DCAC inverter

[hamdi.tn]

Hi everyone,
So i been doing an UPS project, i received my board few days ago, so i had a bit of time of play with it and check some stuff, for the most part it work's fine ( good for me  ) but i have a strange output waveform and am not sure what causing it ( with no load pic , and with SMPS 12W load ).
Schematic attached. Please advice 

[hamdi.tn]

More pic...

[hamdi.tn]

 

[Seekonk]

What part seems strange?  Looks normal for a MSW converter.  Those are just various termination effects.

[hamdi.tn]

Thank you ... finaly someone  it could be me expecting more than reality and since it's basicly my first power thing project i don't really know what my scope is showing me ... but i was expecting a wave form identical for all load and no load, those termination is the strange thing , the command wave is for a modified square wave, my expectation that i shouldn't have any of those termination.
With no load i think some capacitive thing is holding the voltage up ! could it be my scope probe !!! it desapeare when i hookup a resistor on the output.
But those spike on the 12W load is what i don't understand.
Am hoping those picture speak to someone more they do to me. cause am not sure of what am seeing.

[lewis]

The spiky waveform (SMPS 12W load) looks to be caused by ringing in the SMPS's input filter inductor. With a quasi-sine wave you're essentially turning the SMPS on and off very quickly with a fast risetime, and the input filter will ring and generate transient voltages. Try it with SMPSes of different brands and the result will probably be different. If you have an output filter on your UPS not shown on the schematic, this may be the cause.

The open-circuit waveform is caused by some capacitance somewhere, either in the probe you're using to measure this waveform, or in some output filter. A 100K resistor across the output should clean it up. The AC output is essentially floating an open circuit when all the MOSFETs are off.

[hamdi.tn]

There is no output filter at all , will try different loads and post the result
if it's the case of input filter ringing .. can you please suggest a way to eliminate this.

[hamdi.tn]

Hi again
so i been doing some experiment with the thing and it let the magic smoke out several time
But one particular experience is what blow out most of the DC AC part
first i was generating just a 50Hz modified square wave to the driver , but i saw in an ST application note that there DC AC converter use a MSW too but the ON time of the MOS was actually a 16KHz wave once filtered form the 50Hz output.

First am not sure why they need switching the ON time of  the 50Hz signal !!
i tried it out and my mos dead and drivers too.
the board was clearly using much more power and the load connected to it ( an old 300W ATX ) refuse to startup ..
what i did wrong here 

[Zero999]

Without seeing the application note you're talking about, it's difficult to know.

It sounds like it was talking about using PWM and a filter to make a pure sine wave. There are plenty of ways to mess this up.

[hamdi.tn]

AN2794 ,1 kW dual stage DC-AC, Nop i don't think they generate a pure sin it's a MSW

[MagicSmoker]

AN2794 ,1 kW dual stage DC-AC, Nop i don't think they generate a pure sin it's a MSW

Incorrect - that app note describes a pure sine wave inverter. The switches in the H-bridge are switched at 16kHz with the duty cycle following a 50Hz sine wave pattern. The high switching frequency vs. modulation frequency (ie - 16kHz to 50Hz) means you can use a simple/cheap LC filter to integrate the 16kHz PWM'ed square wave output into a decent quality 50Hz sine wave (both the current and voltage waveforms, courtesy of the L and C components, respectively).

The scope shots you posted suggest there is insufficient gate drive for the upper switches in the H-bridge. When using a charge-pump type gate driver IC (like the L6386), this either means the bootstrap capacitor is too small or else the IC can't deliver the necessary peak current for the switch. While this is nominally a good application for a charge-pump type gate driver, it has been my experience that these ICs are notoriously unreliable in practice; that said, I normally design power conversion hardware in the 100kw-2MW power range, so that experience doesn't 100% apply at the 1kW or less level.

[hamdi.tn]

AN2794 ,1 kW dual stage DC-AC, Nop i don't think they generate a pure sin it's a MSW

The scope shots you posted suggest there is insufficient gate drive for the upper switches in the H-bridge. When using a charge-pump type gate driver IC (like the L6386), this either means the bootstrap capacitor is too small or else the IC can't deliver the necessary peak current for the switch. While this is nominally a good application for a charge-pump type gate driver, it has been my experience that these ICs are notoriously unreliable in practice; that said, I normally design power conversion hardware in the 100kw-2MW power range, so that experience doesn't 100% apply at the 1kW or less level.

