Buck converter, controller riding on the switch node

[Klaus Kragelund]

Hi

Well, a typical buck converter with current mode control has a PWM
controller referenced to output ground, a current sense transformer to
bring down the current sense signal and a pulse transformer to level
shift the gate signal.

Another approach is the one from Fairchild with an integrated solution
riding on the switch node:

http://www.electronicsdesign.dk/tmp/AN2544_ViperInBuckMode3.pdf

To be independant on single source parts, I would like to do the same
with the standard PWM controllers like the SG3524, UC1843 etc as shown
in the scematics below:

http://www.electronicsdesign.dk/tmp/HighSideBuck.pdf

I haven't build it yet, but do any of you have experiences with this?
Ofcourse I need to make a seperate plane connected to the SW (switch)
node below the entire PWM controller circuit to combat leakage
currents by the parasitic capacitances/dV/dt.

Anything more to look out for?

Thanks

Klaus

 

[Klaus Kragelund]

Hi

Well, a typical buck converter with current mode control has a PWM
controller referenced to output ground, a current sense transformer to
bring down the current sense signal and a pulse transformer to level
shift the gate signal.

Another approach is the one from Fairchild with an integrated solution
riding on the switch node:

http://www.electronicsdesign.dk/tmp/AN2544_ViperInBuckMode3.pdf

To be independant on single source parts, I would like to do the same
with the standard PWM controllers like the SG3524, UC1843 etc as shown
in the scematics below:

http://www.electronicsdesign.dk/tmp/HighSideBuck.pdf

I haven't build it yet, but do any of you have experiences with this?
Ofcourse I need to make a seperate plane connected to the SW (switch)
node below the entire PWM controller circuit to combat leakage
currents by the parasitic capacitances/dV/dt.

Anything more to look out for?

And bytheway, don't look to closely at components values, they were
without thorough calculations

 

[Fred_Bartoli]

And bytheway, don't look to closely at components values, they were
without thorough calculations

Not a bad idea but to get decent performance from this you'll need to change
D25 for a controled switch (i.e. a mos) so that you really sample the output
voltage and you might see some funny behavior too (think about what's
happening when the ouput overshoots).

 

[[email protected]]

x-no-archive:

And bytheway, don't look to closely at components values, they were
without thorough calculations - Hide quoted text -

- Show quoted text -

I don't see how you addressed the issue that the Vcc and Vout are
referenced to the real ground yet you need to reference them to Sw
ground for the IC.

Mark

 

[Joerg]

Hi

Well, a typical buck converter with current mode control has a PWM
controller referenced to output ground, a current sense transformer to
bring down the current sense signal and a pulse transformer to level
shift the gate signal.

Another approach is the one from Fairchild with an integrated solution
riding on the switch node:

http://www.electronicsdesign.dk/tmp/AN2544_ViperInBuckMode3.pdf

To be independant on single source parts, I would like to do the same
with the standard PWM controllers like the SG3524, UC1843 etc as shown
in the scematics below:

http://www.electronicsdesign.dk/tmp/HighSideBuck.pdf

I haven't build it yet, but do any of you have experiences with this?
Ofcourse I need to make a seperate plane connected to the SW (switch)
node below the entire PWM controller circuit to combat leakage
currents by the parasitic capacitances/dV/dt.

Anything more to look out for?

Since VB and SW (the floating ground) would rapidly swing up and down
with the switch action this can result in major EMC pain.

Maybe there is a way with the traditional concept to level shift with
capacitors if the duty cycle remains in a "normal" range. Of course for
the current sense that would almost have to be done differentially (or
with a transformer).

 

[Klaus Kragelund]

x-no-archive:






I don't see how you addressed the issue that the Vcc and Vout are
referenced to the real ground yet you need to reference them to Sw
ground for the IC.

When the gate is low and the MOSFET turned off, the diode D23
freewheels, so the SW voltage is negative one diode drop. So at that
time the drive for the chip (VB) and the sense voltage (Out/24V) is
transferred via diodes to the high side circuit. Actually, good thing
you asked - Out and 24V should be the same node.

Regards

Klaus

 

[Klaus Kragelund]

"Klaus Kragelund" <[email protected]> a écrit dans le message de






Not a bad idea but to get decent performance from this you'll need to change
D25 for a controled switch (i.e. a mos) so that you really sample the output
voltage and you might see some funny behavior too (think about what's
happening when the ouput overshoots).

