Efficiency of SMPS design.

[Denis Gleeson]

Hi Guys

Ive designed a switched mode PSU around the LM3578A (National Semi).

I need to go from +12V to -55V. My current requirement is 12mA.

I placed a 10k resistor for the load on the output of the switched
mode PSU
and tested the design that Ive created (on a PCB). I find the
following.

INPUT from bench Power supply = 5V @ 160mA = VI watts = 0.8Watts
OUTPUT from the switched mode PSU = -42V @ 4.2mA = VI watts =
0.176Watts

Now to work out the efficiency of this cct I calculate as follows.

0.176
------ * 100 = 22%
0.8

Is this right?
Do I need to source 160mA from 5V to get -42V at 4.2mA?

Is this the way it is or is there a more efficient way to do this
design?
My calculations show a Peak Inductor current of approx. 0.55A so Im
begining to conclude that this is correct.

There must be a better way!

Many thanks for any help.

Denis

________________________________________________________________
Ive designed as per figure 21 of the data sheet.

http://cache.national.com/ds/LM/LM1578A.pdf

According to that figure I have:
R1 = 51k
R2 = 1k
R3 = 0.22
R4 = 860
R5 = 100
L1 = 2 * 220uH in series
C2 = 10uF 65V electrolytic
D1 = 1N5818 (schotky)
C1 = 1.1nF
C3 = 22pF
Q1 = 2N2905

I dont have C4 and R6 connected.

 

[John Popelish]

Hi Guys

Ive designed a switched mode PSU around the LM3578A (National Semi).

I need to go from +12V to -55V. My current requirement is 12mA.

I placed a 10k resistor for the load on the output of the switched
mode PSU
and tested the design that Ive created (on a PCB). I find the
following.

INPUT from bench Power supply = 5V @ 160mA = VI watts = 0.8Watts
OUTPUT from the switched mode PSU = -42V @ 4.2mA = VI watts =
0.176Watts

Now to work out the efficiency of this cct I calculate as follows.

0.176
------ * 100 = 22%
0.8

Is this right?
Do I need to source 160mA from 5V to get -42V at 4.2mA?

Is this the way it is or is there a more efficient way to do this
design?
My calculations show a Peak Inductor current of approx. 0.55A so Im
begining to conclude that this is correct.

There must be a better way!

Many thanks for any help.

Denis

________________________________________________________________
Ive designed as per figure 21 of the data sheet.

http://cache.national.com/ds/LM/LM1578A.pdf

According to that figure I have:
R1 = 51k
R2 = 1k
R3 = 0.22
R4 = 860
R5 = 100
L1 = 2 * 220uH in series
C2 = 10uF 65V electrolytic
D1 = 1N5818 (schotky)
C1 = 1.1nF
C3 = 22pF
Q1 = 2N2905

I dont have C4 and R6 connected.

22% efficiency is abnormally low for this sort of regulator, even
though this configuration is not as efficient as some others. I think
your problem may involve some poor component choices. For instance D1
sees the output voltage plus the input voltage, and your choice is
rated for only 30 volts. A 1N4148 would do better. The PNP
transistor is also a bit pushed for voltage capability at your output
voltage. The parts selected on page 21 are for a +5 to -15 volt
regulator.

 

[Tam/WB2TT]

Hi Guys

Ive designed a switched mode PSU around the LM3578A (National Semi).

I need to go from +12V to -55V. My current requirement is 12mA.

I placed a 10k resistor for the load on the output of the switched
mode PSU
and tested the design that Ive created (on a PCB). I find the
following.

INPUT from bench Power supply = 5V @ 160mA = VI watts = 0.8Watts
OUTPUT from the switched mode PSU = -42V @ 4.2mA = VI watts =
0.176Watts

Now to work out the efficiency of this cct I calculate as follows.

0.176
------ * 100 = 22%
0.8

Is this right?
Do I need to source 160mA from 5V to get -42V at 4.2mA?

Is this the way it is or is there a more efficient way to do this
design?
My calculations show a Peak Inductor current of approx. 0.55A so Im
begining to conclude that this is correct.

