How do you design power supplies?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Lessons learned;
Until now I was thinking the transformer should be specified for idle current - static condition - plus a safety margin. Dynamic performance was then related to output power and speakers efficiency. Dimensioning the transformer according to listener percieved SPL is an obvious practice that I missed so far!

The psu must be beefy. Eliminating sag is major concern. I did some laboratory mouse workaround with PSUDII. My impression is that capacitance alone is not the whole story. It seems that psu impedance is more significant. A final cap after say CRCRC is not going to make big difference being 10μF or 1000μF. Thus, the psu has to be either regulated - like Indianajo's setup - or a simple Pi filter feeding a gain stage with excellent PSRR. The other way around is a constant load that disregards psu Zout.

I would like to put NFB into perspective. I' m aware of the basics and also of the conflict and the "phase shift" argument. Trying to understand all this I made up a theory which you are free to criticize crudely; Phase shift will happen before any NFB is applied due to capacitor/transformer coupled gain stages that will interact with the psu. NFB will substract the modulation of the final stage to the psu from the modulation effect of the psu to the first stage(?) NFB will reduce sag? Or is it reduced sag will improve performance of NFB?
 
Last edited:
Phase shift will happen before any NFB is applied due to capacitor/transformer coupled gain stages that will interact with the psu.
Transformer coupled gain stages went out of fashion about 1966, 6 years after transistors went into production. I own one designed that year.
Capacitor coupling is usually limited to the input to the first stage. Stage to stage capacitor coupling is out of fashion except in beginner transistor manuals. Looking at actual schematics of high volume or popular amp designs should tell you this. Whereas I have four or five transistor texts surplussed by schools that spend a lot of pages & ink on capacitor coupled gain stages, probably because the math is simpler.
The main phase shift calculations that are relevant to amps today have to do with the capacitance of be junctions of bjt's and gate capacitance of fets, at higher frequency. Also the miller capacitor, which is important to power amps. Long chapters written about that one.
Stiffness of the transformer versus output demand depends on the market. Guitar/bass amps are often low power and run near the max all the time, particularly by newby players that feel it is not music unless it hurts. So excess current delivery is not important, cooling capacity is. A lot of these are sold to poor young beginner players and price is very important to making the sale.
Car amps the same thing. The whole CD industry has taken to mastering pop CD's so there is no volume variation. So car amps can be assumed to run near the max power out all the time, as a sign of quality. Loud constant buzzing makes the sale. Switcher supplies make this possible at lower cost than transformers. Also toroids fit under the seat better than e-frame transformers.
Hifi amps can be used to produce music with 55 db (LP's) or 70 db (classical CD's) source, so no hum at the low volume, which is usually where they are run, is important. Then theability to briefly produce the high power, set the sag requirements on the transformer. Cooling is less important than low power guitar amps.
As noted previously for the output capacitors, switcher supplies can produce great power in 30 ms so the output capacitors can be smaller than transformer supplies for classical sources. The AC line is dead near zero volts for 20-30 ms but the rest of the time can produce great energy compared to the usual amp output. 3000 W up amps excepted, the input or output caps have to get over insufficiency of the AC line for more like 50-60 ms.
The whole switcher thing is the wave of now, I'm just getting into it. I've repaired two competent switcher supplies with simple problems, and have been appalled by all the blown up switchers that PC's and their CRT displays go through. My next intellectual challenge is perhaps, repurposing a switcher from a 50" LED TV that produces 55 V @ 3 amps and 160 V @ 6 amps, to produce 55 V from both sections. Useful audio voltages and current. One I paid $0 for, found on garbage day with a broken screen.
 
Last edited:
MagicBus said:
So, what matters mostly is a very good PSU - WRT sag - on the input stage?
You remind me of someone else on this forum, who is always looking for simple recipe-like answers to questions. He has kept up his search for recipes for many many months. Please don't copy him.

'Sag' is what happens to a PSU when the output stage takes more current than the PSU can deliver without suffering significant voltage drop. Input stages may or may not see the sag; it depends on supply rail decoupling details. If they do see the sag, this may or may not affect their gain and it may or may not come through as a subsonic signal; it depends on input stage details.

