Regulated Voltage Supplies

[bob p]

I think the subjective sound quality of a Class-A amp would be better, too. ... I think three pairs cooking along in Class-A would sound so much nicer, especially at the conversation volume you often listen to a hi-fi at.

Steve, I'm inclined to agree with you. I think that if I'm going to throw my wallet at building a serious HiFi amp, the 100W Class A amp would be a good way to go. There are a lot of reasons to go that way:

Pros:
- Sounds great
- Keep it simple (no regulation circuit needed)
- Visually appealing (a dozen 6550 would look like a million bucks)
- Class A1 satisfies even the most snobbish audio critics (who wants to listen to someone say, "Nice amp, but its not Class A.")

Cons:
- heat (not really a "con", as anything less than or equal to one hair-dryer unit is easily acceptable)
- cost of tubes (a dozen 6550 would cost a million bucks)
- hard to find the right PT for the job (the main impediment)

Yikes, feeping creaturism! Looking at what you guys have written, I like the idea of the Class-A setup. You would use a tranny with a lower voltage to give a 400V DC rail (how about a 208 to 480 industrial tranny, but feed 120V into the 208 tap? ...

Let's see what the turns ratio tells us ... (480/208)*120 = 276.9 V

So if we fed 120VAC into a 208 to 480 industrial tranny, we'd get about 277 VAC out. That in turn, would give us about 390 peak DC volts:

Vpeak DC = 1.41 * Sec VAC
Vpeak DC = 1.414 * 277 = 391.7VDC

At first glance, the numbers don't look bad. Plugging this into Duncan's PSUD, it looks like this setup with a FWB with a C-L-C filter and a current tap of about 100mA for bleeder resistors would produce a B+ rail of about 360 VDC at idle.

Things got a little ugly though, when I added a 1.3A current tap to the model (for 1.4A total current tap) to simulate adding the gang of 12 power tubes. Wow, What a sag! The combination of the huge current and the internal resistance of the PSU caused the B+ to sag to 250VDC under load!

To be fair, I'm not sure if the model is accurate as I don't know the real internal resistance figure for the PSU. I left the default value for the PT resistance as-is at 31R. I used 470uF caps and an under-spec 1H/5.75R/1A choke in the CLC filter (best case scenario), and I added a third RC filter and a 40 mA current tap after the power section to simulate the preamp.

The drop of 110V under a 1.4A load implies a PSU internal resistance of 78R. So even if we have a low Z PSU, the HUGE currents that we're dealing with will cause enough sag that the quick calculations on a napkin may not take us where we want to go. We might have to use modeling software instead of calculators to minimize some of the errors. Just in case you use the Duncan's PSUD software, here's my data file that crunches the numbers:

[bob p]

taking this data and working backwards, i increased the voltage of the PT secondary until the DC voltage of the B+ rail approached 400 VDC with the load of the tubes added to the simulation.

as it turns out, a secondary voltage of 400VAC produces an unloaded B+ of about 535VDC, which sags to about 405 when the tubes are hooked up at idle.

i have to admit, i'm not sure how valid these models are, as they're based on an assumption regarding transformer specs that may or may not be valid.

Oh, the numbers were pure fabrications for math illustration. To a certain extent, with a power supply designer hat on, I don't care what the amps pull. You tell me what they pull, I tell you the bad news about what it will cost.

I'll look at your numbers and see if i disagree. Probably I won't.

RG, can you comment on the model? If the voltage does actually sag that much, it actually gets us closer to the 400V target for Class A, which might lessen the thermal requirements that follow the use of a 480VAC transformer. It might make Class A more appealing.

[kg]

bob have you considered going choke input for your PS ?

