I will use pg 5 to 9 of the following to calculate the sw loss , as you recomend....
http://www.ti.com/lit/ml/slup169/slup169.pdf
.....LTspice simulator is pretty true to life.....The guy that designed it, Mike Engelhardt, i am sure, would be personally interested if switching loss calculations using his simulator were very inaccurate.....LTspice is specifically for SMPS's.
Anyway, the switching losses are a lot more than your rule-of-thumb equation, as the link shows.
Also, in a flyback, worst case switching loss is at switch off, as you know, since the leakage stops the fet current building up quickly......the leakage acts like a snubber at turn on.
In the first simulation of a one-switch flyback, i messed up by using a fet with 600v breakdown, and i think it must have been simulating avalanche breakdown and giving high losses....woops, sorry.
http://www.ti.com/lit/ml/slup169/slup169.pdf
.....LTspice simulator is pretty true to life.....The guy that designed it, Mike Engelhardt, i am sure, would be personally interested if switching loss calculations using his simulator were very inaccurate.....LTspice is specifically for SMPS's.
Anyway, the switching losses are a lot more than your rule-of-thumb equation, as the link shows.
Also, in a flyback, worst case switching loss is at switch off, as you know, since the leakage stops the fet current building up quickly......the leakage acts like a snubber at turn on.
In the first simulation of a one-switch flyback, i messed up by using a fet with 600v breakdown, and i think it must have been simulating avalanche breakdown and giving high losses....woops, sorry.
Mag: woops sorry, my dumb assertion that your rule of thumb for calculating switching losses was inaccurate was actually me being dumb......you were in fact right.....
.....i now know this because i just finished calculating switching losses according to slup169 App note by ti.com.
-It took ages to do cuzz the phone kept ringing.
I knew there were two time intervals during transition where energy was dissipated.......what i stupidly did not realise was that the bit where the fet gate is being charged from V(GS)th to V(GS),miller (or vice versa for switch-off) occurs very very quickly and is associated with very little energy loss.
So you were right Mag, and i was totally wrong, my apologies.
One thing i did find though, in the 8 hours it took me to calculate switching loss, was that there are a LARGE amount of FET and circuital parameters which affect the switching loss.....many of which are not stated in the datasheet, or have a wide tolerance.
This now makes me feel that any calculated heatsink size should be doubled at least, to account for the worst case scenario.
One of the unknown parameters is the internal gate drive resistance, which isnt even stated in my datasheet......then theres the fet gate mesh resistance...again not stated.......then there's the VGS(TH) value itself....which is anywhere between 3 to 5V , and varies with temperature.
So switching loss calculations leave me feeling like i am walking on quick-sand.
Maybe this is why the world is going in for LLC resonant supplies?.....because they give you almost zero switching losses....and then you only have to account for the predictable conduction losses.
.....i now know this because i just finished calculating switching losses according to slup169 App note by ti.com.
-It took ages to do cuzz the phone kept ringing.
I knew there were two time intervals during transition where energy was dissipated.......what i stupidly did not realise was that the bit where the fet gate is being charged from V(GS)th to V(GS),miller (or vice versa for switch-off) occurs very very quickly and is associated with very little energy loss.
So you were right Mag, and i was totally wrong, my apologies.
One thing i did find though, in the 8 hours it took me to calculate switching loss, was that there are a LARGE amount of FET and circuital parameters which affect the switching loss.....many of which are not stated in the datasheet, or have a wide tolerance.
This now makes me feel that any calculated heatsink size should be doubled at least, to account for the worst case scenario.
One of the unknown parameters is the internal gate drive resistance, which isnt even stated in my datasheet......then theres the fet gate mesh resistance...again not stated.......then there's the VGS(TH) value itself....which is anywhere between 3 to 5V , and varies with temperature.
So switching loss calculations leave me feeling like i am walking on quick-sand.
Maybe this is why the world is going in for LLC resonant supplies?.....because they give you almost zero switching losses....and then you only have to account for the predictable conduction losses.
Last edited:
That's why a healthy gate drive circuit is always a good idea.