In audio, what matters mostly is finding the right compromise of design details to meet the chosen compromise of performance details. I still feel you are trying to run before you can walk. You will not learn how to design audio PSUs by asking questions on a forum. You need to read, think, build, test, debug and then read some more etc. Each time you think you have got it cracked, some new issue will come along which you had not thought of - indeed you might not have been capable of thinking of it much before you reached that stage of understanding. I remember many years ago reading in my university handbook the title and brief descriptions of undergraduate physics modules. I had just passed A-level and knew a lot about physics (or so I thought) yet there were lots of words there which meant absolutely nothing to me. I could have asked someone what the Hartree-Fock method was - but I could not have understood any reasonable answer (apart from "you will find out in 2 years time").

So try to ask questions about specific issues where you at least understand the question. That means you have a chance to understand the answer. I am not being awkward; I am trying to be helpful.
 
Transformer coupled gain stages went out of fashion about 1966, 6 years after transistors went into production. I own one designed that year.
Capacitor coupling is usually limited to the input to the first stage. Stage to stage capacitor coupling is out of fashion except in beginner transistor manuals. Looking at actual schematics of high volume or popular amp designs should tell you this. Whereas I have four or five transistor texts surplussed by schools that spend a lot of pages & ink on capacitor coupled gain stages, probably because the math is simpler.
The main phase shift calculations that are relevant to amps today have to do with the capacitance of be junctions of bjt's and gate capacitance of fets, at higher frequency. Also the miller capacitor, which is important to power amps. Long chapters written about that one.
Stiffness of the transformer versus output demand depends on the market. Guitar/bass amps are often low power and run near the max all the time, particularly by newby players that feel it is not music unless it hurts. So excess current delivery is not important, cooling capacity is. A lot of these are sold to poor young beginner players and price is very important to making the sale.
Car amps the same thing. The whole CD industry has taken to mastering pop CD's so there is no volume variation. So car amps can be assumed to run near the max power out all the time, as a sign of quality. Loud constant buzzing makes the sale. Switcher supplies make this possible at lower cost than transformers. Also toroids fit under the seat better than e-frame transformers.
Hifi amps can be used to produce music with 55 db (LP's) or 70 db (classical CD's) source, so no hum at the low volume, which is usually where they are run, is important. Then theability to briefly produce the high power, set the sag requirements on the transformer. Cooling is less important than low power guitar amps.
As noted previously for the output capacitors, switcher supplies can produce great power in 30 ms so the output capacitors can be smaller than transformer supplies for classical sources. The AC line is dead near zero volts for 20-30 ms but the rest of the time can produce great energy compared to the usual amp output. 3000 W up amps excepted, the input or output caps have to get over insufficiency of the AC line for more like 50-60 ms.
The whole switcher thing is the wave of now, I'm just getting into it. I've repaired two competent switcher supplies with simple problems, and have been appalled by all the blown up switchers that PC's and their CRT displays go through. My next intellectual challenge is perhaps, repurposing a switcher from a 50" LED TV that produces 55 V @ 3 amps and 160 V @ 6 amps, to produce 55 V from both sections. Useful audio voltages and current. One I paid $0 for, found on garbage day with a broken screen.
I keep forgetting that I' m not in the tubes/valves forum.:) Anyway, I' m going for a transistor amp soon and I keep your example in mind! One thing to ask; What do you think would be the difference in your amps performance if instead of voltage regulation you had current regulation with enough voltage headroom and adequate heatsink of course? Do transistors present a steady current load anyway so that voltage regulation plus large capacitor is more dynamic? Perhaps the ultimate overkill would be a voltage regulator before a current regulator? I don't mean I' m going to build such thing. Just curious. Maybe it's a question that cannot be answered though...
 
You remind me of someone else on this forum, who is always looking for simple recipe-like answers to questions. He has kept up his search for recipes for many many months. Please don't copy him.

'Sag' is what happens to a PSU when the output stage takes more current than the PSU can deliver without suffering significant voltage drop. Input stages may or may not see the sag; it depends on supply rail decoupling details. If they do see the sag, this may or may not affect their gain and it may or may not come through as a subsonic signal; it depends on input stage details.

