Hello Eva
What I mean by this is that I dont push it to 150%, but means that you can push it more close to their limit without the risk of explosion!
I have lot of years of experience in class d and reliability on the field is really important. Lot of user use amplifier in stupid ways (that's the need for stupid proof protection and Mosfet). A cheap 13A IXYS mosfet can be abuse a lot more than a 23A IRF mosfet. That's what I have experiment with the time... And much better, most of the time are more efficient!
That's the breaking point I think between cheap and good products... It's not only if it do the job, but if it can do the job in any environement! Like on a gas generator, on african electricity network, or under tropical rain with 100% humidity when the show must go on! All this event will cause ''commercial'' mosfet to be overdrive, when overdrived ''industrial'' mosfet with electricity spike or humidity leak between drain and gate still work, bad, but still work without faillure! And no, I dont told you to test your amplifier under water.... But certification required a 8 hours test into 40 degree at 95% humidity....Poor Infineon, fairchild and IRF!
That's why, even with the price, I prefered IXYS and APT!
Reliability and effiency are my first concern!
Fredos
What I mean by this is that I dont push it to 150%, but means that you can push it more close to their limit without the risk of explosion!
I have lot of years of experience in class d and reliability on the field is really important. Lot of user use amplifier in stupid ways (that's the need for stupid proof protection and Mosfet). A cheap 13A IXYS mosfet can be abuse a lot more than a 23A IRF mosfet. That's what I have experiment with the time... And much better, most of the time are more efficient!
That's the breaking point I think between cheap and good products... It's not only if it do the job, but if it can do the job in any environement! Like on a gas generator, on african electricity network, or under tropical rain with 100% humidity when the show must go on! All this event will cause ''commercial'' mosfet to be overdrive, when overdrived ''industrial'' mosfet with electricity spike or humidity leak between drain and gate still work, bad, but still work without faillure! And no, I dont told you to test your amplifier under water.... But certification required a 8 hours test into 40 degree at 95% humidity....Poor Infineon, fairchild and IRF!
That's why, even with the price, I prefered IXYS and APT!
Reliability and effiency are my first concern!
Fredos
Hi Fredos
I may be still missing something...
The +/-200V +/-30A amplifier that I mentioned in one of my previous posts is currently being subject to the kind of certification tests that you have mentioned. The own customer is performing these tests as they are quite demanding people. Temperature and humidity conditions are likely to be as you mentioned. Output power is probably 1800W continuous and test periods are likely to be one week long rather than 8 hours.
I'm using Infineon MOSFETs (my favourite for 600V applications), Fairchild IGBTs, On Semiconductor diodes and IR drivers... No failures reported so far. Oh, and the entire amplifier including active PFC power supply is efficient enough to use a single convection-cooled heatsink, without fans. There is only a small fan to move some air inside the case to cool the magnetics and the resistors of the snubbers, and I'm doing my best to get rid of it (because the class-D switching stage itself and the PFC can produce 3KW continous if the magnetics don't overheat).
Furthermore, the previous prototype (no PFC) was also subject to the same endurance tests without failures, and it employed 12 low-cost 600V 10A TO-220 Infineon IGBTs to achieve over 1800W continuous output. However, the customer complained about 88% efficiency and 0.6 power factor, so I had to find a way to get into the 95% range including additional losses from an active PFC stage (it gave me lots of headaches!!)
I suppose that the difference is that some people is skilled enough to get big performance and reliability from small transistors while others just aren't (and will probably never be).
I may be still missing something...
The +/-200V +/-30A amplifier that I mentioned in one of my previous posts is currently being subject to the kind of certification tests that you have mentioned. The own customer is performing these tests as they are quite demanding people. Temperature and humidity conditions are likely to be as you mentioned. Output power is probably 1800W continuous and test periods are likely to be one week long rather than 8 hours.
I'm using Infineon MOSFETs (my favourite for 600V applications), Fairchild IGBTs, On Semiconductor diodes and IR drivers... No failures reported so far. Oh, and the entire amplifier including active PFC power supply is efficient enough to use a single convection-cooled heatsink, without fans. There is only a small fan to move some air inside the case to cool the magnetics and the resistors of the snubbers, and I'm doing my best to get rid of it (because the class-D switching stage itself and the PFC can produce 3KW continous if the magnetics don't overheat).
