Strange problems with ZVS driver.

[Atomillo]

Hello:

Recently, I purchased a couple of ZVS drivers (of which I attach a couple of files), which dont need a center tapped coil. The circuit appears to be the same as the one shown on schematic 1 (http://www.kiblerelectronics.com/bob/app_notes/note11/note11.html).
When you connect the power terminals, and then, turn the power supply on, it inmediately trips and enters current limiting (even at the maximum setting of my supply, 5.2 Amps). If instead, I connect one lead, turn on the supply, and then connect the other lead, the board works perfectly.
In the first case, the gates of both MOSFETs are held at Vcc, turning them on. But, why doesn't the oscillations start? Does the supply have to rise to full voltage on a limited rise time?

Any help is much appreaciated, since I'm puzzled by this behaviour.

[davekni]

Yes, this is normal behavior such ZVS oscillators.  They rely on fast power ramping to kick-start oscillation.  When power ramps up slowly, both FETs are on simultaneously, with Vds and Vgs both at the supply voltage (ignoring inductor resistance).  Oscillation feedback is through diodes.  Neither of the diodes is forward-biased, so no feedback occurs.

BTW, the second schematic in the link has at least one error, 1000uF across one FET.  It also shows separate resonant capacitors to V+ (C2 and C3).  This works, but increases RMS FET current.

[Atomillo]

Thanks Davekni for the reply.

Regarding the schematics, I've only paid attention to the first one, since it seems to be the one I have.
Could you help me understand the need for fast power ramping better? From what I've read, oscillations starts because no two MOSFETs are equal, thus one will turn on faster (presumably, due to gate capacitance or Vths). Is the quick ramping of the power supply required so that these differences become appreciable enough, or is there some other reason I'm unaware?

 

[davekni]

Yes, that first schematic is the most common I've seen.

In most other oscillator circuits, a slight mismatch initiates oscillation.  As you clearly understand, even a 1uV difference is amplified by positive feedback (at the resonant frequency) to grow into full-amplitude oscillations.

These common ZVS circuits don't generate the necessary positive feedback until amplitude is more than one diode forward-drop.  An initial 1uV or even 100mV isn't enough to forward bias the cross-coupling diodes (D3 and D5 in the schematic), so isn't enough to start oscillation.  With a fast supply ramp, voltage across the inductors (L1 and L2) is high enough to forward bias the cross-coupling diodes and allow oscillation to start.  Another way to start oscillation is with a very-high-current supply.  If DC current is high enough at start-up, the inductor's DC wire resistance causes enough voltage drop to forward-bias the diodes.  The latter is the "common wisdom" for these ZVS oscillators - that they need a high-current supply to function.  The third way (my preferred solution) is to use a separate supply for the gate pull-up resistors R2 and R4 that is on before the main supply.  This has an additional advantage of reducing power dissipation in these resistors when using higher voltages for the main supply to L1 and L2.

[Atomillo]

Davekni:

Your response was really insightful. With this knowledge in hand, I connected my scope probe to measure the voltage across one coil and thus one diode (the positive lead connected to the anode and the negative to the cathode). Below are my results. The "Starting Transient" scope shot was taken after briefly touching the positive terminal of the supply with the power supply lead, after which the circuit started to oscillate. I also took shots of the coil voltage when power was applied directly to the circuit: only the EMP noise created by the turn on of the switcher was observed (verified by taking the scope shots with normal voltage and also with voltage set all the way down to zero: both captures were identical). In "steady state", with the supply connected from the start, a voltage of 26mV was measured with a DMM, clearly not enough to turn the diodes on. Tomorrow, I will desolder both pull up resistors in order to connect them to a separated power supply.

Again, thanks for the great explanation!

[davekni]

Nice to hear that my explanation was clear enough!

You must have either a floating power supply and/or floating scope and/or differential probe.  If using a normal probe with floating scope or supply, be careful about connecting the negative "ground" probe lead to high-frequency or sensitive circuit nodes.  In your case, negative to cathode is a good choice, since that's a FET drain, so lower impedance than gate signals.  The waveform appears to be from the cathode of one diode to the anode of the opposite diode (drain to gate of one FET) rather than across a single diode.  If across a single diode, the waveform couldn't go more than a diode-drop positive.  (Unless your scope is set to AC-coupled vertical input mode.)

ZVS oscillators are great for small induction heating projects.  They are very simple, but have other pitfalls too, especially with rapidly changing loads.  I use a ZVS oscillator to drive my Jacob's ladder.  That took a bunch of experimenting and learning to find and patch issues.  Biggest ones after start-up are abruptly-ending load (when arc breaks) and high load where the Q can drop too low for oscillation to continue.  I have many posts on this forum about such ZVS issues.

