CCM boost inductor design

[Piotr Wyderski]

Hi,

I would like to make a low-loss boost PFC converter for the following
parameters: 150kHz switching frequency, 180-250VAC input, 600V output
at 500W. It should operate in CCM with 20% deltaIL, so the inductor
will be pretty large: 1.2mH at 4.6A_peak. Basically, I am considering
two options of winding it: a gapped low-loss ferrite core (3C95 or
similar material) or an alloy powder E core EMS-0432115-060. The latter
is currently my preferred choice due to its great wide-swing saturation
characteristics (~1.8uH@0A and still a nice value of 600uH at a 10A
surge). Unfortunately, it will require 108 turns, so for a bobbin
27mm wide and relatively thin 1mm diameter wire means 4 layers of
windings. I am afraid this can introduce some nasty resonances and
make the AC resistance worse due to the proximity effect.

OTOH, this is a 20% CCM inductor, so the AC component is only about
800mA in the worst case. So, should I consider winding it with litz
wire (7x0.4mm is probably the thickest braid I can fit there) or
ignore the AC component entirely and go to the lowest DCR achievable,
i.e. a 1.2mm solid wire?

I don't think I can obtain a square 1x1mm magnet wire, the
closest purchasable size is 2x1mm, which for sure will not fit.

The alternative is a planar E58 core wound with 4 layers of 2.5mmx1mm
rectangular wire (54 turns in total). But the inductor would be about
2x the size of the powder core one and have a dangerously sharp
saturation curve.

The boost will be based on a SiC device, but I don't want to go into
the MHz switching range in order to have a physically smaller inductor
-- the parameter I optimize is raw efficiency, not power density.
So I see no point in transforming winding losses into switching and core
losses. Any thoughts, please?

Best regards, Piotr

 

[Piotr Wyderski]

bitrex wrote:

Is 1.2 mH the no load or full-load inductance? That sounds like a
mammoth inductor for that spec, even if that's the no-load inductance.

It's at full load. Math is right and you are right as well. CCM boost
inductors just *are* huge. Usually much bigger that the downstream
converters they supply.

Best regards, Piotr

 

[bitrex]

On 7/28/19 10:00 AM, Piotr Wyderski wrote:

Hi,

I would like to make a low-loss boost PFC converter for the following
parameters: 150kHz switching frequency, 180-250VAC input, 600V output
at 500W. It should operate in CCM with 20% deltaIL, so the inductor
will be pretty large: 1.2mH at 4.6A_peak.

Is 1.2 mH the no load or full-load inductance? That sounds like a
mammoth inductor for that spec, even if that's the no-load inductance.
Check math again?

Basically, I am considering

two options of winding it: a gapped low-loss ferrite core (3C95 or
similar material) or an alloy powder E core EMS-0432115-060. The latter
is currently my preferred choice due to its great wide-swing saturation
characteristics (~1.8uH@0A and still a nice value of 600uH at a 10A
surge). Unfortunately, it will require 108 turns, so for a bobbin
27mm wide and relatively thin 1mm diameter wire means 4 layers of
windings. I am afraid this can introduce some nasty resonances and
make the AC resistance worse due to the proximity effect.

OTOH, this is a 20% CCM inductor, so the AC component is only about
800mA in the worst case. So, should I consider winding it with litz
wire (7x0.4mm is probably the thickest braid I can fit there) or
ignore the AC component entirely and go to the lowest DCR achievable,
i.e. a 1.2mm solid wire?

I don't think I can obtain a square 1x1mm magnet wire, the
closest purchasable size is 2x1mm, which for sure will not fit.

The alternative is a planar E58 core wound with 4 layers of 2.5mmx1mm
rectangular wire (54 turns in total). But the inductor would be about
2x the size of the powder core one and have a dangerously sharp
saturation curve.

The boost will be based on a SiC device, but I don't want to go into
the MHz switching range in order to have a physically smaller inductor
-- the parameter I optimize is raw efficiency, not power density.
So I see no point in transforming winding losses into switching and core
losses. Any thoughts, please?

    Best regards, Piotr

 

[John Larkin]

On Sun, 28 Jul 2019 16:00:36 +0200, Piotr Wyderski
<peter.pan@neverland.mil> wrote:

Hi,

I would like to make a low-loss boost PFC converter for the following
parameters: 150kHz switching frequency, 180-250VAC input, 600V output
at 500W. It should operate in CCM with 20% deltaIL, so the inductor
will be pretty large: 1.2mH at 4.6A_peak. Basically, I am considering
two options of winding it: a gapped low-loss ferrite core (3C95 or
similar material) or an alloy powder E core EMS-0432115-060. The latter
is currently my preferred choice due to its great wide-swing saturation
characteristics (~1.8uH@0A and still a nice value of 600uH at a 10A
surge). Unfortunately, it will require 108 turns, so for a bobbin
27mm wide and relatively thin 1mm diameter wire means 4 layers of
windings. I am afraid this can introduce some nasty resonances and
make the AC resistance worse due to the proximity effect.

