SystemD_2kW, any interest for an open design?

....vertical through hole SMD cercaps.. :cool: :D:D

4 Layers:
Well, 4 layers would definitely be more fortunate and allow lowest parasitic inductances, but with two layers you can already achieve good results and realize a mostly continuous ground plane design.
And there is a nice advantages of the 2 layer design (beside cost):
You can easily debug every trace. In a 4 layer PCB all inner layers except full planes are unpleasant.
 
"Mostly continuous" is the issue. Even a small split in the ground plane can increase noise levels by very large amounts unless you are VERY careful with component positioning.
I guess I'm too used to the fact that nearly everything I design for my employer uses 4-layer boards, that apart from the connectors are 95% SMD. It's the easy (but expensive) option of course.
 
I would like to mention that the post-filter NFB topology response can be severely impaired when the opamps with finite GBP are taken into account. Here are open loop signals on the outputs of opamps (post #122).
1)VCVS, gain 1000
2)OPA134
3)TL082
 

Attachments

  • openloop-vcvs.gif
    openloop-vcvs.gif
    12.5 KB · Views: 900
  • openloop-opa134.gif
    openloop-opa134.gif
    13.1 KB · Views: 897
  • openloop-tl082.gif
    openloop-tl082.gif
    12.1 KB · Views: 892
Previous plots were calculated for R104=47k. Attached are simulations for correct value R104=4.7k
1)VCVS, gain 1000
2)OPA134
3)TL082

Charles, what was your target for phase margin?
 

Attachments

  • openloop-vcvs-4.7k.gif
    openloop-vcvs-4.7k.gif
    11.6 KB · Views: 869
  • openloop-opa134-4.7k.gif
    openloop-opa134-4.7k.gif
    12.4 KB · Views: 852
  • openloop-tl082-4.7k.gif
    openloop-tl082-4.7k.gif
    12.3 KB · Views: 48
Hi Dimitri !
Absolutely right. Charles is implementing the D portion in the forward path with OPamps, in order to avoid the caps in the feedback. He does not like feedback caps, because in many designs their values turn very small and it is hard to get types which are welcome in the audio path.

Some insights of my design:
I am not afraid of the caps in the feedback, because I found simple ways to keep the values large enough which allows to get preferable cap types easily.
But I am generally afraid of running the D-portion through the OP amps.
This concern is independent from an implementation with feedback caps or in the forward path.
In case your OP amp is to slow, the oscillation frequency will drop remarkably below the theoretic value . The triangular shape of the carrier will be impacted and consequently distortion will go up.
Furtheron it can be expected that large HF sloping would activate multiple additional distortion mechanisms inside the slow OP amp itself.
For all these reasons I am running the D-portion through the common base arrangement of the BJTs only (post#151).
And already in my configuration a TL082 would be much to slow and an OPA2134 still slightly to slow. The NE5532 is really the limit of acceptable speed. Real measurements with NE5532 showed that in terms of HF accuracy high voltage levels are more critical rather than high output currents.
Consequently I designed the loop on purpose with small HF voltage at the OP amp, but make use of the driving capability to keep the gain behind the OPamp reasonably high. Last but not least I have to drive the NE5532 with low impedances, otherwise its low differential input impedance would divide down the signal and the results would deviate from what the GBW graph in the data sheet makes us think.
Also the shaping network (R220, C209) is a questionable attempt, when using the NE5532. This network is more intended for upgraded builds with LM4562, or LT1364, or any fast OP amp of the DIYers choice.

In most cases of class D amps running the D-portion through the OP amp would demand to use types like LT1364 or AD2652 in order to avoid increased distortion (clocked & self osc.) and frequency drop (self osc.). Even then I am not convinced that it is a fortunate configuration. High gain for high frequencies through a long chain, might be more critical from perspective of disturbances, I guess.
 
Last edited:
The updated schematic now contains tons of provisions for different comparators including LM361.
Supplies upgraded.
Multiple minor adjustments in interaction with ongoing layout.
(Snubbers, CuBars,...)
 

Attachments

  • SystemD_2kW_Power_V0dot2.pdf
    149.6 KB · Views: 337
  • SystemD_2kW_Modulator_V0dot2.pdf
    116.6 KB · Views: 337
  • SystemD_2kW_Supply_V0dot2.pdf
    72.7 KB · Views: 290
Some screen shoots from my ongoing layout activities...

- Back2Back placement of MosFet and series diode ==> short wide CuBar between drain and cathode.
- SMD rail caps between power devices ==> critical loops minimized with front/back position of tracks with high di/dt and return path.
- Freewheeling diodes on the back side ==> no additional track inductances.
-OCP sensing nodes separately snubbered directly at the IRS20957S.
 

Attachments

  • Overview.jpeg
    Overview.jpeg
    264.5 KB · Views: 333
  • Zoom_all_layers.jpeg
    Zoom_all_layers.jpeg
    297.5 KB · Views: 375
  • Zoom_front_layer.jpeg
    Zoom_front_layer.jpeg
    278.1 KB · Views: 313
  • Zoom_back_layer.jpeg
    Zoom_back_layer.jpeg
    272.2 KB · Views: 323
Hi Kanwar,
...your request to add the LM361 is leading to a third comparator outline.
Altogether this is breaking up my GND plane below the comparator.
Nevertheless, I would not expect immediate noise catastrophies - still reasonable GND paths close to the signal available. So I am intending to go for it.
 

