the attachments. What is shown is an X--Y movie clip of current and Voltage in a speaker when connected to a solid state amp banging rail to rail.
When the trace is above mid screen, current is positive flowing through the transistor connected to the + rail. See how some of the time it stays above the mid screen line all the way from the right side of the screen to the left side ? The inductance of the speaker keeps the + side transistor on even when the Voltage goes negative.
Hi Loudthud!
Thanks for your answer!
I have join MEF, but I am having hard time to fine the thread you are talking about. Have you an idea of what keyword should I search, please?
Sorry, I thought I had included the link. I edited the post above but here it is again
Link: https://music-electronics-forum.com/forum/fun-with-computers/36422-testing-zip-files?t=35493
Thanks Loudthud!
It is nice to see real measurement of what append with a scope! :) Is there something else than phase shift between voltage and current due to speaker reactance that I should see here?
To really study this subject, you would need a lab full of equipment. You would need to simultaneously monitor Voltage and current in a transistor, then multiply the two to get power. In a practical sense you would need to look at the instantaneous power over a time scale and compare that to the SOA limits from a transistor data sheet. I say practical because you don't want a 100 W power amp with 20 output transistors.
Attached is the SOA graph from an On Semi power transistor. If you look at the 1 Second line, it stays at 250W out to about 50V. I have extended the 250W line to show the Second Breakdown region in red. The 1 Second line is the one bordering the bottom of the red zone. As the Voltage increases beyond 50V, the allowable power is reduced. The 50V limit for full power is common to many transistors. Some have a lower limit, a few higher up to 80V. Let me know if you find any bypolar transistors that are higher than 80V.
To keep your output transistors out of the red zone, you need to limit their Voltage and that means stacking them in series. Otherwise you need about twice as many transistors to be safe.
MOSFETs don't have Second Breakdown so that is a big incentive to use them.
Edit: Changed pic slightly to make it more clear.
Thanks again Loudthud!
"To really study this subject, you would need a lab full of equipment. You would need to simultaneously monitor Voltage and current in a transistor, then multiply the two to get power. In a practical sense you would need to look at the instantaneous power over a time scale and compare that to the SOA limits from a transistor data sheet. I say practical because you don't want a 100 W power amp with 20 output transistors."
I was trying to have a theorical approch to avoid this, but as you say, this lead to lot of output transistors.
Do you think Spice simulations would differ a lot from reality?
What I find annoying, is what is a practical signal for testing a guitar amp? An hifi amp won't have to produce continuous square waves. But for a guitar amp, for exemple, a larsen with a super fuzz type pedal is a possible situation, and look really more painfull for the amp.
Should I think of it as "How could I suppidly distort the signal before the VI limiter began to conduct?"
Furthermore, what would you use for practical load? Results would vary a lot in fonction of the load used for measurments, with different impedance dips and phase shifts between real speakers.
"Some have a lower limit, a few higher up to 80V. Let me know if you find any bypolar transistors that are higher than 80V."
On SemiNJW21194G / NJW21193G seem to have 100V / 110V limit. (SOA graph in attachment)
https://docs.rs-online.com/420b/0900766b8126cc12.pdf
Quote from: shasam on September 10, 2020, 05:30:09 AM
...
Do you think Spice simulations would differ a lot from reality?
What I find annoying, is what is a practical signal for testing a guitar amp? An hifi amp won't have to produce continuous square waves. But for a guitar amp, for exemple, a larsen with a super fuzz type pedal is a possible situation, and look really more painfull for the amp.
...
The biggest drawback to using simulators is that they will consistently give you an ideal result; that is, any given part will have real-world variations due to many factors, but the simulation will run at exact specs every time. That can lead to disappointing results with real parts, but as long as you're aware of that, it's not so bad.
In LTSpice (and probably others), you can use a sound file as input to your simulation, so I sometimes record a 1 or 2 second clip of a strummed open E, or plucking of a high note as "real-world" input (the clips are short because the simulation with a sound file takes quite a bit more time to process, as you can imagine). Your guitar and pre-amp will NEVER produce a perfect sine wave, so though a sine wave can be used as a useful baseline to see if the circuit is doing what you think it is supposed to do, it won't show you how the circuit will react to an actual guitar signal.