[Atomillo]

The switcher was in fact not grounded. I have grounded and and ungrounded it so many times, now I have just one screw holding the cover in place, so that I can reconnect earth as fast as possible.
 
@!@!@ Writing fast and writing correctly are hardly compatible! The probes were connected to both terminals of one coil, which clearly is not the terminals of the diode itself! My bad, it was one of those things I subconsciously knew to not be the case, but didn't let myself enough time to process it into text.

I will peruse thoroughly all the posts I can find regarding ZVS to be more prepared in the future.

EDIT: Since double posting is not legal, I'm editing this last message.

First of all, desoldering one lead of the resistors and connecting it to the a lower power supply (in my case, the minimum voltage is 4,6V) worked wonders, and the boards reliably turns on and off without any problems. Secondly, I attach a small diagram, where I indicate where I probed to obtain the measurements shown before. As you can see, the +10V pulse measured made it so that the cathode at the diode was almost zero (since it subtracts from the power supply), and thus started oscillation. I also have checked throughfully all the posts made by davekni on ZVS driver, and I have learned a lot, even if I have not fully processed all the info yet.

[davekni]

It took me a while to understand the double-posting rule, and initially I couldn't get edits to work either.  If I understand correctly now, posting a second (or third...) message is allowed when adding significant new information, say from further testing as in your case.  The edit worked just fine too.

Is 4.6V (or 4,6V) the minimum of the primary power supply feeding L1 and L2?  What voltage are you using for the gate pull-up resistor supply?  I'd suggest a gate supply a bit over the 12V of gate clamp zeners to make sure gates are fully turned on.  If running at higher frequencies, I'd also reduce gate pull-up resistance.  470 ohms was sized to handle 48V or 60V input.  At say 15 or 20V gate supply, gate pull-up resistance can be much lower while keeping power reasonable.  Lower gate resistance allows faster switching for higher frequencies.

[Atomillo]

I'm sorry, I did not speak (¿write?) clearly. The 4.5 V are the minimum gate voltage required for turn on. As you say, I normally employ around 15V. I also want to comment some interesting behaviour I have observed.

1.- When onlythe gate supply is turned on, oscillations (despite being of lower amplitude) are still created. If I understand correctly, this is because oscillation start through the diodes, right? But then, why does oscillation always start when powering the circuit from the pull ups and not the ballast inductors? I strongly suspect this is the case because the cathodes of D3 and D5 are initially at ground and not Vcc, but I will really appreciate if you could confirm (or deny) this.

2.- I found the original board would work if a tied one or both of the pull up resistors to Vcc through a long alligator clip. Strangely, this doesn't work if you connect it directly to the Vcc connection on the board, instead of the connector or the power supply. Is the increased inductance the reason why this works? Or it might very well be some bad contact in the alligator clips that created arbitrary current interruptions.

I must admit this has become a lot more interesting side project than I expected originally.

[davekni]

1) Yes, I'd noticed the same behavior when first trying a separate gate supply.  Oscillations start with just gate power.  Power gets to the oscillator through gate pull-up resistors and the diodes.  Because drain/inductor power is provided through the diodes, it is one diode-forward-drop lower than gate voltage.  That's what allows it to start.  The diodes are conducting, so positive feedback works even at 1uV.  When inductors are powered by the same supply, the diodes are not forward biased.

2) I have no specific answer to why clip leads start oscillation.  Best guess is the leads are picking up stray fields, even slight ESD from your body, that's enough to momentarily forward-bias one of the diodes.

Great to hear that you find it interesting.  So do I.  Found this same simple circuit about two years ago, and was immediately taken by its simplicity and efficiency.  I designed a bunch of variations, primarily aimed at avoiding wasted power of the gate pull-up resistors while allowing lower-impedance gate drive for faster switching.  My favorite (and final) version is in my Jacob's ladder project posted here (and several other small undocumented projects).  I think I've shared my next-to-last version as well, using small PFETs to disable gate pull-up resistors when gate voltage is low.  That allows much lower resistance without significant power dissipation.

[Atomillo]

Yes, I have seen (admired!) your Jacob's Ladder. I'm digesting everything now. I've tried to follow your lead and experiment with LTSpice, but obviously in my first try this did not work, and the schematic is such a gigantic mess I would be embarassed if it ever saw the light of day. A couple hours ago I found that you can turn on a grid, and it changed everything.

Once again, thank you very much for all the help and explanations!