Hi, thanks for those info 
I had my doubt about bootstrap cap , but I checked the value , and it seems ok to me.
i check tomorrow if the current is enough to charge up the gate.

however what worry me about those spiky wave , could it be unintentional switching on of high side mosfet ?

[Yansi]

Any images of the pcb/layout? 1kW inverter is not a toy for a beginner. So I hope the board respect the power electronic rules and ratings.

[hamdi.tn]

The application note is 1Kw .. my ups is just 200W

[Yansi]

OK. But still, can we see the pcb please? It'll help us diagnose possible problems.

[lewis]

I normally design power conversion hardware in the 100kw-2MW power range

Wow! That's pretty cool, what sort of stuff?

[hamdi.tn]

OK. But still, can we see the pcb please? It'll help us diagnose possible problems.

This is the drivers and mos part of it

I normally design power conversion hardware in the 100kw-2MW power range

Wow! That's pretty cool, what sort of stuff?

yap 100KW - 2MW is pretty impressive     i guess it can be used from electric car , electric forklift , or huge UPS for data center.

[MagicSmoker]

...
I had my doubt about bootstrap cap , but I checked the value , and it seems ok to me.
i check tomorrow if the current is enough to charge up the gate.

Okay, I actually looked at your schematic and there are a few problems with it but the value of the bootstrap capacitor isn't one of them. Although it is generally a good idea to make turn-off happen faster than turn-on, I would simplify the circuit for now and just use a single 39R resistor in series with each gate (in other words, get rid of D15, D16, D18, D19, R47, R51, R58 & R60, and change the value of R45, R48, R55 & R59 to 39R). The trick of using a reverse-biased Schottky to speed up turn-off is a good one, generally, but the L6388 is only capable of sinking 650mA which means the parallel combination of the turn-on and turn-off resistances should be no lower than ~23R. To stay entirely within the specs of the L6388 but double turn-off speed you could make both turn-on and turn-off resistors 47R. In any event, you absolutely don't want/need R48 & R59.

Also note that the IRF840 datasheet states switching times with a series gate resistance of 9.1R (this value is usually the minimum, btw); driving the gate with 100R will slow down the switching by approximately 10x. That may be okay in your application (switching frequency is less than a few 10's of kHz), but you need to ensure there is sufficient deadtime between switches in each bridge leg to prevent cross-conduction and, regardless, larger values of series gate resistance increase susceptibility to turn on of the supposed-to-be-off switch from Miller current transients.

however what worry me about those spiky wave , could it be unintentional switching on of high side mosfet ?

It could be Miller-induced turn-on of the off switch, yes, or it could be from simple cross-conduction (ie- the next leg turns on before the previous leg has finished turning off), or from excessive inductance and/or insufficient decoupling capacitance in the DC link.

To check for cross conduction you simply measure the DC link current into the H-bridge with no load on its output; it should be 0A, or no more than a few 10's of mA at this power level. To see if Miller current is a problem you need to watch the gate waveform of both switches in each side of the H-bridge (ie - the upper and lower of one side). Finally, check for DC link collapse by watching the DC link with AC coupling on your scope. You might want to use a second channel to watch the gate of a low side switch and use that waveform for triggering.

[MagicSmoker]

I normally design power conversion hardware in the 100kw-2MW power range

Wow! That's pretty cool, what sort of stuff?

These days, switcher locomotive drive systems. GE and Siemens have all but exited the switcher market because the volume is too low, but for my company it's like Christmas come early.

And yeah, it is pretty cool... Except when something blows up, because when that happens at this power level you usually end up on anti-anxiety meds afterwards. 

[hamdi.tn]

...
I had my doubt about bootstrap cap , but I checked the value , and it seems ok to me.
i check tomorrow if the current is enough to charge up the gate.

