SystemD_2kW, any interest for an open design?

SystemD_MD is by far to complex for an open DIY design.
Also I have the feeling that the class D community is heading for higher power.

Since two weeks I am thinking about a DIY friendly design, that could deliver about 2kW into 4R in bridged mode. I am considering to develop this as an open design for DIY class D enthusiasts.

Rough outline of SystemD_2kW:
- Useable as a halfbridge with 1kW/2R
- Useable in bridged mode with 2kW/4R
- High performance and well balanced properties, but of course not as blameless as SystemD_MD
- Simplified implementation of the triangle shaping method known from SystemD_MD
- Instead of mostly discrete, heading for a reasonably simple design
- Selfoscillating, because that's the way to keep component count attractive for DIY
- Hysteresis oscillator, not UcD
- Restrict the really critical parts to MosFets, level shifter and layout

Why a hysteresis oscillator, while the entire classD world is on the UcD trip (or on the IRAUD trip with delta sigma...)?
Compared to UcD the hysteresis oscillator needs just two resistors more and offers the chance to adjust the frequency shift by a dynamic hysteresis control.
At this point I have to highlight, that I am not the only one who has thoughts into this direction. I will post my circuit for the dynamic hysteresis control soon and would love to see postings with different circuits for a dynamic hysteresis control of other enthusiasts, in case you are willing to share it with the DIY community.
 
There are others who can do such designs as well.
Also in the 25W-1200W there are people actively working and for sure there are already valuable contributions including layouts.
I still have not proven that I can design a 2kW class D amp.
The 1.2kW of the bridged MD is the strongest class D amp I have designed so far. Up to now, I just think I could also do 2kW with good quality.


Attached my dynamic hysteresis control.
Please don't be shocked by the LT1711, it is just an avaiable comparator in LT Spice. For the real design I am intending LM160/LM360.
Also please do not wonder about the series connection of a 33V zener and 6.2V zener. LTSpice simply did not offer a 39V zener....

Functionality of the circuit:
At low output voltage Q7 and Q8 are completely OFF, polarity of Ube is reversely biased. No influence on the hysteresis rectangle formed by Rmk1 and R2.
Q9 and Q10 are running at low biasing (theoretically for fundamental circuit no bias needed). When the difference between output voltage and rails becomes smaller than the zener value of 33V+6.2V zeners, then Q10 or Q9 will be driven with increasing currents and pull up or down Q7 or Q8, which in consequence will more and more clamp the hysteresis rectangle. The smaller the magnitude of the rectangle, the higher the operating frequency. It is easy to control the frequency shift this way.
Still a good portion of frequency reduction at high levels is desired, because this is reducing the switching stress of the MosFets. The control circuit shall just avoid really heavy voltage ripple on the speaker signal as it usually happens in ultra simple selfoscillating amps.
In fact this principle is suffering from the parasitic phase shift oscillator mechanism, which will not allow unlimited high frequencies. Transfer function and propagation delay will limit the max. possible switching frequency, even if the hysteresis would be set to zero.
 

Attachments

  • DynamicHysteresisControl-800x800.jpg
    DynamicHysteresisControl-800x800.jpg
    53.3 KB · Views: 6,612
Attached you can see the influence on the intended amp, when running to the clipping limit with a 2kHz sine wave.
Basically also without dynamic control the amp is behaving in an acceptable way, but with dynamic control the ripple remains really nice even during last switching events.

First picture without dynamic control.
Second picture with dynamic control.

Edit: The blue trace is showing the hysteresis rectangle, which has a constant magnitude without dynamic control, but with control its magnitude decreases at high output levels.
 

Attachments

  • withouthysteresiscontrol-800x800.jpg
    withouthysteresiscontrol-800x800.jpg
    59.9 KB · Views: 6,232
  • withhysteresiscontrol-800x800.jpg
    withhysteresiscontrol-800x800.jpg
    57 KB · Views: 5,976
Last edited:
The difference is more obvious, when watching signals which run close to the rail - like large rectangles.
At 150Vpp output voltage the version without dynamic control shows approx 10Vpp switching ripple, while the version with dynamic control is showing approx 4Vpp.
 

Attachments

  • Rect_wo_dynamiccontrol.jpg
    Rect_wo_dynamiccontrol.jpg
    88.1 KB · Views: 5,738
  • Rect_with_dynamiccontrol.jpg
    Rect_with_dynamiccontrol.jpg
    80.9 KB · Views: 5,537
Slightly more than 300 views within two days.
Only one reply from a DIYer.