If you are doing your testing on a "real-world" amp, I would suggest investing in a looper pedal. That way you can strum a few chords into it, start it looping, then jack it directly into the amp and it will repeat until you stop it while you take measurements or gawk at an oscilloscope trace. You could even run it through a pedal of your choice.
QuoteFurthermore, what would you use for practical load? Results would vary a lot in fonction of the load used for measurments, with different impedance dips and phase shifts between real speakers.
There are circuits that simulate a speaker load with varying impedances, etc. that are very close to how a real speaker will affect the operation of the circuit it's connected to. Try this one:
https://www.aikenamps.com/index.php/designing-a-reactive-speaker-load-emulator
Whether you build a real one, or run it through a simulation, I think something like this will work fine. Tonal differences between speakers are often down to the material and construction of the speaker cone itself; the engineering of the power amp is only concerned with the magnet/coil part.
Hi Edvard! Thanks!
What annoy me is, with real world measurments, some real speakers could be worse than the ones I have under my hand now, and input signal could came in wicker shapes that the ones I generate for testing. We can simulate the worse theorical case but this could be really worse than real practical situation so being useless. So I don't know what to really use...
"Tonal differences between speakers are often down to the material and construction of the speaker cone itself; the engineering of the power amp is only concerned with the magnet/coil part."
Could you please elaborate a little more on this? A bass reflex cab would generate other phase shift, why shouldnt we care about this?
I have seen on some websites that a phase shift of 45° it the worse case for output stage (ignoring 90°), as this exemple:
"As discussed in the Audioholics article, the key phase angle (denoted as Greek phi Φ) to keep in mind is +/-45°. This is the point where the amplifier must dissipate double the power compared to just a purely resistive load while the speaker receives only half the power. On either side of that key phase angle, the amplifier demand is reduced. Knowing this allows the amplifier designers to ensure that they can deliver adequate power and maintain the temperature of the devices at safe levels. I've seen my own speakers vary between +/-70° at the extremes with +/-60° being more common."
Douglas Self stated : "Average device dissipation also increases as the load angle increases. A 45 load increases average dissipation by 1.4 times, and a 60 load by 1.8 times.
Peak device dissipation increases more rapidly than average dissipation as the load angle increases.
A 45 load increases power peaks by 2.7 times, and a 60 load by 3.4 times."
From my simulation too (screenshot in attachement), 60° is worse than 45°. Is the first quote false, or am I missing the point?
Likely over my head but I'll ask.
What is it you are trying to design?
Re the last part of the link to SOA by Rod Elliot, (Section5 Protection Schemes)
To my mind that is how you cover the extremes.
I doubt any circuit is totally bullet proof but if it's just a 100Watt guitar amp then considering how many millions of 100watt amps get flogged and survive for years ,,, maybe you are over thinking the design. :-\
Or have I completely missed the point? xP
Phil.
Hi Phatt!
Quote from: phatt on September 18, 2020, 10:19:23 AM
Likely over my head but I'll ask.
What is it you are trying to design?
I am trying to design a discret class B power amp, with CFP output. Something in the 15 -20 Watts range for beginning.
Quote from: phatt on September 18, 2020, 10:19:23 AM
Re the last part of the link to SOA by Rod Elliot, (Section5 Protection Schemes)
To my mind that is how you cover the extremes.
So, I should place the number of output devices to endure most real life situations, and put I-V limiter for protection against anything worse?
Quote from: phatt on September 18, 2020, 10:19:23 AM
I doubt any circuit is totally bullet proof but if it's just a 100Watt guitar amp then considering how many millions of 100watt amps get flogged and survive for years ,,, maybe you are over thinking the design. :-\
Or have I completely missed the point? xP
I am probably over thinking it ;)
Well an LM1875 can easy produce 15~20Watts.