As for temperature... that's what real-world tests are for. In the time the computer finishes number crunching for such a large amount of variables, you can build and test the real thing, and then you won't even need the computer anymore.For me simulation remains a tool to be used to test whether you haven't made any dumb mistakes in your design - i'm all for prototyping and testing. Lots and lots of testing.
Btw, the reason i've given up on a wide range regulated SMPS for audio is because for some reason i had an oscillation at ~4kHz that just didn't want to go away. Everything was calculated right and the real-world parts verified the math and simulation results, there just was something wrong and i couldn't find it for the life of me. I'll come back to it sometime soon tho.
Don't bother too much with these dumb parameters like internal gate resitance, mesh resistance and the like.
There is an interesting article on infineon website called "mastering the art of slowness" or someting like that which makes a comparison of switching losses of different mosfets, with different gate resistors and the like.
I totally agree with th3 un1qu3, please turn off your simulator and turn on your soldering iron and start to build something. This is the only way to make an smps work properly.
If the simulator can predict everything about your circuit why should we do protototyping?
The simulator is good for feedback loops, for protection circuits and the like, forget about using spice to predict things like switching losses which are also sometimes difficult to measure also in the real life.
When your smps is exploded at least 3-4 (in the reality, not in simulation) due to wrong tests or wrong assumptions you will have learnt lot of thing that works in practice independently from what spice is telling you.
There is an interesting article on infineon website called "mastering the art of slowness" or someting like that which makes a comparison of switching losses of different mosfets, with different gate resistors and the like.
I totally agree with th3 un1qu3, please turn off your simulator and turn on your soldering iron and start to build something. This is the only way to make an smps work properly.
If the simulator can predict everything about your circuit why should we do protototyping?
The simulator is good for feedback loops, for protection circuits and the like, forget about using spice to predict things like switching losses which are also sometimes difficult to measure also in the real life.
When your smps is exploded at least 3-4 (in the reality, not in simulation) due to wrong tests or wrong assumptions you will have learnt lot of thing that works in practice independently from what spice is telling you.
http://www.powerint.com/sites/default/files/PDFFiles/rdr203.pdf
...page 26 shows that they never get more than 76% efficiency with their flyback.....whether 10% load or 100% load (100W).
So the 91% efficiency figure seems high?
(......interesting to note was the decrease in efficiency at high Vin...showing that the major loss is in switching loss.)
...page 26 shows that they never get more than 76% efficiency with their flyback.....whether 10% load or 100% load (100W).
So the 91% efficiency figure seems high?
(......interesting to note was the decrease in efficiency at high Vin...showing that the major loss is in switching loss.)
Btw, the reason i've given up on a wide range regulated SMPS for audio is because for some reason i had an oscillation at ~4kHz that just didn't want to go away. Everything was calculated right and the real-world parts verified the math and simulation results, there just was something wrong and i couldn't find it for the life of me. I'll come back to it sometime soon tho.
C input flyback or LC filtered forward converter?
In either case, you most likely forgot a zero in the feedback compensation. This is usually as simple as adding a resistor in series with the feedback compensation capacitor (which is usually wired to turn the error amplifier into an integrator, which has a constant phase shift of 90 degrees and half the phase margin of a proportional controller!). Forward is even worse because the LC filter presents a 180 degree phase shift. A lead-lag network can be used to compensate partially (i.e., in addition to the series resistor, you also add an R+C across the top voltage divider resistor, from the output down to the error amp, to "speed up" the feedback on fast changes).
Re: simulation losses- even with reasonable accuracy (RELTOL = 0.0001, which makes for very slow simulations!), switching losses, power input and output often have no correlation. SPICE does not conserve charge or current because of rounding errors, and the accumulation of these errors results in erroneous products (instantaneous power) and integrals (energy per edge, per cycle).
LTSpice uses MODELs for MOSFETs, so they do not include package parasitics at the very least. Possibly the models have been tuned for realistic spreading resistance, near-threshold capacitance, body diode, etc. characteristics, but on the other hand, almost all manufacturers distribute fairly involved SUBCKTs modeling their transistors. Sometimes package parasitics are included, often not.
