Mmm, seems that eletrolytics are not the most reliable components!
Well, if that was the cause, no problem, I will use higher voltage and better electrolytics, as well as isolate them thermally as much as possible.
But I hope the failure was not in reverse way: the high side mosfet fails short, it causes the driver to fail also and somehow it shorts its Vcc to the positive rail through the mosfet, causing the capacitor to explode. Doesn't seem very feasible. And the lower side mosfet didn't fail either...
Best regards,
Pierre
Well, if that was the cause, no problem, I will use higher voltage and better electrolytics, as well as isolate them thermally as much as possible.
But I hope the failure was not in reverse way: the high side mosfet fails short, it causes the driver to fail also and somehow it shorts its Vcc to the positive rail through the mosfet, causing the capacitor to explode. Doesn't seem very feasible. And the lower side mosfet didn't fail either...
Best regards,
Pierre
Thermally, I doubt you'd really have to worry, for one I doubt you went and mounted it to the heatsink so I'd think that's fine, and if you use a good computer grade cap with a 105 deg C temp rating it will easily remain beneath spec and provide an even longer than rated life span. The main worry is in using one that was made with the knock off electrolyte.
I found one forum at least a thousand posts long or more that started in 2003 and still up to date from yesterday discussing motherboards with this exact issue relating to all manufacturers of them and all the varieties of caps they use.
I found one forum at least a thousand posts long or more that started in 2003 and still up to date from yesterday discussing motherboards with this exact issue relating to all manufacturers of them and all the varieties of caps they use.
Pierre said:
If you're not using high temp caps to start with what are they rated for... 85 deg C as a guess? Would your mosfets survive at that? Possibly, and yes I'd think it could heat the cap over time as well. However a higher voltage ensures that it can safely handle higher surges with ease, if safely regulated you can use a more reasonable margin of safety and instead of going 5 or 10 times higher than you are actually using use a standard of say 20%.
With the regulator you have 16V min to 25V would be alright, that IC likely has overcurrent and over temp limiting/shutdown protection so the odds of the cap having to see full rail voltage is minimal unless your driver fails, and they like to go with mosfets so it would have to be well protected. I'd probably just using something like a 200V cap, at 25V it shouldn't end up being too huge.
Electrolytics are heatet from the inside by ripple and outside by nearby heat sources. Heating will not kill these caps within 5 hours but should be avoided because it is definitely shortening the life expectancy of these caps.
Were you nearby when the failure happened ? Did the amp fail first and then did the cap explode ? Or happened both at the same time ?
Regards
Charles
Were you nearby when the failure happened ? Did the amp fail first and then did the cap explode ? Or happened both at the same time ?
Regards
Charles
Thanks for your interest.
In fact, I wasn't there when it failed. I had two amps connected at the same supply.
When I came, what I saw was one of the amps in protection mode due to undervoltage and the positive rail of the supply with the fuse blown.
When I examined the failed amp, the small cap had exploded: the conductor had unwrapped and near it I found a spark mark in the heatsink, the driver was obviously burnt and the upper side mosfet had its three legs shorted.
The only feasible reason for all this failure chain (IMO) is that the cap failed first, causing a short from Vss+12V to GND (heatsink), producing the spark and feeding the whole 40V to the driver, causing it to fail and take the upper mosfet with it. Either the UVLO protection engaged quickly or the IR2110 driver failed so the lower mosfet didn't have time to turn on so it didn't failed and the negative rail fuse didn't blow.
The inverse sequence could be: failed upper side mosfet shorting its three pins caused the gate to become VCC, causing the driver to fail. But is it possible that this 40V entering on it could have gone to the drivers's supply, raising the 7812 output to 40V and then causing the cap to explode?
BTW: In another amp with a 25V cap instead of 16V, and a different routing that shouldn't produce much heat up from the mosfets, I noticed that the cap was at about 50degs. So it must be due to current ripple. It is clear that low ESR ones should be placed here (as well as good ceramics in parallel).
Best regards,
Pierre
In fact, I wasn't there when it failed. I had two amps connected at the same supply.
When I came, what I saw was one of the amps in protection mode due to undervoltage and the positive rail of the supply with the fuse blown.
When I examined the failed amp, the small cap had exploded: the conductor had unwrapped and near it I found a spark mark in the heatsink, the driver was obviously burnt and the upper side mosfet had its three legs shorted.
The only feasible reason for all this failure chain (IMO) is that the cap failed first, causing a short from Vss+12V to GND (heatsink), producing the spark and feeding the whole 40V to the driver, causing it to fail and take the upper mosfet with it. Either the UVLO protection engaged quickly or the IR2110 driver failed so the lower mosfet didn't have time to turn on so it didn't failed and the negative rail fuse didn't blow.