Furthermore, the previous prototype (no PFC) was also subject to the same endurance tests without failures, and it employed 12 low-cost 600V 10A TO-220 Infineon IGBTs to achieve over 1800W continuous output. However, the customer complained about 88% efficiency and 0.6 power factor, so I had to find a way to get into the 95% range including additional losses from an active PFC stage (it gave me lots of headaches!!)
I suppose that the difference is that some people is skilled enough to get big performance and reliability from small transistors while others just aren't (and will probably never be).
For space purpose, it's better with only 2 outputs...12 is too much for my case size! Anyways, you agree with me that 12 small outputs can be as reliable as 2 big one! And faillure are more probable with 12 outputs than 2! And I'm pretty sure that 12 of your outputs cost near the same as my 2 beast! So topology for topology, I think that we get the same result? No?
About PFC, their is no rule yet here, so I take the train when is there!
Fredos
About PFC, their is no rule yet here, so I take the train when is there!
Fredos
There are reasons other than size or cost to use 12 small co-pack IGBTs with diode in a 400V full bridge rather than 4 big MOSFETs, you may discover some day...
BTW: Big TO-247, TO-218 and TO-3P cases exhibit quite poor performance in fast switching applications because lead inductance (in the 20nH range) resonates with internal drain-source and collector-emitter capacitance, usually in the 10Mhz to 50Mhz range. Lead inductance also imposes a limit on the switching slopes achievable with conventional gate drive circuits. An extra source or emitter lead with its own bonding wire attached to the die would be a great thing. The problem becomes even worse in those big "beast" devices due to increased capacitances and the higher currents being usually conducted.
BTW: Big TO-247, TO-218 and TO-3P cases exhibit quite poor performance in fast switching applications because lead inductance (in the 20nH range) resonates with internal drain-source and collector-emitter capacitance, usually in the 10Mhz to 50Mhz range. Lead inductance also imposes a limit on the switching slopes achievable with conventional gate drive circuits. An extra source or emitter lead with its own bonding wire attached to the die would be a great thing. The problem becomes even worse in those big "beast" devices due to increased capacitances and the higher currents being usually conducted.
again hard, but EXACTLY I dont call 62.5Khz fast switching....Ans most of the time, there is more inductance in the routing of PCB...Specialy with multiple outputs!
Fredos
Fredos
To Fredos
It's really hard to discuss switching electronics with you because you seem to get most of the concepts wrong during most of the time. Switching speed is not necessarily related to frequency. You can sonsider my amplifier as an example of that, since everything is clocked at 48Khz (the PFC is synchronized ) but my turn-on crossover times from 400V to 0V are in the 60ns range and turn-off from 0V to 600V takes 25ns at full load. My switching at 48Khz is probably way faster than yours at 300Khz, and note that my turn-on is intentionally slowed down to reduce EMI and to avoid TO-247 self-resonance, because the magnetic snubbers that I use to limit diode di/dt cause very little current to be yet flowing after 60ns.
I'm taking advantage of more tricks to boost efficiency, but I can't reveal them for obvious reasons. I think that 95% full load efficiencty for an amplifieir including power supply can be considered state of the art.
It's really hard to discuss switching electronics with you because you seem to get most of the concepts wrong during most of the time. Switching speed is not necessarily related to frequency. You can sonsider my amplifier as an example of that, since everything is clocked at 48Khz (the PFC is synchronized ) but my turn-on crossover times from 400V to 0V are in the 60ns range and turn-off from 0V to 600V takes 25ns at full load. My switching at 48Khz is probably way faster than yours at 300Khz, and note that my turn-on is intentionally slowed down to reduce EMI and to avoid TO-247 self-resonance, because the magnetic snubbers that I use to limit diode di/dt cause very little current to be yet flowing after 60ns.
I'm taking advantage of more tricks to boost efficiency, but I can't reveal them for obvious reasons. I think that 95% full load efficiencty for an amplifieir including power supply can be considered state of the art.
fredos said:
sounds like to you the separation point is either price (more expensive = industrial) or failure tolerance (over "capacity" without failure?).
I can hardly see the 1st point: there are expensive devices from all manufacturers and what's the point?
on your second point: there is a difference between rated capacity and true capacity. a device manufacturer may decide to rate a part, for example, at 300v when in fact most of them can take 350v. I don't think logically it is possible to push one over its (true) capacity - that's just violates the definition of capacity.
a prudent design is one where a device isn't pushed over its (rated) capacity. But if you really want to, maybe you should just pick a higher rated part to beging with.
