Why so much interest in something that is inherently flawed and cannot produce the claimed power (or even work)?
I admit that the name of the thread is very attractive, but...
Hello Eva, you said that this circuit does not functional, post some good circuit will work properly ... 4kw RMSem 2 ohms will be very good.
That indicates something to us ... bridge can
Thank you a big hug
I do not bother about good or less good english as long as both sides can understand each other.
But the headache of this thread is not the language.
I do understand that many people here are not much interested in DIY (Do It Yourself), but just in getting a 6kW amp for free, or at least mostly free. Consequently calling 'DI4ME' (Do It for Me).
IMHO this is not a promising spirit for a DIY forum.
If one of the guys or ladies, who have the knowledge, would do it -
this would mean the following.
1. She/he worked on learning this stuff for multipe years. Biggest portion of that time unpaid. Often during night, because during day time most DIYers have to go to work in order to make their living.
2. Designing an amp + PCB + BOM + Getting Started Documentation (just another year, or two...).
3. Publishing in the forum.
4. Giving support on understanding.
5. Giving support on where to get the components and how to check.
6. Giving support on assembly.
7. Spending nights on remote debugging other peoples errors, while being insulted by the same people for publishing something that does not assemble by itself.
- So far just the normal things, but in the mean time there has established a nice topping:
8. Seeing suddenly modules of this design on multiple Web-Sites for sale.
9. Getting forwarded the customers complains....
There is no reason to complain that currently no naive nerd is willing to be the victim.
But the headache of this thread is not the language.
I do understand that many people here are not much interested in DIY (Do It Yourself), but just in getting a 6kW amp for free, or at least mostly free. Consequently calling 'DI4ME' (Do It for Me).
IMHO this is not a promising spirit for a DIY forum.
If one of the guys or ladies, who have the knowledge, would do it -
this would mean the following.
1. She/he worked on learning this stuff for multipe years. Biggest portion of that time unpaid. Often during night, because during day time most DIYers have to go to work in order to make their living.
2. Designing an amp + PCB + BOM + Getting Started Documentation (just another year, or two...).
3. Publishing in the forum.
4. Giving support on understanding.
5. Giving support on where to get the components and how to check.
6. Giving support on assembly.
7. Spending nights on remote debugging other peoples errors, while being insulted by the same people for publishing something that does not assemble by itself.
- So far just the normal things, but in the mean time there has established a nice topping:
8. Seeing suddenly modules of this design on multiple Web-Sites for sale.
9. Getting forwarded the customers complains....
There is no reason to complain that currently no naive nerd is willing to be the victim.
yes,Chocoholic is completely right .
But working with this ic is not easy.
Maybe for people like Eva,Fredos,Phase accurate,and others great in class d ,it could be easy.But i had a lot of troubles with this ic,and if somebody wants to try,pay spacial attention to lay out,also feedback routing,interference in sensitive nodes,feedback noise between audio pre amplifier and switching power stage ,among others.
But working with this ic is not easy.
Maybe for people like Eva,Fredos,Phase accurate,and others great in class d ,it could be easy.But i had a lot of troubles with this ic,and if somebody wants to try,pay spacial attention to lay out,also feedback routing,interference in sensitive nodes,feedback noise between audio pre amplifier and switching power stage ,among others.
Key factor for success and reliable operation is to avoid unpleasant resonances, typically happening in the frequency range between 20MHz and 100MHz (depending on layout and chosen components).
Means optimized layout and optimized snubbers are mandatory.
A fast scope and some patience during optimizing the snubbers.....
Otherwise there is a high chance of phantom tripping of the protection circuits or even killing the driver chips.
Most is said by Eva in posting #52. Furtheron she gave many more detailed hints distributed across the entire class d section.
Means optimized layout and optimized snubbers are mandatory.
A fast scope and some patience during optimizing the snubbers.....
Otherwise there is a high chance of phantom tripping of the protection circuits or even killing the driver chips.
