Maker Pro
Maker Pro

Simple half-bridge switcher 25VDC-75VAC 20kHz+ 90 to 500+ watts

P

P E Schoen

I have built a prototype of a half-bridge switching supply using a
PIC16F1825, an IRS2001 driver, and a transformer that I wound using the
core, bobbin, and wire from a computer power supply probably about 500
watts. The core is 47x47x12mm and N27 ferrite. Here is an image of the
LTSpice simulation and output waveform:

http://enginuitysystems.com/pix/25VDC-75VAC_25kHz_HalfBridge_1Cap.png

And here is a scope shot of the output with 25 VDC and 3.51 amps input, on a
50 ohm WW power resistor. The output voltage reads 67.9 VRMS on my Fluke 45,
so that comes to 87.75W input and 90.6W output.

http://enginuitysystems.com/pix/Scope_0727_800x600.png

Of course that can't be right, and I know that some components are getting
hot, like the current measuring 0.1 ohm resistor, the 180uF bus capacitor,
and the MOSFETs to some degree. But the transformer core and windings did
not get even barely warm. This is about the most I can get from my lab
supply, so to test it further I will need to hook it up to a 48 VDC supply
(it's intended for 4x12V SLA batteries), and I hope to be able to connect a
FWB and capacitor to the output to get about 250 VDC.

The core is similar to an ETD49-25-16:
http://www.epcos.com/inf/80/db/fer_07/etd_49_25_16.pdf

I'm not sure if this core will be sufficient to reach my goal of 500 watts
or more, but I think it is pretty close. Actually I have purchased some
larger cores and bobbins that I think should be able to handle about 1000
watts:
http://www.epcos.com/inf/80/db/fer_07/e_55_28_21.pdf

My next steps will be to see how this transformer works on 48VDC, open
circuit to start with, and then under load. I expect about 150 VAC or 300
Vp-p, so I would need a doubler to get the 250-300 VDC I want. But I could
also rewind the transformer to get twice the output voltage, or I could use
two of these with outputs in series. And I also want to try upping the
frequency, which should get me higher voltage and more power. But the N27 is
characterized at 25kHz, while N87 is at 100kHz.

I had also tried simulations of topologies that used two capacitors across
the bus with the transformer primary from the half-bridge to the center tap,
and another that used three capacitors, with what seems to be a superfluous
capacitor in series with the transformer. But that may be for a resonant
design. The two capacitors across the supply may allow lower voltage ratings
to be used, and also would supply high current high frequency supply that is
now causing high ripple and losses in the electrolytic. It is a high ripple
current low ESR type, but certainly not adequate for the higher power I
want.

I have several 20 uF 100 VAC polypropylene capacitors I got on eBay, and
they seem to have excellent characteristics. I may order more while they are
still available for just a few dollars each. Similar capacitors are $25 each
from Mouser:
http://www.mouser.com/ProductDetail...=sGAEpiMZZMv1cc3ydrPrF7PI1Y6hbN/KixaOZp/5BgY=

BTW, Tim Williams, if you are reading this, I have ordered your book from
Amazon:
http://www.amazon.com/gp/product/075061756X/ref=oh_details_o02_s00_i00?ie=UTF8&psc=1

TIA for any comments and suggestions,

Paul
 
K

Klaus Kragelund

I have built a prototype of a half-bridge switching supply using a

PIC16F1825, an IRS2001 driver, and a transformer that I wound using the

core, bobbin, and wire from a computer power supply probably about 500

watts. The core is 47x47x12mm and N27 ferrite. Here is an image of the

LTSpice simulation and output waveform:



http://enginuitysystems.com/pix/25VDC-75VAC_25kHz_HalfBridge_1Cap.png



And here is a scope shot of the output with 25 VDC and 3.51 amps input, on a

50 ohm WW power resistor. The output voltage reads 67.9 VRMS on my Fluke 45,

so that comes to 87.75W input and 90.6W output.



http://enginuitysystems.com/pix/Scope_0727_800x600.png



Of course that can't be right, and I know that some components are getting

hot, like the current measuring 0.1 ohm resistor, the 180uF bus capacitor,

and the MOSFETs to some degree. But the transformer core and windings did

not get even barely warm. This is about the most I can get from my lab

supply, so to test it further I will need to hook it up to a 48 VDC supply

(it's intended for 4x12V SLA batteries), and I hope to be able to connecta

FWB and capacitor to the output to get about 250 VDC.



