Tek-Tips is the largest IT community on the Internet today!
Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!
-
Congratulations cowski on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!
FET's Issue 3
- Thread starter zacky
- Start date
- Thread starter
- #23
At 6.25 kVA, you are venturing into SCR and TRIAC areas.
FET's with a 200v rating (the minimum I would consider IF
everything else was right), are still premium devices.
At 120vdc in, you will be switching over 60A, even with
a very efficient design. IGBT's with a sufficient voltage
rating are readily available, but suffer from the need to
use emitter balancing similar to BJT's.
<als>
FET's with a 200v rating (the minimum I would consider IF
everything else was right), are still premium devices.
At 120vdc in, you will be switching over 60A, even with
a very efficient design. IGBT's with a sufficient voltage
rating are readily available, but suffer from the need to
use emitter balancing similar to BJT's.
<als>
- Thread starter
- #26
O/k guys plenty of meat here to chew on.
melone, yes most automotive ignition modules certainly do use IGBTs these days in flyback mode. But ignition flyback is a fairly gentle affair !!! First, turn off will be relatively slow due to the low self resonant frequency of a typical automotive ignition coil, and the turn off spike on the primary will be relatively low voltage, perhaps only 200v maximum. Remember it is basically a 12v system being switched.
An ignition system running at 100 Hz cannot really be compared with a 20KHz high voltage switching power supply. Even if the mJ losses per cycle were identical, the total switching loss would be twenty times higher at the higher operating frequency.
zacky, wow that is a lot of power. The average dc input current may be around 50A, and peak current in the switching device four times that, say 200A. Probably closer to 250A in a real supply (worst case).
I think the way I would do it would be to use four separate flyback supplies phased ninety degrees apart. Each would then only see 60 Amps peak current. They could run from a master oscillator and some flip flops to frequency lock the whole thing.
If each flyback supply runs at 20Khz, input and output ripple frequencies would then be 80Khz, but of much lower amplitude than one huge single flyback supply. That will be far easier to filter, and capacitor size, cost, and ripple rating will be much reduced.
Design of the magnetics will be less of a problem too. Skin effect means you cannot just make things bigger, and 250 peak amps is still 250 peak amps. But four independent foil wound flyback transformers (chokes actually) that only see 60 primary amps peak each, is doable. Current sharing between the four supplies will be automatic.
Conduction losses of higher voltage FETs will be a problem for you, but parallel FETs and IGBTs would be one possible way to overcome that. An interesting project to be sure.
Don't overlook the advantages of the diagonal half bridge flyback topology. For reasonably high input voltage, and high power it can be far more reliable. The peak drain voltage on the switching FETs will be hard clamped to the maximum dc input voltage, and that solves a whole lot of problems. The disadvantage is twice the conduction loss, but I think it well worth it for relatively high input voltage.
Ah, this all stirs memories. I used to do a lot of this sort of thing once. Now, being retired, it all seems so very long ago.
melone, yes most automotive ignition modules certainly do use IGBTs these days in flyback mode. But ignition flyback is a fairly gentle affair !!! First, turn off will be relatively slow due to the low self resonant frequency of a typical automotive ignition coil, and the turn off spike on the primary will be relatively low voltage, perhaps only 200v maximum. Remember it is basically a 12v system being switched.
An ignition system running at 100 Hz cannot really be compared with a 20KHz high voltage switching power supply. Even if the mJ losses per cycle were identical, the total switching loss would be twenty times higher at the higher operating frequency.
zacky, wow that is a lot of power. The average dc input current may be around 50A, and peak current in the switching device four times that, say 200A. Probably closer to 250A in a real supply (worst case).
I think the way I would do it would be to use four separate flyback supplies phased ninety degrees apart. Each would then only see 60 Amps peak current. They could run from a master oscillator and some flip flops to frequency lock the whole thing.
If each flyback supply runs at 20Khz, input and output ripple frequencies would then be 80Khz, but of much lower amplitude than one huge single flyback supply. That will be far easier to filter, and capacitor size, cost, and ripple rating will be much reduced.
