My english is not so good, what i want to say - with ucd like feedback i have cracking sound at clipping, but with the pre-clipper band aid everything is ok, 2r-open load; i have tryiet your modulator schematic and at clip with 4r load everything is about the same, switching frequency drop, producing the same cracking sound; then i disconected the load (with 100hz sinewave for testing still clipping) and BooM...
Discrete class d can have some strange behaviours.
You can sometimes get bus pumping causing the power supply to rise above the capacitors rated voltage. Inputting low frequency sine waves is the worst culprit.
Sometimes the clipping can cause Vb to discharge and you lose the top transistor gate drive or even worse the gate drive goers into linear mode and blows up the transistor.
You can sometimes get bus pumping causing the power supply to rise above the capacitors rated voltage. Inputting low frequency sine waves is the worst culprit.
Sometimes the clipping can cause Vb to discharge and you lose the top transistor gate drive or even worse the gate drive goers into linear mode and blows up the transistor.
Hi Ionut,
...as Nigel says there are many possible reasons to struggle with...
In your case you have tried two very different modulators, but found the same issue. This points to an issue in the circuit parts which were the same in both trials.
If your level shifter and/or power stage is in trouble with clipping, then generally loop structures with integrating portions in the loop gain will be more critical. Depending on the issue in the level shifter and/or switching stage this might even lead to defects - fitting to your experiment.
...as Nigel says there are many possible reasons to struggle with...
In your case you have tried two very different modulators, but found the same issue. This points to an issue in the circuit parts which were the same in both trials.
If your level shifter and/or power stage is in trouble with clipping, then generally loop structures with integrating portions in the loop gain will be more critical. Depending on the issue in the level shifter and/or switching stage this might even lead to defects - fitting to your experiment.
Quite interesting.
Could it be layout and GND (no plane), which makes the MAX913 perform so badly? ... would guess that this matters more and more, the faster the device is.
Thanks for this investigation.
/Baldin
The top layer of this bread board is a GND plane, or at least a GND grid.
A completely closed plane might be better -yes.
In any case within a certain environment with some noise the picky comparators are likely to show more jitter than the less picky ones.
Please note that the MAX913 does not perform poor. The input signal is very small and with low dv/dt and for sure contains some noise in itself.
Of course this test situation magnifies the differences on purpose.
In the amp the comparator will operate with signals that show dv/dt of approx one order of magnitude higher. This is the good side.
The bad side is that the switching power stage will generate a more noisy environment than the test set up...
We will see if the found differences turn visible in the amp.
A completely closed plane might be better -yes.
In any case within a certain environment with some noise the picky comparators are likely to show more jitter than the less picky ones.
Please note that the MAX913 does not perform poor. The input signal is very small and with low dv/dt and for sure contains some noise in itself.
Of course this test situation magnifies the differences on purpose.
In the amp the comparator will operate with signals that show dv/dt of approx one order of magnitude higher. This is the good side.
The bad side is that the switching power stage will generate a more noisy environment than the test set up...
We will see if the found differences turn visible in the amp.
Good points.
I have been using MAX913 for some time, and have found it to be way easier to work with than both LT1016 and LT1711. It does require good layout and good decoupling, but it is fast and uses relatively little power.
What you call jitter her couldn't it be fast and precise reaction on changes in the input signal. And if the signal contains noise, you'll see smalle differences in switching times and therefore jitter. If the noise is higher in frequency than the signal, I guess you can place a LP filter to reduce the noise and thereby reduce the jitter.
/Baldin
I have been using MAX913 for some time, and have found it to be way easier to work with than both LT1016 and LT1711. It does require good layout and good decoupling, but it is fast and uses relatively little power.
What you call jitter her couldn't it be fast and precise reaction on changes in the input signal. And if the signal contains noise, you'll see smalle differences in switching times and therefore jitter. If the noise is higher in frequency than the signal, I guess you can place a LP filter to reduce the noise and thereby reduce the jitter.
/Baldin
V1.0 of PCB
One piece of V1.0 ordered.
V1.0 is always the most thrilling version.
Waiting until PCB arrives ==> some weeks.
Building ==> some more weeks.
Debugging ==> don't ask.
Attached two PDF files which show top side (GND plane) and bottom side in order to give some food to the thread.
One piece of V1.0 ordered.
V1.0 is always the most thrilling version.
Waiting until PCB arrives ==> some weeks.
Building ==> some more weeks.
Debugging ==> don't ask.
Attached two PDF files which show top side (GND plane) and bottom side in order to give some food to the thread.
Looks nice
What are the MBR7030 used for?
Why not more SMD components? Would probably make it easier to have a less broken GND layer.
Year always a long wait for PCB ... have just ordered PCBs myself .... but in the oposite direction 5x5cm boards hopefully capable of between 100 and 150W ... let's see
What are the MBR7030 used for?
Why not more SMD components? Would probably make it easier to have a less broken GND layer.
Year always a long wait for PCB
... have just ordered PCBs myself .... but in the oposite direction 5x5cm boards hopefully capable of between 100 and 150W ... let's see
Last edited:
Both MBR7030 avoid that the body diodes will be flooded.
The question about body diodes is one of the oldest in classD and always a hot discussion up to which power level at which switching frequencies it is making sense to use them.
Simulation showed that in this project the IRFP4668 would struggle hard with its body diodes.
Vienna Tom and Workhorse pointed out that already top brands had tried to go this path, but had issues with reliability.
So I stepped into manual calculation and noticed that the simulation was misleading in a forgiving way.