There must be a better way!

Many thanks for any help.

Denis

________________________________________________________________
Ive designed as per figure 21 of the data sheet.

http://cache.national.com/ds/LM/LM1578A.pdf

According to that figure I have:
R1 = 51k
R2 = 1k
R3 = 0.22
R4 = 860
R5 = 100
L1 = 2 * 220uH in series
C2 = 10uF 65V electrolytic
D1 = 1N5818 (schotky)
C1 = 1.1nF
C3 = 22pF
Q1 = 2N2905

I dont have C4 and R6 connected.

I am confused. Are you running this off +5 or +12V. You mention both. At 12
ma load current, you should not expect much efficiency. Also, I suspect your
L is way too small. They use an L in the hundred of uH at .3 A, which says
yours should be a lot bigger at the ~25 KHz switching frequency. Their Fig16
is not clear, but I think the L values bigger than 680 (to the left) are in
milli Henries. I would use a chip that runs at 100 - 300 KHz.
Try it with a 10 mH inductor, and see what the current is.

Tam

 

[Paul Mathews]

Some of the losses in your circuit, e.g., operating current for the
chip, parasitic current losses due to switching, do not vary much with
load. With a light load such as you've used for your test, efficiency
will be low. In fact, with no load, efficiency is always zero
percent.

Paul Mathews

 

[Joerg]

Hi Denis,

That regulator worked like a champ for me. Three thoughts:

A 2N2905 may not be the ticket here. You need a fast FET.

Also, these fast reg chips cannot easily be breadboarded. It must be an
RF circuit type of build, whether that is a prototype or a PCB. No long
traces or wires.

National has a great forum on their site for dessigners of switchers.
They are very responsive and you don't even need password, sign-up or
anything. Select "Power", then "Discussion Forum". Or call and ask for
Fred Embuscado. I wish that all companies were as customer oriented as
National is. Big cheers to them.

Regards, Joerg

 

[Ken Smith]

Hi Denis,

That regulator worked like a champ for me. Three thoughts:

A 2N2905 may not be the ticket here. You need a fast FET.

Also, these fast reg chips cannot easily be breadboarded. It must be an
RF circuit type of build, whether that is a prototype or a PCB. No long
traces or wires.

This depends on what you call easy.

Get some copper clad PCB material, a Dremal(sp) tool, super glue and tin
snips. A modest sized chunk of copper clad is the ground plain. Make a
pattern to mount the chip on a small bit of copper clad. Don't make the
copper clad much bigger than the chip. Solder the chip in place and use
bus wire to hook up the grounds. Add the bypasses right at the chip as
suggested by the maker.

Make up little tiles for the various other components, keeping the tile as
small as the part will allow.. Mount the parts on them. Turn them and
slide them around until you've got what looks like a good design and use
thick bus wire to hook them up.

If something doesn't work right, you can unsolder and move things.

When its is all working, make little paper labels for the parts and put
the references on them. Give the netlist and the board to the layout
artist and tell them "like this".

When you've got a supply circuit with conflicting requirements this can be
easier than repeated layouts.

 

[Denis Gleeson]

Hi Guys

Thanks for the help.

Joerg can you recommend a fast fet for this cct. The data sheet doesnt
propose one so I kept to a transistor.

Do you have a number for Fred Embuscado?

Ok My diode choice is wrong. Im not sure that a 1N4148b is the
solution. Its not a schotky and I dont think it could source the
current required into the coil. Ill find a suitable schotky.

Time for a new Transistor and diode.

The inductor size is confusing me.
I worked out the value as per the application note

http://www.national.com/an/AN/AN-1066.pdf

Ill need to look at the inductor selection some more

thanks.


Bye for now.

Denis

 

[Joerg]

Hi Ken,

Get some copper clad PCB material, a Dremal(sp) tool, super glue and tin
snips. A modest sized chunk of copper clad is the ground plain. Make a
pattern to mount the chip on a small bit of copper clad. Don't make the
copper clad much bigger than the chip. Solder the chip in place and use
bus wire to hook up the grounds. Add the bypasses right at the chip as
suggested by the maker.

