Alright, some more cool scope photos. This time around I followed a different line of reasoning, as when you're trying to limit overshoot at the output, who says that overshoot wasn't present on the input to begin with? So in this case you'd better try to fight the cause rather than the effect. Please refer to the below URL of the schematic these comments are in reference to:
SODFA class-D amplifier, revision 007
http://hardwareanalysis.com/images/...large/11666.gif
Overshoot comparator output
The above image shows what signal the IR2011 MOSFET driver is fed with on pins 5 and 6. It is clear that there's a fair amount of, undesired, overshoot here which we need to get rid off, or at least minimize.
Overshoot integrator output
The above image clearly shows that the integrator itself, 1/2 of the OPA2134, outputs quite some overshoot at pin 7. So what we're looking at here is overshoot that's introduced at an early stage in the amplifier and simply propagates throughout it. This could be crosstalk from the output section of course. For now I'm a bit puzzled on how to reduce or remove the overshoot at the comparator output or whether it is crosstalk or something generated in the amp itself.
Any helpful suggestions are most welcome!
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
SODFA class-D amplifier, revision 007
http://hardwareanalysis.com/images/...large/11666.gif
Overshoot comparator output
The above image shows what signal the IR2011 MOSFET driver is fed with on pins 5 and 6. It is clear that there's a fair amount of, undesired, overshoot here which we need to get rid off, or at least minimize.
Overshoot integrator output
The above image clearly shows that the integrator itself, 1/2 of the OPA2134, outputs quite some overshoot at pin 7. So what we're looking at here is overshoot that's introduced at an early stage in the amplifier and simply propagates throughout it. This could be crosstalk from the output section of course. For now I'm a bit puzzled on how to reduce or remove the overshoot at the comparator output or whether it is crosstalk or something generated in the amp itself.
Any helpful suggestions are most welcome!
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
There you go, now you're getting into it
I'd say that is starting at the mosfets, and propagating through.
Playing with dead time will help (slow down the turn on rise time in a form of soft dead time). Again different mosfets will help... what gate driver voltage? Overdriving them isn't good either, you can reduce that ring like crazy by using just enough drive to get it fully on and nothing more, like say ten volts is a nice round number to have at the gate, and with very efficient switching mosfets, more than enough.
Quick and dirty is a snubber, which also slows rise time.
However the original problem had to do with the filtered output, in which case you only see a spike at the zero crossing of the filtered output.... when your inductor current reverses I believe, a seperate issue.
Regards,
Chris
I think I forgot to add different mosfets could will also help..... it's all about controlling circulating currents and beating down or "tricking" the parasitics into thinking they don't care Welcome to hell.
I'd say that is starting at the mosfets, and propagating through.
Playing with dead time will help (slow down the turn on rise time in a form of soft dead time). Again different mosfets will help... what gate driver voltage? Overdriving them isn't good either, you can reduce that ring like crazy by using just enough drive to get it fully on and nothing more, like say ten volts is a nice round number to have at the gate, and with very efficient switching mosfets, more than enough.
Quick and dirty is a snubber, which also slows rise time.
However the original problem had to do with the filtered output, in which case you only see a spike at the zero crossing of the filtered output.... when your inductor current reverses I believe, a seperate issue.
Regards,
Chris
I think I forgot to add different mosfets could will also help..... it's all about controlling circulating currents and beating down or "tricking" the parasitics into thinking they don't care
Welcome to hell.
Thanks Chris,
But there's isn't much I can do about the drive voltage, the IR2011 mosfet driver handles all that and I guess does a good job at it. Dead time I already played with, changed the R's in series with the gate from 4.7 to 25-ohms in 5-ohm steps, no change whatsoever. I can see that this might be generated at the output and gets propagated through the amplifier though, so a good clean layout should reduce it as well, unfortunately a perf. board isn't the ideal candidate for this
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
But there's isn't much I can do about the drive voltage, the IR2011 mosfet driver handles all that and I guess does a good job at it. Dead time I already played with, changed the R's in series with the gate from 4.7 to 25-ohms in 5-ohm steps, no change whatsoever. I can see that this might be generated at the output and gets propagated through the amplifier though, so a good clean layout should reduce it as well, unfortunately a perf. board isn't the ideal candidate for this
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
SSassen said:Thanks Chris,
But there's isn't much I can do about the drive voltage, the IR2011 mosfet driver handles all that and I guess does a good job at it. Dead time I already played with, changed the R's in series with the gate from 4.7 to 25-ohms in 5-ohm steps, no change whatsoever. I can see that this might be generated at the output and gets propagated through the amplifier though, so a good clean layout should reduce it as well, unfortunately a perf. board isn't the ideal candidate for this
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Don't stop there. Try 50, try 100, up it until the mosfet gets hot, play with it a little.