In audio, what matters mostly is finding the right compromise of design details to meet the chosen compromise of performance details. I still feel you are trying to run before you can walk. You will not learn how to design audio PSUs by asking questions on a forum. You need to read, think, build, test, debug and then read some more etc. Each time you think you have got it cracked, some new issue will come along which you had not thought of - indeed you might not have been capable of thinking of it much before you reached that stage of understanding. I remember many years ago reading in my university handbook the title and brief descriptions of undergraduate physics modules. I had just passed A-level and knew a lot about physics (or so I thought) yet there were lots of words there which meant absolutely nothing to me. I could have asked someone what the Hartree-Fock method was - but I could not have understood any reasonable answer (apart from "you will find out in 2 years time").

So try to ask questions about specific issues where you at least understand the question. That means you have a chance to understand the answer. I am not being awkward; I am trying to be helpful.
I really don't know the member of the forum you mention, not that it matters, but I' ve been following cases that someone put the forum to design his project and then bye bye! Be sure this is not me! You have the right to believe this for me because I asked for ready recipies in the begining of this thread but my intention was to use examples for further discussion.
I' ve built some projects, I tried ready recipies, I realized that many don't work as supposed, I had to troubleshoot ripple noise, ground loops, crosstalking, poor regulation, limitted frequency responce and eventually I came to face IMD, of course:) Not that this is the last I have to deal with, on the contrary. I just thought to start digging! My questions are usually very simplistic and can be answered with a yes or no, although I appreciate analysis. This is not an attempt to run before I can walk. I use the whatever answer to make a safe step.
And no, I don't feel you are being awkward. You have answered with more than a yes or no so you are very helpful:)
 
I forgot you build tube amps. Tube amps still use coupling capacitors & output transformers, they have to. Calculating phase shift is difficult - simulators are good for that. Copying somebody else's old commercial design that works is even better.
If I wanted so solder 50 more wires and increase the size of my board 20% (which wouldn't fit the hole) I could put constant current sources in. Apex AX6 and John Ellis Basic 50 are about the same amp, but first works best with voltage regulation , 2nd has built in current sources. I suspect they would sound the same.
But, I own an 80 v 6.5 A E-frame transformer from dynaco that is shielded for hum reduction. You can't buy those anymore except used. John Ellis Basic 50 currents & resistors would have to be recalculated for the higher voltage, and I doubt his BC556B and BD139/140 transistors would stand the extra heat.
I own a direct coupled no capacitors in sound path Peavey CS800s which specifies 0.02% HD below 240 W/ch. The latter has a "1700 W peak" switcher supply, the ST120 regulates at 520 VA. On my Peavey SP2-XT speakers, using the djoffe modification boards with original fast drivers, they sounded exactly the same at 1/8 to 60W. The AX6 board sound just slightly worse on the highest of frequency sources. Repeating, I can't buy the AX6 specified driver transistors in the USA except from counterfeiters. Further experiments scheduled with MJE15029/30 I can buy.
Like DF96, I think you need to get out and build or repair something. There is a guy on PA forum that owns a amp repair shop in the port of Athens, what's the name Peraius? Take 80 Euro down there and see if he will sell you a nice blown up (good transformer, no burnt smell around t-former) CS800b or c or maybe even a cs800x. An CS800s would be stretching it at first, they are switcher supply, and not for newbies. Or a PV4, or that greek brand he writes about sometimes. Nothing over 75 v rail. Take a backpack, that is how I got my PV-1.3k home on the bus. Or let Sakis reccomend an easy to repair amp to you. He's a good guy. Then get a DVM, an analog VOM with a 20 vac scale, a transistor radio, some salvage speakers & 4 or 10 ohm 200 W resistors, & get to work.
 
Last edited:
New projects to come! And I know a little better than before. I wasn't hopping for that much when I was starting this thread. Thank you very much!

BTW, a little "reverse engineering". I know how it feels when my patients ask for explanations that they would never understand. Not to mention those who know what has to be done and I just have to do it. What is just spare time fun for me may be stressing for you all, no kidding! Let's have a break.

