Assymmetric outputs in half-bridge

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello all.
I am building and testing an half-bridge SMPS for +/-55V output. It works very nice so far, although it has been "only" tested up to 125W, but the mosfets and diodes remaing absolutely cool.
It is running unregulated by the moment, and I only have a problem: one of the outputs puts +50V, but the other puts a bit more, around -57V, with the same load.
I can't figure out is causing that assymmetry, as 4 diodes are the same (MUR1560), the inductors (4.7uH) and capacitance (6600uF+100nF) are identical for both rails.

Another thing I would like to do is to measure the switching node voltage with respect to GND. I tried with 4 47k resistors, taking the output from one (1/4 voltage divider), but the RC constant with the capacitance of my 60MHz probe is enough to filter all the details and obtain a rather roundy waveform that doesn't tell me nothing useful. Could I connect the 10x probe directly or is that a bit too much voltage for a typical oscilloscope?

Are your output inductors coupled? Using two independent output inductor produces poor cross-regulation.

Concerning switching waveform, you can connect your 10x probe to the switching node if it's rated at 400V or more, but be aware that very fast switching transients of high amplitude may cause a bit of partial breakdown in the attenuator that will show as overshoot. I had to live with that problem until I found an affordable 1200V 100:1 probe (there are a lot of good high voltage probes in the market but they are expensive).

You may also build your own high voltage attenuator but you would need to place a high voltaje adjustable capacitor in paralell with the resistors in order to compensate for probe capacitances (that capacitor is the one you can adjust through the calibration screw that most attenuating probes have).
Thanks, Eva.
No, my inductors are independent by the moment.
However, with equal loads and 1 ampere of current flowing, I cannot find the cause that makes the difference in voltages. Can it be an assymmetry in both PWM signals? I don't think so, but... who knows?

Thanks for the advice on the probes also.

Best regards,
Note that 4.7uH is a very small inductor value, so current ripple will be very high (same applies for voltage ripple).

Also, with such a small inductor value, any slight imbalance in duty cycle between each side of the half bridge will cause a severe output voltage imbalance.

You should try a coupled inductor. If you are using iron-powder toroid cores you can unwind your current inductors, stack the cores, and wind them back together in a bifilar fashion. You would get a dual inductor of approx. 4.7uH (actually 2x 9.4uH when exciting only one winding) by keeping the same turn counts. Additional magnet wire would be required, though.

Theoretically, an ideal dual coupled inductor forces energy transfer between both sides until the voltage across them is the same.
Sorry. The inductors are 47uH, not 4.7uH, They are wound on a drum core, so i will have to substitute them.

Unfortunately, this afternoon I was testing the thing a with a bit more duty-cycle, giving around 140W. Everything was quite cool (the diodes were warm but they have no heatsink), when one of the mosfet exploded.

Prior to this, I had measured the switching waveform by directly connecting my 10X probe. There is no visible overshoot, and the tops are flat. But there is ringing (about 10-20% of the total amplitude) at the dead time. I suppost that's quite normal.

The first thing that has come to my mind is that the same problem that causes the voltage imbalance has caused the failure. The core is a ETD44, with 26 primary turns, the secondaries being 13+13 turns.
Half primary is wound, then the secondaries, then the rest of the primary, to minimize inductance. It was not wound with great care, so perhaps the secondary is not absolutely symmetrical.

Can you point to the possible cause of the failure? Thanks Eva, your help is greatly appreciated!

Are you adjusting the duty cycle directly with a potentiometer?

Is your prototype mounted in breadboards or something that could lose contact?

Are you sure that your probe didn't slip and made a short?

Have you implemented some kind of current limiting in the primary side? This avoids a lot of catastrophic failures (one of my prototypes that implemented that survived for some time even with one of the switching devices shorted).

Concerning inductors, you may still turn yours into coupled if you wind half of the desired dual coupled inductor on each core in a bifilar fashion and connect them in series. Be careful with polarities.
Yes, by the moment duty cycle is manually adjusted by means of a potentiometer. Maximum setting is about 44%, with 80KHz sw. frequency, 700ns dead-time.
The prototype is built on a double-sided board, with carefull layout, and I think no false contact or short was produced in that moment.

Some details that may give some clue: the mosfets are driven by a IR2110, with 10 ohm gate resistors and antiparallel schottky diode, mosfet are (were) IRFB11N50. The driver has also died.

There is a current sense transformer, whose output voltage is rectified and goes to the shutdown pin of the SG3535, but I don't think that has worked. There is also a 5A fuse before the bridge, but it hasn't blown (one of the mosfet exploded so no shortcircuit was finally produced :xeye: )

My impression is that, as the transformer is coupled by means of a 1uF/250V capacitor, it shouldn't walk into saturation, and the number of turns limits the max flux to an acceptable level, if I have run the numbers correctly.

Thanks for your help!
Yes, it is possible.
But with an standard connection of the 4 bridge diodes, would that produce an assymmetry in the voltage of both rails or only an error in the output voltage of both of them with respect with the expected output?

Anyway, I have replaced the driver and mosfets, let's see if it doesn't fail anymore and I can start putting heavier loads. I have discovered that one of the primary ends was not correctly soldered at the transformer pins, perhaps it has suddenly disconnected ¿¿¿???

Best regards,
Please verify that the formula I am using for calculation of the flux density vs primary turns is correct:


B is expressed in Gauss
f is expressed in Hz
A is ¿EFFECTIVE AREA? in cm^2 (172mm^2=1.72cm^2 for ETD44) ???

That gives 1100 gauss for 26 primary turns, at 80KHz that's about ok for a 3C90 core, isn't it?

Thanks for the clarifications!
Asymmetric secondaries won't produce asymmetric output voltages in that configuration since each secondary is powering both output rails alternatively.