I'm using 100 ohm gate resistors in my homebrew amp, it is what got the fets cool without a scope at hand.
About the good pcb layout, this is why I don't fully believe when people say their p2p class d is dead silent
They sure can impress though, and is a good test bed before doing a PCB. Better to learn the interactions that way and it may save a few costly revisions later on.Time consuming though
You'll get it, you got this far.
Alright, but I'll wait till monday, these are 55-volt MOSFETs and I'm running at +/- 27-volts already, hence they need just a little nudge to go kaputt, remember I binned three of them already with the rev. 008 schematic. The IRFB23N15D has a lot more reserve, and I can play with those instead, so I'll increase the gate resistors to higher values then. Remember though that they are running cold as is with the current setup.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Remember also that a good part of the ringing you are seeing can be picked up by the oscilloscope probes, so they may not reflect reality with fidelity. I am sure you know that it is good to have a short tip and ground in the probe to minimize noise pickup.
Btw: my comparator outputs and inputs to the IR2113 are quite clean but I have the ringing at the output as well.
Best regards,
Pierre
Btw: my comparator outputs and inputs to the IR2113 are quite clean but I have the ringing at the output as well.
Best regards,
Pierre
Alright, I've taken some of Chris's comments to heart and have added ultra-fast schottky diodes across the MOSFET's which really did clean up their switching behaviour. I've also cleaned up the rest of the design somewhat, and added 1MH inductors, and removed the resistors, in series with the voltage regulators that power the opamps and the comparator, in an attempt to increase the PSRR (power supply rejection ratio). I also put a Zobel at the output to filter out HF and provide a load to the amplifier when no load is connected. I noticed that it ran away without a load attached and clipped hard against V+, destroying the upper MOSFET. Below you'll find the latest version of the schematic which reflects these changes.
SODFA class-D amplifier, revision 009
http://hardwareanalysis.com/images/articles/large/11680.gif
Have a good weekend,
Sander Sassen
http://www.hardwareanalysis.com
Edit: cut 'n pasted wrong url, corrected
SODFA class-D amplifier, revision 009
http://hardwareanalysis.com/images/articles/large/11680.gif
Have a good weekend,
Sander Sassen
http://www.hardwareanalysis.com
Edit: cut 'n pasted wrong url, corrected
I have also re-measured my idle output with other boards and I am quite satisfied with the result: around 200mVpp of ripple, almost sinusoidal, with very small ringing (less than 50-100mVpp), at 260KHz.
After reading your post, Sander, I have also added the MURS120 diodes across the mosfets with no change at all. It seems that your mosfets are not very good for this, I am sure you will notice a big improvement with ther IRFB23N15's (the same ones IRAUDAMP1 uses) and perhaps you won't need the diodes anymore.
BTW, it seems that everyone is running their amps at around 400KHz. I chosen 250-280KHz to minimize switching losses, but perhaps at some cost. I have tried to increase up to 500KHz with no problem (at low power, let's see what happens when pushing it hard). It seemed to me that the background noise (hiss) was noticeable lower (in fact almost unnoticeable ;-). Does this have any logic? (the feedback network remaining the same)
What are the factors that influence the background noise?
(Sorry for this little off-topic question)
After reading your post, Sander, I have also added the MURS120 diodes across the mosfets with no change at all. It seems that your mosfets are not very good for this, I am sure you will notice a big improvement with ther IRFB23N15's (the same ones IRAUDAMP1 uses) and perhaps you won't need the diodes anymore.