The core is similar to an ETD49-25-16:

http://www.epcos.com/inf/80/db/fer_07/etd_49_25_16.pdf



I'm not sure if this core will be sufficient to reach my goal of 500 watts

or more, but I think it is pretty close. Actually I have purchased some

larger cores and bobbins that I think should be able to handle about 1000

watts:

http://www.epcos.com/inf/80/db/fer_07/e_55_28_21.pdf



My next steps will be to see how this transformer works on 48VDC, open

circuit to start with, and then under load. I expect about 150 VAC or 300

Vp-p, so I would need a doubler to get the 250-300 VDC I want. But I could

also rewind the transformer to get twice the output voltage, or I could use

two of these with outputs in series. And I also want to try upping the

frequency, which should get me higher voltage and more power. But the N27is

characterized at 25kHz, while N87 is at 100kHz.



I had also tried simulations of topologies that used two capacitors across

the bus with the transformer primary from the half-bridge to the center tap,

and another that used three capacitors, with what seems to be a superfluous

capacitor in series with the transformer. But that may be for a resonant

design. The two capacitors across the supply may allow lower voltage ratings

to be used, and also would supply high current high frequency supply thatis

now causing high ripple and losses in the electrolytic. It is a high ripple

current low ESR type, but certainly not adequate for the higher power I

want.

The version with 3 caps, two on the supply and one in series with the tap, is less sensitive to voltage variations on the supply. (if you run with 50%duty cycle PWM). The single capacitor solution can lead to stair case saturation of the transformer, since the transient response steady state takes many cycles to complete and the transformer may be saturated before it settles

Cheers

Klaus
 
T

Tim Williams

P E Schoen said:
I have built a prototype of a half-bridge switching supply using a
PIC16F1825, an IRS2001 driver, and a transformer that I wound using the
core, bobbin, and wire from a computer power supply probably about 500
watts. The core is 47x47x12mm and N27 ferrite. Here is an image of the
LTSpice simulation and output waveform:

http://enginuitysystems.com/pix/25VDC-75VAC_25kHz_HalfBridge_1Cap.png

And here is a scope shot of the output with 25 VDC and 3.51 amps input,
on a 50 ohm WW power resistor. The output voltage reads 67.9 VRMS on my
Fluke 45, so that comes to 87.75W input and 90.6W output.

Assuming the resistor is noninductive, which it isn't -- they're usually
dropping off in the low MHz, which would lop off a few harmonics worth of
power, might be enough to account for it.

If you have a current transformer or probe, you can set the scope to
multiply volts and amps, and measure the average of the math signal.
Beware of delay or phase shift on the probes though: not so big a deal for
a resistor, but very important around reactive loads (e.g., measuring the
real power in a resonant tank, where the phase shift is already very
nearly 90 degrees; an extra degree and it can read negative!).
http://enginuitysystems.com/pix/Scope_0727_800x600.png

Of course that can't be right, and I know that some components are
getting hot, like the current measuring 0.1 ohm resistor, the 180uF bus
capacitor, and the MOSFETs to some degree. But the transformer core and
windings did not get even barely warm. This is about the most I can get
from my lab supply, so to test it further I will need to hook it up to a
48 VDC supply (it's intended for 4x12V SLA batteries), and I hope to be
able to connect a FWB and capacitor to the output to get about 250 VDC.

Umm... FWB and *inductor* and capacitor... right?
The core is similar to an ETD49-25-16:
http://www.epcos.com/inf/80/db/fer_07/etd_49_25_16.pdf

I'm not sure if this core will be sufficient to reach my goal of 500
watts or more, but I think it is pretty close. Actually I have purchased
some larger cores and bobbins that I think should be able to handle about
1000 watts:
http://www.epcos.com/inf/80/db/fer_07/e_55_28_21.pdf

If the wire fits, you're already okay: it doesn't appear to be saturating,
so you can run this frequency and voltage all day long (on however many
turns are on there). You just need enough wire to handle the current
without melting.
My next steps will be to see how this transformer works on 48VDC, open
circuit to start with, and then under load. I expect about 150 VAC or 300
Vp-p, so I would need a doubler to get the 250-300 VDC I want.