PS. Hartree-Fock method... Man, hell of a reading!
 
Lessons learned;
Until now I was thinking the transformer should be specified for idle current - static condition - plus a safety margin. Dynamic performance was then related to output power and speakers efficiency. Dimensioning the transformer according to listener percieved SPL is an obvious practice that I missed so far!

just to give you an idea, my power traffo used for my 6C33 20wpc OTL used a traffo with a core capacity of around 1000va and copper capacity of 750va...
it runs on a power of around 350 watts in use...


The psu must be beefy. Eliminating sag is major concern. I did some laboratory mouse workaround with PSUDII. My impression is that capacitance alone is not the whole story. It seems that psu impedance is more significant. A final cap after say CRCRC is not going to make big difference being 10μF or 1000μF. Thus, the psu has to be either regulated - like Indianajo's setup - or a simple Pi filter feeding a gain stage with excellent PSRR. The other way around is a constant load that disregards psu Zout.

sag is inevitable....no matter how small, since your power traffo
will have finite dc resistances in both primary and secondary windings..
so the thing is to be able to design a power traffo that can deliver
the required voltage at a given load current...
remember too that your mains can go up or down
during certain times of the day...

btw, i make all traffos used in my builds and i make them at will...
i am not a commercial winder, i do not offer to sell them....
but i help others make their own...
 
You will not learn how to design audio PSUs by asking questions on a forum. You need to read, think, build, test, debug and then read some more etc. Each time you think you have got it cracked, some new issue will come along which you had not thought of - indeed you might not have been capable of thinking of it much before you reached that stage of understanding.

another gem of wisdom......you will find a lot ot them here....

that is why i am deliberately slow when building something...
think and think and then think some more...
 

PRR

Member
Joined 2003
Paid Member
> I was thinking the transformer should be specified for idle current - static condition - plus a safety margin.

That's like designing the truck to idle in traffic.

If the darn light ever turns green, it will be disappointing.

You can pretty-much get away with this in tube hi-fi amps. We have to idle tubes pretty hot to avoid crossover (and because tubes always run hot, so why be cool?). In hi-fi use the current in the LOUD parts still may not be much above the average.

Over in transistor-land: a modest amp may idle at 0.050A and peak over 2 Amps, a 40:1 ratio. Using a 0.1A transformer it will "play OK" when soft and suck bad when played LOUD.

We may like the caps to hold-up through the longest transient or loud-part. For "nice" music this will be milliSeconds. I have produced a snippet which was MAX volume ALL the way through. This is no longer uncommon. Here's an image from Apple's audio editor docs:

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

link

Here's an "envelope"
envelope.jpg


This appears to be all of Floyd's Echoes:
https://camo.githubusercontent.com/...a2f2f696d6775722e636f6d2f74714b5a386c2e706e67
 
Last edited:
sag is inevitable....no matter how small, since your power traffo
will have finite dc resistances in both primary and secondary windings..
so the thing is to be able to design a power traffo that can deliver
the required voltage at a given load current...
remember too that your mains can go up or down
during certain times of the day...

btw, i make all traffos used in my builds and i make them at will...
i am not a commercial winder, i do not offer to sell them....
but i help others make their own...
It's more clear to me know. Thinking of it from transistors point of view helps me to understand it better.

I' m following your diy transformers thread:up: In almost all my projects I also use custom made iron but I have a local manufacturer to wind them at a reasonable cost. For chokes I let him wind an amount of copper on the bobbin, put it in appropriate laminations and deliver it to me open for final tweaking. Quite fascinating!
 
MagicBus said:
One thing to ask; What do you think would be the difference in your amps performance if instead of voltage regulation you had current regulation with enough voltage headroom and adequate heatsink of course?
Almost all amplifiers are designed to run from voltage supplies, not current supplies.

Do transistors present a steady current load anyway so that voltage regulation plus large capacitor is more dynamic?
Sorry, this is a meaningless question.

Perhaps the ultimate overkill would be a voltage regulator before a current regulator?
A current regulator before a voltage regulator is sometimes used. The converse would not be helpful.
 
> I was thinking the transformer should be specified for idle current - static condition - plus a safety margin.

