Flyback switching power supply for tube amplifiers

[Pawelr98]

Hello

Recently I gained access to a ton of flyback transformers, several dozen.
Well made Friwo transformers with proper internal shielding and thick isolation.

Primary:40T
Aux:8T
Secondary 4x12T
The core is tubular with around 15mm diameter.

Normally the secondaries would be connected in parallel, since the transformer originally was used in a 24V DC power supply.
I have rewired them in series. First 12T tap is used to power the filaments (P series, 300mA heaters are pretty much the only sensible choice) and the rest 3x12T is connected in series, providing the high voltage.

The driver is a basic, oldschool one. UC3845. 8pin DIP chip meant for driving flybacks (45 has a 50% duty limit).

Secondary is nothing but two capacitors and two diodes.
10uF 400V for HV, 100uF 50V for heater.

Drain voltage,primary side, 100V/div

This snubber topology is quite good, with traditional one I would be past 600V most of the time, with the same component values.

100Hz ripple is going to be quite visible, since the primary only has 10uF capacitor.
~70V ripple on primary side. 20V/Div


Secondary ripple, single 10uF capacitor, no LC filter.

3.5V on HV, 1.5V from 100Hz, 2V from switching


Heater, tiny 0.35Vpp, 0.15V 100Hz + 0.2V switching



This power supply provides around 185V HV at something around 70mA and 40-ish V 300mA for two PCL82 heaters (16V 300mA) in series, 40ohms in series to reduce the heater current.
The whole thing is a little dodgy, I made it using 3 PCB's since it evolved over time.
In the near future I plan to design a proper PCB for this and order it from some chinese company.

The amp works fine, a small buzz from the speakers at no signal and that's all.


Flyback is a decent design for tube power supplies, since it can generate a ton of voltage from rather small amount of turns and even the basic CCM regulation is enough to keep up with very high input ripple.
However traditional 6.3V heaters should be avoided, wired in series or powered from another supply. Group regulation for flyback is not really worth it.

[T3sl4co1l]

Shouldn't C3 be a resistor, for fixed current operation?

Regulation is easily added with a TL431 and optoisolator.  The voltage is a bit high -- a bit wasteful for the 1mA bias needed by it -- but that's only a fraction of the 300mA output for example, not a big deal.  (If it were HV only output, I might be tempted to add a secondary side aux winding, to power the TL431.  Or use a TLV431 at lower current, downside is slower opto response.)

The 3845 is made for one-switch forward converters, actually, but of course it works fine for flybacks when >50% duty is not required.

The output waveform shows why I prefer film caps for this application -- electrolytic ESR is usually pretty awful.  For personal projects, at least.

Tim

[TimNJ]

Regarding TL431, you can always use Diodes Inc. AP431s, or ON Semi NCP431. These only need about 50-100uA, though I’d recommend to run them at about 200uA to avoid...weirdness, for lack of a better word.

edit: Ah, I missed the TLV431 recommendation, although I do not know how it compares to the low current 2.5V versions I mentioned.

[langwadt]

Shouldn't C3 be a resistor, for fixed current operation?

Regulation is easily added with a TL431 and optoisolator.  The voltage is a bit high -- a bit wasteful for the 1mA bias needed by it -- but that's only a fraction of the 300mA output for example, not a big deal.  (If it were HV only output, I might be tempted to add a secondary side aux winding, to power the TL431.  Or use a TLV431 at lower current, downside is slower opto response.)

The 3845 is made for one-switch forward converters, actually, but of course it works fine for flybacks when >50% duty is not required.

The output waveform shows why I prefer film caps for this application -- electrolytic ESR is usually pretty awful.  For personal projects, at least.

Tim

is the isolation of an opto needed in this application?

[TimNJ]

Shouldn't C3 be a resistor, for fixed current operation?

Regulation is easily added with a TL431 and optoisolator.  The voltage is a bit high -- a bit wasteful for the 1mA bias needed by it -- but that's only a fraction of the 300mA output for example, not a big deal.  (If it were HV only output, I might be tempted to add a secondary side aux winding, to power the TL431.  Or use a TLV431 at lower current, downside is slower opto response.)

The 3845 is made for one-switch forward converters, actually, but of course it works fine for flybacks when >50% duty is not required.