OTOH, this is a 20% CCM inductor, so the AC component is only about
800mA in the worst case. So, should I consider winding it with litz
wire (7x0.4mm is probably the thickest braid I can fit there) or
ignore the AC component entirely and go to the lowest DCR achievable,
i.e. a 1.2mm solid wire?

I don't think I can obtain a square 1x1mm magnet wire, the
closest purchasable size is 2x1mm, which for sure will not fit.

The alternative is a planar E58 core wound with 4 layers of 2.5mmx1mm
rectangular wire (54 turns in total). But the inductor would be about
2x the size of the powder core one and have a dangerously sharp
saturation curve.

The boost will be based on a SiC device, but I don't want to go into
the MHz switching range in order to have a physically smaller inductor
-- the parameter I optimize is raw efficiency, not power density.
So I see no point in transforming winding losses into switching and core
losses. Any thoughts, please?

Best regards, Piotr

Is it worth designing this? MeanWell and other power supplies are good
and incredibly cheap.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

 

[Winfield Hill]

Piotr Wyderski wrote...

CCM boost inductors just *are* huge.

Maybe break it up into smaller ones in series?


--
Thanks,
- Win

 

[Winfield Hill]

John Larkin wrote...

Is it worth designing this? MeanWell and other power
supplies are good and incredibly cheap.

They don't have 600-volt outputs. Even if you hack
one for its non-isolated PFC bulk-capacitor voltage,
that's only 400 volts. I've long been unhappy about
the scarcity of commercial high-voltage supplies.


--
Thanks,
- Win

 

[Piotr Wyderski]

John Larkin wrote:

Is it worth designing this? MeanWell and other power supplies are good
and incredibly cheap.

I agree, and have recently used their medical-grade insulation PSU for
about 40 bucks in one of my devices. Insane.

But this time the PSU must not contain any electrolytic capacitors and
there are cooling problems. Hence the pretty unusual efficiency
requirements. The 600V bus voltage is also a consequence of these
requirements. The energy stored in a capacitor is proportional to
the square of the voltage -- a 47uF capacitor will suffice. Such 900V
foil capacitors do exist and cost about 10 dollars, which is pretty
cheap if you consider the future service costs.

There already is an FPGA in the system, so using it to implement the
totem-pole bridgeless PFC with synchronous rectification has the added
benefit of great educational value.

Best regards, Piotr

 

[Winfield Hill]

Piotr Wyderski wrote...

L scales up with N^2, B, H and DCR only with N.

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor. So if you can't get a big enough
core, it may be better to use several as large as
you can get; 500 watts is a lot for a single core.


--
Thanks,
- Win

 

[Piotr Wyderski]

Winfield Hill wrote:

> Maybe break it up into smaller ones in series?

This game doesn't pay off. L scales up with N^2, B, H and DCR only with
N. So splitting a 2L monster into L+L means higher total core losses and
ohmic losses. It's better to buy a bigger core or increase the switching
frequency.

Same with interleaving. The complexity is not worth the trouble unless
you have no other option or there are scalability requirements.

Best regards, Piotr

 

[Winfield Hill]

Piotr Wyderski wrote...

-- a 47uF capacitor will suffice. Such 900V foil
capacitors do exist and cost about 10 dollars

Interesting, can you provide some pointers?


--
Thanks,
- Win

 

[Piotr Wyderski]

Winfield Hill wrote:

> Interesting, can you provide some pointers?

I use MKP1848C65090JY5, for example:

https://pl.mouser.com/ProductDetail/Vishay-Roederstein/MKP1848C65090JY5?qs=zDkSFN9STR8qFDSFzHtCiw==

At 600V derating their useful life approaches ~1e6 hours.

There are even bigger units, but the price is a super-linear function of
the capacity. Electrically and economically it is better to buy 2 and
connect them in parallel.

Best regards, Piotr

 

[Jan Panteltje]

On a sunny day (28 Jul 2019 09:36:47 -0700) it happened Winfield Hill
<winfieldhill@yahoo.com> wrote in <qhkiuv0lt8@drn.newsguy.com>:

John Larkin wrote...

Is it worth designing this? MeanWell and other power
supplies are good and incredibly cheap.

They don't have 600-volt outputs. Even if you hack
one for its non-isolated PFC bulk-capacitor voltage,
that's only 400 volts. I've long been unhappy about
the scarcity of commercial high-voltage supplies.