Genomerics approach, which he is trying to patent without having even bothered to try with real components, is clearly based in the following thread that I started two weeks before he filed the provisional patent. On the other hand, my thread shows real oscilloscope captures from a commercial power supply designed by me.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=88740
Fortunately, the approach that I employed (no schematic shown) exhibits superior performance and does not require a coupled inductor. However, I will be no longer publishing any more of my 'tricks' in this forum. I'm sorry for all the DIYers, but I can't risk having more unfair cheaters trying to patent my work without having even bothered to heat their soldering irons...
OK, probably it is based on it, but I don't see any problem with this! His conclusion and solution is quite different.
Let him to be happy with that! I think your soultion is better (due to its simplicity), but his one can work too.
It is high time that somebody tells you that all your "tricks" you post here are nothing more than reinventing hot water. And that is a compliment, since otherwise one might think that you purposely steal and post other's ideas to a bunch of class D amateurs. In the very same thread you have shown is a link to a previous implementation of what you claim to be "your" idea. Sweet Jesus, do you think you invented switchmode design? Next time you post something "original" just check it was not invented and described 20 years ago. Oh, let me guess you can't, you have no access to IEEE articles and you are to cheap to pay for a membership. Please do not bother replying, this is my last post here. This forum has deteriorated low enough.
Eva said:
Honest Eva. I haven't tried to rip off one of your ideas. I hadn't seen your previous posting but I know you are doing this sort of stuff so I thought you might be interested.
I am playing with PFC, as you say just theoretically, and was having problems with power losses in mosfets so turned to IGBTs. I've looked at them before but never really liked them because of the switching losses.
This time I was forced to look harder and I did try paralleling the IGBT with a mosfet to achieve zero current turn off. Unfortunately it did not work, in simulation, for me so I guessed about stored charge and came up with the reverse bias/charge removal idea that is shown on the web page.
I did a patent search to see if there was something similar about and found things like....
paralleled IGBT/Mosfet1
paralleled IGBT/Mosfet2
But nothing that mentioned applying a reverse bias to the CE terminals to remove stored charge so I went and put a patent in. It's not a very good one and I doubt if anything will come of it which is why I published the web page.
Like I say I gave you the link because I know you are working on similar stuff and might be interested. I really did come up with the idea independantly. If you have something that works for you and it is simpler then that's great but please believe me.....
I'm not trying to 'rip you off'.
DNA
Pafi said:
Of course I leave him to be happy with a slightly different solution (which has theoretical background only). Anyway, there are several problems that arise even when simple paralleling is actually tried with real components. For example, at high currents the IGBT may turn on by itself while the MOSFET turns-off and Vce is rising due to reverse transfer capacitance, thus wasting part of the benefit. Another even funnier problem with paralleling is current bouncing action between both devices during turn-on phase, as each one is pulling down the gate of the other through reverse transfer capacitances and lead inductances, which usually results in a nasty oscillator during a few hundred nanoseconds. And finally, there is the classic diode reverse recovery problem.
What I claim is that I'm solving these issues in slightly unique ways, which I can no longer risk revealing. Also, IEEE papers tend to be beautiful from a theoretical point of view, but my experience is that they hardly cover the actual problems that arise when your target is to create a working product for a very demanding client and not just a thesis for university... I'd rather expend the membership money in components and stuff for experimentation...
Genomerics said:
Then please accept my apologies. However, I must still criticise the fact of trying to file a patent on something that you haven't even tried and measured with real components (while somebody else may have it working and will be in trouble the patent is accepted). Hunting patents in that way is a truly unfair thing. Finally, if you accept some advice, stop trusting simulation when switching behaviour of transistors is involved, software is intended for more conceptual work (like control-loop modelling) and it usually gets all low-level switching plots wrong. Software told you that direct paralleling wasn't going to work, while it can actually achieve complete current-tail killing with latencies below 200ns (depending on IGBT choice).
EDIT: Concerning these two Japanese patents that you cited, note that they leave the unexpected problems that I explained unsolved.
No worries......
Regarding patents. I'm not really in a position to defend them both in the legal and moral sense. However if you have commercially sensitive stuff then you can file for a patent or keep it secret and profit from it by using it. The patent route is considered pretty worthless by some people....
