one more try at my input filter question

[David Collier]

Sorry to get repetitive...
I asked a question above, but I'm not sure I got the specificity of
answer I was hoping for. Could I try again in more concrete terms please

I have a source of DC, 12 to 24V say. I have a switching PSU running off
the DC. The sink takes current in bursts at 2us intervals. The 500KHz is
approximate, but reasonably stable and accurate.

see http://www.Micrel.com/_PDF/mic2198.pdf

I'd like to be able to draw 15W in bursts, so assume average input
current of 1 to 2A on full load.

That causes noise on the wire going back to the supply. That stops me
getting CE approval for the box ;-(

I need to kill the noise.

All the PSU chips which run at fixed frequencies say "this makes the
noise easier to filter" - Hmmm. how exactly?

I appreciate that there will be noise at 2us and harmonics at higher
frequencies. We have already fitted a proprietary input filter, which
sadly seems to provide little attenuation as far down as 500KHz.

BNX002-01 http://www.Farnell.com/datasheets/6821.pdf page 14

It doesn't appear that [ after the filter get to them ] the harmonics
are large enough to cause a problem.

Yes I know, shoot the designer who matched that filter with that
switcher. The bullet is on it's way.

So I have just the one problem - noise at 500KHz.

I COULD treat it by just building/fitting a 'bigger/better' filter with
a cut-off frequency well below 500KHz. But I have a PCB artwork already,
and PCBS built, and that would require making new PCBs. I'd like a
retrofit mod if I can work one out.

Now, back when I did exams, I seem to remember circuits which would
exhibit a high impedance at a single frequency. usually consisting of a
cap and inductor.

something like
http://www.allaboutcircuits.com/vol_2/chpt_8/6.html

Could I "fix" the problem by fitting such a cct? Would the components
required fit inside a small building if I did? Should I stop trying to
be clever?

I can buy a 2A-rated inductor at about 33uH
http://www.Coilcraft.com/pdfs/dc1012.pdf

to get the resonant frequency to be 500KHz I'd need a capacitor of

C= 1/( 4 * PI^2 * L * F^2 ) ???

3nF???

Sounds a bit small... maybe I could use a smaller inductor and a larger
cap?

Maybe I should just fit a 1R/inductor in series and change the input cap
to 220uF low ESR instead of 22uF.

Is there a better way, maybe by building a suitable notch-filter circuit
and fitting it into the DC supply lead outside the box for now?

All help appreciated.

David

 

[MK]

Sorry to get repetitive...
I asked a question above, but I'm not sure I got the specificity of
answer I was hoping for. Could I try again in more concrete terms please

I have a source of DC, 12 to 24V say. I have a switching PSU running off
the DC. The sink takes current in bursts at 2us intervals. The 500KHz is
approximate, but reasonably stable and accurate.

see http://www.Micrel.com/_PDF/mic2198.pdf

I'd like to be able to draw 15W in bursts, so assume average input
current of 1 to 2A on full load.

That causes noise on the wire going back to the supply. That stops me
getting CE approval for the box ;-(

I need to kill the noise.

All the PSU chips which run at fixed frequencies say "this makes the
noise easier to filter" - Hmmm. how exactly?

I appreciate that there will be noise at 2us and harmonics at higher
frequencies. We have already fitted a proprietary input filter, which
sadly seems to provide little attenuation as far down as 500KHz.

BNX002-01 http://www.Farnell.com/datasheets/6821.pdf page 14

It doesn't appear that [ after the filter get to them ] the harmonics
are large enough to cause a problem.

Yes I know, shoot the designer who matched that filter with that
switcher. The bullet is on it's way.

So I have just the one problem - noise at 500KHz.

I COULD treat it by just building/fitting a 'bigger/better' filter with
a cut-off frequency well below 500KHz. But I have a PCB artwork already,
and PCBS built, and that would require making new PCBs. I'd like a
retrofit mod if I can work one out.

Now, back when I did exams, I seem to remember circuits which would
exhibit a high impedance at a single frequency. usually consisting of a
cap and inductor.

something like
http://www.allaboutcircuits.com/vol_2/chpt_8/6.html

Could I "fix" the problem by fitting such a cct? Would the components
required fit inside a small building if I did? Should I stop trying to
be clever?