Good point, should sample a shortly after the gate is turned off.
Thats actually how they do it in the integrated device for this
function. Maybe the PWM should be a microcontroller (PSoc) to
incorporate some intelligence into the PWM (softstart, hickup,
sampling, reference etc).

Regards

Klaus

 

[MooseFET]

Hi

Well, a typical buck converter with current mode control has a PWM
controller referenced to output ground, a current sense transformer to
bring down the current sense signal and a pulse transformer to level
shift the gate signal.

Another approach is the one from Fairchild with an integrated solution
riding on the switch node:

http://www.electronicsdesign.dk/tmp/AN2544_ViperInBuckMode3.pdf

To be independant on single source parts, I would like to do the same
with the standard PWM controllers like the SG3524, UC1843 etc as shown
in the scematics below:

http://www.electronicsdesign.dk/tmp/HighSideBuck.pdf

I haven't build it yet, but do any of you have experiences with this?
Ofcourse I need to make a seperate plane connected to the SW (switch)
node below the entire PWM controller circuit to combat leakage
currents by the parasitic capacitances/dV/dt.

Anything more to look out for?

I have done this same thing with the entire switcher controller flying
up and down. The feedback used and op-amp with careful low pass
filtering to get the output voltage level shifted on to the domain
with the huge AC on everything.

Part of the trick is to make your planes look like this:


------------- Top switching parts
------------- Inner 1 interconnect
------------- Flying plane
------------- Nothing in this area
------------- Nothing in this area
------------- Bottom system ground

Around the outside of the area, you can stitch the ground from its
normal layer down to the bottom side in many places. This basically
puts the switching stuff in a box.

You want to soften up the switching edges a little. Any microwaves
created by the circuit will keep looking until they find a way to mess
you up. One problem is with the input transistors of the op-amp.

 

[Joerg]

I have done this same thing with the entire switcher controller flying
up and down. The feedback used and op-amp with careful low pass
filtering to get the output voltage level shifted on to the domain
with the huge AC on everything.

Part of the trick is to make your planes look like this:


------------- Top switching parts
------------- Inner 1 interconnect
------------- Flying plane
------------- Nothing in this area
------------- Nothing in this area
------------- Bottom system ground

Around the outside of the area, you can stitch the ground from its
normal layer down to the bottom side in many places. This basically
puts the switching stuff in a box.

There needs to be something along the lines of system ground above the
flying plane. Either interleaved and well connected ground islands or a
shielded enclosure. It's next to impossible to pass EMC otherwise. BTDT
(or, rather, clients called me with similar designs they had and asked
what to do about their EMC problems).

You want to soften up the switching edges a little. Any microwaves
created by the circuit will keep looking until they find a way to mess
you up. One problem is with the input transistors of the op-amp.

Nibbles away at the efficiency. In the end all this needs to be priced
out. If you must go to 6-layer and shields the savings from not having
transformers might evaporate.

 

[MooseFET]

There needs to be something along the lines of system ground above the
flying plane. Either interleaved and well connected ground islands or a
shielded enclosure. It's next to impossible to pass EMC otherwise. BTDT
(or, rather, clients called me with similar designs they had and asked
what to do about their EMC problems).

The shielded enclosure is from what I've seen about the only way to
go. If you put ground on the upper surface of the PCB, it raises the
capacitance too much. The more capacitance you have the bigger the
current pulses are at the edges. These current pulses look for a way
to get out and usually seem to find one.

Nibbles away at the efficiency. In the end all this needs to be priced
out. If you must go to 6-layer and shields the savings from not having
transformers might evaporate.

The example stack up I gave was from real life. I had other things on
the PCB that drove the layer count. It wasn't just for the flying
plane that I needed to go to that many layers.

BTW: I had a shield over the power section and one over the parts
that did RF and the whole thing was in a metal housing. The
connectors were metal backshelled an the two cables that went in and
out were both shielded.

 

[Joerg]

The shielded enclosure is from what I've seen about the only way to
go. ...

Yep, same experience here. Last month I had to convince a client to do
the same on such a switcher. That didn't exactly create outbursts of joy
with the materials guys because it's a custom part. However, there
just isn't really any other low cost way.

... If you put ground on the upper surface of the PCB, it raises the
capacitance too much. The more capacitance you have the bigger the
current pulses are at the edges. These current pulses look for a way
to get out and usually seem to find one.

