Development of a "reference" class D starting point

[Kenshin]

great!

Originally posted by sovadk
I've build an amp based on the schematis I've posted earlier in this thread (page 55). Additionally short circuit protection has been added, to prevent failure.
It's build of 95% smd components and all hand made, even the pcb.
The losses are quite low. Each mosfet dissapates >1W at zero output and a few watts at full load.
The modulation is done kinda Mueta.

Switcing amp review 9:

[kartino]

Hi Kenshin,

Actually the amp use FDD3682 instead of IRF6665. You see the picture, th mosfet is black not as IRF6665, as Sovadk said before.

Yes, it is great result in speed, and simple of control circuit.

[kartino]

Hi all,

I have MikeB simulation in LTSpice version with P-FET/N-FET. I make some modification such feedback after filter and double mosfet.

But for mosfet I have no library but use as available. I think no problem for simulation.

Please take a look.

Happy Holiday....

[kartino]

This is the file, sorry. You can see that the output is symmetric.

[classd4sure]

Hi Kartino,

This sim doesn't appear to be class d at all? Seems to be far more complex than needed too, especially just for a P/N output stage.

Regards,
Chris

[sovadk]

It does not needs high grade gate drivers, but needs IRF6665... what's the actual rise/fall/dead time? Besides, does it need matching of input pair transistors? (It's a problem common for class D and class AB, but I just doesn't knows it)

I don't think you need IRF6665 or FDD3682. The high side driver has a large delay of 100-150ns, so a MOSFET with a large gate charge won't cause much difference. Mabe it will effect effciency.

The actual rise time for the low and high side driver is arround 50ns, but the propagation delay is ~15ns for the bottom and 100ns for the high side driver. With the delay of the comperator stage, the delay for the whole power comperator is arround 200-250ns. This varies with duty cycle, because the inductor current aids the switching.

I'm using mached transistors for the input stage, but I'm unable to harvest the benefit of this. Since the low side driver is much faster than the high side driver, the input stage always work with a DC offset to compensate for this. This causes distortion, even if you're using a mached pair. This can also give problems with DC supplypumping when a load is attached.
If you can figure out a clever way to mach the delay of the two drivers without affecting rise times and effciency, you'll make me a glad person.

I really can't figure out how the specifications of X1, the small signal MOSFET that controls the high side, affects the propagation delay. I though that a mosfet with a low output capacitance would be the fastet and chose BSS131. The thing is that I have achieve faster propagation with bsh114 from Philips, which has a much larger output capacitance. How come?

[Pierre]

There is something I don't understand, sovadk:
The low side mosfet gate is excited by a NPN/PNP pair, the off voltage being the negative rail (OK), but when the mosfet is to be turned on, the gate-source voltage is 30V! Isn't that too high?
Shoulnd't you connect the collector of Q57 to Vboot instead of GND?

Best regards,
Pierre

[classd4sure]

Hi,

The thing is that I have achieve faster propagation with bsh114 from Philips, which has a much larger output capacitance. How come?

Hmmmmm, given the specs of the two devices that is confusing. I think most likely it could be a case of just another component that doesn't measure/perform as advertised? A detailed comparison would be fun. Or just easier to go with the one that works best but it would be nice to find out why. They're specs are so closely related it's almost as if one was released to compete with the other...

If you find it that slow, try using two mosfets cascaded, the gate of the second mosfet goes to the drain of the first which is tied to the positive rail, the drain of the second mosfet goes to both the load and to the gate of the first creating positive feedback and will snap it on regeneratively once it reaches Vth.

Alternatively I'd work towards making the drivers identical, from the comparator on, ensuring they suffer the same delay all the way through, or as close to it as possible. You're then fully taking advantage of using a dual N output stage. Do you think a BJT based level shifter would have an edge over a mosfet? I've experimented with the fet level shifters and found them to slew very slowly.

As far as offset maybe you can add a pot someplace to null it, or use differential feedback to cancel the effect, matched transistors won't be required.

I think your layout could be improved as well. For instance a dual layer board would have let you place all your SMD components on the opposite side of the PCB, with a ground plane on the other side along with the high power signals/output stage.

You'd improve things further allowing the mosfets to share the same heatsink, which can be salvaged from almost any computer SMPS, along with a few isolators. One mosfet on either side of the heatsink with it bypassed to ground will both shield them from one another and reduce emissions overall.

Also try to keep the mosfet leads as short as possible. That should also reduce EMI and give improved switching behavior.

