Split Phase Output From Full Bridge Inverter

[jbb]

So, now that we've seen a concept schematic we can start to say really confusing things 

The advantage of the 2 stage approach is that the DC DC transformer only handles high frequency and can be quite small & light. We can probably do better than the proposed PWM full bridge converter by moving to a phase shifted full bridge if size is critical ('filter' inductance is integrated into the transformer).

If we modify the DCDC stage to output +-170V, the ground point can be used to split the two output phases.
This corresponds to the last suggestion from T3sl4co1l. The downside is that each output phase will need some separate control elements (e.g. current sensors and current control loops), but that's hard to avoid.

[NiHaoMike]

How many kW are we dealing with? And what kind of loads?

I think it would be better to split the electronic loads from the motor and resistive loads. In particular, you can really reduce the power usage of motor loads with V/Hz scaling.

[jbb]

How many kW are we dealing with? And what kind of loads?

I think it would be better to split the electronic loads from the motor and resistive loads. In particular, you can really reduce the power usage of motor loads with V/Hz scaling.

This is an interesting point. It can be decoded to 'do you have motor loads that could benefit from running on a Variable Frequency Drive (VFD) aka Variable Speed Drive (VSD)?' It's often possible to supply 340V DC power straight into a VFD. This means that the DC to AC output stages can be smaller.

[Zero999]

Hero999, first a big Thank You for providing much better graphics than what I found.

I was under the impression that the idea was to use an h-bridge to convert the DC voltage from that battery to PWM AC and step it up to the mains voltage with a transformer. That will work the transformer still needs to be able to pass 60Hz.

Yes, but why include the extra 60 Hz transformer? Just step it up directly to 2 x 120 VAC with two separate secondaries and filters. Then the outputs can be used freely.

I agree. I didn't mean there needed to be two transformers. You don't even need two secondary windings with separate filters. A single centre tapped secondary winding will do.

My point was, that I thought the original poster needed a light and compact design, but a mains frequency transformer is big and heavy.

Some variations on the output network.

Note that the supplies are already shown as bipolar, so I added a ground reference, implying the supplies are not just plus and minus of a single supply, but a dual, symmetrical supply.

I don't think that was intentional. I can't remember where I got the original schematic from. I think it was for a modified sine inverter. I modified it and posted it on another forum about 10 years ago. If I remember rightly, it was meant for the h-bridge to be run off a 12V battery, hence the transformer. It should have been +12V and 0V, rather than +Vdd and -Vss.

Original


Two notes: 1. The transformer must carry the full low frequency content regardless.  2. LF transformers have high core losses at high frequencies.  Might as well move the filter in front.


The supplies and phases are complementary and symmetrical, so we can divide the LPF in half against ground (or, since the ground terminal is just capacitors, it doesn't care what DC is -- either supply would work as well, if you prefer).  This is identical if the phases are perfectly symmetric, so seems like a lot of component duplication, however in practice they will never perfectly match, and this provides equal common mode and differential mode filtering.  Huge EMI win!


Continuing the symmetry theme, the primary can be CT-grounded without problems*.

*Given that the flux walking constraint now applies to both outputs individually, not just their difference.  That is: DC output current must be very close to zero, for both outputs.

Also, as long as we're putting taps on things, the secondary can be tapped anywhere you like (not just half), getting the split phase that was asked about, or any other voltages for that matter.

Those are very good ideas. Thanks for posting them.

Since it's symmetrical, we can even eliminate the transformer altogether -- assuming isolation is not required.

Now it is especially important for the LPFs to be independent, rather than balanced, because there's no potential filtering benefit from the transformer.

Tim

It also assumes that +Vdd and -Vss are equal to the mains peak voltage, i.e. they're generated by DC:DC converter, otherwise the big and bulky transformer is still needed for voltage conversion.

[Zero999]

Here's what I was thinking of originally. C1 & C2 need to be low impedance enough to ensure the neutral point doesn't shift too much and be capable of carrying the required current. They could be rated to a lower voltage than 340V (I'd go for 250V), as they only see half the DC bus voltage, but balancing resistors would be required.



Note that the DC converter is isolated. The gate drivers would also need to be isolated if the PWM circuit is powered off 12V.

[jbb]

EDIT: removed incorrect screenshot.

That's the right sort of idea, but some refinements are needed to get it to work.

First, we need to make the split DC supply behave nicely when subject to unequal loading.  For this (and because Wrydog mentioned size is a concern) I would suggest something like a phase shifted full bridge converter. (First Last image).