I can buy a 2A-rated inductor at about 33uH
http://www.Coilcraft.com/pdfs/dc1012.pdf

to get the resonant frequency to be 500KHz I'd need a capacitor of

C= 1/( 4 * PI^2 * L * F^2 ) ???

3nF???

Sounds a bit small... maybe I could use a smaller inductor and a larger
cap?

Maybe I should just fit a 1R/inductor in series and change the input cap
to 220uF low ESR instead of 22uF.

Is there a better way, maybe by building a suitable notch-filter circuit
and fitting it into the DC supply lead outside the box for now?

All help appreciated.

David

Hello David,

Have you tried putting the filter the other way round - it's designed to
protect your ciruit from the PSU noise. This may not help with your pcb.
Try a serious ceramic capacitor across the input terminals so the inductors
in the filter do more good - you can get 10uF in 0805 out of the Farnell
catalogue. I think that might go on the back of your board.
You haven't said if the noise is common mode or not (ie is the same on both
input terminals, in which case the capacitor won't help).
Good luck - let us know what happens.

Michael Kellett

www.mkesc.co.uk

 

[Genome]

Sorry to get repetitive...
All help appreciated.

David

Just to be silly, what happens if you use components you already have....
the output inductor and capacitor, and put those at the input to the
converter.

Oh, and you might need to parallel some low value ceramic stuff as well.

DNA

 

[John Popelish]

Sorry to get repetitive...
I asked a question above, but I'm not sure I got the specificity of
answer I was hoping for. Could I try again in more concrete terms please

I have a source of DC, 12 to 24V say. I have a switching PSU running off
the DC. The sink takes current in bursts at 2us intervals. The 500KHz is
approximate, but reasonably stable and accurate.

see http://www.Micrel.com/_PDF/mic2198.pdf

Yes, buck regulators produce a lot more noise in the input side than
the output side, because the input current is interrupted , but the
output current is continuous, because it passes through the inductor,
all the time.

I'd like to be able to draw 15W in bursts, so assume average input
current of 1 to 2A on full load.

So any input filter has to be efficient (have little resistive drop)
while passing at least 2 amperes.

That causes noise on the wire going back to the supply. That stops me
getting CE approval for the box ;-(

I need to kill the noise.

You need a good quality (low equivalent series resistance)capacitor
directly across the input and common terminals, to supply the input
pulses, and a series inductor to allow that capacitor to sag and
recharge each pulse cycle (because that is the only way capacitors
deliver and recover charge), without letting that capacitor ripple
voltage from also producing a similar ripple further up the line.

All the PSU chips which run at fixed frequencies say "this makes the
noise easier to filter" - Hmmm. how exactly?

Beats me.

I appreciate that there will be noise at 2us and harmonics at higher
frequencies. We have already fitted a proprietary input filter, which
sadly seems to provide little attenuation as far down as 500KHz.

BNX002-01 http://www.Farnell.com/datasheets/6821.pdf page 14

That does not have enough inductance to produce approximately a
constant current into the input bypass capacitor. These two
components must have a resonant frequency well below the pulse rate,
if they are going to produce a low pass filter that covers the
fundamental, as well as the harmonics.

It doesn't appear that [ after the filter get to them ] the harmonics
are large enough to cause a problem.

Yes I know, shoot the designer who matched that filter with that
switcher. The bullet is on it's way.

So I have just the one problem - noise at 500KHz.

I COULD treat it by just building/fitting a 'bigger/better' filter with
a cut-off frequency well below 500KHz. But I have a PCB artwork already,
and PCBS built, and that would require making new PCBs. I'd like a
retrofit mod if I can work one out.

Now, back when I did exams, I seem to remember circuits which would
exhibit a high impedance at a single frequency. usually consisting of a
cap and inductor.

something like
http://www.allaboutcircuits.com/vol_2/chpt_8/6.html

Could I "fix" the problem by fitting such a cct? Would the components
required fit inside a small building if I did? Should I stop trying to
be clever?

I can buy a 2A-rated inductor at about 33uH
http://www.Coilcraft.com/pdfs/dc1012.pdf

to get the resonant frequency to be 500KHz I'd need a capacitor of

C= 1/( 4 * PI^2 * L * F^2 ) ???

