RECOMENDATION for 40khz push-pull smps

[alfcoder]

hello!

i'm working on a 40khz push-pull converter, the primer is 12v and i have
3 outputs with minimum 5V at 5A each, the non loaded voltage can be
few volts higher, basically is is a non-regulated psu

i'm studied the great book: switching power supply design by abraham pressman, kaith billings, taylor morey

and i'm become very familiar with the concept and as well the equations
but after i found in my junk a toroid ferrite with a proper size (40mm diameter, 27mm inner diameter, 15mm height) and tested it, i found that
the core getting real hot, it is clearly not from the copper because i only
wound one portion of the toroid and my input current with only the primary
was 650ma, also the copper was cool, i was unable to measure the exact
temp but after a 10-20 minutes the core was so hot that i was unable to hold in my hand (i'm not an psu expert so maybe this temp is ok)

some information about the configuration and calculations, maybe i commited some error here:

my power rating is 75w output => the toroid need to be at least 40mm wide

the measured core al=755nH/N^2

magnetic path is 106mm

cross section area is 97mm^2

i let the flux swing to be +/- 1600 gauss so the calculated turn of the
primary is 2x4 turn

the primary wire diameter need to be at least 1.45mm diameter or i can use
5x 0.6mm diameter wires

the fixed pwm is 80%

so if my calculations are correct, then i can only think that i have some
real bad ferrite material, in this case i would like to know some suitable
ferrite types, if it is possible from the farnell distributor, i've seen they
have a very wide selection, but unfortunatly for me is very hard to find
that material, i mean to know for sure if one will be good in my application
or not and honestly i really don't want to buy 10 or more and try those one
by one

so please, experts, give me some recommended core

many many thanks

david

 

[sawreyrw]

Maybe the core is powdered iron.

 

[busan]

You need spread the windings on all toroid.

 

[savu]

to Alfcoder:

Try 3c90 or 3f3 they are good materials for smps. you can find them at farnell

regards,

 

[odnaizutra]

maybe you can try n27 .

 

[alfcoder]

thank you all, this is a very nice responsive forum

i ordered two toroid with the 3c90 material one is 42mm x 26mm x 13mm
and the other is 40mm x 24mm x 16mm, so in few days we will see

the claimed Al value is much higher then the current one (2690nH and 3500nH)

and my measured was only 755nH, so maybe this is the big difference

as with this full spread winding i'm still not very familiar with, simply because
it looks a bit strange to wind only 8 turn fully on a 4cm toroid and also as
i seen it not take longer path for the magnetic flux to travel then in the
E -core outside legs, but i will try this spreaded winding for sure

so thanks again

david

 

[savu]

thank you all, this is a very nice responsive forum

i ordered two toroid with the 3c90 material one is 42mm x 26mm x 13mm
and the other is 40mm x 24mm x 16mm, so in few days we will see

the claimed Al value is much higher then the current one (2690nH and 3500nH)

and my measured was only 755nH, so maybe this is the big difference

as with this full spread winding i'm still not very familiar with, simply because
it looks a bit strange to wind only 8 turn fully on a 4cm toroid and also as
i seen it not take longer path for the magnetic flux to travel then in the
E -core outside legs, but i will try this spreaded winding for sure

so thanks again

david

spread winding ensures low emi and low skin effect.

regards,

 

[Electrone]

I think you might be experiencing a DC bias on the core, walking it into saturation. One way to tell is if one MOSFET drain is experiencing a much greater turn-off spike than the other. In push pull designs, it is good to use current mode control because the transformer can't easily be capacitor coupled to its MOSFET drive.