The output waveform shows why I prefer film caps for this application -- electrolytic ESR is usually pretty awful.  For personal projects, at least.

Tim

is the isolation of an opto needed in this application?

Assuming the OP is trying to avoid a big 50/60Hz step up transformer (which would indeed provide isolation), then, yes you need opto isolation. If it was some sort of double insulated appliance, maybe not but I think amplifier can almost never fall under this category as there are input/output jacks which are user accessible.

[langwadt]

Shouldn't C3 be a resistor, for fixed current operation?

Regulation is easily added with a TL431 and optoisolator.  The voltage is a bit high -- a bit wasteful for the 1mA bias needed by it -- but that's only a fraction of the 300mA output for example, not a big deal.  (If it were HV only output, I might be tempted to add a secondary side aux winding, to power the TL431.  Or use a TLV431 at lower current, downside is slower opto response.)

The 3845 is made for one-switch forward converters, actually, but of course it works fine for flybacks when >50% duty is not required.

The output waveform shows why I prefer film caps for this application -- electrolytic ESR is usually pretty awful.  For personal projects, at least.

Tim

is the isolation of an opto needed in this application?

Assuming the OP is trying to avoid a big 50/60Hz step up transformer (which would indeed provide isolation), then, yes you need opto isolation. If it was some sort of double insulated appliance, maybe not but I think amplifier can almost never fall under this category as there are input/output jacks which are user accessible.

ah I somehow assumed it was to be powered by a low voltage supply

[Pawelr98]

Shouldn't C3 be a resistor, for fixed current operation?

Regulation is easily added with a TL431 and optoisolator.  The voltage is a bit high -- a bit wasteful for the 1mA bias needed by it -- but that's only a fraction of the 300mA output for example, not a big deal.  (If it were HV only output, I might be tempted to add a secondary side aux winding, to power the TL431.  Or use a TLV431 at lower current, downside is slower opto response.)

The 3845 is made for one-switch forward converters, actually, but of course it works fine for flybacks when >50% duty is not required.

The output waveform shows why I prefer film caps for this application -- electrolytic ESR is usually pretty awful.  For personal projects, at least.

Tim

I made this thing using just parts that I have available + some inexpensive stuff that I could get without any issues (that is no fancy parts).
In 2019 I was doing a practice/internship in a factory that made Friwo power supplies.
Got a ton of parts from a closed product lineup, including those flyback transformers, loads of electrolytics, diodes and inductors.
I could make a small production line with this.


Originally I intended to use primary side voltage regulation, I did browse through some designs.
Well, I found that the typical 1-2 pin resistor (mentioned in documentation for UC384x) rather "redundant" in testing.
Even when it was in operation, it remained fine with just the capacitor compensation.

It runs just on fixed current alone, reaching the set maximum inductor current. There is no other regulation at work here.
Decreasing R2 increases the output voltage a lot. If the voltage feedback was limiting, then the thing wouldn't budge.
The divider on feedback pin of UC3845 is set to keep the voltage at 20V.
This particular mosfet can take 30V on gate.
Most of the time the voltage is only around 12-13V or so, because of R10+R11.
I have encountered the problem of high auxilary voltage when toying around and it remained like that.

As for regulation, tubes don't really need regulation that much.
Tubes can take well over 500V in a cold state and the second grid can have extra filtration that makes the output pentode quite resistant to ripple.
In this particular case the circuit turns on, HV goes up to around 300V (only the filament is loading down the circuit) and then once the tube heats up it goes down to 185V.
Regulated filament current would be nice, since a series resistor (would be working as a shunt) is required anyway, otherwise reaching high enough HV wouldn't be possible.
If I were to set the heater voltage just right (no series resistor with filaments), then the HV would be around the 140-150V mark, which is a little too low (works for some tubes though, like EL82/6P18P).
Ideal would be U-type tubes, operating heaters at 100mA current. Those however are rather rare, though UCL82 (same as PCL82) does exist.

Will do a brand new PCB, this time with everything on a single board, 431 filament current regulation and LC filter for HV.
An extra thing to consider, is that this thing, with HV only and some regulation would be a very small isolation transformer for powering stuff like switching power supplies that can run off DC.