If you do not need a high speed switcher,
then for that sort of HV use an old tube transformer (2 x 240V or something)
to get isolation, add some diode bridge, thyristor for control,
filter choke, filter capacitor.
https://www.ebay.com/itm/132548432418

I make 110 V 60Hz here in the 230 V 50 Hz world using a power audio amp
and an audio amp 230V mains transformer in reverse,
powered by 'sgen' signal generator in Linux, slider for output voltage,
Sometimes you gotta work with the things you have laying about.
Select any frequency within reason, 50 Hz, to 400 Hz should work.

Now if you are smart you feed the output AC via a divider back into the
audio line input, and write those famous few lines of code to adjust the mixer setting
so the output voltage stays constant.

 

[Piotr Wyderski]

Winfield Hill wrote:

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor.

Absolutely correct, but the ACR losses also scale with I_AC^2,
which is pretty low in a decent CCM. 5:1 I_DC-to-I_AC ratio can buy
a lot of power budget to spare on the AC losses. So my working
hypotesis is that only DCR (and unwaned resonances as a higher-level
effect) matter in CCM. I would be happy to have this falsified by
the experts inhabiting this group.

I crave an all-practical book on magnetics similar to your AoE.
What are the quantitive properties of foil winding? How are AC losses
impacted by orthocyclic winding as compared to helical? And so on.
And tons of FEM simulations of current/flux density.

This knowledge can be found in the Internet, but is very scatterd.
Mr. Legg has once sent me a lot useful materials on the behaviour of foils.

So if you can't get a big enough
core, it may be better to use several as large as
you can get;

Oh, definitely. Or start interleaving.

> 500 watts is a lot for a single core.

Just E42/15 KoolMu/MSS, acceptable. Ferrites can't come even close
to this size and there are no discrete gap issues. But the
core losse per unit volume compared to the modern high-performance
ferrites and the number of turns don't look very good. Eh, these
tinky compromises...

Best regards, Piotr

 

[Martin Riddle]

On 28 Jul 2019 10:37:17 -0700, Winfield Hill <winfieldhill@yahoo.com>
wrote:

Piotr Wyderski wrote...

-- a 47uF capacitor will suffice. Such 900V foil
capacitors do exist and cost about 10 dollars

Interesting, can you provide some pointers?

ASC will do HV Foil caps, custom to if you need quantity.
Dearborn has similar foil caps.


Cheers

 

[bitrex]

On 7/28/19 11:39 AM, Piotr Wyderski wrote:

bitrex wrote:

Is 1.2 mH the no load or full-load inductance? That sounds like a
mammoth inductor for that spec, even if that's the no-load inductance.

It's at full load. Math is right and you are right as well. CCM boost
inductors just *are* huge. Usually much bigger that the downstream
converters they supply.

    Best regards, Piotr

<https://www.infineon.com/dgdl/Infineon-ApplicationNote_PFCCCMBoostConverterDesignGuide-AN-v02_00-EN.pdf?fileId=5546d4624a56eed8014a62c75a923b05>

With the full-load requirement equation there I get ~ 1 mH so hmm yes
about right.

You could decrease the required full-load inductance by nearly half if
you could accept a 5% higher deltaIL and run 50kHz higher

 

[Winfield Hill]

Piotr Wyderski wrote...

Winfield Hill wrote:

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor.

Absolutely correct, but the ACR losses also scale with I_AC^2,
which is pretty low in a decent CCM. 5:1 I_DC-to-I_AC ratio can buy
a lot of power budget to spare on the AC losses. So my working
hypotesis is that only DCR (and unwaned resonances as a higher-level
effect) matter in CCM. I would be happy to have this falsified by
the experts inhabiting this group.

I crave an all-practical book on magnetics similar to your AoE.
What are the quantitive properties of foil winding? How are AC losses
impacted by orthocyclic winding as compared to helical? And so on.
And tons of FEM simulations of current/flux density.

This knowledge can be found in the Internet, but is very scatterd.
Mr. Legg has once sent me a lot useful materials on the behaviour of foils.

So if you can't get a big enough
core, it may be better to use several as large as
you can get;

Oh, definitely. Or start interleaving.

500 watts is a lot for a single core.

Just E42/15 KoolMu/MSS, acceptable. Ferrites can't come even close
to this size and there are no discrete gap issues. But the
core losse per unit volume compared to the modern high-performance
ferrites and the number of turns don't look very good. Eh, these
tinky compromises...

We will wait, Piotr, to see your results!


--
Thanks,
- Win

 

[John Larkin]

On Sun, 28 Jul 2019 19:54:07 +0200, Piotr Wyderski
<peter.pan@neverland.mil> wrote:

Winfield Hill wrote:

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor.