Make up little tiles for the various other components, keeping the tile as
small as the part will allow.. Mount the parts on them. Turn them and
slide them around until you've got what looks like a good design and use
thick bus wire to hook them up.

That's actually how I usually do that, too. The only downside with
copper clad is the burrs. Every time you cut yourself with that it tends
to infect a little. After that, even holding a glass of water hurts. And
the dust from the Dremel is a pain as well. I do that outside but on a
windy days that's no fun either.

With "not easy" I meant that it doesn't take much to mess that up. If
you know RF stuff you're fine. But I have seen folks try this on
experimenter's board. Nothing worked but for some reason the AM radio
quit when the circuit was powered up...

Other times I have seen the circuit turn an electrolytic cap into a
space craft. With switchers it can really become dangerous.

When you've got a supply circuit with conflicting requirements this can be
easier than repeated layouts.

Yep, them dang conflicting requirements. Don't we all love these? In med
electronics we call that "feature creep" or "specs on the move".

Regards, Joerg

 

[Ken Smith]

Hi Ken, [... I wrote ..]

Make up little tiles for the various other components, keeping the tile as
small as the part will allow.. Mount the parts on them. Turn them and
slide them around until you've got what looks like a good design and use
thick bus wire to hook them up.

That's actually how I usually do that, too. The only downside with
copper clad is the burrs.

I file the edges. I also have a sheet with pre-made patterns on it. I
just orderd some more.

Every time you cut yourself with that it tends
to infect a little.

Perhaps my skin is just thicker. Copper doesn't need to be infected to be
a problem. Copper is a strongish poison.

After that, even holding a glass of water hurts. And
the dust from the Dremel is a pain as well. I do that outside but on a
windy days that's no fun either.

Safety glasses and don't breath.

With "not easy" I meant that it doesn't take much to mess that up. If
you know RF stuff you're fine. But I have seen folks try this on
experimenter's board. Nothing worked but for some reason the AM radio
quit when the circuit was powered up...

Many many years ago I made a 1000V power supply that ran above 1MHz. It
didn't have to make much current and we wanted it very small so I tried
the then modern idea of going up to 1MHz and using MOSFET switches.

The servo feedback op-amp had to be on a different desk before it would
work. Even with the lower EMI of the semi-resonant design, the RF overran
the op-amps input. We ending up making another part smaller so we had the
room for a more sane design.

Other times I have seen the circuit turn an electrolytic cap into a
space craft. With switchers it can really become dangerous.

You know you are into the limits when you have to heat sink the ceramic
capacitors.

Yep, them dang conflicting requirements. Don't we all love these? In med
electronics we call that "feature creep" or "specs on the move".

That too but what I meant was thing like this:

T1 is 0.25" on a side.
Q1 is a D2-pak and must be near T1's primary winding.
D1 is a D-pak and must be near pins 2 and 3 of T1
D2 is a D2-pak and must be near pins 4 and 5 of T1
D3 is a D-pak and must be near pins 6 and 7 of T1
D4 is a D2-pak and must be near pins 7 and 8 of T1
D5 is a D-pak and must be near pins 9 and 10 of T1
C1 is a 2220 and must be near pins 1 and 7 of T1

Without adding some extra dimensions to the universe it can't be all done.

 

[John Popelish]

Denis Gleeson wrote:
(snip)

Ok My diode choice is wrong. Im not sure that a 1N4148b is the
solution. Its not a schotky and I dont think it could source the
current required into the coil. Ill find a suitable schotky.

Time for a new Transistor and diode.

A 1N4148 may not be the optimal diode, but it can easily handle 250 ma
peaks and 100 volts reverse and is fast (and cheap and common as
dirt). I suggested it as a way to prove that your Schottky breaking
down was the main cause of your efficiency problem. But it is not a
terrible choice, either, at a 10 ma average output current. Its extra
half volt or so of forward drop compared to an 80 volt Schottky diode
lowers the overall efficiency by only a percent or two when the output
voltage is 55 volts.