That's like designing the truck to idle in traffic.

If the darn light ever turns green, it will be disappointing.

You can pretty-much get away with this in tube hi-fi amps. We have to idle tubes pretty hot to avoid crossover (and because tubes always run hot, so why be cool?). In hi-fi use the current in the LOUD parts still may not be much above the average.

Over in transistor-land: a modest amp may idle at 0.050A and peak over 2 Amps, a 40:1 ratio. Using a 0.1A transformer it will "play OK" when soft and suck bad when played LOUD.

We may like the caps to hold-up through the longest transient or loud-part. For "nice" music this will be milliSeconds. I have produced a snippet which was MAX volume ALL the way through. This is no longer uncommon. Here's an image from Apple's audio editor docs:

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

link

Here's an "envelope"
envelope.jpg


This appears to be all of Floyd's Echoes:
https://camo.githubusercontent.com/...a2f2f696d6775722e636f6d2f74714b5a386c2e706e67
I just got a flash and I have to make a question about signal envelope again. I hope I' m not being boring. Back in post #21, envelope's frequency is not 200Hz, but the duration of the 200Hz signal. Generally for music, envelope is expected to be subsonic but the +/-120(100)Hz harmonics may settle right in the low mids. With an ideal flat DC psu there won't be harmonics.
Actually all this is the question. The rest of the post is very clear. Thanks!
 
Almost all amplifiers are designed to run from voltage supplies, not current supplies.


Sorry, this is a meaningless question.


A current regulator before a voltage regulator is sometimes used. The converse would not be helpful.
It takes a couple of dummy questions to understand how things work:)
So, there won't be any amplification without voltage swing. Amplifying devices consume current to swing voltage. There is no way around this. Basics in amplification circuits :eek:
 
>
Over in transistor-land: a modest amp may idle at 0.050A and peak over 2 Amps, a 40:1 ratio. Using a 0.1A transformer it will "play OK" when soft and suck bad when played LOUD.

We may like the caps to hold-up through the longest transient or loud-part. For "nice" music this will be milliSeconds. I have produced a snippet which was MAX volume ALL the way through. This is no longer uncommon. Here's an image from Apple's audio editor docs:

An externally hosted image should be here but it was not working when we last tested it.
We covered this in post 42 to 48.
Wimpy 6.75 80v supply with 3300 uf cap can cover 1/8 watt all day and put out 200W peaks for 6 ms. Which is my definition of "music".
That electronic stuff in waveform 2 is "house" or "techno" which to me has no emotional impact. I own an amp (CS800s) that will put 260W of electronic "music" 24/7 into my 2 speakers, and the speakers have liquid cooled tweeters to deal with it. I didn't design that power supply, but I did for the ST120 amp that will idle at .016 A and possibly peak at 7. Which is a 437:1 current ratio on a 55 db performance of 1812 Overture starting at 1/8 watt.
But I don't want to listen to sample 2. Many of the people that do listen to electronic, house, or techno, at "raves" or "meets" in the UK are on chemical mood alterers, according to stories on BBCnews. I wouldn't expect their ears to be in any condition to appreciate good classical music afterward, either.
As far as current regulation vs voltage regulation, current regulation in the input & vas is very popular on transistor amps posted on diyaudio solid state. Also in commercial amps built on PWB's with discrete parts. The transistors to regulate the current are very cheap, $.07 each for TO92 and maybe $.38 for BD139/140. This current regulation of those two stages give excellent PSRR. I by contrast avoid the cost of keeping Microsoft Windows op system around, about $200 a year IMHO with all the updates required and buying new main boards to keep up with the megacorporate demands. So I can't simulate, produce asc files or eagle files. No pwb. I hand wire my diy boards. so current regulation causes a lot of extra wires, count the difference between honeybadger without the second output pair, and Apex AX6. Then direct coupled no speaker cap amp require a protection circuit to drive $600 speakers economically, another 25 wires and 20 parts for nfet speaker disconnect. See TGM1i. Thus voltage regulation for solid state PSRR , old fashioned as it is in the commercial market since about 1980, is cheaper in hours to the hand builder who isn't a microsoft subscriber.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.