The two legs of the H bridge (M1+M2 and M3+M4) switch at constant frequency and 50% duty cycle.  Control is achieved by varying the phase lag between the two legs of the bridge. This approach gives soft (or at least less hard) switching, it can operate at fairly high frequencies (> 50kHz) with acceptable efficiency.  A split winding output transformer generates a split DC output rail.  Large capacitors C1 and C2 are required to buffer energy over the 60Hz line cycle.

Secondly, we need an output stage.  In general, there will be two phases, each with their own power switches (MOSFET / IGBT), LC output filter and control system (second first image).  Current sensing for each phase will be needed somewhere - on the inductor output is traditional but not the only option.

The exact nature of the output stage is open to debate.  The simplest is a pair of half-bridges (second, third, fourth images) using MOSFETs or IGBTs.  MOSFETs will have lower conduction losses, but their internal body diodes can have large reverse recovery losses and therefore lead to high switching losses.  IGBTs have higher conduction losses, but much lower reverse recovery losses (because the diode is a separate piece of silicon, integrated into the same package).  These power stage both supply +180V or -180V to the output filters (i.e. 2 output voltage levels).

If size is critical, tricks like moving to a three level inverter are possible. (Final Fourth image).  This requires some more switches, but allows for the power stage to output +180V, 0V and -180V.  Because the applied voltages can be controlled to be closer to what you actually want, the output filter inductor can be approximately halved.  Additionally, because the switches are only switching 180V at a go (instead of 360V), it may be possible to increase the switching frequency (and shrink the inductor) without decreasing the efficiency.

[Zero999]

That's the right sort of idea, but some refinements are needed to get it to work.

First, we need to make the split DC supply behave nicely when subject to unequal loading.  For this (and because Wrydog mentioned size is a concern) I would suggest something like a phase shifted full bridge converter. (First image).

I don't understand the variac and isolating part. I presume the low frequency transformers are for test purposes and will be replaced with a switched mode DC:DC converter later?

I don't see why a split DC supply is essential. A half bridge converter will work, if AC coupled to the neutral, as I showed above. Unequal loading won't be a problem, as long as the voltage drop in the filter and switches is minimal.

The only problem I can see with my idea, is that the load needs to draw the same charge on positive and negative cycles, otherwise one of the capacitors will charge up more, causing imbalance. If it's likely such a load will be connected, i.e. load powered by a half wave rectifier, then a split DC power supply is the only way to go.

Note that the above can also be a problem with a transformer: if net flux becomes DC, then the core can saturate, resulting in excessive current draw and overheating.

[jbb]

I don't understand the variac and isolating part. I presume the low frequency transformers are for test purposes and will be replaced with a switched mode DC:DC converter.

Oh dear, that would be the wrong image... I'll replace it later.
Sorry

[T3sl4co1l]

IGBTs are chosen when they have lower conduction losses.  The switching losses are higher, so the switching frequencies go down, too.

For a 340V supply, IGBTs are meh.  For industrial (400VAC+) equipment, they're pretty good, and in large modules (100A+) they're quite good (as you can't really get MOSFETs that large, and they'd be too fast to switch that much current safely anyway, so why not get something cheaper).

In the <400V range (like here), it's better to throw enough MOSFETs at the problem to reduce conduction losses below IGBT levels.  You can even reduce it below diode levels: this prevents body diode forward-bias and subsequent recovery losses.

If you're in a situation where cost is more critical than efficiency and size, you might go more towards IGBTs.

At lower voltages (Vds < 300V), it's no contest.  They don't even make sub-300V IGBTs, no point.   (The ones they do make -- 300 to 400V, 300A+ pulsed, are used for driving plasma displays; or, they were, as PDPs pass out of vogue, now.)

Body diode recovery is kind of hard to avoid, and it is a big problem in a converter that has to handle four quadrant operation (that is, positive or negative, current or voltage, at the same time).  Correct snubber design helps a lot to mitigate recovery.

Tim

[jbb]

What kind of snubbers were you thinking?

A typical commercial product would use hard switching with as little snubbing as possible, but there are a lot of options out there.

[T3sl4co1l]

dI/dt snubbing is for diode recovery.

Tim

[Zero999]

I've just done a calculation on the vales of C1 & C2 in the schematic I posted previously and they'd need to be quite big: 1500µF for a neutral impedance of <1R @ 60Hz and at 250V is quite large. It would be smaller than a transformer, but I think a bipolar DC supply would be more compact, than massive capacitors, even if it does make it more complex.

[jbb]

C1 and C2 will be quite big anyway to handle 60Hz power swings - in a quality scheme, their voltages are allowed to wobble around a bit (i.e. store & release energy) which prevents the 120 Hz power ripple from getting back to the battery bank.

Of course, the size of the cap banks depends on the power rating, so once again:

How many kW are we dealing with? And what kind of loads?