Most is said by Eva in posting #52. Furtheron she gave many more detailed hints distributed across the entire class d section.
about snubber,i had to put 1 rc from vs to +b and other from vs to -b.Also used a rc from +b to -b to help killing stray resonances,since my pcb is not ideal.Now my prototype is working well,using 4 irfb4227 +pnp npn buffer and 2092.+-60volts 2 ohms.But now i will make another pcb to move a track of vs away from ocp circuitry.It was disturbing the protection even tunning the vs snubbers.
IRFB4127 is a newer version. One feature of both IRFB4227 and IRFB4127 is that they are fully on at Vgs=8V, so a high drive voltage is not required, while allowing the gate drive to "clip" at a low voltage gives great control over di/dt during body diode reverse recovery. Transconductance is high, Vgs rise from Id=10A to Id=40A is just 0.5V, resulting in slightly falling di/dt during the recovery process. Higher di/dt at the beginning prevents undesirable charge storage in the diode due to Rds-on current being dumped after gate turn off.
After some trouble with parasitic BJT latchup after recovery (with uncontrolled di/dt well over 1000A/us), I changed gate drive scheme completely and I'm operating them around 650A/us each, nicely symmetrical for low and high side, di/dt was not well controlled previously.
I also solved the lack of di/dt control and timing precision due to the fact that bootstrapped high-side gate drive supply is not regulated and fluctuates with output current, by using zeners at the outputs of IR driver ICs. This is possible because the output stage on these IC use a N-channel MOSFET for the high side whose gate seems to be controlled in a funny way... Good layout with equal parasitic source lead inductances does the rest. Don't ask me how but I get less than 4nH with 5mm lead length (maybe because the front side of the TO-220 cases is facing a ground plane?)
After some trouble with parasitic BJT latchup after recovery (with uncontrolled di/dt well over 1000A/us), I changed gate drive scheme completely and I'm operating them around 650A/us each, nicely symmetrical for low and high side, di/dt was not well controlled previously.
I also solved the lack of di/dt control and timing precision due to the fact that bootstrapped high-side gate drive supply is not regulated and fluctuates with output current, by using zeners at the outputs of IR driver ICs. This is possible because the output stage on these IC use a N-channel MOSFET for the high side whose gate seems to be controlled in a funny way... Good layout with equal parasitic source lead inductances does the rest. Don't ask me how but I get less than 4nH with 5mm lead length (maybe because the front side of the TO-220 cases is facing a ground plane?)
...yep, the driver rails do impact the timing precision.
I am going to use adjustable regulated rails. This allows dead time adjustment by the driver rails and ensures stable operation with the chosen timing.
IRFB4127: Yes they seem to be improved and slightly stronger.
But unfortunately IR again does not specify limits for max dv/dt during body diode recovery.
They just anounce 'enhanced body diode dv/dt and di/dt capability'.
Let's hope that these advertising phrases are based on real component properties.
And let's also hope that this trend from data to phrases, does not evolve any further.
I am going to use adjustable regulated rails. This allows dead time adjustment by the driver rails and ensures stable operation with the chosen timing.
IRFB4127: Yes they seem to be improved and slightly stronger.
But unfortunately IR again does not specify limits for max dv/dt during body diode recovery.
They just anounce 'enhanced body diode dv/dt and di/dt capability'.
Let's hope that these advertising phrases are based on real component properties.
And let's also hope that this trend from data to phrases, does not evolve any further.
See dv/dt rating on page 1 and associated note 3 on bottom of page 2 on IRFB4127 datasheet
You don't need to control each gate drive rail separately, the zener trick and pulse shaping is enough. I use brute force turn off (3A or so per gate), almost brute force turn on (with buffers and clipping/zener trick to control di/dt) and pulse shaping to get rid of dead time (RCD for extension rather than shortening ) on IR driver inputs. I find modern IR driver ICs with built in dead time useless, you always end up with too much dead time. Only drivers with separate H and L inputs are suitable.
Since body diode recovery results in non-linear delay (two slopes) on switching, because above the current level where body diode conduction starts, 30-50% more current has to be developed during the di/dt phase, I use slight dead time for linearization. Try to visualize it... THD now has little dependency on current. 0.05% at 1Khz and 0.15% at 4Khz just below clipping (5.6R load, 170V DC supply), <0.025% at -3dB (in a 125Khz amp that is olny -1dB at 17khz, load independent).