Don Lancaster
At best it just gives me a reference date. If someone else has done similar stuff prior to that then obviously my patent becomes invalid. If they do it later, and arrive at the idea independently, then yes it does seem unfair. However in my case I'm not likely to be chasing them.
It's not actually necessary to physically 'prove' that the idea really does work. There may be some part of patent law that restricts you in that area as well. Again I suppose it seems unfair but if an idea does work then it works. If someone is going to implement it then, as you have said and know, there will other problems.
At the end of the day if I come up with something that is 'new' and worthwhile what should I do with it? It's not as if I haven't invested time and effort in getting there. Is it fair that I give it over to MegaBuck Incorporated without some form of recompense?
That's going to happen anyway.......
I accept some of what you say about modelling but a lot of that is down to experience, how you use the software and how much you believe the results. Certainly in this case I had been using some ST IGBT models and quite frankly they were Rubbish. The IR model I ended up working with was much more believable.
When I have a bench to work at I use Spice as a thinking tool. It's just too easy to waste time prodding about in the guts of something that is misbehaving when what you need to do is get away from it and think. It has also been extremely useful in gaining insight into what might be happening and also learning things I thought I knew. It's good for demonstrating stuff as well...
Snubbing
I haven't added to that but there is something about reverse recovery of the output diodes and leakage inductance in the transformer that results in the voltage spike during diode turn off. I wouldn't have figured it out without modelling it in Spice and seeing what the 'idealised' waveforms were doing. It is cute.
I am biting my tongue here. However I have seen too many people rubbish simulation only to find that they have implemented things that were 'silly' and simulation would have given them some idea of how silly they were being. On the other hand I have also seen people simulate things and get the wrong answer as well......
I use both, believe neither, and question the results from each side until I understand what is going on and what compromises are being or have to be made. Ultimately it's about experience and how you gain it but I do not think it is right to just adopt a hard nosed opinion about simulation and throw it and people who use it in the bin.
No, the Japanese patents don't cover 'problems'. No, mine doesn't cover them either. However I can guess at some of them and, if and when I do build one, will have to sort out the other ones that crop up as well. However you had to have the original problem to think up a solution to it and then discover solutions to the problems that cropped up.
Yes, I believe that using the 'right' IGBT will 'improve' the situation but I'd rather have something that 'solves' it for all cases. I'm not being so bold as to say what I've suggested does but I'll be trying it. You mention problems with direct paralleling but the method I'm proposing is not really direct paralleling so it may overcome some of those problems and throw up ones of its own.
And, yes, I will be trying the resonant diode soft recovery technique. And yes..... I will be modelling it before I implement it to gain some insight as to what is really going on.
Anyway, it's about time for lunch for us English type people and your lot is on the same time line, sort of, so have a tasty one.
DNA
Regarding patents. I'm not really in a position to defend them both in the legal and moral sense. However if you have commercially sensitive stuff then you can file for a patent or keep it secret and profit from it by using it. The patent route is considered pretty worthless by some people....
Don Lancaster
At best it just gives me a reference date. If someone else has done similar stuff prior to that then obviously my patent becomes invalid. If they do it later, and arrive at the idea independently, then yes it does seem unfair. However in my case I'm not likely to be chasing them.
It's not actually necessary to physically 'prove' that the idea really does work. There may be some part of patent law that restricts you in that area as well. Again I suppose it seems unfair but if an idea does work then it works. If someone is going to implement it then, as you have said and know, there will other problems.
At the end of the day if I come up with something that is 'new' and worthwhile what should I do with it? It's not as if I haven't invested time and effort in getting there. Is it fair that I give it over to MegaBuck Incorporated without some form of recompense?
That's going to happen anyway.......
I accept some of what you say about modelling but a lot of that is down to experience, how you use the software and how much you believe the results. Certainly in this case I had been using some ST IGBT models and quite frankly they were Rubbish. The IR model I ended up working with was much more believable.
When I have a bench to work at I use Spice as a thinking tool. It's just too easy to waste time prodding about in the guts of something that is misbehaving when what you need to do is get away from it and think. It has also been extremely useful in gaining insight into what might be happening and also learning things I thought I knew. It's good for demonstrating stuff as well...
Snubbing
I haven't added to that but there is something about reverse recovery of the output diodes and leakage inductance in the transformer that results in the voltage spike during diode turn off. I wouldn't have figured it out without modelling it in Spice and seeing what the 'idealised' waveforms were doing. It is cute.
