Sine wave inverter voltage clipping (transformer saturation?)

[kathmandu]

Every comercial product has the capacitor across the secondary and most of them only use the leakage inductance as a filtering inductor.

I was reluctant to such a topology too but its wide-spreading proves that it actually works.

Btw, maybe the transformer itself (the laminated core) has a filtering effect too (as it's a low frequency core after all)?

with 50mH and 100microF can have a good filter; 50mH at your power level is hard to find / manufacture, is easier to increase capacitor and reduce L.

An unpolarised and high pulse (polypropylene film) 100 uF capacitor is even harder to find (and pricey).

 

[Warpspeed]

Every commercial product has the capacitor across the secondary and most of them only use the leakage inductance as a filtering inductor.

I was reluctant to such a topology too but its wide-spreading proves that it actually works.

This does work, but the devil is in the details !

The capacitor tunes the secondary, usually to about x1.5 the ac output frequency.
Any lower and you might get a giant uncontrollable resonant build-up of no load voltage.
Much higher and you may be cheating yourself on PWM filtering.

Now the problem will be that big capacitor across the secondary, and the large turns ratio of the transformer.
That capacitor is reflected back across the primary and appears as an absolutely HUGE capacitance due to being magnified by the turns ratio.

If your transformer is super efficient with excellent coupling and minimal leakage inductance, there are going to be some incredibly high current spikes in the primary, and your mosfets will be very unhappy.

There absolutely must be sufficient inductance in series with the mosfets to cause the current to ramp up rather than spike at turn on.
Only a very few uH will be needed in a high power inverter to do that, and often the transformer has more built in leakage inductance than that anyway.

But if you are using a cut tape toroid, its something to be aware of. An external choke may significantly reduce the turn on current spikes and reduce switching loss.
With a laminated iron EI transformer these are usually lossy enough and leaky enough to work o/k just by themselves without any help.

 

[kathmandu]

Only a very few uH will be needed in a high power inverter to do that, and often the transformer has more built in leakage inductance than that anyway.

Thanks for your detailed explanations. But how come "a very few uH will be needed" when those calculations lead to a much larger inductance (mH)?

Btw, I'm using a toroidal transformer.

 

[Warpspeed]

The purpose of the choke is to soften the turn on, not to filter out the PWM.

For instance 12 volts with 3uH limits the rate of current rise to 4 amp per microsecond. That gives the mosfets ample time to turn on really hard before the current rapidly builds up.

Think about how a buck regulator works.
What would happen without any inductor, with the switching mosfet trying to directly charge the output capacitor during the on time.
That is what the primary of an efficient transformer looks like, a massive capacitance reflected back from the secondary.
You need a choke, only a small one, to limit the inrush if the transformer cannot do that by itself.

One or two or three turns of welding cable on a large gapped EE core is often about right.

 

[kathmandu]

The purpose of the choke is to soften the turn on, not to filter out the PWM.
One or two or three turns of welding cable on a large gapped EE core is often about right.

Are you saying that you don't need any inductor for filtering?

I've read some discussions on another board (energy matters) where some guys modded their commercial products by replacing the E laminated transformers with toroidal ones.

To decrease the no load current, they indeed have used some choke inductors (like you have described).

Anyway, that idle current (without that choke inductor) is due to high frequency voltage applied to the laminated core (eddy currents are much bigger at higher frequencies).

That's it, by using that choke inductor they actually filtered those HF (btw, they also have used a capacitor across the secondary winding - a small one (0.47 uF) though).

You need a choke, only a small one, to limit the inrush if the transformer cannot do that by itself.

I have no problem with the inrush current as I'm using a PWM duty cycle ramp-up at the inverter start-up.

However, I don't think that such a small choke inductor (few uH) could help avoid inrush current for a bigger toroidal transformer.

Speaking of the experiments made by those guys from the other board, their inverters were starting just fine without that choke inductor (I bet their controllers use a soft-start procedure too).

They just used that choke to reduce the no-load current (simple speaking). But like I've just said, if you apply a HF signal to a LF transformer there will be a lot of core losses due to larger eddy currents build-up.

 

[BradtheRad]

Oustanding simulations! Could you try one more using my real parameters (f = 12.5 kHz, L = 13 uH, leakage inductance of 10 uH)? Thank you very much for your time.

I started to try this, changing the transformer to 13uH primary. I found it admits kA. I guess my simulation is quite different from your topology?