You don't need to control each gate drive rail separately, the zener trick and pulse shaping is enough. I use brute force turn off (3A or so per gate), almost brute force turn on (with buffers and clipping/zener trick to control di/dt) and pulse shaping to get rid of dead time (RCD for extension rather than shortening
) on IR driver inputs. I find modern IR driver ICs with built in dead time useless, you always end up with too much dead time. Only drivers with separate H and L inputs are suitable.Since body diode recovery results in non-linear delay (two slopes) on switching, because above the current level where body diode conduction starts, 30-50% more current has to be developed during the di/dt phase, I use slight dead time for linearization. Try to visualize it... THD now has little dependency on current. 0.05% at 1Khz and 0.15% at 4Khz just below clipping (5.6R load, 170V DC supply), <0.025% at -3dB (in a 125Khz amp that is olny -1dB at 17khz, load independent).
Last edited:
Seems like I am still angry about the data sheet of the IRFB4321, where this information was missing - made me blind...
Yupp the data sheet is looking not so bad.
But... 57V/ns? I would never expect to reach such values at the end of during diode recovery, when operating with such moderate values like 760A/us. Do you???
From my experience, at 760A/us, I would expect resulting values between 5V/ns..10V/ns.
I am not using the IR chips either.
To long dead times, to much dead time skew, to much timing tolerances.
I am using a high speed comparator with inverting and non inverting outputs + high speed isolators + high speed gate drivers. Dead time controlled by adjustment of driver rails, gate drive impedance and my di/dt-limiter (which I am going to integrate in the PCB-tracks).
Why all theses high speed components?
- Minimize timing skew and jitter
- More freedom for 2nd order postfilter feedback loop and optimization regarding stability and avoiding carrier aliasing at high levels...
I am less heading for a mass production product - but more for a special high end high power class D amp. Planning to improve the nice low distortions of my old proto even further, also higher power, and further improved step response. Simply making everything better, even the speaker relay
.Don't ask if this is necessary - it's just my couriosity to see how far I can push things with my homebrew designs.
Let's see how the amp will come within the next 12 months....
Chill out as I said. Already enough time pressure in my regular work.
...must go to bed now...
It's not adaptive, what I mean is that waveforms suggest that a small dead time like 10-20ns actually makes the amplifier more linear
Turn off imposes a delay/current ns/A slope and turn on imposes two ns/A slopes, first one for rds-on current and then other for body diode current which is smaller due to Irr. The former may be used to linearize the latter. Pulse skew is very small thanks to the zeners, even with IR gate drivers.
di/dt in my circuit is around 20V/ns for turn off and around 15V/ns after reverse recovery.
Of course, for high end I would probably be using pulse transformers to produce isolated high side supplies, and one regulator for each gate driver, but in this amplifier there is no room for that. I would probably be using separate optocouplers or transistors for level shifting too, for minimum skew and delay, but optos can't whitstand as high dv/dt as IR drivers (10-15V/ns vs 50V/ns).
You don't really need PCB tracks for di/dt control, minimum inductance is better, so using lead inductance, which is already there, is the best option, I think. If you start adding inductance, the resonant frequency of the power stage is reduced making snubber losses (and delays due to di/dt current transfer periods) higher. The effective turn-on delay that I get is around 1ns per ampere. The resonance that I get is around 100Mhz and is completely tamed with a single RLC snubber on each switching node, which uses the series inductance of a straight axial 0.25W resistor (vertically mounted) plus ~1nH from the 0805 capacitor. This allows to use only 330pF, but it's only effective around the operating supply voltage range (120-200V). For example, at 50V the amplifier also works, but ringing frequency drops to below 80Mhz and damping is not so good, but these are not operating conditions. This is hidden RF engineering that anyone willing to copy the circuit is very likely to overlook (if the layout or the components are changed, or even if TO-220 lead lenghts or mounting style are changed, tuning is lost)
That's why I find so difficult to specify class D amplifiers for DIY, particularly when people wants to use exotic components with high parasitics... There are no big film capacitors in my circuit either, only arrays of small ones whose resulting inductance is so low that I can't measure resonance. Very little EMI filtering is required btw 2nd-order output filter and supply capacitor bank are already very good EMI filters. Mains input and speaker output are together, and a single CM choke and Y capacitor forces them to be at the same RF potential (filtering voltage drops across the ground plane).