 

[alfcoder]

I think you might be experiencing a DC bias on the core, walking it into saturation. One way to tell is if one MOSFET drain is experiencing a much greater turn-off spike than the other. In push pull designs, it is good to use current mode control because the transformer can't easily be capacitor coupled to its MOSFET drive.

yes i have very big drain spikes but they are equal in shape, once i have a
saturation when i was not driven the mosfets with the same square-wave
so i heard a ticking sound from the transformer and many times my 10A
limited psu switched off, but as soon as i fixed the square wave i not
noticed any strange things, so from my limited expertise on smps,
i can say i don't have any saturation, also the non loaded input current
was only 650mA, not a very wild thing

i think this current mode thingy is just complicating the design, also it need
some knowledge to implement it, so i rather use some half-bridge config if
i can not get rid off the spikes

i also ordered some 100v mosfets because the other was only 60v and it is
very on the edge because of these spikes, so now i will have 40v more
space, but i sure will do something with the spikes, but it is very hard
because, it doesn't matter how big capacitor i'm using with whatever
resistor for the snubber, when i'm applying the load the big spike is come
along

so we will see

 

[alfcoder]

spread winding ensures low emi and low skin effect.

regards,

what we should have is a coated copper ribbon for the winding, in an ideal
world

isn't this skin effect is only affected by the frequency? maybe you meant
some proximity effect when on conductor is affecting the neighbor conductor
if they are in close proximity

 

[mag]

Hi Alfcoder,
with Al=755nh/t you will have only 12uH with 4 turns.
At 40khz the magnetizing current will be high just for nothing.
160mT at that frequency is a reasonable choice but if your core is an iron
powder type or other distributed gap cored like kool-mu (and I think it is due to the low AL value) the core losses are much higher than plain ferrite as
3C90, N87 and similar materials. The core losses makes the core heat up.

Try this test: let the psu working at no load and check if the core heats up, if it heats up it is for sure core losses due to the wrong ferrite material.

How did you wind your transformer? In push-pull converters you need to keep the primary to primary leakage inductance as low as possible to reduce spikes on mosfets.

If you need 4+4 turns with 5x0.6mm wires you have to wind 4 turns of
10x0.6mm wires in parallel and then connect the wires ends in the proper way to build your transformer (bifilar winding). The same applies for the secondary side, otherwise you will have huge spikes on the output diodes.
Spread the primary and secondary winding over the full toroid to reduce primary to secondary leakage inductance.

BTW: 40mm toroid for only 75W is huge, you can probably put a kW out of it. Keep in mind that a very big core has also bigger core losses at a given frequency and Bmax, so using a core much bigger than what is really needed is just a waste of power and efficiency.

Probalby your brigde is unbalanced but the core is an iron powder type, so it can sustain a small DC bias without saturating.

You can use also an ETD, RM, E shape core, much easy to wind and much easier to order. You can also introduce a small gap in those core if it is needed for curing some small imbalance.

The snubber resistor (it is usually better to put 2 RC, one for each primary half) does quite nothing on the spike, it is there to damp the ringing at turn-off.

When you say 40kHz is your oscillator running at 40kHz or are your mosfets operating at 40kHz? Controllers like SG3525 and similars divide the oscillator frequency by 2. Check your mosfet on-time it should be 1/(2*fosc)*0.8.
If your fosc=40kHz your on time will be 10us --> B=306mT NOT OK!
If your fosc=80kHz your on time will be 5us --> B=153mT OK!

please check this

ciao

-marco

 

[Electrone]

I recall a particular push pull circuit that I was playing around with that also suffered from a hot core. You might consider adding more turns to each winding. I can't recall if I had time to try anything like that.

 

[alfcoder]

Hi Alfcoder,
with Al=755nh/t you will have only 12uH with 4 turns.
At 40khz the magnetizing current will be high just for nothing.
160mT at that frequency is a reasonable choice but if your core is an iron
powder type or other distributed gap cored like kool-mu (and I think it is due to the low AL value) the core losses are much higher than plain ferrite as
3C90, N87 and similar materials. The core losses makes the core heat up.

Try this test: let the psu working at no load and check if the core heats up, if it heats up it is for sure core losses due to the wrong ferrite material.

How did you wind your transformer? In push-pull converters you need to keep the primary to primary leakage inductance as low as possible to reduce spikes on mosfets.

If you need 4+4 turns with 5x0.6mm wires you have to wind 4 turns of
10x0.6mm wires in parallel and then connect the wires ends in the proper way to build your transformer (bifilar winding). The same applies for the secondary side, otherwise you will have huge spikes on the output diodes.
Spread the primary and secondary winding over the full toroid to reduce primary to secondary leakage inductance.