[Don Symes]

If anyone's interested, Fred Nachbaur is regulating the +420V rail in his dogzilla amp. (only 70mA, though)

[bob p]

bob have you considered going choke input for your PS ?

yes, its good to think of that. i had focused on L-C inputs in all of my early designs because of their "superior" voltage regulation characteristics compared to C-L-C inputs. even choke inputs weren't good enough for the high current application, as the chokes that were required added enough internal resistance to the PSU that the 100W Class AB1 amp still sagged way too much under load. this thread on regulation was started to get help in addressing the deficiencies of the choke input in that application.

now that transformer ideas from Steve and R.G. have released some of the current constraints imposed by the PT iron, its possible to look at a 100W Class A1 design. the problem remains though, of how to supply B+ voltage that doesn't sag out of spec under high current demands. the only difference is that now our current demands are high and constant instead of widely swinging.

Let's look at the choke idea and crunch some numbers. Because a Class A1 amp doesn't have very significant current swing, the choke calculations are pretty easy. Since Imin and Imax are almost exactly the same for a Class A1 amp, Iopt = Imin = Imax = 1.4A. Now all we need is the correct value for L at a 400VDC operating voltage. Looking at the nomograms in Crowhurst & Cooper, here's how the numbers work out:

Imin  = 1.4 A
Imax  = 1.4 A
Iopt  = 1.4 A
MaxDC = 400 V
Lopt  =  30 mH

30mH is a pretty small inductance. I would guess that it would be fine for smoothing ripple, but I wouldn't expect it to store enough energy to alter the B+ appreciably.

Looking at my available catalogs, none of the Hammonds are up to spec. They're either deficient in current handling or voltage rating, so I'd need to find another source. (If anyone can recommend a source, please do!)

The closest (but no cigar) Hammonds are:

193R: 300mH, 1000mA, 3.50, 1000V max DC 
193T:  50mH, 2000mA, 0.70R, 300V max DC

Just for fun, I've plugged these chokes into Duncan using an L-C-L-C filter. The small inductance of the 193T doesn't effect the B+ much, as the results look pretty much like a C-L-C filter. The 193R has "appreciable" internal resistance that causes the B+ to sag more. With a 480VAC secondary the no load/load figures are:

193R: 579 / 370 
193T: 625 / 485

If you split the PSU rail into two parallel rails (one per channel), you could run the 193R within specs. Then the B+ numbers are 580/450, and the amp now has 4 chokes at 5 lb each, or 20 lb. of chokes.

Then there's still that problem that adding more iron adds internal resistance to the PSU, and any internal resistance in the PSU causies huge voltage sags under load. This means that you have to know the REAL figures for internal resistance, or B+ is going to be unpredictably effected when 1.4A of current gets pulled across it. Any small miscalculation in Z makes the models potentially VERY inaccurate.

From my position as an armchair PSUD, it looks like adding iron (chokes) to solve an iron problem (wrong PT) isn't the best approach. Instead of adding 20 lb of chokes to cover up the fact that I don't have the right PT, it would be better to just spec the right PT for the job.

Perhaps, going back to regulation might be an answer to the problem. Granted, at first it appears that a 480 VDC secondary provides too much voltage for the Class A application. But after crunching the numbers, the B+ really sags because of PSU internal resistance. Perhaps the answer is to use a regulation circuit with the 480VAC transformer in the Class A design. But because I'm such an amateur at this, I really need the expertise of someone who really knows what he's doing.

It would really help to know the internal Z of these big industrial PTs.

R.G., for a 100W Class A1 amp that requires 1.4A at 400VDC, how would you do the supply?