Just google it, there are tons of circuits out there using that chip, a few to be found here on SSguitar
The LM1875 delivers 20watts into a 4Ω or 8Ω load
from ±25Vsupplies.
Used within specs they are perfectly happy.
As long as using ONLY 8Ω load you can run them from ±30V for even more watts. ~30W would be the limit.
For such a small amp there would be little benefit from designing discrete stuff. Unless of course you just want the experience or learn how to blow stuff up. ;)
I sent many perfectly good transistors to magic smoke heaven learning how to build amps. :-X
Phil.
I would build a tried & true amplifier.
Attached is a simple low power amplifier.
Tube Works TD 742.
Build it.
Study it.
Blow it up. :lmao:
Thanks!
I have already built little amps, chipamp and class B amp drived by an ampli op (but simplier than this TD 742).
My real goal is to learn how to design a discret one, not just having a new amp.
I know I could buy PCB with TDAxxxx, or used ss guitar amp for cheap. I could etch Randy Slone PCB too, and follow his schematic.
I just want to learn how to design it myself. ;)
OK I now understand. :tu:
As you have already Found ESP site then dig deep into Rods fine web pages and you will find a lot of gems. :tu:
Be aware that Guitar Amps are so far away from perfect that they bare little resemblance to the Audio Fools magic smoke dealers.
The last thing you want for guitar sound is HiFi flat bandwidth so steer clear blameless amp stuff.
ODrive guitar through a dead flat EQ with 20/25khZ bandwidth is the fastest way to annoy the punters.
Some of the greatest guitar sounds ever heard were produced by what would now be considered a very bad design. :lmao:
As A friend of mine found out when he spent a fortune on a custom build by a hifi teck. He eventually came to me unhappy so I built him an amp from a pile of El crap for 1/10th of what he had wasted and he was blown away by the result.
As for design help if you can find a copy of "Art of Electronics" it's well written for those of us who are less gifted at complex maths. :duh
As noted in the ESP page you quoted, anything running below +/-35VDC is not likely to give much trouble and that covers a basic amp of 40~50Watts so 15 ~20 Watts is fairly safe.
Just wack a few ideas on a bread board and use a Light bulb limiter and see what you can come up with.
I'll try and dig up some old pictures of my testing circuits to give you ideas.
Phil.
Thanks Phatt!
How would you limit the bandwidth please? With bigger Miller capacitor? Or in feedback network? I was thinking to limit it in the preamp section.
Quote from: phatt on September 25, 2020, 04:04:47 AM
Some of the greatest guitar sounds ever heard were produced by what would now be considered a very bad design. :lmao:
I am aware of that, but without going to digital, i prefer take a modern approach, with tone and distorsion independant of volume control (OK, speakers and ears are still going to react differently with volume change ;) )
Quote from: phatt on September 25, 2020, 04:04:47 AM
As for design help if you can find a copy of "Art of Electronics" it's well written for those of us who are less gifted at complex maths. :duh
Thanks! I have something similar, in french (my native langage), but this one look more complete.
Quote from: phatt on September 25, 2020, 04:04:47 AM
As noted in the ESP page you quoted, anything running below +/-35VDC is not likely to give much trouble and that covers a basic amp of 40~50Watts so 15 ~20 Watts is fairly safe.
That's the reason why I begin with this, I already have collect bigger transformers for future projects :)
Quote from: phatt on September 25, 2020, 04:04:47 AM
I'll try and dig up some old pictures of my testing circuits to give you ideas.
I would really appreciate it, thanks a lot!
Hi Shasam,
If you want the indepth analysis there are far greater minds here that can explain the finer points.
I was just a frustrated muso guy who got fed up with the crap that was being sold that never delivered so I bent my brain and read a whole lot of books and started building my own gear.
Most of my knowledge is just basic *Rule of thumb* thoughts I have gleaned from books and later wonderful people (via the WWW) like Rod Elliot and many other great people who have shared there knowledge.