 

[R.Legg]

Hi Guys

Thanks for the help.

Joerg can you recommend a fast fet for this cct. The data sheet doesnt
propose one so I kept to a transistor.

Do you have a number for Fred Embuscado?

Forum link -

http://wwwd.national.com/national/PowerMB.nsf/

Web Bench -

http://www.national.com/appinfo/webench/scripts/my_webench.cgi/

RL

 

[Tam/WB2TT]

Hi Guys

Thanks for the help.

Joerg can you recommend a fast fet for this cct. The data sheet doesnt
propose one so I kept to a transistor.

Do you have a number for Fred Embuscado?

Ok My diode choice is wrong. Im not sure that a 1N4148b is the
solution. Its not a schotky and I dont think it could source the
current required into the coil. Ill find a suitable schotky.

Time for a new Transistor and diode.

The inductor size is confusing me.
I worked out the value as per the application note

http://www.national.com/an/AN/AN-1066.pdf

Ill need to look at the inductor selection some more

thanks.


Bye for now.

Denis

Joerg <[email protected]> wrote in message

Denis,

I fired up my SWCAD simulator (running open loop) with their Fig21 valuess
(5V in 12V out), and came up with the following:

1. Their 150 uH inductor is too small, and causes discontinuous operation at
the stated 50 KHz and 40 ma load current.

2. Your 440 uH inductor works just fine AT THE RATED 40 MA CURRENT. I
measured an efficiency of about 90%. But this is from the collector of the
2N2905 to DC out only.

3. When I changed the load to 10K, the thing went off into lala land, and
the efficiency dropped to 12%. So, I made the inductor bigger. 5 mH was the
smallest value that worked, and gave an efficiency of 86%. It also had a
huge overshoot, which probably would not happen in closed loop operation.

Tam

 

[Joerg]

Hi Denis,

Joerg can you recommend a fast fet for this cct. The data sheet doesnt
propose one so I kept to a transistor.

Not out of the blue since my switchers almost always need N-channel
(BSS123, 2N7002 for small switchers). I did use the BSS84 a few times
but that one is only 50V I believe so it won't work here. The ON Semi
MTD6 series is high voltage but not SOT23. You'd have to plough through
the parametric tables here.

Do you have a number for Fred Embuscado?

I just use the NS 1-800 number. They'll connect you. But first I'd post
in their forum. You will normally get a same day answer.

Ok My diode choice is wrong. Im not sure that a 1N4148b is the
solution. Its not a schotky and I dont think it could source the
current required into the coil. Ill find a suitable schotky.

The MURS series from ON Semi is great, and small packages. Cheap, too
. A fast regular diode can sometimes be used when the switcher is
always in discontinuous mode but I'd recommend a Schottky even then.

Time for a new Transistor and diode.

The inductor size is confusing me.
I worked out the value as per the application note

And watch saturation. Scope the current at the sense resistor and see
that it cuts off at the desired trip point. A sudden steep rise would
indicate core saturation and then the efficiency is toast. The current
must ramp linearly.

Regards, Joerg

 

[Joerg]

Hi Tam,

2. Your 440 uH inductor works just fine AT THE RATED 40 MA CURRENT. I
measured an efficiency of about 90%. But this is from the collector of the
2N2905 to DC out only.

A 2N2905 may not be so good here. Since it is being operated as a
saturated switch it would be kind of slow.

3. When I changed the load to 10K, the thing went off into lala land, and
the efficiency dropped to 12%. So, I made the inductor bigger. 5 mH was the
smallest value that worked, and gave an efficiency of 86%. It also had a
huge overshoot, which probably would not happen in closed loop operation.

Hmm, a smaller inductor would just cause discontinuous mode (DCM) which
should be fine. I never had any efficiency drop there since all switcher
enter DCM on light loads.

Tam, since you are a ham radio operator I have another question.
Slightly off topic and I'll post separately. Topic "FAA wants us to
remove trees, antennas etc."