[kg]

bob, are you really intending to throw off nearly 600W at all times from this amp? that's a lot of plate input power...

as far as adding chokes increasing effective DCR, also remember one of the biggest benefits of a choke input--much lower peak to average secondary current ratios. in other words, instead of charging the input caps up in very short, very high intensity bursts of current, the choke input's "flywheel effect" will average out those periods during which the power supply diodes are forward conducting. the current is delivered in longer duration, shorter amplitude pulses.

the end result of all of this is that the L input filter will show much less voltage drop than one would experience out of an "equivalent" C input power supply (ie, two designs, one L input with a higher secondary V, one C input with a lower secondary V, both have same secondary DCR, etc--the L input will show much better inherent regulation and less drop under load). the downside is weight, cost, and the need to always ensure that you have the minimum bleed current (not an issue with a class A design). although it starts off at 1.4*Vsec, when you hit a C input filter hard, it will tend to drop voltage rapidly down to about 0.9*Vsec. this just doesn't happen with a proper L input filter, as it STARTS at 0.9*Vsec.

ken

[bob p]

bob, are you really intending to throw off nearly 600W at all times from this amp? that's a lot of plate input power...

thanks for your help. i'm not sure where you got the 600W figure. my figure was 520, but i guess those numbers are close enough as long as we're both thinking about the same thing. the answer to your question is yes. Class A1 isn't very thermally efficient, and i think about 5:1 is the price that you have to pay for Class A1 power, and when you're talking about 100W of Class A1 power per channel, you're talking about an amp that is going to be a MONSTER.

as far as adding chokes increasing effective DCR, also remember one of the biggest benefits of a choke input--much lower peak to average secondary current ratios. in other words, instead of charging the input caps up in very short, very high intensity bursts of current, the choke input's "flywheel effect" will average out those periods during which the power supply diodes are forward conducting. the current is delivered in longer duration, shorter amplitude pulses.

the end result of all of this is that the L input filter will show much less voltage drop than one would experience out of an "equivalent" C input power supply (ie, two designs, one L input with a higher secondary V, one C input with a lower secondary V, both have same secondary DCR, etc--the L input will show much better inherent regulation and less drop under load). the downside is weight, cost, and the need to always ensure that you have the minimum bleed current (not an issue with a class A design). although it starts off at 1.4*Vsec, when you hit a C input filter hard, it will tend to drop voltage rapidly down to about 0.9*Vsec. this just doesn't happen with a proper L input filter, as it STARTS at 0.9*Vsec.

ken

regarding the input filters - i understand the figures you're referring to -- they're the same ones that i have in my electronics books, and they're the same ones that i've been using all along. i've run into an interesting problem though -- the rules of thumb that come from the equations are not always valid. at least that's what the computer models are telling me.

having run through what seems like thousands of model iterations, the models are telling me that the equations only hold true when the inductance of the choke is significant enough that it can store a large amount of energy relative to the current demands of the circuit. for example, modeling a L-C-L-C filter should be very different from modeling a C-L-C filter. that's true most of the time, and in those cases the models show what the equations would predict.

OTOH , I have performed successive approximations of the LCLC model while decreasing the value of the first inductor. When I do that, the choke input filter essentially approaches the behavior of a CLC filter. when i thought about it, there really wasn't any reason to be surprised by this., even though its not what the rules of thumb that we like to rely on would have predicted.

the problem is that easily obtainable off the shelf parts give the designer the choice of a large inductance or a large current carrying capability, but not both. it would appear that in this high current situation the rules of thumb are not valid, because the only chokes that are capable of handling the current have "insignificant" inductance, and when the L value of the first inductor becomes as small as 30 - 50 mH, the LCLC filter essentially behaves like a CLC filter:

Filter Type    Values           DC idle     DC load
  C-L-C          470-5-470      640         507
L-C-L-C     .050-470-5-470      622         478
L-C-L-C     .300-470-5-470      575         362  
L-C-L-C        1-470-5-470      508         355 
L-C-L-C        5-470-5-470      415         296

from what i've read, this makes complete sense, as a choke stops acting like a choke under conditions of saturation. high current demands force us to use low inductance chokes, and low inductance chokes saturate earlier than high inductance chokes, so we're back to looking like a CLC filter.

the other alternative is that the modeling software just produces invalid results. that's also a possibility, as (per Crowhurst) there's never been a satisfactory mathematical model that fully explains the behavior of chokes. this raises the question of whether i may be operating in the range where even the best mathematical models break down.

this leaves me in a lurch. i am at a loss to determine whether the modeling software is accurate or not. so far, it seems that nobody else can answer this question either. i'm hoping that someone who really knows this stuff can crunch some numbers and shed some light on whether or not the models are valid. if Duncan's PSUD is a defective design tool for this application, it would be really helpful to know that.

thanks again.

