Class D driver (IRS20214) low duty cycle issues

Den torsdag den 12. februar 2015 kl. 00.28.09 UTC+1 skrev Joerg:

can you change the behavior by changing the deadtime, OC etc. to get an idea what is happening?

-Lasse

An audio part is probably intended to center on 50% duty cycle. So 2% is almost railed. I'd expect that the deadtime and delay blocks get confused near the duty cycle limits.

One thing that bothers me about all-digital audio is that the main V+ supply gets multiplied by duty cycle to drive the load, so, without analog feedback, supply ripple goes right into the load.

OC doesn't change anything. Dead time shifts make it move slightly which was my clue that it could be internal feedback, with the internal VS rail (the flying node) talking into the input.

The behavior is oscillatory, when touching the input you can trigger it partially on and off. But only partially, there will always be a jitter which causes the current to be fairly high. When the input is set to any duty cycle other than 2% the class D stage with the FETs draws only about 50mA from the HV rail. At 2% duty cycle this jumps to 80-150mA. By touching the inut pin you can make it switch between 80mA and 150mA but it will never drop to 50mA like it is everywhere else. Unless I move the duty cycle away from 2%, then it will immediately become quiet and drop to 50mA.

Very possible that there is an undocumented hiccup condition in the deadtime logic. That would be bad. The datasheet does not say anything about a minimum duty cycle though and in that case it should. That there is a maximum is normal because of the bootstrap capacitor although than can normally be circumvented using a flying supply.

Not in my case because I have a loop around the whole thing. That muffles out anything in the audio range. High frequency noise is inductively filtered out. Of course right now I have the loop unhooked to get to the ground of this 2% duty cycle issue.

These class D stages pack an incredible punch without becoming toasty. This is for a servo control application and I am tempted to hang a big fat speaker onto it and play Stairway to Heaven. It should make the rafters shake but I guess then the neighbors won't like me anymore. You could probably simulate an earthquake with it.

if you have time to wait maybe this will help by giving you a comparison (it's $20)

I was considering the IRS2029 initially but it won't work well because it doesn't easily do DC and also can't do 0-150V.

A data sheet is not a Catholic Confessional. Quite the opposite.

In the good old days it used to be. I'll ask the mfg but these days I have little hope that they know a good answer. Hopefully they won't make me go through the rigamaroo of some engineer fan club membership.

I don't remember those days! Ancient National data sheets were notorious for hiding hints of bugs in obscure application circuits and

I'll ask the mfg but these days I

Support is usually bad. I had an Analog Devices support guy clarify the trim behavior of one of the ADR-series references. He was wrong about everything, including the polarity. Had to ECO the board to get it right.

(I pointed out to him some obvious errors on the data sheet, and they're still there. Many old ADI data sheets are still rev 0. Recently ADI asked ME to correct one of them.)

In some cases, I'm confident that hundreds of people have individually discovered and battled a bug that could have been honestly announced in the data sheet.

Don't get me started on Xilinx.

Have you tried adjusting the dead time, gate series resistors, or putting a resistor in series with the bootstrap capacitor?

Check and check. No change. The deadtime moves the spurious cycle a little which, of course, was expected. But it won't make it go away even a bit.

Haven't tried that. It would be really weird if the IC needed that. I think there may be something screwed up in the deadtime control inside like what John was saying.

What's your PWM frequency, a 2% duty-cycle pulse is how wide?

A powerful and relatively painless way to make very short PWM pulses is to use a phase-shift PWM. We're talking H-bridge, driving both sides of a floating load. Each half-bridge is driven with a 50% duty cycle square wave, and the relative phase of the two square waves is shifted either way to create a bipolarity output. Examples of phase-shift controller ICs: UCC3895, UCC3875, UCC28950, LM5046 and LTC3722. I used the UCC3895 with my favorite HIP4081A driver as the basis for a precision 600kHz 500W resonant-transformer driver. RIS-514.

Some switching power supply chips recommend a resistor in series with the bootstrap capacitor if EMI, glitching, or ringing is seen anywhere in the circuit. It's being charged and discharged with a square wave so it has to radiate a small amount of power somewhere.

It's 300kHz so it'll be 66nsec. However, if I lower the PWM to 100kHz the effect is the same and also at 2%. The chip seems not to low duty cycles.

In this case the load has to be grounded on one side which makes it challenging. I don't know what's so difficult about generating a clean duty cycle from 0% to 95%. Or even 100% if you accept that the part goes into upper rail UVLO when using a bootstrap capacitor. I've done it with discretes before, but do we really have to build everything out of discrete parts in this day and age?

So now I am wondering if any of the ZVS half-bridge or LLC half-bridge ICs are any better. I should be able to press one into service by ripping open the loop and driving the comp pin. Of course, I'd have to hand out sick bags before starting the design review :-)

This one has no such recommendation. It should not be an issue as long as you don't use an overkill-size diode for the bootstrap.