Even if the models included parasitics, it's still up to the user to include realistic estimates of layout parasitics. Few even attempt crude versions, let alone representative models (there is software which takes a PCB layout, actually computes the electromagnetic fields around the traces and generates a SPICE model, but it's very expensive, and the generated models are probably overly elaborate for casual usage). Particularly at high switching speeds, a tight layout can actually cause destruction of the active devices!
Tim
LC filtered half bridge. Nope, hadn't forgotten anything and checked it seven times. It worked fine with the output caps hanging by wires outside and the controller ground brought with its own wire to the junction of the caps. Once i put everything inside the case all hell broke loose.
It was likely a ground loop somewhere, caused by my laziness to make a whole board from scratch and just hacking the crap out of ATX PSUs.
I see, though this is another reason why the great current mode flyback is really the SMPS for audio......because firstly the average power is low, so we are in the flyback domain, but , not only that, feedback compensation with a current mode flyback is fsar easier than with a bridge or forward with an LC in its output
The flyback of this post is a BCM flyback, and the control of such things is particularly easy
The flyback of this post is a BCM flyback, and the control of such things is particularly easy
Hi All,
yesterday I had some time to measure accurately my SMPS.
In attached you will find a pdf with all the measurement I have done.
The SMPS has been measured directly connected to the 230V mains
and connecting its output to an electronic load.
I think that the results are not bad at all, please comment.
ciao
-marco
yesterday I had some time to measure accurately my SMPS.
In attached you will find a pdf with all the measurement I have done.
The SMPS has been measured directly connected to the 230V mains
and connecting its output to an electronic load.
I think that the results are not bad at all, please comment.
ciao
-marco
Interestingly we see the miller plateau in the gate signal at switch on....but we do not see it at switch off....this suggests that the lower fet switches off after the upper fet switches off....so the upper fet takes all the switch-off switching loss......and that means the upper fet will be getting the hottest.
also, i see resonations in your fet current just after switch off......that will also be seen in the diode current........have you properly snubbed the output diodes?
What kind of fet transition times are you getting..(Vds at switching point)......it looks like around 200ns for low fet switch on.........so that sounds ok, i am sure you know if too fast then its said to be bad for emc.
The following offline flyback is only 80% efficient, so yours is doing very well indeed.
http://www.powerint.com/sites/default/files/PDFFiles/di134.pdf
also, i see resonations in your fet current just after switch off......that will also be seen in the diode current........have you properly snubbed the output diodes?
What kind of fet transition times are you getting..(Vds at switching point)......it looks like around 200ns for low fet switch on.........so that sounds ok, i am sure you know if too fast then its said to be bad for emc.
The following offline flyback is only 80% efficient, so yours is doing very well indeed.
http://www.powerint.com/sites/default/files/PDFFiles/di134.pdf
You see the Miller plateau at turn on because the gate turn on resistor is 100r while the turn off resistor is only 5r6. The miller plateau is also present at turn off but it is so short that with the timebase I have measured it you don't notice it.
The ringing on the curret is due to parasitic resonances, it is unlikely to be present also on the other side of the transformer.
Btw: this smps is for sure not the best one invented but you can not compare it to power intergration s**t....
Power integration is quite good for low power stuff with their tiny switch and link switch family where you don't need high efficiency but just a small and reliable supply.
Of course you can boost a smps desinged properly for 10w to 100w or more, but don't expect high efficiency from that stuff.
Did you notice which kind of rdson have their crappy integrated mosfets? We are speaking of at least 10x the rdson of the ones I used.
The ringing on the curret is due to parasitic resonances, it is unlikely to be present also on the other side of the transformer.
Btw: this smps is for sure not the best one invented but you can not compare it to power intergration s**t....
Power integration is quite good for low power stuff with their tiny switch and link switch family where you don't need high efficiency but just a small and reliable supply.
Of course you can boost a smps desinged properly for 10w to 100w or more, but don't expect high efficiency from that stuff.
Did you notice which kind of rdson have their crappy integrated mosfets? We are speaking of at least 10x the rdson of the ones I used.
The hard part about flybacks is damping those little resonances. You can burn a lot trying. When the transistor turns on, the secondary rings (parallel resonant, voltage ringing), the transistor sees series resonance (current ringing). When the transistor turns off, the opposite happens. Best you can do is put an R+C across each winding (in this case, secondary is shown), which burns power from the switching edge as well as the resonance. Alternate option is an L || R in series with either winding (thus dampening the series resonance when that side is conducting). You could put ferrite beads on the diodes, but don't be surprised if they get ridiculously hot!