BTW, it seems that everyone is running their amps at around 400KHz. I chosen 250-280KHz to minimize switching losses, but perhaps at some cost. I have tried to increase up to 500KHz with no problem (at low power, let's see what happens when pushing it hard). It seemed to me that the background noise (hiss) was noticeable lower (in fact almost unnoticeable ;-). Does this have any logic? (the feedback network remaining the same)
What are the factors that influence the background noise?
(Sorry for this little off-topic question)
I suspect the possibility that there is a type of feedback occurring. It could involve switching transients feeding back with delay and causing multiple triggering of the input logic of the IR2011 in rapid succession.
You may also want to set Vcc on the IR2011 to 15v so that during peak positive output of the amp when the low time may be be very short, the upper capacitor, C20, will maintain a charge of at least 11.3v when it draws charge off of the lower one, C19. When it does so, it temporarily could take C19 from 15 down to 12 volts. But, that worse case scenario of the bootstrap capacitor dropping the lower one to 12v could happen only if the bootstrap one is zero when it draws charge from the lower one.
However, if during clipping, the lower MOSFET does not turn on for at least a very short time, the top capacitor could run out of steam during a long bass note, as Chris mentioned earlier, especially since the gain of the speaker drive rises with lowering frequency. With C20 being .22uF, I came up with the approximation that the bootstrap supply can hold out for a 75hz note in the worse case and about a 35hz note typically. That conclusion is based on the data sheet figure that the bootstrap supply quiescent current is typically 90ua and 210ua in the worse case.
You may also want to set Vcc on the IR2011 to 15v so that during peak positive output of the amp when the low time may be be very short, the upper capacitor, C20, will maintain a charge of at least 11.3v when it draws charge off of the lower one, C19. When it does so, it temporarily could take C19 from 15 down to 12 volts. But, that worse case scenario of the bootstrap capacitor dropping the lower one to 12v could happen only if the bootstrap one is zero when it draws charge from the lower one.
However, if during clipping, the lower MOSFET does not turn on for at least a very short time, the top capacitor could run out of steam during a long bass note, as Chris mentioned earlier, especially since the gain of the speaker drive rises with lowering frequency. With C20 being .22uF, I came up with the approximation that the bootstrap supply can hold out for a 75hz note in the worse case and about a 35hz note typically. That conclusion is based on the data sheet figure that the bootstrap supply quiescent current is typically 90ua and 210ua in the worse case.
Pierre said:
There's a number of factors which determine the effectiveness of that paralleled diode. Truth is it isn't a good measure on it's own and the extra blocking diode should be used in order to render it 100% effective.
Doubling Fs or so I would think should lower noise until at least EMI swamps everything.
You've pushed up switching noise much further where it can more effectively be filtered out.
Sanders may we see more screens of the progress please? I have no scope as I keep saying so it helps me test my own theory as well. That was but one idea of many to try, like I wasnt' trying to say increasing the gate resistor was a sure thing, just thought it might be worth trying to go even higher at least until you do notice some change one way or the other.
If you have a problem smoking fets while testing don't be shy to save your good ones you have on order for later refinements, I've had excellent sucess with industrial type mosfets for initial testing.
Regards,
Chris
Thanks Chris,
I'm on my last two, hence I'm waiting till monday with further experimenting (got 10 beefier ones coming) or else I'll probably be able to junk the perf board when they do burn out, it has pads dropping off already and soldering in another pair of MOSFETs may mean I'll have to populate another.
In the meantime I've drafted up a simpler version of the schematic I posted before (ver. 008) that does away with the IR2011 altogether, what do you think? Would this work? I only have a calculator at hand, I guess I should really learn how to simulate using LTspice properly.
simple_class_d_010
http://hardwareanalysis.com/images/articles/large/11685.gif
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
I'm on my last two, hence I'm waiting till monday with further experimenting (got 10 beefier ones coming) or else I'll probably be able to junk the perf board when they do burn out, it has pads dropping off already and soldering in another pair of MOSFETs may mean I'll have to populate another.