Design of the magnetics will be less of a problem too. Skin effect means you cannot just make things bigger, and 250 peak amps is still 250 peak amps. But four independent foil wound flyback transformers (chokes actually) that only see 60 primary amps peak each, is doable. Current sharing between the four supplies will be automatic.
Conduction losses of higher voltage FETs will be a problem for you, but parallel FETs and IGBTs would be one possible way to overcome that. An interesting project to be sure.
Don't overlook the advantages of the diagonal half bridge flyback topology. For reasonably high input voltage, and high power it can be far more reliable. The peak drain voltage on the switching FETs will be hard clamped to the maximum dc input voltage, and that solves a whole lot of problems. The disadvantage is twice the conduction loss, but I think it well worth it for relatively high input voltage.
Ah, this all stirs memories. I used to do a lot of this sort of thing once. Now, being retired, it all seems so very long ago.
So warp... You just run them all as independent supplies each putting out 250VDC but phase locked?
You say that sharing is automatic...
This is because each one will be at it's output voltage peak at a different time?
You don't need any diode isolation between them? Or is it a give by their output rectifiers?
You say that sharing is automatic...
This is because each one will be at it's output voltage peak at a different time?
You don't need any diode isolation between them? Or is it a give by their output rectifiers?
One way to do it would be to use a standard dual output push pull PWM control chip, each output could drive a separate independent flyback supply. These would operate alternately, with each having an "on" duty cycle variable from zero to half the total period. They would then operate 180 degrees out of phase if you like.
So while one is charging up, the other is dumping into the output capacitor. They will load share exactly because the conduction times will be identical, and the load dumps will be into the same output capacitor.
This whole thing could be duplicated to make a total of four independent supplies all operating in phase quadrature. Each of the two PWM control chips need to be synchronised to a master oscillator so they will run ninety degrees out of phase. That will then give you 0, 90, 180, and 270 degree outputs.
The concept can be expanded to have any number of phases. But the beauty of it is that the total supply and load currents become much smoother. Provided all the "on" times are similar, all the supplies will load share exactly. Getting similar "on" times is just a case of paralleling the control voltages to the PWM chips.
There may also be a certain amount of redundancy if one circuit fails.
So while one is charging up, the other is dumping into the output capacitor. They will load share exactly because the conduction times will be identical, and the load dumps will be into the same output capacitor.
This whole thing could be duplicated to make a total of four independent supplies all operating in phase quadrature. Each of the two PWM control chips need to be synchronised to a master oscillator so they will run ninety degrees out of phase. That will then give you 0, 90, 180, and 270 degree outputs.
The concept can be expanded to have any number of phases. But the beauty of it is that the total supply and load currents become much smoother. Provided all the "on" times are similar, all the supplies will load share exactly. Getting similar "on" times is just a case of paralleling the control voltages to the PWM chips.
There may also be a certain amount of redundancy if one circuit fails.
That is sweet! I like it. You could save bucks and size of final output caps and get faster response, current pulse size would drop a lot generating far less EMI and reducing ripple currents. You essentially get to build one smaller cheaper unit, debug it without as much fanfare, then step and repeat. Very nice.
Where does power factor correction come into all of this?
Or are the flyback converters just operating off the raw 120Hz bridge rectified AC as a sort of do it yourself pf correction circuit?
This sort of power level is way out of my league,so I've just got to ask
(curiosity killed the cat etc).
Or are the flyback converters just operating off the raw 120Hz bridge rectified AC as a sort of do it yourself pf correction circuit?