You certainly don't want to put a cap-input doubler on there, but there is
such a thing as a choke-input doubler. You need four diodes, two chokes
and two caps. You already lose half Vpp in rectification (half wave rect.
each side, one positive, one negative), so it doesn't get you anything
over a FWB.

Alternately, you can put the choke on the primary side, but this is more
difficult to arrange on a half bridge. Full bridge (current sourced
inverter) and PP (ala Royer oscillator) are in semi-common use. But then
you need an extra front end chopper to generate the constant current.

Beware also, as turns go up, so does parasitic capacitance, and thus
turn-on transients (primary current spikes, secondary voltage spikes).
This can be improved with winding design (optimal sectioning), and
addressed with snubbers (but not really fixed).
But I could also rewind the transformer to get twice the output voltage,
or I could use two of these with outputs in series. And I also want to
try upping the frequency, which should get me higher voltage and more
power. But the N27 is characterized at 25kHz, while N87 is at 100kHz.

N27 is a lowish frequency material, so just keep the Bmax lower than for
N87. I would think in that size, 0.2T would be fine for N87, maybe 0.1T
for N27, should run plenty cool. Notice BTW that means doubling the
number of turns.
I had also tried simulations of topologies that used two capacitors
across the bus with the transformer primary from the half-bridge to the
center tap,

This is the most common, and the most effective.

Actually that's a lie. For a decade or two, the three-cap layout
dominated AT and ATX power supplies. And by quantity, that's up there in
the millions. But the reason is, the filter caps were already split
because of voltage doubling (the 120/240V switch wires the circuit for
full wave doubling at 120, or regular FWB at 240); for the purposes of the
circuit, it's a +/-160V supply, so they only needed a 1uF (typically) to
"ground" (the center tap between them).

But when the supply isn't already split, you generally want to split it
with equal capacitance above and below.
I have several 20 uF 100 VAC polypropylene capacitors I got on eBay, and
they seem to have excellent characteristics. I may order more while they
are still available for just a few dollars each. Similar capacitors are
$25 each from Mouser:
http://www.mouser.com/ProductDetail...=sGAEpiMZZMv1cc3ydrPrF7PI1Y6hbN/KixaOZp/5BgY=

Ah, 935Cs, good stuff, no problems there. Do mind the stray inductance:
they're simply long, so there's nothing you can do about the first
10-20nH. Which can be advantageous, but important to keep in mind
regardless.
BTW, Tim Williams, if you are reading this, I have ordered your book from
Amazon:
http://www.amazon.com/gp/product/075061756X/ref=oh_details_o02_s00_i00?ie=UTF8&psc=1

I don't know if that particular Tim reads this newsgroup, but if he did
I'm sure he'd appreciate it. ;)

Umm... this:
http://webpages.charter.net/dawill/tmoranwms/TimWilliamses.html
is rather old, so I'm sure half the links have died, and all the search
rankings have changed. But kind of amusing nonetheless.

We're no John Smith, but it's still a pretty common name. Not sure if
there's any correlation to electronics, but it seems to me I've already
met a Jim Williams (service engineer, retired), not the more famous (and
recently passed) one; a few other Tims (not to mention the other Tim W.
here), and a few other Williamses.

OTOH, Paul is a pretty common name, but I would guess there aren't many
Schoens outside Europe. :)

Tim
 
P

P E Schoen

"Tim Williams" wrote in message
Assuming the resistor is noninductive, which it isn't -- they're usually
dropping off in the low MHz, which would lop off a few harmonics worth of
power, might be enough to account for it.

It very well could be reactive at 20kHz, so the V*A would be more than the
actual power. I just measured it to be 114.9 uH at 10kHz, or about 15 ohms
reactive at 20kHz. That makes it about 52 ohms impedance. Thus the output is
about 88.3 watts but that is still 100.6% efficiency. The simulation shows
about 90%. From the heat I detect I would estimate about 5 watts of losses
or about 94%. The best measure of efficiency will be when I add a rectifier
bridge, capacitor, and inductor. I may need to use SiC Shottky diodes like
these:
http://www.mouser.com/ProductDetail...=sGAEpiMZZMtQ8nqTKtFS/Cwtife2N73ILNibCxuSGE0=
If you have a current transformer or probe, you can set the scope to
multiply volts and amps, and measure the average of the math signal.
Beware of delay or phase shift on the probes though: not so big a deal
for a resistor, but very important around reactive loads (e.g.,
measuring the real power in a resonant tank, where the phase shift
is already very nearly 90 degrees; an extra degree and it can read
negative!).