PS - admittedly, there is a little bit of hidden apples-to-oranges artifact going on in that table, as each choke has a different value for R, so the supply DCR changes somewhat from iteration to iteration. that makes it impossible to perform a valid model that shows only the effects of L, because in the real world L and R are inexorably intertwined and we have two variables that change simultaneously.

[kg]

well, let's see..

you want 1a4 at 400vdc.

critical inductance will be

400vdc / 1400ma = 0.28571428571428571428571428571429H

so say 300mH.

you could achieve that with two 159Y in parallel:
0.6H, 750ma, 11 ohms, 500VDC. EACH $22 (angela).

dcr would be ~5r5, you'd have 1a5 current capacity and 0.3h effective inductance

or if you're rich and strong, two 193u in series:
193U .2H, 2000ma, 1.7 ohms, 300VDC. EACH $35.

dcr would be ~3r4, you'd have 2a current capacity and 0.4h effective inductance

obviously the latter solution is a bit "better" on paper (higher i and higher l) but also more $ and more #.

personally i've never designed a choke input power supply any other way, and they've always worked for me on first power up. don't take it as a slam on PSUD--i just always did it on paper using the "rules of thumb." i've never done cascaded LCLC or anything like that--if i needed regulation that tight i went to active solid state series pass regs.

one thing i will point out is that hammond chokes used as L inputs are pretty noisy bastards... a separate supply chassis is practically a must.

hth
ken

[bob p]

one thing i will point out is that hammond chokes used as L inputs are pretty noisy bastards... a separate supply chassis is practically a must.

D'Oh! I had been focusing on the enclosed Hammond 193 series, and I hadn't even given a thought to the open 159 series.

When you mention noisy, are you referring to audible humming radiating out into the room, EMI radiating out into the chassis, or both?

or if you're rich and strong, two 193u in series:
193U .2H, 2000ma, 1.7 ohms, 300VDC. EACH $35.

dcr would be ~3r4, you'd have 2a current capacity and 0.4h effective inductance

does that 300VDC rating on the 193u bother you at all?

personally i've never designed a choke input power supply any other way, and they've always worked for me on first power up. don't take it as a slam on PSUD--i just always did it on paper using the "rules of thumb." i've never done cascaded LCLC or anything like that--if i needed regulation that tight i went to active solid state series pass regs.

one of the problems that i may be having with PSUD is that i'm assuming that the internal resistance of the PT is 31R (default PSUD value). if that value turns out to be unrealistically high for the industrial control transformer i'm thinking about using, then my worries about DCR will go away. i hope that's the case, as with the choke recommendations you've made, PSUD tells me that the B+ is falling to 360VDC. If the real impedance is significantly lower, that would help a LOT. 'd really like to get a better handle on this before committing to the iron.

Does anyone have a clue what kind of impedance those transformers actually have? Unfortunately, the spec sheets don't have that information on them.

thanks again for your help, Ken.

[Steve Conner]

I believe these industrial transformers are specced with a "regulation" somewhere between 5% and 10% with the smaller ones being worse.

Regulation means the difference between full load and no load output voltages. So a 480V transformer, 1kVA, with 5% regulation has an internal impedance of about 12 ohms referred to the 480V side. 10% regulation would give 24 ohms.

Judging by KG's comments on noise, you're probably looking at a choke and transformer rack in the basement anyway, with a third rail leading up to your amps. ;-)

I always ignored those DC voltage ratings on chokes completely. I used tube amp smoothing chokes on the 10kV DC power supply for my old Tesla coil. The way I see it, if you float the cores, they're not going to know. I actually used them in a radar-like DC resonant charge circuit, though, which probably way exceeded any AC voltage rating they might have had too... I just put four in a series string on a well insulated base and hoped for the best.