I am biting my tongue here. However I have seen too many people rubbish simulation only to find that they have implemented things that were 'silly' and simulation would have given them some idea of how silly they were being. On the other hand I have also seen people simulate things and get the wrong answer as well......
I use both, believe neither, and question the results from each side until I understand what is going on and what compromises are being or have to be made. Ultimately it's about experience and how you gain it but I do not think it is right to just adopt a hard nosed opinion about simulation and throw it and people who use it in the bin.
No, the Japanese patents don't cover 'problems'. No, mine doesn't cover them either. However I can guess at some of them and, if and when I do build one, will have to sort out the other ones that crop up as well. However you had to have the original problem to think up a solution to it and then discover solutions to the problems that cropped up.
Yes, I believe that using the 'right' IGBT will 'improve' the situation but I'd rather have something that 'solves' it for all cases. I'm not being so bold as to say what I've suggested does but I'll be trying it. You mention problems with direct paralleling but the method I'm proposing is not really direct paralleling so it may overcome some of those problems and throw up ones of its own.
And, yes, I will be trying the resonant diode soft recovery technique. And yes..... I will be modelling it before I implement it to gain some insight as to what is really going on.
Anyway, it's about time for lunch for us English type people and your lot is on the same time line, sort of, so have a tasty one.
DNA
...I am really wondering how to measure such high efficiencies...
If I look to the specs of normal equipment then I am finding accurracies like +/-1% for normal DC measurements. For AC at 50 Hz also, but already for AC power where we have to take into account phase shift and harmonics I am happy if I can measure the power with +/- 3% overal accuracy with professional equipment (forget my hobby instruments at home). And if things are coming to amp output with several kHz..., then power is really hard to measure high accuracy. Even if we assume perfectly resistive load and just measure the voltage, you usually can expect a power tolerance of again +/-2%...
So even if professional equipment is telling us 95% efficiency from AC power input towards amp output... we must consider that it is 95% +/-5%... means: Something between 90% and 100%.
Or do I miss something?
On the other hand it is better not look to the accuracy specs of measurement equipment and better not to do a measurement error analysis, because it's time consuming and usually frustrating...
So for my DIY projects I have simplified things:
If it is small and cool, then efficacy is OK. 😀
If I look to the specs of normal equipment then I am finding accurracies like +/-1% for normal DC measurements. For AC at 50 Hz also, but already for AC power where we have to take into account phase shift and harmonics I am happy if I can measure the power with +/- 3% overal accuracy with professional equipment (forget my hobby instruments at home). And if things are coming to amp output with several kHz..., then power is really hard to measure high accuracy. Even if we assume perfectly resistive load and just measure the voltage, you usually can expect a power tolerance of again +/-2%...
So even if professional equipment is telling us 95% efficiency from AC power input towards amp output... we must consider that it is 95% +/-5%... means: Something between 90% and 100%.
Or do I miss something?
On the other hand it is better not look to the accuracy specs of measurement equipment and better not to do a measurement error analysis, because it's time consuming and usually frustrating...
So for my DIY projects I have simplified things:
If it is small and cool, then efficacy is OK. 😀
ChocoHolic said:
why bother with measuring? I thought it has been a tradition for some on this forum to just deem anything from others to be of a low enough efficiency. does 30% ring a bell for you?
...well, if a class D amp shows only 30% efficacy... YES, then some alarms bells would ring ...
And in fact finding out if you are around 30% or around 90% is easy to measure.
But I am bothering, because this thread here is discussing if it is 91% or 94%. Unfortunately there is quite a good chance that exactly one and the same proto would show 89% when measured with equipment A , but when measured with equipment B it would show 96%....
And in fact finding out if you are around 30% or around 90% is easy to measure.
But I am bothering, because this thread here is discussing if it is 91% or 94%. Unfortunately there is quite a good chance that exactly one and the same proto would show 89% when measured with equipment A , but when measured with equipment B it would show 96%....
Jaka,
".....this is my last post here. This forum has deteriorated low enough"
Right because of that it will be really a pity to lose you, please think about!
Regards
Heinz!
".....this is my last post here. This forum has deteriorated low enough"
Right because of that it will be really a pity to lose you, please think about!
Regards
Heinz!
- Status
- Not open for further replies.