My rules of thumb;
1/ KISS (keep it simple stupid)
Don't get caught up with teck heads who want to analyze the black holes in amp design. i.e Current FB is it worth it? Well some claim a big difference others say nay,, only you can make that call xP
My observation,,,It likely helps if the Amp is working at high volumes.
At bedroom levels I doubt if any ears could hear a dramatic difference.
You have to use your nose (Intuition) to decide if something tecky is worth the extra effort (& often expense).
Yes all done in preamp,, I don't even use a guitar amp, lol.
It's an old SS Laney keyboard amp,, all the mojo tricks are in my pedal board so the Amp is just the power driver.
I have 3 Valve Amps but I use the SS Laney when playing live, it's just so versatile as I can play my El Acoustic, Strat or keyboards all through one Amp. 8|
I've lost count of all the circuits I've built in the last ~35 years and I've built a lot of disasters. :-[ over that time I've designed about 4 success circuits that work really well (Some are on this site)
The Laney rebuild I'm posting for you to show the BBoard testing setup of a power amp
It's very basic and likely not an ideal bench test setup but well worth the effort.
As you can see in the pics the laney only had a small Tx and when using the Keyboards the supply could not keep up. I rebuilt the entire PSU and power amp section. The 3 Ch preamp section is basic but functional. (Hey it does have a Tank Reverb which is nice)
You can see in the pic the difference from the original back panel and the rebuild.
I've added the Schematic for you as that may help to understand the pictures.
This uses over rated devices as I've found with my basic rule of thumb knowledge. :P
As long as the transistors are way above what is needed then less chance of failure.
(i.e the 2SC5200 is a 250V-15A-150W device) They are the 3 main things to look out for.
As explained in Art of Electronics the whole idea of these DC coupled amp designs is to keep the DC Q point of *Output & the bases of Q1,Q2 at close to Zero as possible*,as I've noted on my Schem.
The Voltage and Temp measurements were taken from test circuit not a simulation.
Current limit is done with a trick little device on the output,a PTC (Poly Thermal Capacitor or just PolyFuse)
Hifi geeks might not like them but saves all the extra PCB space.
Short circuit testing was done with amp at full power on the test circuit pictured about 20 times for at least 10 Seconds.
The Poly Fuse will trip at ~110Watts, The colour will darken when they get close to trip.
Absolutely brilliant little device. :dbtu: :dbtu: :dbtu: :dbtu:
Pleae note;; This was 5 Years back I may have slightly changed something since then,, but it's all very close to how it was built.
Re bandwidth,
I used 250pF at C6 to help limit HiFreq. (Use 250V cap)
C4 sets the low Freq roll off,, smaller values can help roll off excess bass.
100uF there is likely of little use for guitar freq.
Have fun,, Phil.
(https://i.postimg.cc/858PYSsP/Laney49-V-Rails.png) (https://postimages.org/)
(https://i.postimg.cc/4dRRczsc/a-Laney-Mod-12-15-001e.jpg) (https://postimg.cc/t1kcQn1R)
(https://i.postimg.cc/rm2CLpNc/a-Laney-Mod-12-15-004e.jpg) (https://postimg.cc/mcVFYBn6)
Oppsy forgot the Test board pic. :-[
Probably not the ideal test setup but when on a budget one has to be inventive.
Most of this is recycled El stuff I've collected over time.
As I've done a few times, I did again test to see if Current FB was worth the fuss but just did not hear any magic so gave it a mis. It's those two 5watt Resistors on the side, near the Grey output jack.
Not in picture but I used a light bulb limiter for every new tweak,, It saved me a couple of times during this experiment. ;)
Phil.
(https://i.postimg.cc/LXFQgYYB/Laney49-Vtest-Mod-002.jpg) (https://postimages.org/)
(https://i.postimg.cc/0QgQXv2j/Laney-KD50-Mod-002.jpg) (https://postimages.org/)
Thanks a lot Phatt!
I try to une rules of thumb too, but different ones look to give really different results sometimes (for heatsinks calculation for exemple, I find results from simple to double with different rules of thumb I have found).