 

[kathmandu]

I started to try this, changing the transformer to 13uH primary. I found it admits kA. I guess my simulation is quite different from your topology?

Thank you for your time! It wasn't my but @Warpspeed oppinion that a choke inductor could limit the inrush current that much. I was just curious if a smaller inductor (13 uH) could do any filtering of the switching frequency (12.5 kHz).

 

[Warpspeed]

I've read some discussions on another board (energy matters) where some guys modded their commercial products by replacing the E laminated transformers with toroidal ones.

To decrease the no load current, they indeed have used some choke inductors (like you have described).

Yes that is right.
And if you read through that thread again I was significantly involved in that thread, especially in the design of that choke.

Anyway, that idle current (without that choke inductor) is due to high frequency voltage applied to the laminated core (eddy currents are much bigger at higher frequencies).

Are you sure about that ?
Work out the Xl of only a few uH it is nothing.
All it does is slow the initial rise time of current, its a turn on snubber.

That's it, by using that choke inductor they actually filtered those HF (btw, they also have used a capacitor across the secondary winding - a small one (0.47 uF) though).

That choke is too small to filter anything, it hardly changes the voltage waveform, it only knocks the corner off the current waveform.

I have no problem with the inrush current as I'm using a PWM duty cycle ramp-up at the inverter start-up.

That is not the inrush I am talking about.
Every time one of the mosfets turns on, its looking into almost a dead short, without some inductance to limit the rate of current rise.
It does it every switching cycle, not just when you first apply power.

However, I don't think that such a small choke inductor (few uH) could help avoid inrush current for a bigger toroidal transformer.

Those toroids have significant flux doubling problems too, but its not that.
Its the humongous capacitor wired straight across the secondary to do the filtering that when multiplied by turns ratio squared is enough to frighten a mosfet to death.

Speaking of the experiments made by those guys from the other board, their inverters were starting just fine without that choke inductor (I bet their controllers use a soft-start procedure too).

They just used that choke to reduce the no-load current (simple speaking). But like I've just said, if you apply a HF signal to a LF transformer there will be a lot of core losses due to larger eddy currents build-up.

As I remember, I posted some loss curves for cut tape toroids to put some numbers onto what others had already discovered.
Adding the choke actually improved it further, and do you know why ?

Because even with nothing connected to the inverter output that humongous capacitor was still there causing a massive current spike at every PWM cycle, loaded or unloaded.

And I was one of "those guys", and I posted over there also as Warpspeed.

 

[kathmandu]

Hey, nice to hear you were involved too (that thread is way to big to read all those posts). Then how do you explain that they (you) didn't use any other series inductor for filtering purpose (if that choke was for avoiding switching inrush current only)?

Its the humongous capacitor wired straight across the secondary to do the filtering that when multiplied by turns ratio squared is enough to frighten a mosfet to death.

It wasn't just a 0.47 uF capacitor? Multiplied by 64 it's 30 uF, indeed. It does the filtering alone? There's no inductor involved at all?

Those toroids have significant flux doubling problems too, but its not that.

Could you elaborate about that, please?

And if you read through that thread again I was significantly involved in that thread, especially in the design of that choke.

Could you point me to the exact page of interest? Thank you very much for your time.

 

[Warpspeed]

That "little" 0.47uF is magnified to something like 47uF, and as you say, assuming a turns ratio of something like 24 volts to 240 volts.

Cannot remember what the switching frequency was, but I recall it was fairly high.
Trying to charge and discharge 47uF between +24v and -24v at maybe 30 to 50 Khz requires quite a bit of energy. And it's all a massive current spike as that sucker charges and discharges alternately each way.

Its still a lot of wasted energy, but at least with the small choke the current spike is much reduced, and remember conduction losses are proportional to current squared.

The thread is a very long one, and as it progressed, we knew the choke helped a lot, but there were various theories about why. I pointed out that the inductance was very small and all it could do is act as a turn on snubber, not as a filter.
That was generally accepted. That choke was very difficult to wind as it was, and I suggested using a larger UI core instead of the original EE cores. Bigger but much easier to wind with thick welding cable. There were pictures of that posted.

Now the original iron transformers were lossy because of high magnetising current, eddy current may have been a factor, but those iron transformers still draw a lot of current even when excited with a 50 Hz sine wave, without any high frequency PWM.