This and the monologue of my last thread is telling me that there is just little interest in class D DIY.
Also it tells me that there are around 20 knowledgeable class D freaks watching the thread, but not willing to share anything, because of the 280 CopyCats that lurk around the thread in silent spy mode .
In history there have always been these technically ignorant guys who only want to suck&sell the design, but it is sad they became powerful enough to kill the DIY-spirit in the class D section. :mad::mad:

Not sure whether I will completely stop the thread
- or continue, but restrict the design details to DIYers who I know more or less.
 
And one day i made this :cool:

One shouldn't be pessimistic. I know copy cats are everywhere and are hard bent to making big monies.:)

And its very sad and serious issue that people who don't know anything technically , ripe the fruits from the hardwork done by others.
 

Attachments

  • 1shx6u.jpg
    1shx6u.jpg
    139.7 KB · Views: 3,280
Last edited:
  • Like
Reactions: 1 user
Markus, the problem with your thread is it is too high technical level.
Adjusting hysteresis window to get tighter switching frequencies range is scary. I've tried modulating hysteresis window (for PLL-ing phases in multiphase) and I'm still scared too. This was strainghtforward dynamic adjustment of comparator supplies.
It is not unusual that jazz piano virtuoso gets less audience then a boysband if you know what I mean.
From my side, I think I can contribute to this open design in certain parts of modulator circuitry, switching part and layout if you'd wish.

Regards,
Adam
 
Last edited:
Hi Kanvar,
I am loving your rookie attempt!
The second one is looking great and might outperform your earlier work,
but it does not have this special DIY feeling like your first. :cool:

Discrete Comparator:
Hm, you are also a fan of this?
I detected the integrated comparators and their jitter mechanisms as one of
the key contributors to poor noise figures. And because of this I build the comparator in discrete for SystemD_MD.
The downside is that this is one of the most effective ways to boost the component count and by this reducing the chances for other DIYers to get their build running.
In the mean time I understood and learned how to tame the jitter a little bit.
I can implant a LM360 in the SystemD MD and still derive reasonably good noise figures, about 10db worse than with my discrete comparator.

Basically my intension of this amp is to design it in a DIY-friendly way.
That's why I intended to go for an integrated comparator.
Also for the output devices I am intending to live with the short comings of the TO247 and use IRFP4668 instead of paralleling multiple IRFB4227...
Do you think I should change this overall direction?
 
Just remember that UCD is fully discrete and easily made by many Diyers here.

Yes , i love discrete comparators, because even i have seen the SNR levels increase[less jitter, less noise] with them and using chip solutions annoys you in results.

Its always better to use a single device rather than paralleling, I know TO-220 lead inductance is lower than TO-247 but then more devices makes the layout more critical which has adverse impact on DIYing.

One can make both versions, for users wanting to step into more depth, discrete version could do, for easy doers the chip comp will be sufficient.
 
From my understanding there is a fundamental difference in the comparator requirements between UcD and hysteresis oscillators.
For UcD you need a chain of comparator and power stage which has to have a long delay in order avoid unsaint high switching frequencies. Also low comparator gain is helpful.
In the inteded circuit I dislike the hidden intrinsic parasitic phase shift self oscillator, because it impacts the intended mode of operation. So I prefer to have a chain with low delay and exact switching points, which is making the comparator design more extensive....

Nevertheless the idea to think about options to enable both is good.
In case the LM360 would become annoying, it should be possible to design a discrete comparator that fits to the DIL8 socket of the LM360 plus two wires for +/-12V.

Attached a simplified schematic of the amp from which the earlier simulations were derived.
I already have a sim with more detailed power stage including IRFP4668, but so far I have no reasonable option to post larger schematics in a readable way...

The overall structure is very similar to System_MD.
- table E3 can be any reasonably fast OP amp, which is stable at gains >3.
Design goal: Allow to use 1/2 NE5532.
- Q3, Q4 provide the virtual GND for the D portion of the feedback
- table E1 should be the LM360
- table E2 is the level shifter (IRS20957) and BJT gate buffer and IRFP4668
- In front of the shown amp the second half of the NE5532 will be put as an differential input amp to enable symmetric inputs and easy amp bridging.
- Upgrading the circuit with better OPamps possible.

The connection point for the hyteresis control circuit links to the previously shown circuit.

With these values the amp has an idle frequency of 315kHz in simulation.
I am also considering higher frequencies, but from theory the switching losses in case of a short circuit with small input signals, which can run the amp with unreduced frequency close to the overcurrent shut down, but just not triggering it - are already at the limits of the SOA of the IRFP4668.
2kW, 80V rails and an overcurrent protection around 60A demand their tribute...
 

Attachments

  • SystemD_2kW_SimplifiedSchematic_Shrinked.jpg
    SystemD_2kW_SimplifiedSchematic_Shrinked.jpg
    109.7 KB · Views: 2,781
  • Like
Reactions: 1 user
Choco,

I did a lame attempt on discrete comparator...