" ... and usually seem to find one"

Well said. EMI is like water or synthetic oil. You seal over here and
then ... phssss ... it starts to ooze out over there.

[...]

 

[MooseFET]

The shielded enclosure is from what I've seen about the only way to
go. ...

Yep, same experience here. Last month I had to convince a client to do
the same on such a switcher. That didn't exactly create outbursts of joy
  with the materials guys because it's a custom part. However, there
just isn't really any other low cost way.

 ... If you put ground on the upper surface of the PCB, it raises the
capacitance too much.  The more capacitance you have the bigger the
current pulses are at the edges.  These current pulses look for a way
to get out and usually seem to find one.

" ... and usually seem to find one"

Well said. EMI is like water or synthetic oil. You seal over here and
then ... phssss ... it starts to ooze out over there.

[...]

I always say that it is like light and conductors are like ideal
mirrors. The light bounces around until it finds a hole.

RF beads (lossy inductors) help because they are black.

RF can even get out of a sealed box if the box is driven with an RF
current. This isn't much like light or oil.

 

[Joerg]

MooseFET wrote:

[...]

... If you put ground on the upper surface of the PCB, it raises the
capacitance too much. The more capacitance you have the bigger the
current pulses are at the edges. These current pulses look for a way
to get out and usually seem to find one.

" ... and usually seem to find one"

Well said. EMI is like water or synthetic oil. You seal over here and
then ... phssss ... it starts to ooze out over there.

[...]

I always say that it is like light and conductors are like ideal
mirrors. The light bounces around until it finds a hole.

RF beads (lossy inductors) help because they are black.

Dang, some of mine are blue ;-)

RF can even get out of a sealed box if the box is driven with an RF
current. This isn't much like light or oil.

Not if the box is a really good conductor and not very large compared to
the wavelength. Same happens with oil. If you put a strain on the gear
box it'll leak a little more

 

[Klaus Kragelund]

Yep, same experience here. Last month I had to convince a client to do
the same on such a switcher. That didn't exactly create outbursts of joy
  with the materials guys because it's a custom part. However, there
just isn't really any other low cost way.

The initial idea of mine was expanded from the use of the NCP101X
switchers, which are integrated switcher referenced to the switch
node. I did do a 5W design with almost the same PCB stackup as
MooseFet described, but the enclosure was not shielded and we had no
problems passing the EMC. Thats why I would try it for a bit large
power rating (35W). I think the area of the plane will be almost the
same and the rise/fall times also, so I would expect to see around the
same CM noise as the 5W design.

(as far as I remember we used a full ground plane and cutouts in the
inner layers and routing only on the flying layer surrounded by a
small plane)

Regards

Klaus

Regards

Klaus

 

[MooseFET]

MooseFET wrote:

[...]

 ... If you put ground on the upper surface of the PCB, it raises the
capacitance too much.  The more capacitance you have the bigger the
current pulses are at the edges.  These current pulses look for a way
to get out and usually seem to find one.
" ... and usually seem to find one"
Well said. EMI is like water or synthetic oil. You seal over here and
then ... phssss ... it starts to ooze out over there.
[...]

I always say that it is like light and conductors are like ideal
mirrors.  The light bounces around until it finds a hole.

RF beads (lossy inductors) help because they are black.

Dang, some of mine are blue ;-)

Are you sure someone didn't sell you Real Fancy beads?

Not if the box is a really good conductor and not very large compared to
the wavelength. Same happens with oil. If you put a strain on the gear
box it'll leak a little more

A lot depends on how much current is driven into the box. In one case
it was a lot of amps at every multiple of 150KHz. It was a rack
mounted power supply. I had to mount it with plastic hardware to keep
the currents in the case from going through the rack.

 

[Joerg]

The initial idea of mine was expanded from the use of the NCP101X
switchers, which are integrated switcher referenced to the switch
node. I did do a 5W design with almost the same PCB stackup as
MooseFet described, but the enclosure was not shielded and we had no
problems passing the EMC. Thats why I would try it for a bit large
power rating (35W). I think the area of the plane will be almost the
same and the rise/fall times also, so I would expect to see around the
same CM noise as the 5W design.

Congratulations, running something like a NCP1014 with flying ground is
not that easy. AFAIR those operate under 150kHz and thus under the radar
screen with the fundamental but the harmonics can easily get you.
Increased power can increase the conducted readings because it always
finds its way out the mains side via capacitances but if you had 15dB
margins there you'd probably be ok.