[Tom Phillips]

I always ignored those DC voltage ratings on chokes completely. I used tube amp smoothing chokes on the 10kV DC power supply for my old Tesla coil. The way I see it, if you float the cores, they're not going to know....

There is also a trick I saw in an old Ham radio manual that put the choke in the negative leg. Works just fine for a choke input filter.
Tom

[Tom Phillips]

I always ignored those DC voltage ratings on chokes completely. I used tube amp smoothing chokes on the 10kV DC power supply for my old Tesla coil. The way I see it, if you float the cores, they're not going to know....

There is also a trick I saw in an old Ham radio manual that put the choke in the negative leg. Works just fine for a choke input filter.
Tom

[kg]

D'Oh! I had been focusing on the enclosed Hammond 193 series, and I hadn't even given a thought to the open 159 series.

When you mention noisy, are you referring to audible humming radiating out into the room, EMI radiating out into the chassis, or both?

audible hum, mostly. seems hammond run the iron cores themselves pretty hot magnetically speaking and the flux is high. nothing wrong with that per se, but there's a degree of magentostriction in the hammonds that just cannot be stopped, and the high flux just makes it worse. at least this was my experience with a couple of the higher henry (10-20) range units.

i got around the issue by using some soft urethane washers i scavenged from somewhere (a copy machine i think) to float the cores from the power supply chassis, which in the case of the hifi amp was also separated from the main circuit chassis via a 6' umbilical.

does that 300VDC rating on the 193u bother you at all?

nahhhhh..



as mentioned above, you've got a couple of options open to you, maybe even both: float the core, and put it in the ground side of the rectifier circuit.

even then hammond iron is NOTORIOUSLY conservatively rated. iirc the hifi amp was running a 300v choke at around 400, and no special precautions were taken. runs acoustically noisy but just fine..

one of the problems that i may be having with PSUD is that i'm assuming that the internal resistance of the PT is 31R (default PSUD value). if that value turns out to be unrealistically high for the industrial control transformer i'm thinking about using, then my worries about DCR will go away. i hope that's the case, as with the choke recommendations you've made, PSUD tells me that the B+ is falling to 360VDC. If the real impedance is significantly lower, that would help a LOT. 'd really like to get a better handle on this before committing to the iron.

the 440vac 1k2va bigass EI industrial tranny i have in the basement specs out at around 10r total secondary DCR, fwiw..

ken

[bob p]

I have a question about running the chokes in parallel: Do the potential problems of running PTs in parallel also apply to running chokes in parallel?

I'm familiar with the principle that if the voltages and impedances in the PTs aren't equal, a circulating current would exist in the closed network between the two transformers. That causes excess heat and short transformer life. To avoid this problem, impedance values for transformers need to be within 7.5% of one another. (For example, if Transformer A has an impedance value of 4%, Transformer B has to have an impedance value between 3.7% and 4.3%.

Hammond specs their chokes as having 15% tolerance for L and Z. To be on the safe side, would I have to assure that I get chokes that are matched for Z?

[bob p]

Judging by KG's comments on noise, you're probably looking at a choke and transformer rack in the basement anyway, with a third rail leading up to your amps. ;-)

Believe it or not, I have contingency plans in effect for doing just that. I'm thinking about building the beast into a computer server rack that would have doors and fans. The PSU would go on the bottom, and the amps up above.

Another option is to use separate PSU chassis and an armored flexible conduit as an umbilical cord to the amps. If I have to, I'll isolate the PSU in another room and run supply line feeds through solid conduit to a junction box in the living room, and then run flexible conduit umbilicals from there.

I'm hoping that neoprene washers and a rack mount will do the trick, maybe with some fireproof sound insulation that doesn't get in the way of air flow.