I don't have such a probe and my scope does not have those math functions.
and windings did not get even barely warm. This is about the most I
Umm... FWB and *inductor* and capacitor... right?

Yup. Something like this:
http://enginuitysystems.com/pix/48V-320V_DCDC_HalfBridge_1Cap.png
If the wire fits, you're already okay: it doesn't appear to be saturating,
so you can run this frequency and voltage all day long (on however
many turns are on there). You just need enough wire to handle the
current without melting.
You certainly don't want to put a cap-input doubler on there, but
there is such a thing as a choke-input doubler. You need four
diodes, two chokes and two caps. You already lose half Vpp in
rectification (half wave rect. each side, one positive, one negative),
so it doesn't get you anything over a FWB.

I think the advantage of a FWCT is two versus four diodes, which makes a big
difference with expensive SiC devices, and it has less forward drop losses,
but I think it requires higher voltage devices for the same DC output.
However, transients can also produce higher voltage with a FWB, as shown
here:
http://enginuitysystems.com/pix/48V-320V_DCDC_HalfBridge_1Cap-1.png
Alternately, you can put the choke on the primary side, but this is more
difficult to arrange on a half bridge. Full bridge (current sourced
inverter) and PP (ala Royer oscillator) are in semi-common use. But
then you need an extra front end chopper to generate the constant current.
Beware also, as turns go up, so does parasitic capacitance, and thus
turn-on transients (primary current spikes, secondary voltage spikes).
This can be improved with winding design (optimal sectioning), and
addressed with snubbers (but not really fixed).

That's too much for this design. I'm looking for something inexpensive and
reasonably efficient. 90-95% would be fine.
N27 is a lowish frequency material, so just keep the Bmax lower
than for N87. I would think in that size, 0.2T would be fine for
N87, maybe 0.1T for N27, should run plenty cool. Notice BTW
that means doubling the number of turns.

I'll see if it saturates at 48 VDC. I could easily go to 30 or 40 kHz. This
transformer was mostly a proof of concept and some practice in winding. The
primary has four conductors in parallel. The entire tranny was salvaged from
a trashed computer PSU.
This is the most common, and the most effective.
Actually that's a lie. For a decade or two, the three-cap layout
dominated AT and ATX power supplies. And by quantity, that's up
there in the millions. But the reason is, the filter caps were
already split because of voltage doubling (the 120/240V switch
wires the circuit for full wave doubling at 120, or regular FWB
at 240); for the purposes of the circuit, it's a +/-160V supply,
so they only needed a 1uF (typically) to "ground" (the center
tap between them).
But when the supply isn't already split, you generally want to
split it with equal capacitance above and below.

I'll have to try simulations with two 20 uF capacitors in series or in
parallel and see what works better. The criteria is the current through each
capacitor.
Ah, 935Cs, good stuff, no problems there. Do mind the stray inductance:
they're simply long, so there's nothing you can do about the first
10-20nH. Which can be advantageous, but important to keep in mind
regardless.

These are only similar. The actual devices are ASC USS5, which are special
order surplus:
http://www.ebay.com/itm/221306488445

I made an offer of $1.25 each for 50 pieces and it was accepted. They seem
to be really awesome:
C: 20.315 uF
D: 0.0008
Z: 0.785 ohms at 10 kHz
L: -12.49 uH
I don't know if that particular Tim reads this newsgroup, but if he
did I'm sure he'd appreciate it. ;)

Oops! I thought you mentioned a book you wrote? Oh, well, hopefully it will
be good reading. :)
Umm... this:
http://webpages.charter.net/dawill/tmoranwms/TimWilliamses.html
is rather old, so I'm sure half the links have died, and all the search
rankings have changed. But kind of amusing nonetheless.
We're no John Smith, but it's still a pretty common name. Not sure if
there's any correlation to electronics, but it seems to me I've already
met a Jim Williams (service engineer, retired), not the more famous (and
recently passed) one; a few other Tims (not to mention the
other Tim W. here), and a few other Williamses.
OTOH, Paul is a pretty common name, but I would guess there aren't many
Schoens outside Europe. :)