Not a lame attempt at all. I did not see this posting earlier, because I was writing my posting at the same time.
I cannot see all details of the circuit, but it looks like a typical straight forward design which allows reasonably results.
From my findings symmetric delays and rise times can be achieved best by an a fully symmetric approach... again more components.
Most simple a fully symmetric input stage of such an comparator can be achieved by using JFets instead of BJTs, because their control characteristic allow simple self biasing.
 
...
Adjusting hysteresis window to get tighter switching frequencies range is scary. ...
From my side, I think I can contribute to this open design in certain parts of modulator circuitry, switching part and layout if you'd wish.
..

Hi Adam,
of course I welcome your input !
It might bring me to the trouble to decide whether I should try your ideas or my ideas and I might not always find the time to elaborate in proper detail on both.
Why is a dynamic hysteresis scary?
...such discussion is what I enjoy....
In case we have two parallel ideas, which we both would love to see in reality - it might be even cooler to derive two design versions. You are living in EU, so exchange even of components and PCBs should be possible in an easy way...

Btw: What happened with your genious bridgeable fullbridge?


...back to the discrete Comparator:
Attached a sketch of a simple discrete approach, which should just satisfy the needs and still would be possible with reasonable component count.
It has similarities to the design in MD, but in MD I used a MosFet input which does not allow for easy self biasing....
So the attached version should be more DIYable (except obsolete components).
 

Attachments

  • SimpleDiscreteComparator.JPG
    SimpleDiscreteComparator.JPG
    90.3 KB · Views: 2,562
  • Like
Reactions: 1 user
Great going Choco

I think a Hysteresis modulator with a CCS control [i have seen this approach somewhere] is best option in order to mitigate any effects of voltage fluctuations. This will give more precision and control over how it actually performs.

Nevertheless the idea to think about options to enable both is good.
In case the LM360 would become annoying, it should be possible to design a discrete comparator that fits to the DIL8 socket of the LM360 plus two wires for +/-12V.

Attached a simplified schematic of the amp from which the earlier simulations were derived.
I already have a sim with more detailed power stage including IRFP4668, but so far I have no reasonable option to post larger schematics in a readable way...

Supply voltage limit for LM360 is recommended at +/-6V max not +/-12V, for that you need LM361. Also the IRFP4668 in bridge format is capable of more than 4KW in 2 ohms music[been there done that] :D

Not a lame attempt at all. I did not see this posting earlier, because I was writing my posting at the same time.
I cannot see all details of the circuit, but it looks like a typical straight forward design which allows reasonably results.
From my findings symmetric delays and rise times can be achieved best by an a fully symmetric approach... again more components.
Most simple a fully symmetric input stage of such an comparator can be achieved by using JFets instead of BJTs, because their control characteristic allow simple self biasing.

I agree with you that a symmetric section is always needed for symmetric slewrate. Will be doing some more sims in short while.

back to the discrete Comparator:
Attached a sketch of a simple discrete approach, which should just satisfy the needs and still would be possible with reasonable component count.
It has similarities to the design in MD, but in MD I used a MosFet input which does not allow for easy self biasing....
So the attached version should be more DIYable (except obsolete components).

Don't you think matching of JFETs is little critical, or one can get away with that in case of comparators.?
 
12V-Misunderstanding:
I am planning to run the LM360 with +/-5V, output stage with 5V only, of course.
But the discrete solution will need the +/-12V rails (or something like this),
which are not available on the DIL8 socket of the LM360, so there the two additional wires for the discrete plugin would be necessary.

SOA:
Of course the IRFP4668 can handle more than 4kW, especially when used at lower frequencies. And of course you will find plenty designs that do not consider worst case situations and also no theoretical SOA check at all.
When looking to selfoscillating class D amps, then the critical situation will never be normal operation, because you only have the high currents at high modulation levels where you have low switching frequencies. This is definitely a very fortunate nature of the selfoscillating designs.
But when looking to the described worst case scenario:
-Output running into a shorted wire ==> i.e. 50m Ohms
-Input signal small, but enough to go for approx. 3V output into 50m Ohms
Then you will still have the high switching frequency and 60Amax and almost no rail sagging ==> That's the worst case scenario.
Well, so far the theory. In practise I came to the conclusion that the IR devices are tough ones and making full use of the SOA, instead of keeping hypochondric margins, never caused any issues. In fact on my workbench they even took overload situations moderately beyond the theoretic SOA considerations with a smile....
Coming from this and fitting to your comment, I might give a chance to my 400kHz version, because it simply offers noticeable better performance.

JFET-matching:
You are right, massively unmatched devices might cause issues.
But I think it should be easy fit them in a sufficient way.
We do not need offset perfection, just a reasonable bias situation when the output is its linear region.