A common base flyback power supply with two transformers and a DIAC?

[Feliciano]

I pry-opened a damaged power brick to see whether it could be an easy fix or scavenge for parts. To my surprise, I found unexpeted real fuse, inrush-limiter NTC, and a PI filter, followed by a topology I'm not familiar with: something like a NPN placed on common-base between two transformers (looking like the top one providing feedback and the bottom one being the main transformer) instead of the typical optocoupler plus TL431, and the driving pulses through a DIAC?

I attach the circuit I obtained, to ask whether somebody has seen something similar, and highlight the key points to finish the reverse engineering of this. 

[T3sl4co1l]

Oh interesting... I've done a few discrete things myself but never with common-base.

What are the inductances -- measured?  k?  Phasing correct?  The P:S ratio or k_L3L4 is particularly unbelievable, but other than that, it would probably work.

I think operation is an unregulated forward converter: primary turns on, same phase appears on L4 (so the dot drops below GND), D10 conducts, D9 conducts momentarily (T2's k will almost certainly be lower than given, for which D9 provides a path until the leakage inductance charges up), and C4 gets charged to some ratio of primary voltage.  This is more like a transformer-coupled charge pump than a converter, since the voltage on C4 is obtained straight from C2, PSRR is nil (Vout proportional to Vin), and efficiency is worse and worse at low Vo (being especially dangerous during startup and fault conditions).  The transformer ratios need to be slightly mismatched (T2 lower than T1) to deliver the excess emitter current required to maintain saturation; it latches on until inductances charge up enough to break conduction, at which point it turns off and positive feedback accelerates this; flyback is clamped by D7.  (Evidently, T2 discharges through C3+R6, D8 blocking avalanche breakdown of Q1.  C5+R5 also help, though being almost the same value as R3+R6 they're probably to snub T1.)

Maybe helpful as a comparison, an LED driver I built some time ago:



This uses a saturation commutated, self-driving switch function.  The Seven transistors act as a variable frequency oscillator and error amp, regulating output current (comparing to a Vbe).  The oscillator has a one-shot sort of function, switching current into the 60t primary, which then turns on the power transistor, and load current through the feedback winding maintains it on, at a forced hFE(sat) = 15t/3t = 5.  This is a proper buck regulator, so the catch diode and inductor (the #26 powdered iron didn't work out, I later changed that to a ferrite core -- never updated the schematic) filter the output into LEDs, and no output filter cap is needed because the ripple current only costs a bit of efficiency.  And it runs at full brightness all the time, no adjustable range, so, no problem with discontinuous current flow or anything.

If you made some transformations on this circuit, I think it might end up looking similar to your example.  First, this is a buck converter; it needs to be made an isolated forward converter instead.  Make the choke into a primary, ground the one end, and add a catch diode for flyback (reset flux).  Given the single supply, this should be a CT winding with the far end going to a diode to +160V.  Also, the transistor is high side, just swap that around, so now the transformer CT is +160V and transistor is at GND side.  Move the feedback winding to the secondary side, adjusting the ratio accordingly: if the power transformer is say 12:1 (for a ~12V output), and an hFE(sat) = 5 is desired, then a 5:12 ratio is needed for the secondary feedback winding to the base drive winding.

The DIAC serves to kick it on; apparently it's free-running, probably the ringdown (after the catch diode turns off) is poorly damped and it turns on for another go pretty consistently.  The DIAC is only needed to kick it into action at all.  In this circuit, of course the regulator / oscillator serves the same function, and then some (because the buck topology permits regulation).

The above example, could also be considered a half-wave Royer converter: the transistor turns on until magnetizing current in the drive transformer rises faster than load current, thus reducing hFE(sat) and eventually turning it off.  In particular, a small transformer is chosen (indicated turns on a 0.37" #43 ferrite toroid), which saturates, i.e. the magnetizing current increases suddenly and massively after sufficient on-time (here, about 10us).  Saturation costs a bit more core loss, but this is in the smaller drive transformer where it's okay, meanwhile the sharpness greatly improves switching efficiency.  (A Royer can also be done with a saturating power transformer, but this costs both core loss (the power transformer itself is running up to saturation) and switching loss (basically the saturation current forcibly yanks the oscillator out of saturation, forcing commutation).)

This might be a relevant mechanism in your case as well; depends on the core size in the two transformers.

Oh, finally, the D5 into C3 thing, is commonly seen in CFL converters as well: these use a full-wave (half bridge) current-feedback topology, of a very similar sort but the resonant load provides commutation, no need for saturable transformer hijinx; but it doesn't start up, so there's an RC charging network into a DIAC, and a clamp diode that simply keeps it discharged while the bridge is switching.  So, as long as Q1 is saturating from time to time, D5 keeps C3 discharged, keeping Q2 from turning on.

Tim

[Feliciano]

Thanks for your reply, Tim,

Indeed makes sense to think on this like an isolated charge pump. The DIAC path I explain bellow reinforces that.

The original transformers measurements were in-circuit, just to grasp the working principle. So to proper analyze this I now removed the transformers and measured the right inductances, the phasing was right, and certainly the values were quite different as you guessed. To avoid confusions for other readers, I updated the previous image.

And for the Ks, I don't know, I haven't seen the T2 core used as a Transformer (normally I've seen it as a line filter).

I think operation is an unregulated forward converter: primary turns on, same phase appears on L4 (so the dot drops below GND), D10 conducts, D9 conducts momentarily (T2's k will almost certainly be lower than given, for which D9 provides a path until the leakage inductance charges up), and C4 gets charged to some ratio of primary voltage. 

This is the part I don't get. If the primary you refer is L5, at startup L5 is discharged so Q1 should be off, therefore the only way to charge that transformer (to positive) would it be throught the very low current R4R1C3R6R3 network, and discharge it through the higher current path R3R6C3 when the DIAC kicks in. Is the resulting stored energy on L5 high enough to set dot voltage negative enough to turn on Q1 and D8 to reinforce (or take over) the cycle? I suppose it has to be that way, but in my simulation it doesn't generates strong enough current and/or wide-enough pulses to keep the operating point.

BTW: Interesting design of your LED driver, somehow alike the one we're discussing. Shall B and E of my Q1 be feeded from T1, it would be easier to analyze.

[T3sl4co1l]

k can be measured by shorting the other windings and measuring leakage inductance, then dividing by that winding's magnetizing inductance, subtracting from 1, and taking the square root (that is, k = sqrt(1 - LL/Lm)).

Mind that if you're using just whatever L-meter, the numbers may be inconsistent; testing at too low a frequency, or the magnetizing inductance at low voltages, can have inaccurate results.  Best results are probably using a signal generator at known frequency, resonating with a known cap if applicable (select C to give a reasonable Fo say 10-500kHz?), and plugging the numbers into the respective formulas e.g. https://seventransistorlabs.com/Calc/RLC.html  Preferably, measuring voltage with a scope, also checking voltages on other windings to read the turns ratio directly.


The primary I meant as the main winding from +V, L3.  Very little power should be going through T2 I would think, but that would be interesting if it has some feedback or feedforward value.  That's the thing about these sorts of very simple, highly coupled circuits: it's hard to figure out what's doing what, to where.  I mean, this isn't too bad, it should just be some manner of oscillator, and probably a rather square one (blocking oscillator) at that, but even just in that space, there's a lot of weird behavior to figure out.

The ratios as shown, still seem rather peculiar: the power transformer is around 8:1 turns ratio, yet feedback ~1:10, less than unity loop gain so it shouldn't latch on.

What if... could D5 be relevant from cycle to cycle?  If it's got some positive feedback (just not >=100% worth), that effectively increases the current gain of the common-base configuration, and then maybe C3+R6 act as positive feedback just like in an RC timer multivibrator, with just a single transistor?

Probably not though, the RC value is so much lower impedance than the bias supplied to it (100s kohm), and there's no other path for bias into that network.  Now I kinda wonder what this sort of arrangement might do without the diode though, hmm...

Tim

[xavier60]

The position of D9 doesn't look right. It would make more sense with A to the start of L6 and K to the finish of L4. Yet the schematic seems to agree with the PCB photo.
On time is possibly controlled by the saturation of T2.

Edit: L6 isn't being used a an output filter inductor because it's being clamped by D9. The output from L4 is being hard rectified.

[Feliciano]

k can be measured by shorting the other windings and measuring leakage inductance

I had forgotten the relationship among M and k. Thanks for the reminder.

I don't have a high-end LCR meter. I'm using a couple of ATmegas component testers (because different ranges, and running OSHW firmware), which even callibrated cannot be considered precision equipments for Inductors, as the measurements are made with some DC pulses.

With this I now calculated k, for T1 gives me around .999, and for T2 around .976 (way lower than I was assuming, as you thought).

The primary I meant as the main winding from +V, L3.  Very little power should be going through T2 I would think, but that would be interesting if it has some feedback or feedforward value.

I see, but still at startup L5 is discharged, so Q1 and D8 are off, hence no current flowing through L3 until the dot of L5 becomes negative enough after being discharged through the DIAC impulse (1us/1ms).

And if you guys notice the big 3A/60V D10 (for a 1A power brick) and the thick 3-turns secondary of T2, it hints the real power transformer is T2, so T1 should be used for some kind of feedback and/or a filter, and

The output from L4 is being hard rectified.

I suppose by that we are talking that D7 clamps L4 from generating counter-current.

And about D5: I though it was part of a snubber for T2, but wondering why to connect it to the collector instead of Vcc or the high-side ground, is should be to loop that current into T2 trying to keep Q1 and D8 on.

Indeed a piece of work, isn't it?

On the other hand, there is the posibility the original UTC MJE13003L could have been another thing or having a different pinout, because I replaced it with an XW E13003L along with the burned R3, and it shorted the new Q1 and burned L1. And I haven't found any other damage in this circuit, which is now almost totally dismounted.

[xavier60]

D7, FR107 isn't a super fast diode. After it has finished conducting primary magnetizing current back to HVDC, its reverse recovery flow could be putting enough energy back into the transformer to cause enough primary forward swing to help Q1 turn on again for the next forward stroke.

[Feliciano]

Yes, the only Super Fast is D9, there are no Ultra Fast Rectifiers, and the only Schottky Diodes are D8 and D10. And trully D5 and D7 can inject some mA into the primary, to be used assuming the circuit is already oscillating.

And for all these reasons I've been wondering why such an odd topology (unless my schematic is wrong, although I double-checked it before posting).