This sort of power level is way out of my league,so I've just got to ask
(curiosity killed the cat etc).zacky; your "bulky" commutation components could be as simple as a resistor/cap combination, possibly in parallel
with a diode/zener pair or MOV. The higher the switching
frequency, the smaller the surrounding components will be,
but the semiconductors must be up-rated accordingly.
smoked; current sharing between paralleled supplies is NOT
automatic. Some designs have in-built sensor loops to allow
feedback to paralleled supplies, and there are other schemes
to prevent current-hogging. It is almost never as simple as
hooking multiples and expecting the sources to be balanced,
whether analog or switching. (Analogs are more forgiving).
zeitghost; PFC and harmonic reduction is a whole 'nother
subject, and is of great concern in line-operated switchers.
Much of the same problems encountered with SCR/TRIAC input
drives and supplies are reflected in SMPS's - the conduction
period of the input device(s) are not generally the full
cycle, and the results are often ugly and hard to suppress.
<als>
with a diode/zener pair or MOV. The higher the switching
frequency, the smaller the surrounding components will be,
but the semiconductors must be up-rated accordingly.
smoked; current sharing between paralleled supplies is NOT
automatic. Some designs have in-built sensor loops to allow
feedback to paralleled supplies, and there are other schemes
to prevent current-hogging. It is almost never as simple as
hooking multiples and expecting the sources to be balanced,
whether analog or switching. (Analogs are more forgiving).
zeitghost; PFC and harmonic reduction is a whole 'nother
subject, and is of great concern in line-operated switchers.
Much of the same problems encountered with SCR/TRIAC input
drives and supplies are reflected in SMPS's - the conduction
period of the input device(s) are not generally the full
cycle, and the results are often ugly and hard to suppress.
<als>
- Thread starter
- #35
I'm a little curious why a conventional full or half bridge can't be employed for this application? The magnetics would be used far more effectively, and the magnetics are going to be expensive. Generally the higher you can push the frequency the better from the point of view minimising size and cost of the wound components, providing you can design a winding to suit. At the frequency you're proposing, use of one of the high saturation ferrites such as 3C85 should be possible which will also help keep the core size down. You could probably get up as high as 100kHz, but the winding design will become trickier. You're almost certainly going to be using tape windings at these frequencies and current ratings.
The suggestion to use multiple phase-locked modules is a very good one - well worth investigating further.
----------------------------------
Your body might be a temple. Mine is an amusement park...
The suggestion to use multiple phase-locked modules is a very good one - well worth investigating further.
----------------------------------
Scotty, what you say about the full bridge forward converter being the traditional route to very high power is true. Theoretically it makes the most efficient use of the magnetics, but it is not usually that simple to put into practice.
There are problems of flux doubling and staircase saturation to think about in the ungapped magnetics. Current mode control is not easy to apply because the input current is a squarewave, and if magnetising current is made deliberately low, it will almost be perfectly square. So protecting it from overload is not a simple task.
I have seen a lot of high powered forward converters run for hours and abused in all sorts of imaginative ways, and then spontaneously go *BANG* at quite low load, leaving a pile of smoking parts, with little clue as to why it actually failed.
Another difficulty is the requirement for an output choke/capacitor combination which makes designing the frequency/phase characteristics of the control loop a far a from trivial task.
On the other hand flyback supplies run in current mode are beautiful things and very easy to apply. Control loop design is simple, and current mode operation is inherently protective. The only real disadvantage of flyback is the horribly high peak cyclic current drawn from the source. But multiple phased flyback supplies can solve that particular disadvantage rather neatly.
The easiest way to get the phasing is from a Johnston (twisted ring) counter. Two flip flops give you four 90 degree phases, and three flip flops six sixty degree phases, and so on.
There are problems of flux doubling and staircase saturation to think about in the ungapped magnetics. Current mode control is not easy to apply because the input current is a squarewave, and if magnetising current is made deliberately low, it will almost be perfectly square. So protecting it from overload is not a simple task.
I have seen a lot of high powered forward converters run for hours and abused in all sorts of imaginative ways, and then spontaneously go *BANG* at quite low load, leaving a pile of smoking parts, with little clue as to why it actually failed.
Another difficulty is the requirement for an output choke/capacitor combination which makes designing the frequency/phase characteristics of the control loop a far a from trivial task.