When the transformer has low leakage inductance, the impedance of the ringing is very low, so your snubber must dissipate a lot of the switching edge; when the transformer has high leakage, the ringing is very powerful and dissipates a lot of power. No free lunch.
With some additional Cds to dampen the dV/dt, series supply inductance to dampen the dI/dt, and a snubber to absorb what remains, you could speed up switching and probably push it to 94% efficiency. Make it a quasi-resonant ("lossless") snubber and you might push that up to 97%. After that, we're talking incremental gains, and when you're down to saving a watt here and there, you pretty soon get to the point where your controller, feedback, auxiliary supplies, stuff like that all start to look big in comparison, way down in the diminishing returns. Too bad all that extra hardware also takes up a lot of space. Over 90% is a fine stopping point on the 500W level.
Tim
When the transformer has low leakage inductance, the impedance of the ringing is very low, so your snubber must dissipate a lot of the switching edge; when the transformer has high leakage, the ringing is very powerful and dissipates a lot of power. No free lunch.
With some additional Cds to dampen the dV/dt, series supply inductance to dampen the dI/dt, and a snubber to absorb what remains, you could speed up switching and probably push it to 94% efficiency. Make it a quasi-resonant ("lossless") snubber and you might push that up to 97%. After that, we're talking incremental gains, and when you're down to saving a watt here and there, you pretty soon get to the point where your controller, feedback, auxiliary supplies, stuff like that all start to look big in comparison, way down in the diminishing returns. Too bad all that extra hardware also takes up a lot of space. Over 90% is a fine stopping point on the 500W level.
Tim
Well,
I get a little bit worried when I look at the transient response.
The repsonse when the load is decreased, is OK.
The response when the load is applied needs some more investigation to find an explanation of the behaviour. The "thickness" of the trace usually means that there is an oscillation in the regulation loop.
But this might be coupled to the topology.
I get a little bit worried when I look at the transient response.
The repsonse when the load is decreased, is OK.
The response when the load is applied needs some more investigation to find an explanation of the behaviour. The "thickness" of the trace usually means that there is an oscillation in the regulation loop.
But this might be coupled to the topology.
I've seen that before, that thickness you speak of...one of the powerint.com puzzlers explains it......
i though it was a characteristic of voltage mode control but i must be wrong.
Anyway, i am sure Mag, that you know any current pscillation in the primary whatsoever, will be reflected to the secondary, and appear there
i though it was a characteristic of voltage mode control but i must be wrong.
Anyway, i am sure Mag, that you know any current pscillation in the primary whatsoever, will be reflected to the secondary, and appear there
Hi,
If you are into resonant converters for Supplying audio amplifiers then the attached two simulations of an LLC resonant converter are worth seeing.
(they are .txt, but if you change them to .asc you can run them in the free simulator from LTspice)
I dont know if you have noticed it, but virtually every single company that currently sells offline SMPS for audio such as Connexelectronic, hypex, abletek etc etc ....all do LLC resonant type SMPS like in the simulations i attach.
One is with vin = 181 VDC and the other at vin = 373VDC.
...these correspond to the high and low levels of the primary bus when a mains voltage doubler link is usable with a 90-265VAC smps, where the load is +/-40V 4A.
I am wondering why none of these manufacturers do QR flybacks?
-must be a reason
If you are into resonant converters for Supplying audio amplifiers then the attached two simulations of an LLC resonant converter are worth seeing.
(they are .txt, but if you change them to .asc you can run them in the free simulator from LTspice)
I dont know if you have noticed it, but virtually every single company that currently sells offline SMPS for audio such as Connexelectronic, hypex, abletek etc etc ....all do LLC resonant type SMPS like in the simulations i attach.
One is with vin = 181 VDC and the other at vin = 373VDC.
...these correspond to the high and low levels of the primary bus when a mains voltage doubler link is usable with a 90-265VAC smps, where the load is +/-40V 4A.
I am wondering why none of these manufacturers do QR flybacks?
-must be a reason
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.