Design of a High-ish Voltage Power Supply
24V to 500V 50W flyback converter

board sm

Legalese

If you don't know what you are doing, these power supplies can be LETHAL! Be very careful, taking the usual high voltage precautions.

Intro

I had a need for some high-ish voltage power supplies. I needed four voltages: +/- 200V, and +/- 500V. Current is about 100mA max, so up to 20 to 50W each.

I looked at many different architectures. AC-DC and DC-DC. AC-DC was rejected due to regulatory requirements and the need for fully custom magnetics. A single, big custom transform could provide all 4 voltages though.

For DC-DC, I looked at Royer / Baxandall types. Royer's are for low power.  Here is a German language page on low power Royers. It looks OK from what I can understand.

I read all the Jim Williams CCFL backlight power supply stuff. These are basically Baxandall/ Royer with an extra Linear Tech DC-DC for regulation. Not suitable for the 50W that I need.

Baxandall looks good for higher voltages because it is a resonant design with zero-voltage switching.  A simple push-pull design in the sub-100W power range. High voltage transformers have high inductance windings and so high inter-winding capacitance. So they have high-Q resonances. If you try to drive fast square waves through these transformers, you'll be fighting these resonances.
http://www.sophia-electronica.com/Baxandall_parallel-resonant_Class-D_oscillator1.htm  Nice paper, with an LTSpice simulation. But, output is not regulated. It would require a fair amount of work to regulate. I recommend Bandaxall for higher voltage and unregulated applications.

A simple boost DC-DC? 500V at 50W is a big boost converter with a pretty big inductor. One advantage of boost is that you can get a decent negative output with a simple 2 caps and 2 diodes. I built one of these for my day job that was 6V  to +/-100V to power an ultrasound pulser. It worked great at a few watts. But 50W? 500V? +/- 500V? Nah.

I looked at other resonant types, but too hard to design.

I looked at push-pull, but the leakage inductance hurts you. And requires full-custom magnetics. Beefy snubbers are needed. And an output inductor to provide 500V is a beast.

I looked for eval boards from TI, Linear Tech, ADI, etc. and found none that are even close. In fact I couldn't find any simple DC-DC flyback boards.

I looked at off-the-shelf modules, but they are very expensive, ~$1000 per supply. I need 4.

Then I found some cheap Chinese DC-DCs that --almost-- do the job. About $10 each.  Here are three that I used. I call them Small, Large, and Small bipolar. They claim up to 400 or 450V, 40W. I bought a few to evaluate, and to reverse-engineer. Maybe use their magnetics as a first pass.

These supplies use an 8-pin controller, similar to UC3845. They use flyback transformers to convert 24V input to 100 - 400V. The primary is current-mode driven with a single FET. The secondary high voltage winding is a single flyback winding. The output circuit is a simple high-voltage diode and a 450V 10uF cap. The feedback voltage is measured with three 1206 resistors in series. I reverse-engineered much of it. Here is the Small Bipolar one and its schematic.

small
The Small Unipolar one is the same schematic and PCB. It has a different output connector (2 pins), one less output diode and cap, and the transformer has one less output winding.

small sch

The Small ones have a separate over-voltage shutdown, over current shutdown, digital shutdown, and current mode compensation circuit.
Big but but: The output is not stable at all. It turns on hard, then shuts off, and repeats this annoying behavior forever. The output with a ~40mA load varies about 20V p-p. After disabling all the shutdown circuits and messing with the loop filter, it still oscillated. I finally traced the problem down to poor layout of the current sense circuit. I reworked a few and got them to be quiet.

There are a few simple rules for current-mode converter layout:
  1. Connect the controller IC to the power circuit ground only at the current sense resistor ground side.
  2. Put the capacitor of the RC current filter close to pin 3
  3. Keep the oscillator RC parts close to pin 4.
The Small ones violated 1) and 2). I reworked the current sense RC filter by putting the cap on top of the IC, from pins 3 (CS) to 5 (GND). That helped a lot.

I measured the current shunt resistors on both amps, and they were quite low, about 10m Ohms. On the Large, a resistive wire was used. On the Small, a small resistive plate was used. The large ones are a considerably simpler design. They are thru-hole, and use a single +12V to +24V supply, They drive the FET at 12 to 24V, which is not ideal. 10-15V is better. Since the current threshold on the UX384x controllers is 1.0V, this means that the current threshold was about 100A, far too high. I purchased some 0.050 and 0.025 ohm 1W resistors for both supplies: thru-hole for the Large, and SMT 2512 for the small.  Both worked much better with 0.025 ohm resistors.

I measured the cross-regulation of the Small Bipolar one. It is pretty terrible. Without a load, the - Out increases a lot. Even with a load it varies a lot.

Also I fried many components: output diodes, output caps, FETs, controller ICs. Initially I scavenged broken ones to fix them. But I fried so many parts I needed to buy extra parts.

Here is the Large one.
large
These are a bit easier to deal with, being a simpler thru-hole design. Here is the Large one's reverse-engineered schematic.

LargeSch

Note the emitter-follower transistor Q3 in the voltage feedback circuit. This isn't really needed since the feedback node impedance on the trimpot is 6K max. Plus it introduces a few volts of temperature-dependent voltage drift.  The blue LED is a nice touch, but I'm not sure what it does to the control loop. There is a start-up circuit in the input ground path consisting of a second TO-220 FET Q2 and the fuse. Not sure why, and it introduces a voltage drop in the ground path. I bypassed the FET (S-D short) and no problem.