The inverse sequence could be: failed upper side mosfet shorting its three pins caused the gate to become VCC, causing the driver to fail. But is it possible that this 40V entering on it could have gone to the drivers's supply, raising the 7812 output to 40V and then causing the cap to explode?
BTW: In another amp with a 25V cap instead of 16V, and a different routing that shouldn't produce much heat up from the mosfets, I noticed that the cap was at about 50degs. So it must be due to current ripple. It is clear that low ESR ones should be placed here (as well as good ceramics in parallel).
Best regards,
Pierre
I haven't tried it out so far (I am somewhat reluctant to use patented ideas sometimes) but I would take the the ratio such that the voltage is slightly larger than the typical diode forward voltage. This could get quite tricky in some cases however.
E.g. if the coil has 20 windings and the your supply voltage is +- 40 Volts you will achieve a transformed voltage of 2 volts peak by the use of just one "secondary winding" for this "autotransformer" topology. So you would have to "kill" the excess voltage by the series resistor.
I am not sure at the momenbt but the resistor might best be dimensioned empirically.
Regards
Charles
E.g. if the coil has 20 windings and the your supply voltage is +- 40 Volts you will achieve a transformed voltage of 2 volts peak by the use of just one "secondary winding" for this "autotransformer" topology. So you would have to "kill" the excess voltage by the series resistor.
I am not sure at the momenbt but the resistor might best be dimensioned empirically.
Regards
Charles
Hey, guys. Have you seen the new LCAudio ZapPulse 2.3? They use a varistor at the output to protect from sudden spikes produced by load removal at high power levels.
What do you think about these things? I suppose it is connected from the mosfet's output to GND, to clamp any peaks from the coil, right?
Does anyone know what value does it have from the photo?
Have a look at:
http://www.lcaudio.com/zp23-2.jpg
Seems an interesting idea to protect mosfets, isn't it?
BTW: Still accumulating hours of reliable operation at high power with my small nice IRF640's!
What do you think about these things? I suppose it is connected from the mosfet's output to GND, to clamp any peaks from the coil, right?
Does anyone know what value does it have from the photo?
Have a look at:
http://www.lcaudio.com/zp23-2.jpg
Seems an interesting idea to protect mosfets, isn't it?
BTW: Still accumulating hours of reliable operation at high power with my small nice IRF640's!
Hi,
I think Lars is the one to answer why he uses varistor. I only think that mosfets are in no danger because of interrupted load. Parasitic diodes do not allow mosfet voltages to be higher than rail capacitor voltage. Flyback current will rather charge output filter capacitor which has considerably smaller value than bulk rail capacitor.
Best regards,
Jaka Racman
I think Lars is the one to answer why he uses varistor. I only think that mosfets are in no danger because of interrupted load. Parasitic diodes do not allow mosfet voltages to be higher than rail capacitor voltage. Flyback current will rather charge output filter capacitor which has considerably smaller value than bulk rail capacitor.
Best regards,
Jaka Racman
Maybe it's limiting the filtered output voltage in case of LC series resonance?
When the load is not connected, the output filter is a LC series resonant tank with high Q. If a resonance occurs, high voltage
will be send out from the speaker terminals -- very dangerous.
The resonance could happen during these two circumstances (but not limited to these):
1) Class-D without post-LPF feedback, driven by a input signal with resonance frequency.
2) Class-D with a FAULT post-LPF feedback loop and oscillating at resonance frequency. (This is likely to happen in a DIY one.)
I had once did the resonance experiment and got 150V output from single 24V power supply, yet no any device failed. If powered by +/- 50V , the output terminals could easily kills!
Since the experiment, all my class-D design has clamp diode at the LPF output.
When the load is not connected, the output filter is a LC series resonant tank with high Q. If a resonance occurs, high voltage
will be send out from the speaker terminals -- very dangerous.The resonance could happen during these two circumstances (but not limited to these):
1) Class-D without post-LPF feedback, driven by a input signal with resonance frequency.
2) Class-D with a FAULT post-LPF feedback loop and oscillating at resonance frequency. (This is likely to happen in a DIY one.)
I had once did the resonance experiment and got 150V output from single 24V power supply, yet no any device failed. If powered by +/- 50V , the output terminals could easily kills!
Since the experiment, all my class-D design has clamp diode at the LPF output.
Hi everyone,
Just want to share something that happened to our amps 5 years ago.
As we use only N-channel Mosfets only.
Once our amp was attached to high frequency piezo horns and after just 1 hour of playing music near the clipping it went into sudden brust of flames and all the mosfets were destroyed instantly. This was happening with almost our all amps at that time. So i decided to perform a check on gate to source voltage spikes. We at that time use only one zener of 12V at the driver side before gate resistors, and what we see was certainly strange. the waveform of AC signal before the gate resistor was never exceeding 5 VRMS whereas the waveform after gate resistor was jumping to almost 15Volts into spikes. Then to eliminate this we first increase the driver current to 55mA to drive 10 pairs of mosfets and connected the gate to source legs of each mosfets with seperate zeners of 9V each, and till now this problem never arises. The miller effect was the Culprit of turn on the mosfets thereby self charging the gate of upper mosfets to max and eventually destroy them in no time.
regards
Kanwar
Just want to share something that happened to our amps 5 years ago.