I won't try current feedback for this one. Like you said I try to keep it simple (and I still find it not so simple...).
Maybe for a next project.
I am not familiar with polyfuse, I need to learn more about them, but this look really interesting! :cheesy:
I have noticed you use different heatsinks for the driver transistors. Do you find the thermal tracking to be better like this?
I am going to use a 2*15V transformer, and NJW0281G (NPN) / NJW0302G (PNP) output transistors, in CFP configuration, for 8Ω output.
Datasheet can be found here : https://docs.rs-online.com/41ca/0900766b813e9eb0.pdf
For now, my calculations are :
Total DC power supply = 2 * (Veff *√2 - 2Vdiode) = 2 * (15 * √2 – 2) = 38,5V
Output power = (power supply - loss)² / (8*Rload) = (38,5 – 5)² / (8*8 ) = 17W (I found similar result in LTSpice)
Output peak current = (power supply - loss) / (2*Rload) = (38,5 – 5) / (2*8 ) = 2,1A
Output peak power = (power supply - loss) * Output peak current = (38,5 – 5) * 2,1 = 70,35W
With Rod Eliott rules-of-thumb previously quoted :
Quote
"Having discounted the idea of any 'rules-of-thumb', I'm going to give you one anyway . Let's assume that you want to deliver 100W into 8 ohms, so you need a power supply with ±42V rails (I'm going to ignore losses here). The amp must also be able to drive nominal 4 ohm loads, so expect the minimum impedance to be 3 ohms. Worst case (resistive load) dissipation is therefore ...
I = V / 2 / R = 21 / 3 = 7 Amps
P = V / 2 * I = 21 * 8 = 168 Watts (peak)
This accounts for the resistive part of the load, and as we saw above, the reactive part of the load causes dissipation to double. Just like second breakdown, we aren't interested in the average dissipation - this influences the size of heatsink needed, but not the transistor's safe area. Therefore, Ppeak will be ...
Ppeak = P * 2 = 168 * 2 = 336 Watts"
(from https://sound-au.com/soa.htm)
That give me : I = (Vrail / 2) / R = (19.25 / 2) / 8 = 1.203125A
Ppeak_resistive = Vrail / 2 * I = (19.25 / 2) * 1.203125 = 11.58...W(peak)
Ppeak_reactive = Ppeak_resistive * 2 = 23.16...W(peak)
Max dissipation for NJW0281G / NJW0302G is 150W at 25°, 0W at 150°
Derating = 150W / (150° - 25°) = 1.2W/°C
Maximum_case_temperature = (Pmax_at_25° - Ppeak_reactive) / derating + 25°
= (150 - 23.17) / 1.2 +25 = 130°C
I have use the formula given here for average worse dissipation : https://www.updatemydynaco.com/documents/Class_B_Amplifier_Dissipation_Calculations.pdf
Pdiss_worst_case = Vps² / (19.75 * RL) = 38,5² / (19,75 * 8 ) = 9,38...W
I don't really understand the calculation here. I have use it because it give me the worse result of all the calculation I have read online. What do you think about it?
Heatsink_max_thermal_resistance = (Maximum_case_temperature - Ambiant_temperature) / Pdiss_worst_case = (130 - 50) / 9.5 = 8.42... °C/W
I don't really know what to use for ambiant temperature, I have use 50°C here. I will use bigger heatsink to limit the max temperature lower than 130°C too.
I tried to make it as clear as possible. What do you think about it, please? Could you spot some mistakes? Should I take bigger margin?
Thanks a lot!
The output transistors that you picked are way overkill for the amp that you are designing.
Yeah, they will work.
But still.
Way over the top.
Yep Mr Jazz beat me to it,, Agree way over the top. Maybe keep those for a bigger build.