Turn off imposes a delay/current ns/A slope and turn on imposes two ns/A slopes, first one for rds-on current and then other for body diode current which is smaller due to Irr. The former may be used to linearize the latter. Pulse skew is very small thanks to the zeners, even with IR gate drivers.
di/dt in my circuit is around 20V/ns for turn off and around 15V/ns after reverse recovery.
Of course, for high end I would probably be using pulse transformers to produce isolated high side supplies, and one regulator for each gate driver, but in this amplifier there is no room for that. I would probably be using separate optocouplers or transistors for level shifting too, for minimum skew and delay, but optos can't whitstand as high dv/dt as IR drivers (10-15V/ns vs 50V/ns).
You don't really need PCB tracks for di/dt control, minimum inductance is better, so using lead inductance, which is already there, is the best option, I think. If you start adding inductance, the resonant frequency of the power stage is reduced making snubber losses (and delays due to di/dt current transfer periods) higher. The effective turn-on delay that I get is around 1ns per ampere. The resonance that I get is around 100Mhz and is completely tamed with a single RLC snubber on each switching node, which uses the series inductance of a straight axial 0.25W resistor (vertically mounted) plus ~1nH from the 0805 capacitor. This allows to use only 330pF, but it's only effective around the operating supply voltage range (120-200V). For example, at 50V the amplifier also works, but ringing frequency drops to below 80Mhz and damping is not so good, but these are not operating conditions. This is hidden RF engineering that anyone willing to copy the circuit is very likely to overlook (if the layout or the components are changed, or even if TO-220 lead lenghts or mounting style are changed, tuning is lost)
That's why I find so difficult to specify class D amplifiers for DIY, particularly when people wants to use exotic components with high parasitics... There are no big film capacitors in my circuit either, only arrays of small ones whose resulting inductance is so low that I can't measure resonance. Very little EMI filtering is required btw
2nd-order output filter and supply capacitor bank are already very good EMI filters. Mains input and speaker output are together, and a single CM choke and Y capacitor forces them to be at the same RF potential (filtering voltage drops across the ground plane).
Last edited:
That's exactly consideration #2 of my floating driver supply. But already 20V are more than enough!
Eva:
With the natural lead inductance and a threshold voltage around 5V and a driver voltage around 10V, I am getting a natural di/dt limitation around 1000A/us. In order to slow down I could reduce the driver voltage (but I would like to keep this skrew for dead time adjustment), or by gate drive impedance, or by adding a 5mm-10mm copper track inductor on the PCB.
In the end I will probably end up in a combination of all.
Looking for the best trade off....
My optos are guaranted for 15kV/us, which I am considering acceptable.
Two separate HCPL 9030 or same in SMD.
Additional an anti shoot through locking circuit with another fast opto, but different brand/technology - reducing the risk that signal drive and shoot through protection happen to be disturbed at the same time.
Hm, in your circuit you find 15...20kV/us, that's more than what I observed in my old proto. I will have to keep an eye on that.
For gate drive I prefer the optos vs. pulse transformers, because pulse transformer gate drive circuits are headache at extreme duty cycles.
Pulse transformer for isolated supplies:
This or a syncronized mini smps (just the luxury version of a simple pulse transformer supply) is option #1 of my floating driver supply.
Snubbering:
Yes the RCL-snubbers are cool !
Especially when going to higher frequencies, you can reach the same or even better damping like with RC - but with less capacitance. Means less snubber losses. I am chasing my old proto with IRFB4321 at 1MHz snubbered with 2x680pF in series + few Ohms and some few nH. It's always fun again to see the ringing disapearing by increasing the snubber L
!!!It seems to be best for the snubbers, when connected as close as possible to the switching node itself (means close to the die).
I tend to couple the snubber between the drain copper plate and shortest allowed connection point at the source lead.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.