Efficient Magnetics Design for Offline Resonant Push-Pull (Royer) Converter

[Odysseus]

I have built a functional off-line (US 120V) power converter that uses a LC resonant push-pull switching topology, but I am struggling to keep losses at a reasonable level in the resonating inductor/transformer. The target frequency of the LC tank is 100kHz, and requires a 10nF resonator cap to keep the frequency variation across load conditions within 10%, indicating a target inductance of ~250uH.  With a bridge rectified AC-DC input, the DC input is max 185V and thus the peak (not p-p) voltage on the LC tank is ~580V. Currently, I have built an inductor (secondary/transformer to be added later) from an EPCOS N87-ETD34 core with a 1mm air gap, 42 total turns of AWG 26 wire in a single layer, and the inductance value is spot on.  However, the measured losses in this setup approach 13W no load, breaking 70C within a few minutes, when the losses I've (apparently incorrectly) calculated should be closer to 3-4W. This calculation attempted to account for skin effect and core loss from the relevant core material datasheets.

Obviously, the winding window is being underutilized, but the point is that I don't know where the extra losses are coming from.  To further confuse matters, this is an 2x improvement upon a 4-filar, 4 layer winding configuration, same wire and # of turns. What gives? Proximity effect could be blamed for the paradoxical improvement, but the math just doesn't work out with the current setup.  My only guess is fringing field near the gap, but the current single layer is spaced 2-mm from the center post with some cardboard.

 

[Jay_Diddy_B]

Hi,
Can you share the schematic?

Have you tried building lower voltage, lower power versions?

Regards,

Jay_Diddy_B

[T3sl4co1l]

You neglected to mention the total power level.  Is this 10W or 1kW??

What part of what things are getting to 70C?  Is that spot heating, or is everything thrown in an enclosure and it's the air inside heating up?

The magnetic field divergence is exceptionally intense near the air gap; turns near the center of the bobbin experience extreme eddy currents, even of very fine wire.  Accordingly, layers at different heights will experience slightly different voltages; layers connected in parallel won't work as well as good old Litz, which is the best approach for high-ripple, high-Q inductors.

Tim

[Odysseus]

Hi,
Can you share the schematic?

Have you tried building lower voltage, lower power versions?

Regards,

Jay_Diddy_B

It is almost exactly this, just omit Q3 and D1 and the gate pull up resistors have a separate low voltage supply.

Yes, I initially tested a version using a 30V DC supply and a higher turns ratio, with a consequently higher frequency variation.  It used a different core, a Ferroxcube UR42/21/12-3C8 with a very small air gap, as I had it handy.  I got the EPCOS core because it was available on mouser with all the associated hardware and in various gap sizes.

You neglected to mention the total power level.  Is this 10W or 1kW??

What part of what things are getting to 70C?  Is that spot heating, or is everything thrown in an enclosure and it's the air inside heating up?

The magnetic field divergence is exceptionally intense near the air gap; turns near the center of the bobbin experience extreme eddy currents, even of very fine wire.  Accordingly, layers at different heights will experience slightly different voltages; layers connected in parallel won't work as well as good old Litz, which is the best approach for high-ripple, high-Q inductors.

Tim

The target power delivered to the load is around 75W continuous in the worst case. These measurements were taken without a secondary (no load).  I measured the core temperature by placing a thermocouple in the unused winding window, far from the middle of the center-post.  The transformer is becoming uniformly hot and is placed open-air on my bench.

I am going to try using an appropriate guage litz wire, and also try increasing the gap size and turns to reduce the flux density.  I too am suspecting the fringing field near the gap. I can detect 1-2% increase in power consumption just by placing the thermocouple near the gap.  Would a distributed gap help in this case?

When I get home, I'll post the math I am using to predict the core losses, and maybe someone can point out an error, assumption, or omission that I am making.  I would really like to be able to design on paper and build something that performs reasonably close to specifications the first iteration.  This is my first foray into actual magnetics design, and a lot has already been demystified for me.