In the meantime I've drafted up a simpler version of the schematic I posted before (ver. 008) that does away with the IR2011 altogether, what do you think? Would this work? I only have a calculator at hand, I guess I should really learn how to simulate using LTspice properly.
simple_class_d_010
http://hardwareanalysis.com/images/articles/large/11685.gif
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Oh and Chris,
I'd be happy to do any measurements on this design, or the UcD's if you're wondering about something. I've learned a lot from your posts in this forum and your replies here have also helped me move forward with this design, so I'd be happy to return the favor.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
I'd be happy to do any measurements on this design, or the UcD's if you're wondering about something. I've learned a lot from your posts in this forum and your replies here have also helped me move forward with this design, so I'd be happy to return the favor.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
SSassen said:
Hi,
It does seem workable at a quick glance. I'll try and simulate it for you but may take some time I'm pretty slow getting around in LTspice still and can't get around to it until later.
I dont' usually see such output stages driven by a differential output so it would include the offset error of it in this way, but that is such a high end op amp I'd imagine any error would easily be ignored.
The FET's you ordered likely won't be more rugged. Better suitable to the job for sure, but I've found "rugged" to be the old industrial work horses that seem to really take a beating even though they are far less efficient and harder to drive, they're also cheaper and easier to obtain than the latest high end ones. Just another option
Regards,
Chris
PS: Thanks for kind words
PPS: Good seeing you around again Sub.
I agree with Chris.
From my experience, older mosfets are more reliable. I started with NTP35n15 (theroretically very big and rudged), but they failed quite often.
Then I started a thread (Mosfet reliability in Class-D amplifiers) which opened a very informative discussion about the many factors that influence reliability. I think the more important ones (apart from Vds rating and max. Id, of course), are reverse capacitance and diode dV/dt ratings.
My best experience so far is the old nice IRF640, which has a good balance between speed and Rds(on), and seems to be very rudged! They are also very cheap and easy to find. But there seems to be huge differences between manufacturers, so I always choose IR.
Best regards,
Pierre
From my experience, older mosfets are more reliable. I started with NTP35n15 (theroretically very big and rudged), but they failed quite often.
Then I started a thread (Mosfet reliability in Class-D amplifiers) which opened a very informative discussion about the many factors that influence reliability. I think the more important ones (apart from Vds rating and max. Id, of course), are reverse capacitance and diode dV/dt ratings.
My best experience so far is the old nice IRF640
, which has a good balance between speed and Rds(on), and seems to be very rudged! They are also very cheap and easy to find. But there seems to be huge differences between manufacturers, so I always choose IR.Best regards,
Pierre
Pierre said:I agree with Chris.
From my experience, older mosfets are more reliable. I started with NTP35n15 (theroretically very big and rudged), but they failed quite often.
Then I started a thread (Mosfet reliability in Class-D amplifiers) which opened a very informative discussion about the many factors that influence reliability. I think the more important ones (apart from Vds rating and max. Id, of course), are reverse capacitance and diode dV/dt ratings.
My best experience so far is the old nice IRF640
Best regards,
Pierre
Pierre I have the feeling the reason for the older/slow fets taking more abuse is because they take more to turn so they're perhaps less responsive to the miller effect and circulating currents.
I clearly remember they ran alot hotter and took that abuse for a looong time, while the high speed switching ones we love just burn instantly under less than ideal conditions.
Regards,
Chris
Hello Sander,
Switching amplifiers can generate a lot of common mode impulse noise. This can make taking clean oscilloscope measurements problematic, especially low level measurements like the triangle wave on your integrator opamp. Take all measurements relative to a ground local to the measured signal. Test for intruding common mode noise by touching the probe tip directly to the point where the probe ground is connected to the circuit being measured. Theoretically all signal should be shorted out, but in many cases almost all of the high frequency impulse noise may still be present, even when measuring "ground".
How to Nearly Eliminate All Common Mode Measurement Noise.
Note: what I am about to state may seem silly or too much trouble, but if you want clean measurements, carefully follow this advice.