Output imbalance is probably due to a duty cycle asymmetry issue that the DC blocking capacitor compensates with some offset, thus yielding asymmetric volt*second product to the output inductors.

On the other hand, I've been using SG3525A for several years and some units have proved to be quite asymmetric, particularly when a low value discharge resistor is employed and at low duty cycles. This behaviour suggests that the IC is actually suffering from internal ground loops, but that shouldn't be an issue when coupled output inductors are employed.

Note that the asumption that duty cycle will be perfectly matched between sides of the bridge can't just be made because there are several unpredictable sources of mismatching and the circuit must cope with them.
Thanks, Eva. That matches with what I thought: differences in secondary turns cannot produce different in output voltages in each rail.
I hope that won't cause major problems like transformer walking into saturation (you have worried me saying that the coupling cap can create offset due to different duty cycles)

I will measure the duty-cycle of both SG3525 outputs carefully to see if there are any differences for various duty-cycles. If I find them, I will try with different discharge res. values to see if that's corrected.

Your experience is unvaluable!

Best regards,
Done: the SG3525 outputs (with the mains disconnected and no mosfets, that shouldn't change anything, anyway), have duty-cycles that differ only in a decimal, for example, 30 and 30.1%.
The greatest difference is at max. duty cycle: 44% and 44.3%. That kind of error may be due to my Tektronix DSO, however. In any case, that should produce a difference of less than 1V, while I have observed up to 7-8V unbalance.

Now I have to check the gate-source waveforms of both mosfets just in case there is something weird in the high side or whatever.

so... having an assymmetric secondary (12 vs 13 turns, for example), shouldn't have any effect other than output voltages different to calculated (but both the same), right? I was thinking in re-winding my transformer, but if that's true, I think it doesn't worth the pain.

Pierre said:
Please verify that the formula I am using for calculation of the flux density vs primary turns is correct:


B is expressed in Gauss
f is expressed in Hz
A is ¿EFFECTIVE AREA? in cm^2 (172mm^2=1.72cm^2 for ETD44) ???

That gives 1100 gauss for 26 primary turns, at 80KHz that's about ok for a 3C90 core, isn't it?

Thanks for the clarifications!

flux swing (delta)B=(V*t)/(N*A)
t= maximum pulse width in us
A=efective core area in mm^2

I got flux swing of 220mT wich is same as your 1100 gauss peak. i prefer this equation as it is easier to remember for me and uses easy units without coefficients :)

110mT on 3C90 @80khz should be ok.
The DC blocking capacitor charges in order to compensate for asymmetric duty cycles and prevent saturation, but that correction offset, together with the asymmetric duty cycle that it's compensating for, are a source of output imbalance. The DC blocking capacitor will prevent any kind of saturation as long as fixed duty cycle or voltage control is employed (it will cause a lot of trouble with current control, though).
This is starting to resemble a chat! ;-)
Well, then it seems that everything is about right, and that mounting a coupled inductor and paying attention to the duty cycle difference will solve these problems.
I will try to post some photos of the primary switching waveform for you to see it: I am only a bit worried about what's happening the portion of time that both mosfets are off, where you can see some damped ringing (I have to measure the freq. but it is several MHz for sure), with around 80Vp-p maximum amplitude. I don't know if that's dangerous or causes dissipation, or if it should be corrected before going to higher power testing.

Best regards
Unfortunately, my mosfets have exploded again, as soon as I have increased output power from aroun 100W to about 170W. Now I had 2sk2141, 6A/500V, more than enough for that power.

I attach a figure of the switching waveform when it worked (100W output, worked fine and everything cool), with around 20% duty cycle. It is something strange, isn't it? Y scale is 50V/div, X scale is 2us/div

What can be the cause of the violent explosion (they weren't hot, in fact they exploded just when I plugged the supply)? I am quite puzzled.


  • primary_voltage_100w.jpg
    53.7 KB · Views: 294
This has been a very interesting discussion so far! (Sorry, Pierre, for all the smoke...)

I would perhaps suspect the MOSFET drive circuitry and check that out once again. The IR2110 type driver parts are very picky to set up correctly. I've tried to use them before and ended up going with a transformer drive which is much easier to deal with (to me, at least). If the IR part is not bypassed REALLY well and everything with nice tight loops, you may be getting ringing above or below the Vgs limit of the FETs(usually around +/-30V). The layout problems (if there are any) and ringing will increase when you start to increase the load.

If your FET's blew at turn on, maybe you had FET cross-conduction because of the driver chip not establishing a stable operating point yet. If I recall correctly, though, the chip has UVLO, so would not have turned on without an adequate voltage. Strange indeed...

Something to think about...

I'm curious about the ringing in your picture, too, and hope someone (Eva ;) ) can shed some light on that.

The waveform looks right. It shows discontinuous mode operation. Output inductors are charged during the on time, then they discharge during the part in wich voltage rests at center (inductors are effectively shorting the secondaries), and then the transformer is left free so it tends to reset from the previous cycle but the discharged inductors in the secondary side clamp the voltage.

As carvinguy said, this may be an issue with gate drive and resonance peaks exceeding maximum Vgs rating, thus punctuting metal oxide gate isolation and inmediately destroying the device (and the gate driver IC).

It may be also an issue with continuous conduction mode and secondary side diode reverse recovery, since according to the 100W waveform 170W semms like a bit past the continuous mode boundary. Note that if there is a gate drive issue, it may arise only in these circumstances.

You will need to connect oscilloscope probe directly at MOSFET terminals and build an improvised common-mode filter for the probe like the ones shown in the picture in order to get an accurate picture of gate drive waveforms.

An externally hosted image should be here but it was not working when we last tested it.
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.