3nF???

Sounds a bit small... maybe I could use a smaller inductor and a larger
cap?

Maybe I should just fit a 1R/inductor in series and change the input cap
to 220uF low ESR instead of 22uF.

That would help a lot. The data sheets always show the smallest
capacitor that might work, to make the application look very cheap and
small. But the ESR of the capacitor is as important as the value. I
like to parallel a 1 uF stacked film or X5R multilayer ceramic,
e.g.
http://www.panasonic.com/industrial/components/pdf/abd0000ce8.pdf
http://www.panasonic.com/industrial/components/pdf/abj0000ce4.pdf
(for the higher harmonics), in parallel with a low ESR electrolytic,
rated for the full ripple current, at least.
e.g.
http://www.panasonic.com/industrial/components/pdf/ABA0000CE108.pdf
http://www.chemi-con.com/files/PXA.pdf
Note that he capacitance may have to be quite a bit higher than 22 uF
to get to a ripple current rating above an ampere.

Then select an inductor rated for at least 2 amperes to smooth the
current to that bypass capacitor pair.
e.g.
CDRH127/LDNP-680MC (look in Digikey) 68 uHy, 2.6 amperes
http://www.alfamag.com/PDF/InductorsChokes/Power Inductors & Chokes/SW_150kHz_Jul-03.pdf

The drum core type you asked about is famous for spraying magnetic
field to its surroundings, and radiating noise, this way. This
warning also applies to the inductor in the output filter of the
regulator.

 

[Terry Given]

Sorry to get repetitive...
I asked a question above, but I'm not sure I got the specificity of
answer I was hoping for. Could I try again in more concrete terms please

I have a source of DC, 12 to 24V say. I have a switching PSU running
off the DC. The sink takes current in bursts at 2us intervals. The
500KHz is approximate, but reasonably stable and accurate.

see http://www.Micrel.com/_PDF/mic2198.pdf

Yes, buck regulators produce a lot more noise in the input side than the
output side, because the input current is interrupted , but the output
current is continuous, because it passes through the inductor, all the
time.

I'd like to be able to draw 15W in bursts, so assume average input
current of 1 to 2A on full load.

So any input filter has to be efficient (have little resistive drop)
while passing at least 2 amperes.

That causes noise on the wire going back to the supply. That stops me
getting CE approval for the box ;-(

I need to kill the noise.

You need a good quality (low equivalent series resistance)capacitor
directly across the input and common terminals, to supply the input
pulses, and a series inductor to allow that capacitor to sag and
recharge each pulse cycle (because that is the only way capacitors
deliver and recover charge), without letting that capacitor ripple
voltage from also producing a similar ripple further up the line.

All the PSU chips which run at fixed frequencies say "this makes the
noise easier to filter" - Hmmm. how exactly?

Beats me.

I appreciate that there will be noise at 2us and harmonics at higher
frequencies. We have already fitted a proprietary input filter, which
sadly seems to provide little attenuation as far down as 500KHz.
BNX002-01 http://www.Farnell.com/datasheets/6821.pdf page 14

That does not have enough inductance to produce approximately a constant
current into the input bypass capacitor. These two components must have
a resonant frequency well below the pulse rate, if they are going to
produce a low pass filter that covers the fundamental, as well as the
harmonics.

It doesn't appear that [ after the filter get to them ] the harmonics
are large enough to cause a problem.

Yes I know, shoot the designer who matched that filter with that
switcher. The bullet is on it's way.

So I have just the one problem - noise at 500KHz.

I COULD treat it by just building/fitting a 'bigger/better' filter
with a cut-off frequency well below 500KHz. But I have a PCB artwork
already, and PCBS built, and that would require making new PCBs. I'd
like a retrofit mod if I can work one out.

Now, back when I did exams, I seem to remember circuits which would
exhibit a high impedance at a single frequency. usually consisting of
a cap and inductor.
something like
http://www.allaboutcircuits.com/vol_2/chpt_8/6.html

Could I "fix" the problem by fitting such a cct? Would the components
required fit inside a small building if I did? Should I stop trying to
be clever?
I can buy a 2A-rated inductor at about 33uH
http://www.Coilcraft.com/pdfs/dc1012.pdf

to get the resonant frequency to be 500KHz I'd need a capacitor of
C= 1/( 4 * PI^2 * L * F^2 ) ???