Regards, Joerg

 

[Tam/WB2TT]

Hi Tam,

A 2N2905 may not be so good here. Since it is being operated as a
saturated switch it would be kind of slow.

The 2N2905 works OK speed wise. Problem is the base current. During startup
the current pulses are hundreds of ma; so, if I want the transistor to
saturate, I need lots of base current. Once steady state is reached, almost
half the supply current comes out of the base. I reduced the drive; now the
transistor is current limited during startup, which may be good. I think a P
channel FET is the preferred device.

WOW! I just looked at Fig21, and Denis' parts list. He is running 50 ma base
current (3 x what I had) into the 2N2905. Denis, quick, change R5 to 1K and
see what happens.

Hmm, a smaller inductor would just cause discontinuous mode (DCM) which
should be fine. I never had any efficiency drop there since all switcher
enter DCM on light loads.

I think you are right, but I am running open loop. So, I can't compensate
for the DCM. Still, I don't think you want to design for that mode in normal
operation; the inductor current goes very high. He should at least try the
bigger inductor. But the small molded chokes won't quite make it current
wise

Tam, since you are a ham radio operator I have another question.
Slightly off topic and I'll post separately. Topic "FAA wants us to
remove trees, antennas etc."

I saw that. I would have assumed they would pay for it. Whole thing seems
strange.

Regards, Joerg

I tried the version that Denis really wants: 12V in 55V out, f=50 KHz.
Smallest inductor that worked was 3.9 mH (I picked Rdc=6 Ohms). Worked well,
but I still think a small P channel would be better.

Tam

 

[Joerg]

Hi Tam,

WOW! I just looked at Fig21, and Denis' parts list. He is running 50 ma base
current (3 x what I had) into the 2N2905. Denis, quick, change R5 to 1K and
see what happens.

Ooops. Maybe a good idea to throw in a new LM3578 as well because it
might be damaged now. It can take that as surge but I doubt it likes it
for the complete cycle time.

I saw that. I would have assumed they would pay for it. Whole thing seems
strange.

Indeed it is. I searched the web up and down to find similar incidents
but no luck. There just has to be a parallel case somewhere. At many
airports such as this one they'd have to move mountains to make them
fully compliant with the regs (7:1 transitional surface etc.). I don't
mind the noise of the old warbirds taking off with a roar that makes the
walls shake. They are a pleasure to see flying. But now that they want
to stiff dozens of household to pay big time for their fun the climate
is souring fast out here.

Regards, Joerg

 

[Tam/WB2TT]

Hi Tam,

Ooops. Maybe a good idea to throw in a new LM3578 as well because it
might be damaged now. It can take that as surge but I doubt it likes it
for the complete cycle time.

Shouldn't be a problem for the chip. Fig 21 drives even more current. Spec
appears to be 750 ma. Occurs to me a FET might have problems too, unless it
has very low gate input capacitance. There is no active pullup on the chip.
So, it depends on a gate to VCC resistor to turn the FET off. If you make
the reistor low, you are drawing ON state current again. The FET will be
mostly on.

Tam

 

[Joerg]

Hi Tam,

Actually this chip can drive tremendous loads. That always amazes me
considering that I can hardly even see these tiny pins. It does have an
active pull-up, they spec 4.5ohms for the low side and 16ohms for the
high side. That is pretty good for such a chip.

So I believe you are right, it should be able to survive such stresses
at the drive pin.

Regards, Joerg

 

[Rich Grise]

That's actually how I usually do that, too. The only downside with
copper clad is the burrs. Every time you cut yourself with that it tends
to infect a little. After that, even holding a glass of water hurts. And
the dust from the Dremel is a pain as well. I do that outside but on a
windy days that's no fun either.

I've seen copper-clad .032" thick epoxy-glass, and I'd think you could cut
that on a paper cutter.

But either way, it sounds cool.

Cheers!
Rich

 

[Ken Smith]

I've seen copper-clad .032" thick epoxy-glass, and I'd think you could cut
that on a paper cutter.

Yes you can use a paper cutter. You have to buy one for the purpose and
replace it now and then. The glass eats the edge of the blade.