BTW: 40mm toroid for only 75W is huge, you can probably put a kW out of it. Keep in mind that a very big core has also bigger core losses at a given frequency and Bmax, so using a core much bigger than what is really needed is just a waste of power and efficiency.

Probalby your brigde is unbalanced but the core is an iron powder type, so it can sustain a small DC bias without saturating.

You can use also an ETD, RM, E shape core, much easy to wind and much easier to order. You can also introduce a small gap in those core if it is needed for curing some small imbalance.

The snubber resistor (it is usually better to put 2 RC, one for each primary half) does quite nothing on the spike, it is there to damp the ringing at turn-off.

When you say 40kHz is your oscillator running at 40kHz or are your mosfets operating at 40kHz? Controllers like SG3525 and similars divide the oscillator frequency by 2. Check your mosfet on-time it should be 1/(2*fosc)*0.8.
If your fosc=40kHz your on time will be 10us --> B=306mT NOT OK!
If your fosc=80kHz your on time will be 5us --> B=153mT OK!

please check this

ciao

-marco

my ton_max is 10us, i'm not using any ic, but i controlling this from
software, also i measured this, i think with ton 10us you will get 308mT,
but this value is a delta, so it means that we go to plus 154mT and
negative 154mT so we are in the recommended 320mT delta limit, but
if i'm misunderstand something here please let me be corrected

the serial connected r-c passive dissipative snubber is sure damping the
oscillation, but the begin of this oscillation is the big spike itself, so you
can damp that also, in my experiment i was a bit trouble with it, but at
least the spike amplitude was less and less as i increased the capacitor
(i would like to point here that we need to tune the resistor for the rc
subber, i done it with a watt rated potenciometer and i clearly seen
how one point was optimal and the next one is again not, i guess i
overcompensated it)

i'm really not an expert on the smps, i only read that up to 200w you need
to use 40mm diameter toroid, and because there was no other step where
my 75w is fitted more, and also i imagined my 4 double winding needs
i was sure that this diameter will be fine for me

ok now my fine result:

the new cores are arrived from farnell and i wound one with this special
winding process, just for sure and for my biggest surprise the big spike
and the oscillations dissapered, i can only see a nice square wave on the
drains, sure the gate resistors are set for the not sharp turn-on, also
my idle current went down to 110ma from 650ma, and the core is at room
temperature

i guess this 110ma idle is the price i paying for the "big" toroid

so the lesson to learn here is never ever try to use a toroid from the junk

i will put some report here later on when i'm moving forward with my psu

and many many thanks you guys, for helping me on this new field

david

 

[alfcoder]

hi all!

after my fight with this strong push-pull drain pulse ended with my give up
i turned to the half-bridge configuration, first i considered myself lucky
because i found a very compact half-bridge driver ic from the irf: IRS2153D
this one is only need one timing capacitor (10n) and a potenciometer (2.2k)
to function as a 50% half-bridge driver, sure the output current is way too
low, at least what i use to use, but if one is careful with the mosfets then
this config should work very well, so i took my professional ferrite toroid
and my fresh transformer winding skill and made one converter to see how
well it is working and to measure some parameter, below you can see the
result, the output was rectified with only one way (one diode) because
i don't want to use any splitted winding anymore also i use this trick
with the 50hz transformers, so the transformer can rest in half cycle

as we can see from the table, the result is very wrong, low efficiency
the loaded voltage is not 6v, the frequency is more like 20khz then 40khz
so i'm wonder why it is so bad, what i'm doing wrong?

if any expert out there should point me some good direction it would be
very help-full i don't say it is not fun to try out 10-20 more config
because i kinda like it, but maybe i can use my time on some more
important tasks then this

my theory about the result is that because in the half-bridge configuration
we only use the half of the total input supply this lead to the wrong
performance because in this case the primer voltage is only 6V and we
know that more the primer voltage, better the efficiency

if this config will also fail i will move to the full-bridge config, but then it
will be a very funny design because i only need 75w but at this point
i really don't care

many thanks

david

An externally hosted image should be here but it was not working when we last tested it.