As we use only N-channel Mosfets only.
Once our amp was attached to high frequency piezo horns and after just 1 hour of playing music near the clipping it went into sudden brust of flames and all the mosfets were destroyed instantly. This was happening with almost our all amps at that time. So i decided to perform a check on gate to source voltage spikes. We at that time use only one zener of 12V at the driver side before gate resistors, and what we see was certainly strange. the waveform of AC signal before the gate resistor was never exceeding 5 VRMS whereas the waveform after gate resistor was jumping to almost 15Volts into spikes. Then to eliminate this we first increase the driver current to 55mA to drive 10 pairs of mosfets and connected the gate to source legs of each mosfets with seperate zeners of 9V each, and till now this problem never arises. The miller effect was the Culprit of turn on the mosfets thereby self charging the gate of upper mosfets to max and eventually destroy them in no time.
regards
Kanwar
Perhaps not the best thread for posting this but I haven't found the one I started ("help with feedback").
Well, my amplifiers have proven to be quite reliable by now once I substituted the mosfets and rerouted the pcb.
Now I have integrated two amplifiers in a case with a common power supply, etc.
The mosfets are screwed to a heatsink for each module, that is isolated from the module's GND but connected to the chassis, that is ultimately connected to GND via the RCA input connectors.
I have noticed two annoying effects:
a) There is a whistle when two modules are running. It has stopped when I have connected the heatsink to each module's ground. I think it is due to capacitive coupling of the switching node to the heatsink via the mosfet isolator, so it has become a big antenna, and to put it worse, connected to the other module via the chassis, so there is strong coupling between both switching nodes.
What is best: connect the heatsink to GND in the module? But then, my earth will become the same node as GND, is that safe/good?
Perhaps connecting the heatsink to the amp's ground with a capacitor is better, right?
b) I will talk about the other problem later... ;-)
Thanks!
Pierre
Well, my amplifiers have proven to be quite reliable by now once I substituted the mosfets and rerouted the pcb.
Now I have integrated two amplifiers in a case with a common power supply, etc.
The mosfets are screwed to a heatsink for each module, that is isolated from the module's GND but connected to the chassis, that is ultimately connected to GND via the RCA input connectors.
I have noticed two annoying effects:
a) There is a whistle when two modules are running. It has stopped when I have connected the heatsink to each module's ground. I think it is due to capacitive coupling of the switching node to the heatsink via the mosfet isolator, so it has become a big antenna, and to put it worse, connected to the other module via the chassis, so there is strong coupling between both switching nodes.
What is best: connect the heatsink to GND in the module? But then, my earth will become the same node as GND, is that safe/good?
Perhaps connecting the heatsink to the amp's ground with a capacitor is better, right?
b) I will talk about the other problem later... ;-)
Thanks!
Pierre
All
This has been a most enlightening thread and I have read it all. Almost a tutorial in fact. Many thanks to all the contributors.
I'd like to add a simple contribution on snubber design. In my experience the formulas in the following link result in RC component values that are usually effective on first try. I often find the C value can be reduced, sometimes by as much as 50%.
http://www.ridleyengineering.com/snubber.htm
These formulae assume you know one of the parasitics, C or L
C can be found by measuring the frequency of the ringing and then temporarily adding an external Cext across the existing C (usually FET source-drain) such that the ringing frequency is halved. Then:
f/2 = 1/(2*pi*sqrt(L(C+Cext);
so that approximately, C = Cext.
I have found that the 'TI trick' with RC snubber from top FET drain to bottom FET source, with V+ power fed through a small inductor(as described by Jaka in this thread) works well with snubbers designed with this method. When finding Cext, bridge the cap across first the top FET and then the bottom FET and use the one that shows the biggest difference in ringing frequency to find Cext.
Fitting RC snubbers across the FETs themselves may well damp the ringing most effectively, but this is always done at the cost of slowing the switching time which will increase distortion. But sometimes this may be preferable to a blindingly fast OPS that's under high repetitive voltage and/or current stress that can bring failure; if not today, in six months time.
Choice of components is quite critical: for RC snubbers you must use non-inductive resistors and I find ceramic caps are easily the best. Jaka Racman has previously posted on this thread on the inductance of foil wound caps. NB!!
Layout is even more critical. I find best results are obtained when the RC snubber is connected as close as possible to the FET pins. Any snubber trace more than 6mm is too long.