Absolutely correct, but the ACR losses also scale with I_AC^2,
which is pretty low in a decent CCM. 5:1 I_DC-to-I_AC ratio can buy
a lot of power budget to spare on the AC losses. So my working
hypotesis is that only DCR (and unwaned resonances as a higher-level
effect) matter in CCM. I would be happy to have this falsified by
the experts inhabiting this group.

I crave an all-practical book on magnetics similar to your AoE.
What are the quantitive properties of foil winding? How are AC losses
impacted by orthocyclic winding as compared to helical? And so on.
And tons of FEM simulations of current/flux density.

This knowledge can be found in the Internet, but is very scatterd.
Mr. Legg has once sent me a lot useful materials on the behaviour of foils.

So if you can't get a big enough
core, it may be better to use several as large as
you can get;

Oh, definitely. Or start interleaving.

500 watts is a lot for a single core.

Just E42/15 KoolMu/MSS, acceptable. Ferrites can't come even close
to this size and there are no discrete gap issues. But the
core losse per unit volume compared to the modern high-performance
ferrites and the number of turns don't look very good. Eh, these
tinky compromises...

Best regards, Piotr

KoolMu is great stuff. Other people are now making equivalents. I was
burning the paint off powdered iron toroids, and CoolMu came to the
rescue.

Don't you just need X pounds of core and copper for a given
application? One big inductor or several small ones?

The only difference, seems to me, might be surface area/cooling, which
would favor several small ones.


--

John Larkin Highland Technology, Inc

lunatic fringe electronics

 

[Winfield Hill]

Piotr Wyderski wrote...

Winfield Hill wrote:

Interesting, can you provide some pointers?

I use MKP1848C65090JY5, for example:

https://pl.mouser.com/ProductDetail/Vishay-Roederstein/MKP1848C65090JY5?qs=zDkSFN9STR8qFDSFzHtCiw==

At 600V derating their useful life approaches ~1e6 hours.

There are even bigger units, but the price is a super-linear
function of the capacity. Electrically and economically it
is better to buy 2 and connect them in parallel.

That's a very nice choice, I like it. Thanks!


--
Thanks,
- Win

 

[bitrex]

On 7/28/19 1:05 PM, Winfield Hill wrote:

Piotr Wyderski wrote...

L scales up with N^2, B, H and DCR only with N.

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor. So if you can't get a big enough
core, it may be better to use several as large as
you can get; 500 watts is a lot for a single core.

If you use toroids there's no law that you can't glomp two toroids
together, like stacking donuts, to get more effective core volume and
winding surface area, it's done all the time

 

[bitrex]

On 7/28/19 5:24 PM, John Larkin wrote:

On Sun, 28 Jul 2019 19:54:07 +0200, Piotr Wyderski
peter.pan@neverland.mil> wrote:

Winfield Hill wrote:

DCR may scale with N, but multi-layer proximity
effect losses scale more like N^2, or even
stronger. Likewise core losses can scale by a
squared factor.

Absolutely correct, but the ACR losses also scale with I_AC^2,
which is pretty low in a decent CCM. 5:1 I_DC-to-I_AC ratio can buy
a lot of power budget to spare on the AC losses. So my working
hypotesis is that only DCR (and unwaned resonances as a higher-level
effect) matter in CCM. I would be happy to have this falsified by
the experts inhabiting this group.

I crave an all-practical book on magnetics similar to your AoE.
What are the quantitive properties of foil winding? How are AC losses
impacted by orthocyclic winding as compared to helical? And so on.
And tons of FEM simulations of current/flux density.

This knowledge can be found in the Internet, but is very scatterd.
Mr. Legg has once sent me a lot useful materials on the behaviour of foils.

So if you can't get a big enough
core, it may be better to use several as large as
you can get;

Oh, definitely. Or start interleaving.

500 watts is a lot for a single core.

Just E42/15 KoolMu/MSS, acceptable. Ferrites can't come even close
to this size and there are no discrete gap issues. But the
core losse per unit volume compared to the modern high-performance
ferrites and the number of turns don't look very good. Eh, these
tinky compromises...

Best regards, Piotr



KoolMu is great stuff. Other people are now making equivalents. I was
burning the paint off powdered iron toroids, and CoolMu came to the
rescue.

Don't you just need X pounds of core and copper for a given
application? One big inductor or several small ones?

The only difference, seems to me, might be surface area/cooling, which
would favor several small ones.

All the PC power supplies with boost mode PFC that I've seen, of about
500 watts max output, use a single large toroid, maybe bifiliar wound
with two or three windings of magnet wire in parallel.

See e.g. time index 13:00

<https://youtu.be/S7SFR9-1Z48?t=793>

The specs for this design don't seem that much different so idk why need
to go to two coils. 500W isn't an absurd amount of power. The difference
in efficiency between 20% current ripple spec and 25% will probably be
like 98% vs 95%...is this important