The grain oriented tape wound toroids have much more inductance per turn, much higher permeability, and the inductive reactants is much higher.
So under no load the idling current will be lower.
Everyone was pretty excited about that, and with good reason.

One of the problems with these toroids is they hold their magnetism rather well when turned off. (high remnant flux). When power is applied again, they can easily saturate on the first half cycle, but a soft start will fix that.
If you google "flux doubling" it explains that phenomenon.

All the filtering was done by that 0.47uF which effectively shorted out all the higher PWM switching frequencies.
Interestingly the secondary inductance of the transformer resonates with that capacitor only slightly higher than the mains output frequency, so the XL and XC tend to almost balance, and although the resulting circulating current may be reasonably high, is fairly efficient, and it does not significantly load the inverter.

Its a very long thread, and it was some time ago. It would take me a long time to read through it all again. I can pretty much remember how it all went, but cannot point to specific posts or pages.

We all of us contributed many ideas, it was a terrific example of how a bunch of guys can motivate each other, kick ideas around, and come up with some fairly original thinking. All jolly good fun, and I think we all learned a lot from it. I sure did.

For those with time on their hands, and real stamina, here is a link to that 124 page monster thread.
**broken link removed**

I came into it a bit late, only joined the thread on page 23

 

[kathmandu]

Thanks for sharing your memories! I'll try to read that thread from the beginning and I hope to clarify my doubts.

Anyway, I still can't make my inverter work properly. After reading your post, I've just made few more tests with the output filter.

For now I'm using a sendust toroid inductor (100 uH, 50 A) in series with the primary winding.

If I'm using just a 0.47 uF capacitor across the output (secondary winding), the LF transformer is humming pretty loud (though the wave form looks clean enough).

If I further add few 2.2 uF capacitors (for a total of three) the humming sound starts to fade. Even after adding those 3 x 2.2 uH (and the existing 0.47 uH) there's still a humming noise from my LF (toroidal) transformer.

Anyway, from time to time (as the load changes), that humming sound disappear almost completely then it comes back.

What could be the problem? I really want it to be silent (not for the noise sake but for its health). Seems like there are some parasitic oscillations build-up depending on the current load (or at no load).

 

[Warpspeed]

If you have an audio oscillator and oscilloscope, measure the resonant frequency of your transformer with the capacitor fitted across the secondary.

It does not need to be perfect, but an Fr of about x1.5 mains frequency would be about right. Say 75Hz for a 50 Hz inverter and 90 Hz for a 60Hz inverter.

As to the hum it may just be a characteristic of the particular transformer, difficult to say.

 

[kathmandu]

I've read that thread up to page 43 or so.. I've seen what those discussions were about but I guess you guys were wrong.

Every time one of the mosfets turns on, its looking into almost a dead short, without some inductance to limit the rate of current rise.
It does it every switching cycle, not just when you first apply power.

Every time a switch closes, it sees an _inductor_ (the primary winding of the transformer) not a _short_. There's a different approach of an inductor in AC versus DC environments.

According to a basic physics law, an inductor will always deny any fast change of the current flowing through it. That's it, there is a strong reaction from that inductor against fast current ramp-up.

If we're talking about the primary winding of such a (toroidal) transformer, it could be in a range of tens of mH (thus much larger than those tens of uH of a ferrite choke).

At the usual switching frequency (5-20 kHz), that transformer's primary winding is anything but a dead short.

So it's not the Mosfets are in danger at the switching frequency but the transformer itself. As you know, the losses due to eddy currents, core and conductor skin effect increase with frequency thus this is where that huge "idle power" goes.

There is an abnormally condition when you just put a capacitor across the secondary (or the primary for that matter) without adding a series inductor. You're actually shunting that primary/secondary hence now the Mosfets truely see a short.

But that's not how things suppose to work. If you want to add an output filter (LC) then add both the series inductor and the parallel capacitor. If not, just don't use any of them.

A Mosfet will feel happy to switch such a big inductor (tranformer primary); just don't short that inductor with a capacitor!

That's being said, that choke is no rocket science but the missing link of a classical LC filter.

 

[Warpspeed]

A transformer reflects whatever is on the secondary back into the primary modified by either the turns ratio, or turns ratio squared if its an impedance.

If the secondary is completely open circuit, then yes the primary does look like an inductor.

But what if you dead short the secondary ?
Does the primary still look like a high impedance inductor ?

What if you connect a very large capacitor right across the secondary ?
Will not whatever drives the primary be looking straight into a large highly capacitive load ?