Okay, I actually looked at your schematic and there are a few problems with it but the value of the bootstrap capacitor isn't one of them. Although it is generally a good idea to make turn-off happen faster than turn-on, I would simplify the circuit for now and just use a single 39R resistor in series with each gate (in other words, get rid of D15, D16, D18, D19, R47, R51, R58 & R60, and change the value of R45, R48, R55 & R59 to 39R). The trick of using a reverse-biased Schottky to speed up turn-off is a good one, generally, but the L6388 is only capable of sinking 650mA which means the parallel combination of the turn-on and turn-off resistances should be no lower than ~23R. To stay entirely within the specs of the L6388 but double turn-off speed you could make both turn-on and turn-off resistors 47R. In any event, you absolutely don't want/need R48 & R59.

Also note that the IRF840 datasheet states switching times with a series gate resistance of 9.1R (this value is usually the minimum, btw); driving the gate with 100R will slow down the switching by approximately 10x. That may be okay in your application (switching frequency is less than a few 10's of kHz), but you need to ensure there is sufficient deadtime between switches in each bridge leg to prevent cross-conduction and, regardless, larger values of series gate resistance increase susceptibility to turn on of the supposed-to-be-off switch from Miller current transients.

however what worry me about those spiky wave , could it be unintentional switching on of high side mosfet ?

It could be Miller-induced turn-on of the off switch, yes, or it could be from simple cross-conduction (ie- the next leg turns on before the previous leg has finished turning off), or from excessive inductance and/or insufficient decoupling capacitance in the DC link.

To check for cross conduction you simply measure the DC link current into the H-bridge with no load on its output; it should be 0A, or no more than a few 10's of mA at this power level. To see if Miller current is a problem you need to watch the gate waveform of both switches in each side of the H-bridge (ie - the upper and lower of one side). Finally, check for DC link collapse by watching the DC link with AC coupling on your scope. You might want to use a second channel to watch the gate of a low side switch and use that waveform for triggering.

thanks a lot  for all of this  
i will check all that once i get replacement components i post what i get

[hamdi.tn]

Hi, again, long time since i started this topic and i finally find a bit if time to do an other test run.
so i did what magicsmoker suggested and modified the schematics, and changed boot cap value cause it wasn't enough,  it didn't change a single thing.

i explain
with no load the output keep on falling slowly until the low side MOS short it to ground.
i observed gate waves on scope, am totally sure that the gate can see  the 100Khz switching . but the H-bridge doesn't ,i tried different frequency, different wave form ( true sin, modified square ) i only get what am posting on the first post, a square wave if no load is connected.
and by the end of the day, the driver is dead once again with no clear reason.

---

i return to a simpler circuit by using an L268 H-bridge, driven by same 100Khz waves i can totally see the switching on the H-bridge, and after adding an LC filter a quite nice sin wave is out ... but i don't understand why it simply collapse when i connect a load, even if the load is not that demanding (observed with inductive loads) .
i finally killed it too ,when connecting a 24VAC contractor 

am getting my ass kicked on this project and am not sure what am doing right or wrong, what piss me off the most is that the full circuit is taken of an application note, i used a similar component and not getting any results.
help please 

[mich41]

The 84W trace looks nasty. After low-side turn-off, there is a fast rise to roughly 0.6*VCC, then brief fall to 0.4, back to 0.6 and gradual fall to 0.5, until the high side turns on.

This dip to 0.4*VCC really really looks as if the low side switch was turning back on for a moment. If you verified that Miller effect doesn't distort gate waveforms (I hope you probed them at the FET, not at the driver IC...), maybe the FETs turn on due to excessive Vds rise rate disturbing their parasitic BJTs? See "dv/dt capability" here for details. The rise rate will naturally be fast if load current is high and/or the load is inductive.

You may need to drive the FETs slightly softer.

[hamdi.tn]

am sorry but am not quite following what you wrote, please explain

[hamdi.tn]

To explain a bit more what am going through
I soldered a chip on a prototype board as shown on sch.png, just a half bridge with low voltages ,the goal is just to test the chip
now what am getting is what i suppose to be a correct wave form on the gates both high and low side ( it's a 100Khz pulse blocs alternate at 50Hz) but the out pin is just not returning to 0V until the low side switch on.

[Phoenix]

but the out pin is just not returning to 0V until the low side switch on.

It looks like you don't have a load connected.