I decided that instead of fixing their crummy design, I would build my own. Perhaps I would initially use their transformers, so I wouldn't have to design and wind my own. The circuit is pretty basic, and the layout is carefully done. 
The board should  have:
Just for the heck of it, I purchased a few off-the-shelf flyback transformers. These are medium size, and designed for +350V input and  +12V output. I was hoping to use them wired backwards for +24V to +500V. I built up the first prototype with one of these transformers, and it worked extremely well.

HiV

PCB

sch


I didn't know exactly which transformer I would use, and wanted to be flexible. Fortunately many transformers including the Large, the Small, and the Wurth all use 5mm pin spacing and 25mm row-to-row spacing. As do some commercial E-E core bobbins. Very handy.

The off-the-shelf transformers are Wurth 750855240. Specs are:
 I laid out a board, ordered from PcbWay, the boards arrived, and I assembled one. I got the low-voltage PWM side working, then installed a Wurth transformer, high-voltage output Diode and Cap, and the feedback. After a bit of debug, it worked perfectly! The oscillator is 100KHz, and because the UC3845 has a 2:1 divider, the output is 50KHz and the output max duty cycle is 50%, ideal for a flyback converter. A single 0.025 ohm 2512 shunt works perfectly. No instability, no oscillation from +200V up to +500V, at any load.

I built up a second one, configured for negative output, and it also worked perfectly. I built 2 more and have all I need for now. All the annoying issues of the Chinese ones were fixed!

High voltage output diodes are thru-hole FR207, 1KV, 2A. I use two in series for 500V output. At 500V, These see ~1000V reverse voltage, which is marginal for a 1000V diode  These diodes have Trr of 500nS. The UF4007 has a much better Trr of 75 nS. They are in a slightly smaller package and only 1A. I've had no trouble with the FR207. Either should work fine.

board

Voltage Feedback Issue

Note that there is an op-amp that can be used for optionally buffering or inverting the high voltage feedback. It is required to invert the negative versions.  On the positive version, I was not using the buffer.  I found that without the buffer, the output dropped about -7.5V from 400V (1.9%), with a load variation from 10K (40mA) to 5K ohms (80mA).

With R9, the feedback resistor changed from 100K to 300K, the output variation reduced to -2.5V, so the variation is about inverse to R3. There are 4 resistors that set the gain: 600K (3 x 200K), the trimpot RV3 + R13, R10 10K, and R9 100K. As the load increases, the duty-cycle increases and the voltage on pin 1 COMP increases, causing a bit less load on the 600K / 5K divider, causing a voltage drop. Adding the buffer and changing R9 reduced the variation to -1.6V (-0.4%). Good enough.

Design
 Output variation from +400V
with 10K to 5k load change:
40 to 80mA
Original, no buffer, R9=100K
 -7.5V ( -1.9%)
No buffer, R9 = 300K
 -2.5V (-0.6%)
Buffer, R9 = 300K
 -1.6V (-0.4%)

So if you care about minimizing load variation, use the op-amp buffer. This exercise also explains the Large Chinese power supply's use of an NPN buffer (emitter follower) here. But the NPN causes a significant temperature drift due to Vbe -2mV/C tempco. Not a problem with an op-amp.

Multiple outputs

I tried to use the other primary tap as a possible low voltage output. The main output would provide the critical +500V, the tap could provide the less critical +200V. But the cross-regulation of the tap output is pretty bad, changing up to +/-50V depending on loads.

There are several contributing factors to this. One is the output voltage spikes. These are mainly caused by transformer overshoot. This overshoot is typically caused by transformer leakage inductance, and by the output diode turn-on time. 

A snubber on the primary should help. Also faster output diodes. To be continued....

Isolated output?

My outputs are ground referenced, so no worries about isolation, common-mode, etc. The common pin of the output is just grounded.

This supply could be isolated. It would need a TL431 and an opto-isolator. Powering the few milliamps that this circuit needs is a challenge when all you have is +400V. Maybe combine a bleeder resistor and an output LED.... Or us the auxillary winding. Could come in handy for other low voltage needs.

Possible Improvements

The flexible transformer footprint isn't really necessary, the Wurth transformers work very well, so can be wired up permanently.  One could always cut traces if necessary to accommodate another transformer.  The Aux low-voltage winding could come in handy. But if you have +24V, and the secondary is grounded, you don't really need it.

I used a DIP for the controller IC since I have toasted a few of the Chinese controllers. DIPs can be socketed, and are easier to rework.  I never toasted one on 5 new boards (so far), so a SO-8 would be OK.

The board should have an input fuse, About 3A for the +24V. Next rev. Both of the Chinese ones have input fuses.

The output capacitors are running at their rated 500V (not good!). But 600V and higher 10uF caps are expensive and hard to find. It could use 2 lower Voltage (300V or so) caps in series, with balancing (~1Meg) resistors.

Since the FET does not get hot, maybe use a more convenient, PCB-mounted device, like a D-PAK or D2-PAK. .

The D-PAK 78M12 is overkill. Can use a smaller SO-223.

The board definitely needs a nice, bright LED to indicate when it it ON.

The board should have an over-voltage shutdown circuit, set for about 525V. Maybe a separate voltage monitor output as part of the over-voltage circuit. The Small Chinese one has it, the Large one does not.

My current application needs both +500 and +200V. Currently I use 2 boards. There is a second, lower voltage tap on the transformer,  so maybe it could provide the +200V.

The board is larger than necessary. By changing to the SOIC controller, removing the proto area, and optimizing the layout, it could be smaller.

Could use a DAC for setting the output voltage. And ADC / DPM for reading current, voltage.... Now we're getting crazy....

Applications

I think these would be good for driving tube circuits. Even for power amps up to 100mA.
Enjoy!



   Dave's Home Page
Last Updated: 7/24/2024