 

[kathmandu]

That's exactly what I was saying: the (toroidal) transformer itself it's not a problem for the Mosfets. The problems arrive when you short its primary (or secondary) with a capacitor.

The inverter (from a Mosfet point of view) works happily with a bare transformer as a load (or an inductor for that matter).

On that thread, people were fascinated that a "simple" magic like adding a choke could save that inverter. Well, behind that magic, there was a dummy capacitor thrown across the output.

Once again, if someone wants to save that inverter (actually, the transformer) he just need to put BOTH the choke inductor and the capacitor in place.

 

[FvM]

It easy to agree that you shouldn't use a filter capacitor without appropriate series inductance.

I was reviewing this thread lately and noticed that you never mentioned the applied PWM algorithm, if it's state-of-the-art three-level switching and if so, what's the implemented dead time. A sine pwm driven H-bridge might introduce some harmonics due to inverter non-ideality, but usually with a level far below the amount seen in the first posts. So there could be a different reason.

 

[kathmandu]

I'm using an unipolar PWM scheme.

During a (50 Hz) half wave, the low-side Mosfet of one bridge's leg is conducting (keeping the load to ground) while the other leg's switches are driven using complementary PWM signals. After zero crossing, the legs are reversed.

I've tried a switching frequency of 10 kHz, 12.5 kHz and 16 kHz. The dead time was constant (around 0.8 us).

My problems (at the begining of this thread) were due to long & thin wires used to connect the inverter to the battery bank.

The reason for that voltage clipping was the insufficient filtering capacitor (I was in the process of testing various L and C values) thus there was a large voltage drop across the series (choke) inductor.

Do you recommend a specific topology for that LC filter? Is there a problem if the capacitor (much smaller) is placed across the secondary?

 

[FvM]

Unipolar sounds good and shouldn't involve too much distortions if dead time is chosen appropriately (no too high). The optimal value depends on the gate driver circuit and it's delay skew. I would rather expect a value of 100 to 200 ns for MOSFET.

When placing the filter capacitor at the secondary, you have the transformer leakage available as additional series inductance. If the output filter quality factor is too high, you may want to split the output capacitance and use a series resistor for a part of it.

 

[kathmandu]

Unipolar sounds good and shouldn't involve too much distortions if dead time is chosen appropriately (no too high). The optimal value depends on the gate driver circuit and it's delay skew. I would rather expect a value of 100 to 200 ns for MOSFET.

I think I'll decrease the deadtime then. It was 300 ns at the begining but I had some trouble with that optocoupler output inverter circuit (if you remember the thread) hence I have increased it to 800 ns to compensate signal slow rise/fall times.

What's the effect of a too long deadtime? How the output wave form looks like (or what else should give any clue about that)?

If the output filter quality factor is too high, you may want to split the output capacitance and use a series resistor for a part of it.

Looks like some type of snubber (RC)?

 

[Warpspeed]

Returning to the thread on the other Forum,

The toroidal transformers they were using were salvaged from a large batch of 1.5 Kw Chinese grid tie inverters that had to junked for scrap value by the importer due to not being compliant with very strict Australian design rules for grid tie.

I obtained several of these myself, and was curios to see how the PWM output filtering problem was handled. These particular inverters used a full bridge of four IGBTs switching at 20 Khz from a +200 volt dc rail with about 2,400uF of reservoir capacitance.

The IGBTs are coupled to the transformer primary through a series 2.7mH 30 Amp silicon steel UU cored choke. The transformer ratio is 135v to 240v 50 Hz, and it had a 5uF capacitor directly across the 240v secondary, and no other output filtering.

Some testing showed that the 5uF and the inductance of the secondary resonated at about 82 Hz. That seemed strange until I realised that is very roughly about x1.5 the mid way between 50 Hz and 60 Hz, and doing that would prevent a big resonant build up of energy when running unloaded at either frequency.

As the resonant circulating current tries to build up each half cycle, the next half cycle is mostly out of phase and tends to damp out any energy build up when otherwise unloaded.

But its also close enough to 50/60 Hz that the inductive and capacitive reactances will not be too different and tend to cancel out and not present a big parasitic Va load across the output.

I never ran one of these inverters up myself, as at the time I did not have a high enough current 200v dc source to power it up with.
But I am told the output waveform is a very clean sinusoid.

Anyhow, that is how at least one commercial high voltage PWM system has been put together.