On the other hand flyback supplies run in current mode are beautiful things and very easy to apply. Control loop design is simple, and current mode operation is inherently protective. The only real disadvantage of flyback is the horribly high peak cyclic current drawn from the source. But multiple phased flyback supplies can solve that particular disadvantage rather neatly.
The easiest way to get the phasing is from a Johnston (twisted ring) counter. Two flip flops give you four 90 degree phases, and three flip flops six sixty degree phases, and so on.
- Thread starter
- #38
I have not looked at all this for a very long time, but basically it depends upon the availability of the magnetics and the switching devices.
For 1.5Kw of transmitted power, that would be 75mJ per cycle at 20Khz. So you need to design a choke that will store 75mJ worth of energy, (assuming that the flyback supply will be discontinuous). That is rather a lot, but the larger U cores, or U and I cores should be well up to the job. Inverter grade ferrite being the material of choice.
Just design the number of turns and wire gauge as for a normal transformer using Faraday's laws, and then gap it to reduce the inductance down to the reqired inductance.
Foil windings may be best, because skin effect will limit you to about 2mm thickness of whatever you decide to use anyway.
The choice of switching device looks a lot brighter these days than it did 15 years ago. And the wider choice of control and driver chips makes a high powered flyback design more feasible too.
I would start off and do a trial design of either a 1Kw module (six required) or a 1.5Kw module (four required), or even a 3Kw module (two required) and see what sort of numbers come up. Then look at the costs and availability of parts, and then decide which way to go from there.
Probably the biggest headache will be getting ferrite sample cores to play with. Ordering several thousand is easy, getting a single sample pair is hell. All you need to do is get one module fully tested, and then you can scale up the whole thing by using as many modules as you need.
In the end it will take X number of ferrite cores, and Y number of power transistors. It really does not matter if you build one huge single flyback inverter, or several smaller modular ones. Going to a number of uniform smaller modules has a number of advantages.
For 1.5Kw of transmitted power, that would be 75mJ per cycle at 20Khz. So you need to design a choke that will store 75mJ worth of energy, (assuming that the flyback supply will be discontinuous). That is rather a lot, but the larger U cores, or U and I cores should be well up to the job. Inverter grade ferrite being the material of choice.
Just design the number of turns and wire gauge as for a normal transformer using Faraday's laws, and then gap it to reduce the inductance down to the reqired inductance.
Foil windings may be best, because skin effect will limit you to about 2mm thickness of whatever you decide to use anyway.
The choice of switching device looks a lot brighter these days than it did 15 years ago. And the wider choice of control and driver chips makes a high powered flyback design more feasible too.
I would start off and do a trial design of either a 1Kw module (six required) or a 1.5Kw module (four required), or even a 3Kw module (two required) and see what sort of numbers come up. Then look at the costs and availability of parts, and then decide which way to go from there.
Probably the biggest headache will be getting ferrite sample cores to play with. Ordering several thousand is easy, getting a single sample pair is hell. All you need to do is get one module fully tested, and then you can scale up the whole thing by using as many modules as you need.
In the end it will take X number of ferrite cores, and Y number of power transistors. It really does not matter if you build one huge single flyback inverter, or several smaller modular ones. Going to a number of uniform smaller modules has a number of advantages.
Warpspeed,
Excellent points you raise in your last couple of posts, a lot of real world experience in there which I suspect Zacky has yet to meet. Well worth a star. How long were you in the power converter design game for? Longer than I was, I suspect - I had a major career change about 8 years ago into an industry that wasn't leaving England's shores bound for China.
----------------------------------
Your body might be a temple. Mine is an amusement park...
Excellent points you raise in your last couple of posts, a lot of real world experience in there which I suspect Zacky has yet to meet. Well worth a star. How long were you in the power converter design game for? Longer than I was, I suspect - I had a major career change about 8 years ago into an industry that wasn't leaving England's shores bound for China.
brought a wry smile to my face. I have similar recollections!
----------------------------------
- Status