Altogether a clear indication to spend the additional components.
I have chosen a low amount of SMD components, because many DIYers complain about SMD. So this design is a trade off.
Intended for high performance, high power - still DIYable for experienced enthusiasts who like the design.
Of course not a beginner project, because of it's complexity and involved high voltages.
In case you want to build it, please wait until the design is ready and has reached a sufficient level of maturity.
If you read this thread from the beginning you will find the evolution and many design considerations.
The question about body diodes is one of the oldest in classD and always a hot discussion up to which power level at which switching frequencies it is making sense to use them.
Simulation showed that in this project the IRFP4668 would struggle hard with its body diodes.
Vienna Tom and Workhorse pointed out that already top brands had tried to go this path, but had issues with reliability.
So I stepped into manual calculation and noticed that the simulation was misleading in a forgiving way.
Altogether a clear indication to spend the additional components.
I have chosen a low amount of SMD components, because many DIYers complain about SMD. So this design is a trade off.
Intended for high performance, high power - still DIYable for experienced enthusiasts who like the design.
Of course not a beginner project, because of it's complexity and involved high voltages.
In case you want to build it, please wait until the design is ready and has reached a sufficient level of maturity.
If you read this thread from the beginning you will find the evolution and many design considerations.
@MoschFet:
SIC are great, but I am not aware of types for reasonable voltages.
More in the voltage range of 600V and above with pretty high Rdson.
I am intending to set my overcurrent protection above 50A, consequently not just the body diode but also the Rdson is important.
My personal guess is that over the next 5 years GaN will evolve as most fortunate for normal class D applications. You never know. GaN was already promoted 3 years ago, but then they noticed that the dynamic Rdson was more than an order of magnitude higher than the static values which had been used as design criteria and promoted by the semiconductor design guys. In the mean time they seem to have overcome this issue and I am hoping that not to much further surprises will come up.
@Tom:
MBR40250 or V30200:
Both are perfectly fine to me.
From simulation including PCB parasitics the 200V types are easily enough, fitting to your statement.
I am settling a file with alternatively usable components for the power diodes, power MosFets and comparators. The file will show if it is already tested or not and which adjustments might be necessary to fit (i.e. OCP settings with other MosFets).
SIC are great, but I am not aware of types for reasonable voltages.
More in the voltage range of 600V and above with pretty high Rdson.
I am intending to set my overcurrent protection above 50A, consequently not just the body diode but also the Rdson is important.
My personal guess is that over the next 5 years GaN will evolve as most fortunate for normal class D applications. You never know. GaN was already promoted 3 years ago, but then they noticed that the dynamic Rdson was more than an order of magnitude higher than the static values which had been used as design criteria and promoted by the semiconductor design guys. In the mean time they seem to have overcome this issue and I am hoping that not to much further surprises will come up.
@Tom:
MBR40250 or V30200:
Both are perfectly fine to me.
From simulation including PCB parasitics the 200V types are easily enough, fitting to your statement.
I am settling a file with alternatively usable components for the power diodes, power MosFets and comparators. The file will show if it is already tested or not and which adjustments might be necessary to fit (i.e. OCP settings with other MosFets).
...here the updated schematic.
Compared to the older schematics it has improved low noise voltage references for the modulator without adding tons of components and a faster output filter and accordingly optimized loop gain parametrization.
==> Resulting in improved large signal step response into low impedances.
C201, C202 are kept NIP, but most likely I will have to place 47pF to avoid catching RF noise. Further changes can be expected when I start building and debugging...
Keeping you updated.
Compared to the older schematics it has improved low noise voltage references for the modulator without adding tons of components and a faster output filter and accordingly optimized loop gain parametrization.
==> Resulting in improved large signal step response into low impedances.
C201, C202 are kept NIP, but most likely I will have to place 47pF to avoid catching RF noise. Further changes can be expected when I start building and debugging...
Keeping you updated.
@IR:
IMHO the only high quality solution.
...just compare the effective parasitic series inductance of one huge MKP with LS22.5 vs 6 small paralleled MKP with LS5 each.
A single large cap will show at least one order of magnitude more parasitic inductance vs the six small paralleled ones.
Also Mundorf, Jantzen etc. cannot overcome physics.
I am using them as input caps and for cross overs, but it would not be clever for the output filter of a class D amp.
Another ignorant approach would follow the spirit of the age by and place a completely overdriven and distorting SMD XR7.
Very hip these days, because it's 'small and cheap'.
@Tom:
Absolutely right. Many thanks. It is not caused by your PDF viewer. I rechecked it in KiCad.
The schematic editor also does not display this dot, but the ratsnest has the connection. ...software wonders everywhere.... Fortunately the gerber data seems to be fine.
IMHO the only high quality solution.
...just compare the effective parasitic series inductance of one huge MKP with LS22.5 vs 6 small paralleled MKP with LS5 each.
A single large cap will show at least one order of magnitude more parasitic inductance vs the six small paralleled ones.
Also Mundorf, Jantzen etc. cannot overcome physics.
I am using them as input caps and for cross overs, but it would not be clever for the output filter of a class D amp.
Another ignorant approach would follow the spirit of the age by and place a completely overdriven and distorting SMD XR7.
Very hip these days, because it's 'small and cheap'.
@Tom:
Absolutely right. Many thanks. It is not caused by your PDF viewer. I rechecked it in KiCad.
The schematic editor also does not display this dot, but the ratsnest has the connection. ...software wonders everywhere.... Fortunately the gerber data seems to be fine.