And technically speaking, this is a step down design, since the mains are 230V, the primary DC side is over 300V.
I do have some experience with powering tubes directly from mains (RF stuff), however I really wanted to try and make a proper supply, that could replace the expensive mains transformer.
That was the main idea here. If I can make a compact, inexpensive supply that can get around the "audiophile tax" so to speak (transformers "dedicated" for tubes are rather expensive, yet I can get a brand new "non tube" transformer for less than half the price).

One question to ask (since I'm just a self learning amateur), why is the snubber design, as shown in my schematic, not really that common ?
I did a lot of testing and couldn't bring down the overshoot using the "traditional" design.
Then I looked through some materials on snubber design and found this thing
https://rudys.typepad.com/files/snubber-e-book-complete.pdf
Figure 4-45 page 123
I rewired the thing and was amazed at how good it works.
C4 sits at around 400-450V and slowly discharges into the main tank capacitor through R8.

Also some extra notes.
For a normal amplifier that uses 6.3V filament tubes, a half-bridge is a lot better design.
First of all, voltage multipliers can be used, which makes HV a lot easier to obtain.
Then you don't need any regulation at all. An IRS2166 can be used to make a current limited PSU with APFC.
IR2153 works fine too (except for the absent current limit part), it's my second design, that I'm going to use for a bigger amp.

[T3sl4co1l]

I could make a small production line with this.
https://obrazki.elektroda.pl/6470480300_1588807910.jpg

Hah you weren't kidding, nice haul.

I scored myself a lifetime supply of the building blocks, a bunch of years ago.  First the cores, from a surplus auction; later I finally found a supplier of bobbins:



EE33 size, good for about 100-200W depending on topology.  Of course, then I need the magnet wire, and the yellow tape, and so on... but that's no problem.  For one-offs, I mean.  Winding may be relaxing, but it's still tedious...

Originally I intended to use primary side voltage regulation, I did browse through some designs.
Well, I found that the typical 1-2 pin resistor (mentioned in documentation for UC384x) rather "redundant" in testing.
Even when it was in operation, it remained fine with just the capacitor compensation.

Well, you're just operating in saturation (of the error amp) -- no need for a voltage divider or "compensation" capacitor, just strap VFB to GND, or perhaps COMP to VREF.

What the resistor would do, is set DC feedback so that a lower V(COMP) can be set, I guess to save on ISENSE voltage for instance.

Regulated filament current would be nice, since a series resistor (would be working as a shunt) is required anyway, otherwise reaching high enough HV wouldn't be possible.
If I were to set the heater voltage just right (no series resistor with filaments), then the HV would be around the 140-150V mark, which is a little too low (works for some tubes though, like EL82/6P18P).
Ideal would be U-type tubes, operating heaters at 100mA current. Those however are rather rare, though UCL82 (same as PCL82) does exist.

Would also make class AB operation possible.  I take it this is a stereo SE amp, so that's not important in this case?  Could be attractive for a later project, of course.

One question to ask (since I'm just a self learning amateur), why is the snubber design, as shown in my schematic, not really that common ?
I did a lot of testing and couldn't bring down the overshoot using the "traditional" design.
Then I looked through some materials on snubber design and found this thing
https://rudys.typepad.com/files/snubber-e-book-complete.pdf
Figure 4-45 page 123
I rewired the thing and was amazed at how good it works.
C4 sits at around 400-450V and slowly discharges into the main tank capacitor through R8.

It's common enough.  I find the dV/dt snubber more often beneficial, as it can solve two or three problems at once: switching loss (taking load off the transistor soon after it turns off, while drain voltage is still rising), peak voltage (when leakage inductance is small enough), and damping ringdown (when C > 2.5 (Coss+Cp) and R = sqrt(Lpri / (Coss+Cp))).

Sometimes both are helpful.  When leakage is high, it would take a rather beefy dV/dt snubber to absorb its spike, costing efficiency, especially at light load.  Sometimes an R+C damper, or two, is desirable to control ringing from both LL and Lp ringdown.



Here's an example from a 12V to 150V module.  The 10n + (15 || BYV27-100) is the dV/dt snubber.