3nF???

Sounds a bit small... maybe I could use a smaller inductor and a
larger cap?

Maybe I should just fit a 1R/inductor in series and change the input
cap to 220uF low ESR instead of 22uF.

That would help a lot. The data sheets always show the smallest
capacitor that might work, to make the application look very cheap and
small. But the ESR of the capacitor is as important as the value. I
like to parallel a 1 uF stacked film or X5R multilayer ceramic,
e.g.
http://www.panasonic.com/industrial/components/pdf/abd0000ce8.pdf
http://www.panasonic.com/industrial/components/pdf/abj0000ce4.pdf
(for the higher harmonics), in parallel with a low ESR electrolytic,
rated for the full ripple current, at least.
e.g.
http://www.panasonic.com/industrial/components/pdf/ABA0000CE108.pdf
http://www.chemi-con.com/files/PXA.pdf
Note that he capacitance may have to be quite a bit higher than 22 uF to
get to a ripple current rating above an ampere.

Then select an inductor rated for at least 2 amperes to smooth the
current to that bypass capacitor pair.
e.g.
CDRH127/LDNP-680MC (look in Digikey) 68 uHy, 2.6 amperes
http://www.alfamag.com/PDF/InductorsChokes/Power Inductors & Chokes/SW_150kHz_Jul-03.pdf


The drum core type you asked about is famous for spraying magnetic field
to its surroundings, and radiating noise, this way. This warning also
applies to the inductor in the output filter of the regulator.

drum core = never, ever use. always have a closed magnetic circuit,
pref. toroid or pot core.


Go further:

- ensure your input LC filter is critically damped, even at no load
(filter resonances can be a real PITA)
- ensure the impedance seen looking from the smps into the filter (C in
parallel with L) is LESS THAN the magnitude of the smps impedance,
Vin^2/Pin. This is to prevent the input filter from causing closed-loop
oscillations in an otherwise stable smps, as smps have negative input
impedances (raise Vin, Iin drops). hence the issue with filter resonance
(parallel L-C resonance gives infinite impedance)

that said, its actually pretty easy to design the input filter. you know
the shape of the smps input current, so you can Fourier transform it
into its harmonics. you also know the emi spec you must meet. So pick a
cap that turns the evil current into not much wobbly DC volts, and an
inductor that turns the wobbly volts into even less wobbly amps, which
feed into 50R (in the LISN) and must be less than the conducted
emissions limits.

I normally do this in spice, as its easy to take into account the cap
and L parasitics, which are important at EMI frequencies, and far faster
than a DIY mathematical analysis.


HTH

Cheers
Terry

 

[Ken Smith]

David Collier wrote: [...]

All the PSU chips which run at fixed frequencies say "this makes the
noise easier to filter" - Hmmm. how exactly?

Beats me.

To the degree that you work at only one running frequency figuring out the
filter is easier.

If the PWM chip had a frequency that varied all over the place, the same
marketing person would find a way to say "spread spectrum" in the data
sheet.

[...]

If this is a big complex PCB with lots of other stuff on it, I suggest you
consider this:

If you can identify the footprint of some part that you can use more or
less as a connector, you can make a small sub-assembly that plugs on there
to carry the better filter.

 

[Ken Smith]

That causes noise on the wire going back to the supply. That stops me
getting CE approval for the box ;-(

Is it noise raditation from the wire or conducted noise on the wire that
you are concerned with. In the past, I have gone so fars as to run power
supply lines as shielded wire or even coax to reduce radiation.

[....]

Now, back when I did exams, I seem to remember circuits which would
exhibit a high impedance at a single frequency. usually consisting of a
cap and inductor.

Parallel tuned circuits in the filter are seldom a good idea. The AC
current in the inductor is large so you need a big core. If you've got a
big core, you could make a larger inductor at a lower ripple current.

Can you place an inductor right at the power entry to the PCB? Some types
of inductors, such as "pot cores" are easy to mount to the sheet metal.
If you've got the room, this may help.

They make PCB mounting coil formers for pot cores and a few otehr types.
You could use one of these to hold a small PCB onto the POT core and
chassis. This will give you room for more capacitors.