Attachments

  • triplecomp.jpeg
    triplecomp.jpeg
    839 KB · Views: 299
KiCad does not allow to set different clearances, as a function of position, for a filled area.
Nevertheless the necessary structures below the comparator would be OK for most PCB manufacturers even with 70µm copper.
So I added manually a GND grid.

Edit: The track thickness and and clearance setting for the grid is 0.35mm.
 

Attachments

  • triplecomp_grid.jpeg
    triplecomp_grid.jpeg
    821.2 KB · Views: 758
Last edited:
Multi Outline of Comparators

Today I set up a small jig for testing the intended multi outline before going on with the layout.
Also this is helpful for comparing the comparators.
The jig is build with identical routing as intended for the amp.
Attached some pictures showing the different positions of LM306 or MAX913 or LM361.
All comparators are working correctly with the intended routing.
The electrical testing was done with input signals of +/-10mV and a frequency of 2MHz.
LM306 is the slowest and shows slightly more self disturbance when driven with rectangles, but for a so called 'scrappy' comparator it is really good.

Besides measurements with rectangles, I also measured with +/-10mV triangles.
And there we can see, why I am stating that the fastest high end comparators are not the best choice for classD. When driven with small and slowly creeping signals these ultrafast high gain devices start to show serious jitter.
MAX913 performs great with rectangular input signal of reasonable size, but for input signals with low dv/dt it generates by far more jitter than a LM306 or LM361.
I am curious if we will see impact from this effect on the noise figures of the amp.

Overall the LM361 seems to perform very well.
My thanks to Kanwar that you pushed me to add it, despite its unpleasant size.

Edit:
All measurements were done with 10:1 probes.
Upper trace is the input voltage of the comparator.
Lower trace is the output voltage of the comparator.
 

Attachments

  • MAX913tri_853x640.jpg
    MAX913tri_853x640.jpg
    508.7 KB · Views: 109
  • LM361tri_853x640.jpg
    LM361tri_853x640.jpg
    456.6 KB · Views: 112
  • LM306tri_853x640.jpg
    LM306tri_853x640.jpg
    512.2 KB · Views: 102
  • MAX913rect_853x640.jpg
    MAX913rect_853x640.jpg
    515.1 KB · Views: 103
  • LM361rect_853x640.jpg
    LM361rect_853x640.jpg
    502.7 KB · Views: 115
  • LM306rect_853x640.jpg
    LM306rect_853x640.jpg
    496.8 KB · Views: 478
  • MAX913jig_853x640.jpg
    MAX913jig_853x640.jpg
    448.9 KB · Views: 524
  • LM361jig_853x640.jpg
    LM361jig_853x640.jpg
    441.6 KB · Views: 578
  • LM306jig_640x853.jpg
    LM306jig_640x853.jpg
    442.7 KB · Views: 648
Last edited:
Power stage Irfb4332+ir2110, +/-107v, 4r dummy load, with load works ok, but not superior to ucd at clip, frequency drop the same.
This is what i have in hand for a little experiment (ucd-like builded about 3 years ago, but works ok even in 2r, offcourse-with power supply sag):
 

Attachments

  • mod1.jpg
    mod1.jpg
    82.5 KB · Views: 302
  • mod2.jpg
    mod2.jpg
    87.4 KB · Views: 320
  • ucd.jpg
    ucd.jpg
    90.3 KB · Views: 354
Hi Ionut,
your posting does not show any key information.
You are stating that you tried a class D amp as 3D pin2pin construction and it was not better than an older design with a GND plane construction.
Considering that the layout makes 50% of the design of a class D amp, your findings are not surprising.

Regarding circuitry I can only guess you tried one of the modulator schemes of this thread with a completely different power stage and higher rails.
If done right, 3D pin2pin can operate in a satisfying way also for class D.
But I have to disappoint you - in the moment my priority is clearly to bring forward my design rather than debugging yours.

I understood that you did some comparative measurements before your experimental set up died.
If you could post details of your experiment and screenshoots of the measurements it might be interesting to have a look at.

Well, there is still your UcD amp.
IMHO UcD is a reasonable benchmark.
I have no UcD amp on hand, but we could use measurements from yours to set the mark.
I.e.
- Screen shots from clipping ( 2R and no load )
- Screen shots from the output signal of a 10kHz sine wave at high output levels, say 10% below clipping
- Screen shots from load dump
- THD measurements
- 10kHz rectangle into 2R and no load
 
Please don't get me wrong, i hope that your modulator design is better, because actually i have to counteract the cracking sound at clip with a pre-clipper tracking power supplies and i feel that is not right. This cracking sound happened to be worst when feeding 2-3 way loudspeakers and almost imperceptible from feeding a woofer only, even running to 2khz. I think that frequency drop at clip inside the audible range of a HF driver.
I donot have a good scope at hand, only one very old. The waveforms looks like in this post: http://www.diyaudio.com/forums/class-d/195243-hard-clipping-classd.html
 
Hm, the wave forms shown in your thread are looking much better than your verbal description. Your words are pointing more to the direction that the amp falls into interuptive operation... - which is not easy to catch without a storage scope. ...but already with a cheap USB DSO it should be possible to find.
I guess you already followed Jens' hint and tried larger caps for the bootstrap and floating supply?
Edit: I see - you tried...
 
Last edited: