Hi, Mike,
I'm not experienced at all in classD, still learning. But from my previous experiment, IRF640 works better than IRF640N. Have you tried IRF540 and IRF9540 (without N)?
I also found something strange in my experiment. The inductor, if the position is changed or the distance to the pcb is changed, the switching frequency shown on osciloscope also changed. Do you have the same experience? Seems the radiation from the inductor can change the frequency, depending on placement or distance from the pcb.
I'm not experienced at all in classD, still learning. But from my previous experiment, IRF640 works better than IRF640N. Have you tried IRF540 and IRF9540 (without N)?
I also found something strange in my experiment. The inductor, if the position is changed or the distance to the pcb is changed, the switching frequency shown on osciloscope also changed. Do you have the same experience? Seems the radiation from the inductor can change the frequency, depending on placement or distance from the pcb.
classd4sure said:
Very diplomatic said !!!
I will try this evening new driverstage....
i will use CFP for closing, and single transistor for opening.
subwo1, i'd really like to increase currents in the levelshifter, but these
transistors are already very warm... But looking at the scopeshots, i see
that i have already above 12volts Vgs, so i will replace the 1k with 680ohms,
still giving above 8volts.
lumanauw, as i take feedback before coil, there should be no influence
from there to the oscillationfreq. If you take feedback after coil, any
change in inductance of course changes the freq.
The only thing changing freq in my amp was supplyvoltage, but only
moderate. When switching off this freq goes down to ~100khz, then
the amp breaks down.
The oscillation seemed rockstable...
Hey , that's really funny, "rockstable oscillation"... ClassAB is rockstable
if not oscillating at all !
Mike
Heh - what diod I say about those 1k in the predriver
Also, someone advised to use IRF540/9540 without the N... I think I mentioned that too ;-) as capacitances are different.
IRF640/9540 have better complementarity, including to a degree the capacitances, currents, Rdson.
I would say your surmise of Cgd 'millering' a ~70V voltage swing back to the gate will make a notch at Vtreshod.
Maybe you could increase the 150 ohm resistor in the VAS (slowdown should not be a problem since you are slowing it down with a capacitor anyway), and make a resistive divider out of the 1k resistors to compensate for the larger swing of the predriver. This would give you more current for switch-off, I think.
Also, someone advised to use IRF540/9540 without the N... I think I mentioned that too ;-) as capacitances are different.
IRF640/9540 have better complementarity, including to a degree the capacitances, currents, Rdson.
I would say your surmise of Cgd 'millering' a ~70V voltage swing back to the gate will make a notch at Vtreshod.
Maybe you could increase the 150 ohm resistor in the VAS (slowdown should not be a problem since you are slowing it down with a capacitor anyway), and make a resistive divider out of the 1k resistors to compensate for the larger swing of the predriver. This would give you more current for switch-off, I think.
I feel good in this thread, people making a pretty good cooperation Mike.
This is the "spirit"...very good!
Someone told about less capacitance components, maybe can be helpfull, the resistor change can be fast checked too.
I thank you, in advance, related some sample produced and let people know the download FTP adress if you can.
The method is not perfect, we know, but is the best we can do to have some "idea" of the sonics, and having another already known amplifier as reference, we can, sometimes perceive some details....not easy, with our computer amplifiers destructing all, but i am using good headphones and good amplifier...i can have some quality idea...having an edit "reference" sample together.
I will be waiting, as i know you are busy...no need to run, if you prefer to develop more before the comparison recording....no problems.
This is an enormous cooperation you are giving to us, as some discrete D amplifier is something very interesting to DIY community, i am sending you again, my deep and sincere congratulations!.
regards,
Carlos
This is the "spirit"...very good!
Someone told about less capacitance components, maybe can be helpfull, the resistor change can be fast checked too.
I thank you, in advance, related some sample produced and let people know the download FTP adress if you can.
The method is not perfect, we know, but is the best we can do to have some "idea" of the sonics, and having another already known amplifier as reference, we can, sometimes perceive some details....not easy, with our computer amplifiers destructing all, but i am using good headphones and good amplifier...i can have some quality idea...having an edit "reference" sample together.
I will be waiting, as i know you are busy...no need to run, if you prefer to develop more before the comparison recording....no problems.
This is an enormous cooperation you are giving to us, as some discrete D amplifier is something very interesting to DIY community, i am sending you again, my deep and sincere congratulations!.
regards,
Carlos
Hi !
Carlos, i thought it would be interesting to the DIY-community !
Ilimz, as you can see on the scopeshots, for the pchannel you see
one big notch/overshoot when voltage rises, going far above +vcc,
and the same nearly compensated by the drivers when voltage
goes down... At least this was the most logical thing to me...
I overlaid all 3 scopeshots in photoshop, and this scenario fitted best...
I also think that my mosfets are not best choice... I ordered irf530/9530
2 weeks ago, but reichelt simply does not deliver...
Mike
Carlos, i thought it would be interesting to the DIY-community !
Ilimz, as you can see on the scopeshots, for the pchannel you see
one big notch/overshoot when voltage rises, going far above +vcc,
and the same nearly compensated by the drivers when voltage
goes down... At least this was the most logical thing to me...
I overlaid all 3 scopeshots in photoshop, and this scenario fitted best...
I also think that my mosfets are not best choice... I ordered irf530/9530
2 weeks ago, but reichelt simply does not deliver...
Mike
Re: I have listen many amplifiers with perfect square waves, blameless was one of them
Excuse the stupid question, I'm not an expert in class D amps. But why not using bipolar transistors? I remember that they have less junction capacitance... Excuse me if it is obvius, I just don't know...
destroyer X said:
Excuse the stupid question, I'm not an expert in class D amps. But why not using bipolar transistors? I remember that they have less junction capacitance... Excuse me if it is obvius, I just don't know...
I think speed is the problem, this amplifier have to produce
Square waves, with variable width, in a reason of 308 thousand cicles each second, and power transistors will not work very well in that frequency, the wave will be reduced in size and will be distorted.
As i can imagine...if wrong, please correct me.
Those FETs are almost perfect switches to that speed, and new ones are reducing the inter electrode capacitance to avoid problems when switching under hi frequency conditions.
Those informs, if not misunderstood by me, came from MikeB, when i asked him the same question in direct mailing.
regards,
Carlos
Square waves, with variable width, in a reason of 308 thousand cicles each second, and power transistors will not work very well in that frequency, the wave will be reduced in size and will be distorted.
As i can imagine...if wrong, please correct me.
Those FETs are almost perfect switches to that speed, and new ones are reducing the inter electrode capacitance to avoid problems when switching under hi frequency conditions.
Those informs, if not misunderstood by me, came from MikeB, when i asked him the same question in direct mailing.
regards,
Carlos
So the problem is speed? But I recall btjs with speeds up to 100MHz and high power... maybe it's just a cost choiche... probably mosfets for switching are less expensive than similar performance btjs... and what about transmitter tubes for a class D amp? That's an idea 
They're powerful, fast...
They're powerful, fast...
Oh, bjts beeing able to supply more than 10amperes have "slightly"
higher capacitances... And reach at least ft of 50mhz. The transients
for these squarewaves are equivalent to frequencies above 25mhz,
giving a currentgain of 2 (hfe)... This makes these devices useless !
Also, a mosfet costs around 60cents, a powerbjt 5$...
And transistors don't like it too much to be reversebiased.
There are many reasons why mosfets are far superior for ClassD.
Even these "small" mosfets can work with peakcurrents up to 100A !
Modern mosfets have inputcapacity below 500pf and outputcapacity
of 40pf ! A mjl3281 has outputcapacity of 600pf !
Mike
higher capacitances... And reach at least ft of 50mhz. The transients
for these squarewaves are equivalent to frequencies above 25mhz,
giving a currentgain of 2 (hfe)... This makes these devices useless !
Also, a mosfet costs around 60cents, a powerbjt 5$...
And transistors don't like it too much to be reversebiased.
There are many reasons why mosfets are far superior for ClassD.
Even these "small" mosfets can work with peakcurrents up to 100A !
Modern mosfets have inputcapacity below 500pf and outputcapacity
of 40pf ! A mjl3281 has outputcapacity of 600pf !
Mike
MikeB said:
Hi MikeB, the change you mentioned should work from two angles. because it means there is less voltage the gate must traverse to reach the threshold region besides the steeper slope in getting there. I rather like your predriver, such an approach is generally avoided in linear amps due to instability, but that aspect is no problem here.
Your general approach has some advantages over the source follower circuit I have simulated, but which may not quite be what I seek. You mentioned folded cascode. That is the method that I was actually using since it is much faster than the straight resistor pull-down and slightly simpler than a current mirror.
I would suggest replacing the last transistor emitter follower pair with something like the IRF9952. It has the smallest surface mount package I tolerate, plus only about a 1 volt gate threshold specification. I would separate the gates with a resistor, driving the p-ch through the resistor for turn-on and the N-ch directly for turnoff of the IRF9540n. Both would be used in the source follower configuration. It would get your IRF9540n gate below the threshold voltage quickly, but based on your earlier comment, the problem may still persist. Best Regards.
After recalling that the IRF9952 is rated for 30v, I think an even better idea could be to use an IRF7207 for turning off the IRF9540n and your PNP transistor for turning it on. I cannot get the PDF reader to work, but that SO-8 packaged MOSFET is rated for 20v, which does not matter for turn-off since you don't want to enhance the 9540 gate that much anyway. The lower voltage device has less input capacitance. You could do same approach but opposite polarity for the IRF540n if you wish. Then, I'd suggest using the IRF1902 for turn-off and an NPN transistor for turn-on. The good thing about the single MOSFET in the SO-8 package is that it is easier to handle p2p-wise than a dual device.
OK, looks like I had my gate turn-off MOSFETs mixed up. The N-channel IRF1902 would turn off the P-Ch output device, the IRF9540n, while the P-ch IRF7207 would turn off the N-ch output, the IRF540N. Those gate turn-off MOSFETs would be used as source followers for ease. They are rated as having gate threshold voltage figures of .7v for the IRF1902 and -.7v for the IRF7207. The key is to try to use the lightest and easiest to drive SO-8 packaged MOSFET for discharging the enhanced gates of the output devices. But from the results MikeB was getting with the IRF9540n turning off so much slower than the IRF540n, only the IRF9540n may need the extra oomph of being turned off with a MOSFET.
MikeB, just wanted to say that it may appear I am nagging a lot, but I am trying to help because I find your design very nice. I love the idea of a good class D amp made out of readily available parts.
I do have some experience with designing switching power supplies, which in some cases by definition drive an inductive load. MOSFET turn-off is one of the most crytical parts of the design as you are turning off a MOSFET while it's conducting the highest current - so you want to do it fast.
Keep in mind that Cgd is not a constant. If you look at the datasheet, you may want to be mindful of the accumulated charge diagram, it has a flat point. That tends to produce exactly the notches you have on your P-MOSFET. Of course, this is not the only thing that will do it
Secondly, in your design you have essentially a current source driving the gate drivers in the switch-on phase, while there is a resistor driving them in the switch off phase. Simplified, whatever current you feed your driver, gets scaled up on it's output. This means that on turn-on, you have a linear slope, and on turn-off you have exponential decay. This in itself will make turn-off slower, and relatively speaking it will be the slower sompared to current source of same magnitude as initial decay slope, the higher the output voltage of the driver. In your design, turn on current is higher than turn-off (at Vgs=threshold, switching on has approx Vgst+2xVbe on the resistor, at switching on Vgst-2xVbe) even discounting the different slopes. Couple that with reverse transfer capacitance effects and you get a notch. The standard approach in eg. flyback SMPS if you want to be cheap is to switch on with a resistor, and switch off with a current source (transistor). IRF9540N also has 6 times the reverse transfer capacitance of the IRF540 (240Pf), which is certainly a clue why you get this effect on the P MOSFET. Regular 9540 has 140pF so about 3.5x as large.
In a class D design you want this to be as symetrical as possible, makes dead time setting easyer. I think you will have a very hard time finding a P-MOSFET with 40pF of reverse transfer capacitance for the currents you need. There are purely physical reasons for this which i will not go into here.
I also have some experience designing class A and AB amps with MOSFETs, and even there you want matched Ciss, Coss and Crss because these non-linear capacitances add distortion, so at least you want to cancel out the even harmonics. I have found that there are certain combinations of P and N MOSFETs that match FAR better in most things than the 'obvious' XXX/9XXX matches, for Vdsmax ~~2x higher on the N parts. Have a look at the datasheets for IRF9540/640 (no N at the end for either) to see what I mean.
Best of luck with the design, I am looking forward to your further posts with great anticipation, and thanks for your patience in reading my nags
I do have some experience with designing switching power supplies, which in some cases by definition drive an inductive load. MOSFET turn-off is one of the most crytical parts of the design as you are turning off a MOSFET while it's conducting the highest current - so you want to do it fast.
Keep in mind that Cgd is not a constant. If you look at the datasheet, you may want to be mindful of the accumulated charge diagram, it has a flat point. That tends to produce exactly the notches you have on your P-MOSFET. Of course, this is not the only thing that will do it
Secondly, in your design you have essentially a current source driving the gate drivers in the switch-on phase, while there is a resistor driving them in the switch off phase. Simplified, whatever current you feed your driver, gets scaled up on it's output. This means that on turn-on, you have a linear slope, and on turn-off you have exponential decay. This in itself will make turn-off slower, and relatively speaking it will be the slower sompared to current source of same magnitude as initial decay slope, the higher the output voltage of the driver. In your design, turn on current is higher than turn-off (at Vgs=threshold, switching on has approx Vgst+2xVbe on the resistor, at switching on Vgst-2xVbe) even discounting the different slopes. Couple that with reverse transfer capacitance effects and you get a notch. The standard approach in eg. flyback SMPS if you want to be cheap is to switch on with a resistor, and switch off with a current source (transistor). IRF9540N also has 6 times the reverse transfer capacitance of the IRF540 (240Pf), which is certainly a clue why you get this effect on the P MOSFET. Regular 9540 has 140pF so about 3.5x as large.
In a class D design you want this to be as symetrical as possible, makes dead time setting easyer. I think you will have a very hard time finding a P-MOSFET with 40pF of reverse transfer capacitance for the currents you need. There are purely physical reasons for this which i will not go into here.
I also have some experience designing class A and AB amps with MOSFETs, and even there you want matched Ciss, Coss and Crss because these non-linear capacitances add distortion, so at least you want to cancel out the even harmonics. I have found that there are certain combinations of P and N MOSFETs that match FAR better in most things than the 'obvious' XXX/9XXX matches, for Vdsmax ~~2x higher on the N parts. Have a look at the datasheets for IRF9540/640 (no N at the end for either) to see what I mean.
Best of luck with the design, I am looking forward to your further posts with great anticipation, and thanks for your patience in reading my nags
Hi !
Subwo1, these devices would surely solve the problems !
But, i wanted to stay with "normal" devices, as i want to make some
easy to build and cheap ClassD.
Ilimzn, i did not perceive your comments like "nagging"... Damn it, needed
babelfish to check the meaning...
The 9540n might be a very bad choice, but had to use what i have...
Obviously you were all right, turnofftime is all !
Yesterday i replaced the driverstage, using a single transistor for
opening mosfets, and a cfp for closing. I attached the bc-emitter
diretcly to gate, a 150ohm bewteen collector and supplyvoltage.
To this 150ohm i connected a complementary bjt, emitter/base to this
150ohm, collector to gate.
The notch in the p-channel-vgs is nearly gone and heating up of the
p-channel is gone. Also the outputwaveform looks a bit better.
But, the n-channel-vgs-waveform is now some "M", i absolutely have
no idea how this is possible... The rising and falling looks normal,
just between the edges the voltage drops ~5volts. The voltage
at base to opening-transistor is absolutely ok. How can get Vbe
above several volts without the bjt conducting ? In this time it
does not supply any current... I already replaced this transistor...
I will try to make scopeshots this evening.
Mike
Subwo1, these devices would surely solve the problems !
But, i wanted to stay with "normal" devices, as i want to make some
easy to build and cheap ClassD.
Ilimzn, i did not perceive your comments like "nagging"... Damn it, needed
babelfish to check the meaning...
The 9540n might be a very bad choice, but had to use what i have...
Obviously you were all right, turnofftime is all !
Yesterday i replaced the driverstage, using a single transistor for
opening mosfets, and a cfp for closing. I attached the bc-emitter
diretcly to gate, a 150ohm bewteen collector and supplyvoltage.
To this 150ohm i connected a complementary bjt, emitter/base to this
150ohm, collector to gate.
The notch in the p-channel-vgs is nearly gone and heating up of the
p-channel is gone. Also the outputwaveform looks a bit better.
But, the n-channel-vgs-waveform is now some "M", i absolutely have
no idea how this is possible... The rising and falling looks normal,
just between the edges the voltage drops ~5volts. The voltage
at base to opening-transistor is absolutely ok. How can get Vbe
above several volts without the bjt conducting ? In this time it
does not supply any current... I already replaced this transistor...
I will try to make scopeshots this evening.
Mike
Very interesting! Thanks for providing us with your experiences.
Secondly, I would suggest to change the output coil too. Bending it to a toroid would not help very much IMHO, because of the missing "flux collecting" core. Therefore the stray field would remain very strong and tend to go outwards through the windings, not comparable to a ferrite core toroid.
To prevent the skin effect I used enamelled (?) wire (dt: HF-Litze). But this is useful, when winded on a core only.
Best regards and good luck, Timo
May be this happens due to the "same" Cgd of the N-channel FET as mentioned before? While the BJT does not deliver current to the gate, the gate charge coming from the rise up due to the opening of the PFET (and therefore the rise of the pre filter output voltage) will change more slowly. But this would describe the first "peak" only... On the other hand it would make me more happy, if I'm sure, the gate is hard driven all(!) the time.
Secondly, I would suggest to change the output coil too. Bending it to a toroid would not help very much IMHO, because of the missing "flux collecting" core. Therefore the stray field would remain very strong and tend to go outwards through the windings, not comparable to a ferrite core toroid.
To prevent the skin effect I used enamelled (?) wire (dt: HF-Litze). But this is useful, when winded on a core only.
Best regards and good luck, Timo
I perceive you as a very cooperative and non selfish man
As you put all details, you want deep an truly that people understand what you mean, you make all the needed efforts for that.
I feel very good watching those things happening, and happened in the past a couple of times.
But, the most common, is not to show the secrets!...well, i do not like that, but i can understand when the guy have own business, and those secrets make part of the whole thing.
I could read people going in details till certain point.... and them...stopped...keeping precious informs that are impeachment to absorb their advanced knowledge.
This way, i think class D will make justice with those guys, as all the things they keept hidden are not more needed to be known.
regards,
Carlos
As you put all details, you want deep an truly that people understand what you mean, you make all the needed efforts for that.
I feel very good watching those things happening, and happened in the past a couple of times.
But, the most common, is not to show the secrets!...well, i do not like that, but i can understand when the guy have own business, and those secrets make part of the whole thing.
I could read people going in details till certain point.... and them...stopped...keeping precious informs that are impeachment to absorb their advanced knowledge.
This way, i think class D will make justice with those guys, as all the things they keept hidden are not more needed to be known.
regards,
Carlos
Hi !
Timo, the problem is, that this effect happens in the middle of the cycle,
there are no other transients that could have some effect like this...
I consider that the 2n5551 was simply overloaded ? some 2nd breakdown ?
Carlos, yes, i find it annoying how often i read these 2 words:
"Confidential", "Patented"...
Yesterday i replaced the mosfets with irf530/sfp9530...
(irf530 from ST, sfp9530 from Fairchild)
All mentioned notches are gone, even the "M" for the nch-vgs...
I removed the caps in vas and levelshifter, they are no longer needed.
The mosfets get handwarm at most ! Seems that the 9540n was a very
bad choice !
Rise/fall-time at output (before coil) measured as ~20ns.
BUT, the sound seemed to have lost some quality...
And, another device i recognized heating up... The fuses ! ???
Or the fuseholder ? They burned my fingometer, how is this possible ?
Is the coil inducting some weird currents inside the metal ?
Have to replace the coil soon !
Mike, still blowing at his fingometer...
Timo, the problem is, that this effect happens in the middle of the cycle,
there are no other transients that could have some effect like this...
I consider that the 2n5551 was simply overloaded ? some 2nd breakdown ?
Carlos, yes, i find it annoying how often i read these 2 words:
"Confidential", "Patented"...
Yesterday i replaced the mosfets with irf530/sfp9530...
(irf530 from ST, sfp9530 from Fairchild)
All mentioned notches are gone, even the "M" for the nch-vgs...
I removed the caps in vas and levelshifter, they are no longer needed.
The mosfets get handwarm at most ! Seems that the 9540n was a very
bad choice !
Rise/fall-time at output (before coil) measured as ~20ns.
BUT, the sound seemed to have lost some quality...
And, another device i recognized heating up... The fuses ! ???
Or the fuseholder ? They burned my fingometer, how is this possible ?
Is the coil inducting some weird currents inside the metal ?
Have to replace the coil soon !
Mike, still blowing at his fingometer...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.