SMD snubber components are preferable but difficult to find in the right values. I've found that making a 'stack' of 2512 1.5W SMD high surge power resistors for the right R and P value works quite well. The Vishay TO-220 package 'non inductive' resistors, and the supposedly dedicated 5W film 'snubber resistors' on ceramic subtrate have too much lead inductance and are too bulky. Watch the voltage rating on SMD ceramic caps!
When examining switching waveforms for ringing, the need for proper probe connections has been amply covered in this thread. What's not been addressed is the need for a really fast scope. I will show this by way of an example:
Suppose you have built an OPS which has very tight power routing and 20nH stray inductance. And your FETs have, say, 250pF of effective capacitance, that gives you a ringing frequency of 71MHz. You're not going to see this on a 20MHz scope. You need at very least a 150MHz scope for doing this kind of work.
This has been a most enlightening thread and I have read it all. Almost a tutorial in fact. Many thanks to all the contributors.
I'd like to add a simple contribution on snubber design. In my experience the formulas in the following link result in RC component values that are usually effective on first try. I often find the C value can be reduced, sometimes by as much as 50%.
http://www.ridleyengineering.com/snubber.htm
These formulae assume you know one of the parasitics, C or L
C can be found by measuring the frequency of the ringing and then temporarily adding an external Cext across the existing C (usually FET source-drain) such that the ringing frequency is halved. Then:
f/2 = 1/(2*pi*sqrt(L(C+Cext);
so that approximately, C = Cext.
I have found that the 'TI trick' with RC snubber from top FET drain to bottom FET source, with V+ power fed through a small inductor(as described by Jaka in this thread) works well with snubbers designed with this method. When finding Cext, bridge the cap across first the top FET and then the bottom FET and use the one that shows the biggest difference in ringing frequency to find Cext.
Fitting RC snubbers across the FETs themselves may well damp the ringing most effectively, but this is always done at the cost of slowing the switching time which will increase distortion. But sometimes this may be preferable to a blindingly fast OPS that's under high repetitive voltage and/or current stress that can bring failure; if not today, in six months time.
Choice of components is quite critical: for RC snubbers you must use non-inductive resistors and I find ceramic caps are easily the best. Jaka Racman has previously posted on this thread on the inductance of foil wound caps. NB!!
Layout is even more critical. I find best results are obtained when the RC snubber is connected as close as possible to the FET pins. Any snubber trace more than 6mm is too long.
SMD snubber components are preferable but difficult to find in the right values. I've found that making a 'stack' of 2512 1.5W SMD high surge power resistors for the right R and P value works quite well. The Vishay TO-220 package 'non inductive' resistors, and the supposedly dedicated 5W film 'snubber resistors' on ceramic subtrate have too much lead inductance and are too bulky. Watch the voltage rating on SMD ceramic caps!
When examining switching waveforms for ringing, the need for proper probe connections has been amply covered in this thread. What's not been addressed is the need for a really fast scope. I will show this by way of an example:
Suppose you have built an OPS which has very tight power routing and 20nH stray inductance. And your FETs have, say, 250pF of effective capacitance, that gives you a ringing frequency of 71MHz. You're not going to see this on a 20MHz scope. You need at very least a 150MHz scope for doing this kind of work.
Hello all.
Continuing with the mosfets issue, just a comment:
I am using IRF640 mosfets, marked with a "H" (they must be from Harris), not from IRF, and working perfect.
Yesterday I replaced a couple with IRF640 fron another manufacturer, ST (from the datasheet they seem equivalent to IRF640N), and they run veery bad
: they overheat a lot at idle, and the switching waveform has a lot of ringing. I assume that they are slow and produce a lot of shoot-through.
How can that be? Aren't they exactly the same?
I have to buy some IRF640N from IR to check them, because now I am worried about mosfet disparity.
BTW: I have mounted a couple of IRFB38N20 and they seem OK by the moment, no heat at idle, 45ns rise/fall time (it was about 20ns with IRF640 "H"). Let's see how they work under load.
Best regards,
Pierre
Continuing with the mosfets issue, just a comment:
I am using IRF640 mosfets, marked with a "H" (they must be from Harris), not from IRF, and working perfect.
Yesterday I replaced a couple with IRF640 fron another manufacturer, ST (from the datasheet they seem equivalent to IRF640N), and they run veery bad
: they overheat a lot at idle, and the switching waveform has a lot of ringing. I assume that they are slow and produce a lot of shoot-through.How can that be? Aren't they exactly the same?
I have to buy some IRF640N from IR to check them, because now I am worried about mosfet disparity.
BTW: I have mounted a couple of IRFB38N20 and they seem OK by the moment, no heat at idle, 45ns rise/fall time (it was about 20ns with IRF640 "H"). Let's see how they work under load.
Best regards,
Pierre
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.