Step 1. Get yourself a ferite core through which you can pass three to five turns of your 'scope probe's cable (the coax). Some of the larger, split-and-polished, clip-on anti-EMI test beads work nicely for this; so do ungapped ring cores with center holes large enough to pass the 'scope probe connector. Try to get a core with a cross section of at least half a square centimeter. Wrap the turns so that the core ends up next to the oscilloscope rather than the probe. Now rerun the ground noise pollution test from before. Common mode noise should be reduced by an order of magnitude or more.
Step 2. If more noise reduction is still required or desired, throw away the probe's ground clip and probe tip cover so that its probe tip can be plugged directly into the circuit under measurement. Commercial probe tip connectors are available for this purpose or you can make your own. The simplest DIY method is to solder in two short lengths of stout, tinned bus wire, one each to the measurement point and the nearest ground related directly to the circuit being measured. Keeping lengths as short as possible, wrap a curly cue of the bus wire around the bare 'scope probe's tip and wrap several turns of the ground bus wire around the 'scope probe's ground ring (right above its tip). Clip off the excess wire and you should be able to leave these in place and repeatedly unplug and replug the 'scope probe as needed. If these little "springs" seem too flimsy, I have found that some types of Euro style fuse clips work just right for making the ground connection to the ring of the 'scope probe.
Step 3. Any remaining noise is probably real and is indicative of layout problems such as improper mixing of noisy and quiet grounds, running sensitive low level signal lines too close or in parallel with the high level switched ones, and/or excessive loop areas involving switched high currents.
Regards -- analogspiceman
Switching amplifiers can generate a lot of common mode impulse noise. This can make taking clean oscilloscope measurements problematic, especially low level measurements like the triangle wave on your integrator opamp. Take all measurements relative to a ground local to the measured signal. Test for intruding common mode noise by touching the probe tip directly to the point where the probe ground is connected to the circuit being measured. Theoretically all signal should be shorted out, but in many cases almost all of the high frequency impulse noise may still be present, even when measuring "ground".
How to Nearly Eliminate All Common Mode Measurement Noise. Note: what I am about to state may seem silly or too much trouble, but if you want clean measurements, carefully follow this advice.
Step 1. Get yourself a ferite core through which you can pass three to five turns of your 'scope probe's cable (the coax). Some of the larger, split-and-polished, clip-on anti-EMI test beads work nicely for this; so do ungapped ring cores with center holes large enough to pass the 'scope probe connector. Try to get a core with a cross section of at least half a square centimeter. Wrap the turns so that the core ends up next to the oscilloscope rather than the probe. Now rerun the ground noise pollution test from before. Common mode noise should be reduced by an order of magnitude or more.
Step 2. If more noise reduction is still required or desired, throw away the probe's ground clip and probe tip cover so that its probe tip can be plugged directly into the circuit under measurement. Commercial probe tip connectors are available for this purpose or you can make your own. The simplest DIY method is to solder in two short lengths of stout, tinned bus wire, one each to the measurement point and the nearest ground related directly to the circuit being measured. Keeping lengths as short as possible, wrap a curly cue of the bus wire around the bare 'scope probe's tip and wrap several turns of the ground bus wire around the 'scope probe's ground ring (right above its tip). Clip off the excess wire and you should be able to leave these in place and repeatedly unplug and replug the 'scope probe as needed. If these little "springs" seem too flimsy, I have found that some types of Euro style fuse clips work just right for making the ground connection to the ring of the 'scope probe.
Step 3. Any remaining noise is probably real and is indicative of layout problems such as improper mixing of noisy and quiet grounds, running sensitive low level signal lines too close or in parallel with the high level switched ones, and/or excessive loop areas involving switched high currents.
Regards -- analogspiceman
analogspiceman said:
Spiceman you're a goldmine.
SODFA "Self Oscillating Digital Feedback Amplifier", don't blame Sanders for the name.
It came from a German fellow, here's a rough translation:
"The designation grew 1998 in unawareness of the two transducer
principles next-mentioned on my own muck.
Characteristic: Linear reversal integrator also comparator along-coupled by the exit
of the performance level."
The unawarness he claims is in reference to Bruno's SODA and the other I have no idea.
Maybe the name makes more sense in german, I don't know.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.