Drain waveform:
https://www.seventransistorlabs.com/Images/Snubber_Wave.jpg
https://www.seventransistorlabs.com/Images/Snubber_Rising.jpg

I think the R+C is there to dampen the bounce as it's going up, and also just after, as it settles down.  Because the impedance is quite low here (a few ohms -- something like 30V and 10A peak?), stray inductance matters a lot, and the inductance of the snubber (diode and capacitor in series) rings with the transistor's capacitance.  The R+C doesn't have much less stray inductance, but it helps.

https://www.seventransistorlabs.com/Images/Snubber_103Z.jpg

Here you can see the 15 ohm and 10nF (green disk) on the left, and the side of the diode hiding behind a vertical 2.2 ohm resistor (behind the red and black wires).  The transistor's leads are surface mounted to the board, which is carved up to make pads (who needs "printed" circuit boards when you have a utility blade?).  On the far right, the 2.2 + 10nF damping network.

Also, one of those EE33 core sets in the background.

For a normal amplifier that uses 6.3V filament tubes, a half-bridge is a lot better design.
First of all, voltage multipliers can be used, which makes HV a lot easier to obtain.
Then you don't need any regulation at all. An IRS2166 can be used to make a current limited PSU with APFC.
IR2153 works fine too (except for the absent current limit part), it's my second design, that I'm going to use for a bigger amp.

Yeah, it's a bit of a pain -- low voltages prefer forward converters, especially full-wave.  High voltages prefer flyback or resonant converters.  Well, low voltage would be fine with resonant too.  You almost want to do two separate converters.

Multipliers aren't a good idea, BTW: tying one to a flyback supply, makes a combination charge pump and flyback.  Tying one to an inverter, makes only a charge pump.

What I mean by "charge pump" is, when the transistor switches on, it dumps into a capacitor -- worse than hard switching, just raw unlimited inrush.

In steady state, this may be fine, but it's disastrous under startup and fault conditions.  It's also self-defeating, when current limiting is used -- a peak-current-mode controller like UC3842 will just sit there at a low duty cycle.

To make it practical, some series inductance is needed, to limit switching dI/dt, and deliver useful charge before current limiting fires.

It is, however, a fine way to use a resonant converter -- in that case the series inductance is part of the topology, and the rectifier simply manifests as some equivalent resistance at the switching frequency, who cares what conduction angle it has.

The most common way to get more voltage from a flyback, is to simply stack outputs -- a single say 2000V winding might have too much stray capacitance (Cp), but if we divide it into four equal windings, then each 500V winding has 1/4th the capacitance, and 1/16th the impedance -- so that capacitance is driven that much more strongly.  Each winding has one rectifier diode and filter cap, and their DC outputs are simply stacked.  (They can also be stacked without a cap each, especially if each winding is a single layer on top of each other: in that case, there's no AC voltage between layers and Cp actually helps instead!)

Indeed, this is the method used in CRT flyback transformers -- with sweep rates up to 100kHz, using a combined high voltage and deflection supply, they basically obtained low-MHz bandwidth along with voltages up to 30kV!

Tim

[Pawelr98]

As for the half-bridge operating with voltage multipliers.

Well, I rewound a core from some random ATX power supply.
Primary was 20T+20T which is pretty much a standard for those cores.

Secondary with 45T was wound.
150V (primary caps charged to something around 300-ish V) square wave going into primary.
170V is just too low for high power tubes. With many of those tubes you want 250-300V or more.
A doubler is still fine-ish for the most part. Especially considering ~40kHz switching frequency.
There is going to be around 300V DC using such a doubler with some LC filters.
Otherwise I would have to wind 90T which is taking up the precious space and gets labour intensive.

The same trick can be used for the heaters.
A doubler means full wave rectification from a single winding (no CT winding) and two diodes.
And it makes sense for bigger amps, since you get 12.6V (2x6.3V actually) and the wiring can be thinner, with heaters connected in series, instead of parallel.

A standalone flyback for HV with another supply for heaters (say a 5V usb charger bumped to 6.3V) is one of the cheaper options.

However for now I'm going to try and develop the new PCB for the flyback.

[Pawelr98]

The new design.

C11 is a 470nF 275Vac MKP62 capacitor, is it good enough for this application ? I'm still trying to work with parts I already have.
It should (hopefully) have lower ESR and the L2 inductor should slowly charge C9, bringing down the ripple to more reasonable levels.
As for snubbers C12 R21 should reduce EMI from the main high voltage secondary diode, while R20 C13 should reduce ringing.