When I was a kid the only Schoens in the phone book were my father, uncle,
and grandparents. When I entered Johns Hopkins, there was a young woman in
their directory who was a Schoen, but she was Jewish and from the DC area. I
dated her but it didn't go anywhere. Later, as I was in line at the JHU
bookstore, I watched the fellow in front of me sign his name on a check:
Paul E. Schoen! I was so floored that I just watched him walk off and I
never met him. But I did Google him and he was/is an engineer working on
carbon nanotubes and such.
http://academic.research.microsoft.com/Author/50371984/paul-e-schoen

Paul E. Schoen
US Naval Research Laboratory
Publications: 66 | Citations: 256
Fields: Polymer Materials, Nuclear Magnetic Resonance, Polymer Chemistry
View FAQ about top research areas and Fields of study

Collaborated with 163 co-authors from 1976 to 2006 | Cited by 751 authors

I actually called and talked to him some time ago. He said he knew of other
Paul Schoens as well, and said one of them was a poet. Actually, that would
be yours truly:
http://peschoen.com/earlypoems.txt
http://peschoen.com/kpoems.htm

Whew! Long post :0

Paul
 
T

Tim Williams

P E Schoen said:
It very well could be reactive at 20kHz, so the V*A would be more than
the
actual power. I just measured it to be 114.9 uH at 10kHz, or about 15
ohms reactive at 20kHz.

That seems awfully high, I guess that's series equivalent? I bet the
measurement changes with frequency. I would guess an average 50 ohm 100W
tubular resistor to be in the range of 5-10uH by construction. It isn't
wound with copper, is it? (Or... perhaps constantan?)
The best measure of efficiency will be when I add a rectifier
bridge, capacitor, and inductor. I may need to use SiC Shottky diodes
like these:
http://www.mouser.com/ProductDetail...=sGAEpiMZZMtQ8nqTKtFS/Cwtife2N73ILNibCxuSGE0=

Yeah, downside is you need $20 of them. They also have higher voltage
drop on account of the high internal resistance. It might be the
difference between needing a heatsink (or needing a bigger one), which
could be worthwhile in the end. But I would recommend this more if it
were at, say, 200kHz rather than 20, where the switching and recovery
losses will be problems. Down here anything will do (e.g., UF5407?).
I don't have such a probe and my scope does not have those math
functions.

Well, you'll have to *rectify* that situation then, huh!?

...Which is what you're planning on..

But really, a current probe is easy: if you have some usefully sized
ferrite toroids (say FT87-W or better), put down a hundred turns and hook
it to a 1 ohm resistor. Bam, 100:1 current probe: 0.01 trans-ohms (100A
load = 1V measured). It'll be good out to a few megs, where you get
squigglies from resonant modes.

Only downside is it's not clamp-on, but that's what soldering irons are
for.

Speaking of capacitance, might not hurt to have that inductor sectioned as
well -- I wound this for a class D tube amplifier:
http://webpages.charter.net/dawill/tmoranwms/Elec_Compound10.jpg
Of course, 300V, 0.1A and 120kHz is much more demanding in that regard
than 300V, 1.6A and 20kHz, so use your judgement to decide how much it
needs to be.

Offhand, I'm guessing if you go for off-the-shelf parts, you'll have a
hard time finding a single 10mH ~2A choke, in which case that takes care
of itself -- put a bunch in series and you're fine by default.
I think the advantage of a FWCT is two versus four diodes, which makes a
big difference with expensive SiC devices, and it has less forward drop
losses, but I think it requires higher voltage devices for the same DC
output. However, transients can also produce higher voltage with a FWB,
as shown here:
http://enginuitysystems.com/pix/48V-320V_DCDC_HalfBridge_1Cap-1.png

At this voltage and impedance, I'd be concerned that making a bigger
winding would lead to even higher voltage spikes. You get less coupling
(there's more winding to interleave, and both halves should be interleaved
with each other, and with the primary!), and you waste slightly more
copper (since each half of the secondary is used 1/2 the time, the RMS is
1/sqrt(2) the average output, so you waste (1 - 1/sqrt(2)) x 100% worth of
copper).

FWCT was of course massively popular back in the day, but tube rects were
expensive. Like, a few bucks. Which says something about how cheap
transformers were to wind. (Imagine if we could purchase transformer
variants for less than the cost of diodes! Well, if you're trying to
avoid the schottky FWB, I suppose that would still be true, eh?)

Tim
 
Top