I salute you for you ability to deciyfur the maths I'm just painting by number most of the time as I'm way too old to go back to school now. xP
Others here will correct me if I'm wrong but all that maths assumes the Transformer is an infinite current source and the tX is the size of a car battery. :lmao:
Meantime back in the real world your 30vct Tx supply will dramatically sag long before the circuit can pull that much current.
i.e. Put your meter across the car battery while someone starts it. That 12~14 Volts will instantly drop back to 7~9volts depending on the health of the battery. Current goes through the roof and voltage sags,, same goes for Amplifiers. 8|
If you are well versed in sims you can probably setup full load conditions for a more realistic outcome.
For me understanding power supplies was likely the hardest thing to get to grips with. :duh :loco
Maybe look at what has already been built? Some cheap amps only use the case as heat sink.
Look at my Laney pic where I've shown the original back panel for comparison. Note the heat sink is only alloy angle bolted to the back panel.
That amp ran from +/-37Volt rails and rated at 40~50Watts. (more like 30ish watts clean to my ears)
Even with only that meager heat sinking it never got burn finger hot.
That design was very basic and yet performed well for 5~6 years before I rebuilt the power amp.
(just to get more clean headroom for My Synth)
Re Q on predrivers (Q5&6) on separate heat sinks.
That was just a safety backup while testing, they were getting hot until I found why.
Q4 (the Voltage stage) actually runs hotter as it's basically got the full supply across it.
In the final build Q4 has a heat tab added, hidden in pic.
Q3 (BD139) is the bias that needs to track the heat sink temperature and it can be seen in the pic.
The power devices are over speced but as I had these well why spend money.
Hope it helps, Phil.
Thanks!
You are right, those transistors are oversized.
I have redone the calculations for BD243 / BD244.
(datasheet here : https://www.onsemi.com/pub/Collateral/BD243B-D.PDF)
I got this :
Derating = 0,52W/°C
Max dissip 25°C = 65W
Maximum_case_temperature = (Pmax_at_25° - Ppeak_reactive) / derating + 25° = (65-23,17) / 0,52 +25 = 105,45°C
Pdiss_worst_case = Vps² / (19.75 * RL) = 38,5² / (19,75 * 8 ) = 9,38...W
Heatsink_max_thermal_resistance = (Maximum_case_temperature - Ambiant_temperature) / Pdiss_worst_case = (105,5 - 50) / 9.5 = 5,85 °C/W
Quote from: phatt on September 29, 2020, 09:38:05 AM
Others here will correct me if I'm wrong but all that maths assumes the Transformer is an infinite current source and the tX is the size of a car battery. :lmao:
Meantime back in the real world your 30vct Tx supply will dramatically sag long before the circuit can pull that much current.
Maybe there is something that I dont understand, but this voltage should be under load.
With no load, I have 44.9V (mesured) from Vcc to Vee.
It is true that I have negligected ripple, that would reduce the usable voltage.
Well we are assuming you have 30VAC ctap Tx and assuming you are using a full wave bridge then your NO LOAD Voltage is going to be 42VDC but that will vary depending on the VA rating of the Tx, the mains voltage at the time of day, the Diode drop through the Rectifier, the size of the filtercaps.
So with Amp connected and power on that 44.9VDC you read will now be under idle which is only going to *Lightly LOAD* the supply Voltage,,,,Maybe a couple of volts drop.
(the hotter you bias the more idle current will flow and voltage will drop even more)
But running at half to full power the current draw will drain the filter caps and the voltage will sag. Then your 45Volts will read more like 37~35Volts,, (Again depends on VA of Tx used and as above)
Ripple is then bigger as the Filter caps try to keep up.
Bass is the thing that drags down the supply.
Tiss why I rebuilt my laney,, sustained notes on my keyboard would really drag down the voltage and bass notes just collapsed. :grr
Tiss why guitar amps only need 40watts while bass amps need 400W. 8|
Phil.
Hi! I have been away from the workbench for the last two weeks, sorry for my late answer.
I have finally run some tests :cheesy:
Just before clipping level, with a 8 ohm load (resistive), the power supply sag to +/- 19.8V, with 35V ptp output (approximately, I have only measured it with the scope).