 

[alfcoder]

after a few configuration i was able to found one for my requirements:

the toroid core was:
FERROXCUBE - TX40/24/16-3C90 - FERRITE, TOROID, 40MM, 3C90

Ferrite Grade:3C90; Ae Effective Cross Section Area:125mm²; Effective Magnetic Path Length:96.3mm; External Diameter:40.25mm; External Length / Height:16.4mm; Inductance Factor, Al:3500nH

the output was rectified in two-way, for this i turned the output coil twice

primary turn: 3
secondary turn: 2x4

at minimum load:

in: 12v@58ma
out: 7.248v@36ma


at maximum load:
in: 12v@3.856a
out: 6v@6a

eff: 77.8%

the frequency was tuned to 25.8khz


this is a very good result, and a very good hint that yes we can design
with a very few external component a switch-mode power supply
(the only important component is the half-bridge driver IRS2153D)

and yes, the "simple" push-pull configuration is a very bad idea

this unregulated nature of the psu in many-many cases is not a problem
(in my case the loaded and the non loaded voltage difference is only 1.2v!)


but now my big question

how can i improve this psu???

the goal is the high efficiency, smaller voltage drop, maybe higher frequency

right now i'm thinking about to use a smaller toroid, and also to try out
some other turn rates

so if you know any improvements, please tell me/us


ps. i feel that in the end we will have a very good and very easy building
block in our hands

 

[Slobban]

why did you choose a half bridge (or initially push pull) for this type of design? I mean with this low input voltage you will not have to worry about avalanching semiconductors. I would guess that a flyback design would be cheaper and simpler, however a single switch forward would proberly give you better performance.

for the half bridge to have the same performance on the transistors (if we are talkning about on-losses) they would need to be way better, than in a single switch case, since they carry half the voltage, and twice the current (P = I^2 * R). I think that you should atleast investigate moving to a single switch forward. ofcource a 2 switch forward is not bad either, the transformer will be simpler, but for these voltage level its not really justified (the high side driver is hard to supply since the transformer is in series, a bootstrap is not always guaranteed to work).


what kind of caps are you using for the halfbridge? if you are using good polyprophylene types than this would not be noticable but if you are using electrolythics, then you might be loosing a substaincial amount in them (bad ESR).

another thing you could investigate on you halfbridge, if not done already, is to tweak gate resistors. if you switch way to fast or way to slow it would show up in the efficiency measurement. faster diodes or slow down switching could improve efficiency a tad (but we probably dont talk about several percent). or you may be able to increase the switching speed with your current diodes, and gain efficiency?

//Slobban

 

[alfcoder]

why did you choose a half bridge (or initially push pull) for this type of design? I mean with this low input voltage you will not have to worry about avalanching semiconductors. I would guess that a flyback design would be cheaper and simpler, however a single switch forward would proberly give you better performance.

for the half bridge to have the same performance on the transistors (if we are talkning about on-losses) they would need to be way better, than in a single switch case, since they carry half the voltage, and twice the current (P = I^2 * R). I think that you should atleast investigate moving to a single switch forward. ofcource a 2 switch forward is not bad either, the transformer will be simpler, but for these voltage level its not really justified (the high side driver is hard to supply since the transformer is in series, a bootstrap is not always guaranteed to work).


what kind of caps are you using for the halfbridge? if you are using good polyprophylene types than this would not be noticable but if you are using electrolythics, then you might be loosing a substaincial amount in them (bad ESR).

another thing you could investigate on you halfbridge, if not done already, is to tweak gate resistors. if you switch way to fast or way to slow it would show up in the efficiency measurement. faster diodes or slow down switching could improve efficiency a tad (but we probably dont talk about several percent). or you may be able to increase the switching speed with your current diodes, and gain efficiency?