The heatsink touching the sleeve of that cap looks like a possible source of noise. If you didn't have protection on it already I'd be alot more worried about it, causing a possible short.

The thing is, I'm seeing those mosfets + heatsinks towering over all the nice SMD work like a pair of twin tower RF generators. Even if you can't use dual layer some clever heatsinking arrangment could shield the SMD from them, somewhat.

BTW, FDP3672, very nice !

Regards,
Chris

[sovadk]

to Pierre
D35 and R41 clamps the voltage to the negative rail + 11V, so the gate drive voltage is well within range.

They're specs are so closely related it's almost as if one was released to compete with the other..

I don't see that. BSH114 has a 1A, 100V, Coss=21pF rating and BSS131 on the other hand has a 0.1A, 240V, Coss=8pF rating. In my view BSS131 should be capable of turning off much faster than BSH114, which dosn't seam to be the case.

Do you think a BJT based level shifter would have an edge over a mosfet? I've experimented with the fet level shifters and found them to slew very slowly.

I've tryed with a BJT and it was slower turning off than the mosfets. We're talking ~1us for the BJT and 100ns for the MOSFET. The BJT was a high voltage one with low output capacitance. I don't remember the part number or the exact specs, so I wont generalize about the results I got.

I've also played arround with R22 (the 130ohm resistor). Making this 390ohm dosn't affect things significant besides the power dissipation. I my view 390ohm should give a time constant three times slower than 130ohm.

About gound plane, a two side board and common MOSFET heatsink: What is see from IRL mesurements is no different from what I get in simulations. This suggests that the reason why the high side driver is slow when driving the output high, has something to do with the circuit design, choice of components, etc. and not the layout.

About FDP3672. Yes they have a lower Ron which is nice, but the output capacitance is larger, which will give a larger idle loss (P=U^2*f*C*0.5). I found FDP3682 would give the lowest loss @ full load (8ohm) in a 200W amplifier.
You have Ron losses, switching losses and capacitance charge losses to thing about when choosing your MOSFET's.
When LC-audio's amplifiers dissapate 20W in idle, it can only be because of the high capacitance charge losses, due to high supply voltage, choice of low Ron old generation MOSFET's and low deadtime. Correct me if I got it wrong, it's only a consideration of mine.

[classd4sure]

Hi,

I was looking at all the other specs as well, and how they match up given the same Vds etc. Not so far from one another in that area, they seemed very similar. But yeah one did have a bit of an edge, you say it performs less well, I don't find that hard to believe, would like to see how well each compares to their data sheets.

Sounds like the BJT you tried was hard in saturation and that's a simple enough fix over using a mosfet which leaves you with little alternatives. Perhaps your level shifting mosfet is far more prone to miller effect than the BJT would be. In most cases I've seen video amplifier/high voltage/medium power BJT's used for the level shifting.

I guess I didn't read your posts closely enough to know if this delay you speak of is caused by a time/propagation delay or just a reduced slew rate which you can certainly expect (and have to try and keep minimal).

I see now it's a time delay, isn't it?

I might be wrong, but it's looking pretty simple to me while I look at your schematic now.

You have to think what's turning on the mosfet and what's turning on the BJT? Voltage / current? They're "thresholds" are very different, yet you're driving them identically, hence the big delay. Rather than operate it as a switch, why not as an amplifier? You've many options here. Once you've got that fixed up you should have no problems removing R39.

I know you didn't ask for layout advice and no that doens't relate to the delay at all, I didn't mean to imply it did, those were just further observations on my part and I think all good ones... something to think about for the next one that's all. Your SMD work is real nice though.

About the FDP3672, it got my amp rocking. "Optimized efficiency at high frequencies". I've tried them both btw. There's just so much more to the story aside from Coss, I really think you put too much emphasis on it, sure it's a factor, one of over a dozen, and they all matter in their own way.

I think the little bit of dissipation caused by a slightly higher Coss is negligeable between the two compared to everything else. That's all icing on the cake though.

As per the Zap and its 20W idle dissipation, I think it mostly has to do with increased idle current, which helps lower THD.

Also, I'm unable to spot your reference to D35 and R41, or is that D36 and R47? That still does nothing to clamp the lower driver's rail though. Pierre is correct, Q57 collector needs to go to Vboot. Symmetry is a key thing in all aspects of the output/driver stage, that's why we prefer dual N channels etc. I think you're real close with this though and it won't take much to nail it all down.

Regards,
Chris