[Jay_Diddy_B]

Hi,

I have put together an LTspice model of the Royer oscillator including the Chan core model. The LTspice model gives us some insight in the circuit operation.

LTspice model



The model is power with 170V dc to represent rectified and smoothed 120 vac.

LTspice Results



Reactive power circulates around the circuit formed by the transformer primary and the resonant capacitor. By measuring the reactive power in the capacitor we have 2.3A x 370V = 850VA. There is 850VA of reactive power circulating around the resonant circuit. This is there if the circuit is loaded or unloaded.



The Chan core model allows the flux to be measured using a single turn winding and integrating the voltage. The model shows that the flux excursion is 210mT peak (420mT peak-peak) You can look up the core losses knowing the frequency and the flux excursion.



I have attached a zip file containing the LTspice model.

Regards,

Jay_Diddy_B

[T3sl4co1l]

Wow, is it actually 850VA?  I'd say you're lucky to get what you have -- assuming all inductor losses, 10W loss is a Q of 85!

I would think aiming for a Q as low as 1 (under rated load) would be fine for this circuit.  In other words, 8 times the inductance used in the simulation above; which, if that's correct for the chosen core (I haven't reviewed the parameters, I'll take your word for it, for now), means you need about 1/8th the core gap.  Time to get out the wet-or-dry sandpaper!

Tim

[Odysseus]

...awesome LTSpice simulation...

Thanks! 

That Bp value is close to what I had predicted using Bp = Vp/(2*pi*f*N*Ae) = 530 V / (2*pi*100 kHz*42*97.1 mm^2)=206 mT.  The ETD34 datasheet only says that this will result in a core loss < 4 W.  However, using the N87 material datasheet, this flux density swing at 100kHz results in about 375 kW/m^3 of power density, and thus 375 kW/m^3*7630 mm^3 = 2.86 W.

So predicted core losses are somewhere between 3 and 4 W.

I measured the DC resistance of the winding using my Keithley 197A to be 0.395 ohm.  At 100kHz, the conductor height to skin depth ratio is ~1.7, and based on http://en.wikipedia.org/wiki/Proximity_effect_(electromagnetism)#Dowell_method_for_determination_of_losses the Rac/Rdc ratio of a single layer is around 1.5, giving an AC resistance of roughly 0.590 Ohms.  With 2.3 A of RMS circuilating current, that gives a predicted copper loss of ~3.13W.

So, I'll place the total predicted loss around 7W.  I repeated the power measurement just now, and at 170 VDC in, it draws 73 mA, giving 12.4 W.  So I guess I'll chalk up the remaining loss to the eddy currents induced by the fringing field.

Wow, is it actually 850VA?  I'd say you're lucky to get what you have -- assuming all inductor losses, 10W loss is a Q of 85!

I would think aiming for a Q as low as 1 (under rated load) would be fine for this circuit.  In other words, 8 times the inductance used in the simulation above; which, if that's correct for the chosen core (I haven't reviewed the parameters, I'll take your word for it, for now), means you need about 1/8th the core gap.  Time to get out the wet-or-dry sandpaper!

Tim

I can buy a 0.5mm gap core from mouser.  Don't know why I didn't in the first place, instead I bought two identical 1mm sets.  Losses should be proportional to Bp^2, so that should drop the loss by about four, right? I'll grab the 0.2mm gap while I'm at it.

[T3sl4co1l]

Same turns, same cross section = same Bmax -- but the copper and eddy current losses will be so much lower.  You can increase turns to compensate for the core loss, if you like.

Tim

[Odysseus]

I finally settled on a sufficient design.  It looks like it was a combination of fringing field and proximity effect losses, as switching to Litz wire for the primary greatly improved the efficiency.  I also increased the inductance by a factor of two to ~500uH.  Anyway, for all those curious, I've attached a scan of the final schematic with annotations about the design.

Take note of the voltage multiplier section. It's not the usual circuit, and has some important benefits, namely a floating output that can be freely grounded at either terminal.