//Slobban

yes it is not easy to select a proper converter topology, the fly-back is
problematic because the high current and also because it is working like
some king of current mode, so if i not put any load to it then i will have
real high voltages or a burnt mosfet, the forward is only acceptable if it is
driven with a dual mosfets but yes, the upper is hard and complicated to
drive, the push-pull is real ugly because of the drain spikes, so the only
converter left is the half-bridge or the full-bridge, if we consider that very
easy irf half-bridge driver ic, i mentioned before, then simply because of
this, the half-bridge is the winner, also the easy spike canceling makes it
ideal to use, also that i think we really need to rectifier the output in two
way mode, when i wanted to do it in only one way i got very bad results

i agree every converter has a problem, in half bridge the half of the supply
and the double current, as you pointed out in a very cleaver way, and sure
the supply splitter capacitors need to be a low esr type, simply because in
one cycle it got the current from the supply then it give it out to the coil

today i did some experiment and now i'm at the 80% region and the
switching frequency was 35khz with an etd 29x16x10 core, the material
is unknown and the loaded and non-loaded voltage difference was 1.2v

the drain spikes was filtered with a simple tvs diode and the core was only
a bit warm after the very long 25w usage

so these results are not so bad, but still i feel that something can be done
more, like for example why i got this 1.2v drop, why it is not only 0.5v?
(yes i know the supply is unregulated, but still), then why the frequency
always so low, why it can not be say 80khz? is it real hard to go for 90%
efficiency?

so more experiment will come

 

[Slobban]

for efficiency resons having a low switching frequency is beneficial. in most hardswitching application the switching losses are usually dominating the losses, meaning increased frequency-> increased losses. so if efficiency is the target, then you should try to have as low frequency as acceptable, ofcource the penelty is size on magnetics and caps.

if you need to find the voltage drop, you have to measure to find it. I mean it has to be either the primary switches, transformer, sec diodes or inductor.

are you running the output inductor in continous mode? If it's not regulated, then I would guess you have to, right? when using a coil in continous mode without regulation you would experience some low frequency ringing during load steps. you would also experience some slight variations in the output voltage.

the continous conduction mode is an equlibrium between inductor charge and inductor discharge. for the system to be able to increase the current, the output voltage must drop momentarily to be able to charge the inductor with higher voltage (thus the low freqeuncy ringing). in the end you will end up with a voltage slightly less than before the load step. this is partly because of the nature of CCM, but mostly inductor parasitics, like increased core loss, increased copper loss.

increasing the inductance is one way to get it better, but you would also need to increase the cross section area of the core, and the wire size (it would also show up on the efficiency). but a non regulated system will never be as good as a regulated one.

1,3v voltage drop at 6A, this gives the system an output inpedance of ~200mOhm.

I would say this is the wire resistance in transformer, output coil and RdsOn combined with switching losses (caused by a finit switching speed) that causes the system to have this impedance.

//Slobban

 

[mag]

Hi Alfcoder,
why did you choose the half bridge topology?
It is not suitable at all at low voltage input.
Al the current in the transformer primary must flow trough the primary
electrolytics; with 75W 12V input voltage I assume that the transformer primary current will be something around 10A. Check is your input electrolytics
can sustain that ripple current otherwise they can vent or be overheated and damaged.
80% efficiency with this design I think is quite good, don't expect much more from a half bridge at 12V. You will have a lot of losses in the mosfets, transformer primary resistance and primary electrolytics (ESR) due to the current doubling.

Push-pull topology is the best suited for low voltage, moderate to high power operation. With push-pull you will have half of the current in the transformer and mosfet compared to half bridge. The main problem of push-pull is the voltage doubling on the primary mosfets and the difficulty to clamp the overvoltage spike.
In any case your input is 12V so your mosfets must be able to sustain at least 24V+overvoltage spike. Normally 55V-60V mosfets are used at 12V to have some safety margin.
Don't care too much to that spike, with the RC snubber you can damp it but at the expence of efficiency (big cap --> big losses in the resistor). Tune your snubber just to remove the ringing and keep the spike.
If you select 55V mosfet and the spike is higher than this value it means that your transformer is wound in a really bad way (high leakage) and to clamp the spike you need a very heavy snubber.

Also a flyback can be adeguate at those power level. It is true that at no load the output tend to rise when not regulated, but winding a low leakage transformer and regulating on the primary side the output voltage can be mainained quite constant. Check for example the FAN400 datasheet.

ciao

-marco