EMC issue with single board computer

[Mike Warren]

I am using an industrial 5.25" SBC and am having some issues with
EMC. All the problem frequencies are harmonics of the 12.288Mhz
bit clock on the audio chip.

Since high speed digital is not an area I have any experience with
I am hoping someone here will be able to advise me if what I am
seeing is normal or not.

The reference design for the audio chip (ALC655) shows a 22 ohm
series resistor and 22pF capacitor to ground on the bit clock line
but the SBC manufacturer has just used a 33 ohm series resistor.
This resistor is about 50mm from the audio chip.

There is no room for me to fit a capacitor and even if I could I would
want to move the resistor so it was close to the pin of the audio chip.
This is also very difficult since the track disappears into the inner
layers of the PCB right at the chip.

Here are some pictures of what I see on the bit clock line. It seems to
me that the ringing is excessive.

Waveform
http://web.aanet.com.au/pics/M20060815123741-web.jpg

Spectrum
http://web.aanet.com.au/pics/M20060815125554-web.jpg

The supplier responded to my request for help by supplying their
CE and FCC reports which don't show any of my problem frequencies
at all. If I were a cynical type, I might think they had switched the audio
chip off for the test. ;-)

The only options I can see open to me at the moment is the either
switch to another SBC or add an extra shield to the SBC inside
my metal case.

Any suggestions will be appreciated.

-Mike

 

[Eeyore]

I am using an industrial 5.25" SBC and am having some issues with
EMC. All the problem frequencies are harmonics of the 12.288Mhz
bit clock on the audio chip.

Since high speed digital is not an area I have any experience with
I am hoping someone here will be able to advise me if what I am
seeing is normal or not.

The reference design for the audio chip (ALC655) shows a 22 ohm
series resistor and 22pF capacitor to ground on the bit clock line
but the SBC manufacturer has just used a 33 ohm series resistor.
This resistor is about 50mm from the audio chip.

There is no room for me to fit a capacitor and even if I could I would
want to move the resistor so it was close to the pin of the audio chip.
This is also very difficult since the track disappears into the inner
layers of the PCB right at the chip.

Here are some pictures of what I see on the bit clock line. It seems to
me that the ringing is excessive.

Waveform
http://web.aanet.com.au/pics/M20060815123741-web.jpg

Spectrum
http://web.aanet.com.au/pics/M20060815125554-web.jpg

The supplier responded to my request for help by supplying their
CE and FCC reports which don't show any of my problem frequencies
at all. If I were a cynical type, I might think they had switched the audio
chip off for the test. ;-)

The only options I can see open to me at the moment is the either
switch to another SBC or add an extra shield to the SBC inside
my metal case.

What's the source of the 12.288MHz clock ?

If you have a metal case, your emissions must be down to some kind of poor
grounding IME. Where exactly is it getting out ?

Graham

 

[Mike Warren]

What's the source of the 12.288MHz clock ?

The audio chip.

If you have a metal case, your emissions must be down to some kind of
poor grounding IME.

I think that as well but don't know what I can do differently. The board is
mounted straight to the chassis, as is everything else.

The SBC only has 4 mount points, one on each corner. It might be worth
looking at what changes if I don't attach it to the chassis.

Where exactly is it getting out ?

Everywhere it can. The worst area is through the LAN cable. I come out
of the RJ45 on the SBC with a short cable which goes to a socket on my
I/O PCB. This is the same technique used by the board manufacturer
in their sample kit.

Here is a picture of the prototype.
http://web.aanet.com.au/pics/2600-d.jpg

I've tried shielding various cables but nothing makes more than about
1dB of difference. Switching the audio chip off in the CMOS makes all
my problems go away.

-Mike

 

[Dave]

Firstly, know ye that EMC is a very black art !

What exactly are you measuring ? It is important to note the difference
between near and far field. Near field means putting your SA probe directly
on a cable or board. Unfortunately, what you see doesn't tell you much about
far field measurements, where the test receiver is placed a considerable
distance from the 'apparatus'. Far field is what you have to pass for EMC
tests. The bad news is that far field tests are difficult to do without a
quiet site or anechoic chamber, plus a suitable amplifier for your SA.

If far field measurements show that you have a problem, then the fun starts
....

Dave

 

[Mike Warren]

Firstly, know ye that EMC is a very black art !

What exactly are you measuring ? It is important to note the
difference between near and far field. Near field means putting your
SA probe directly on a cable or board. Unfortunately, what you see
doesn't tell you much about far field measurements, where the test
receiver is placed a considerable distance from the 'apparatus'. Far
field is what you have to pass for EMC tests. The bad news is that
far field tests are difficult to do without a quiet site or anechoic
chamber, plus a suitable amplifier for your SA.
If far field measurements show that you have a problem, then the fun
starts ...

I'm using a near field probe kit to localise the source of problem
frequencies discovered during a pre-compliance test in an anechoic
room.

One of the probes is an amplified antenna which I place 1M from
the machine.

Although this isn't calibrated, I see some similarities to the proper
test.

The thing that bothers me is that *all* my problem emissions come
from the same source and yet the SBC has many other potential
sources of problems.

If I didn't need the audio, I would just switch it off in the CMOS config
and be happy.

-Mike

 

[Dave]

I'd guess that the xtal waveform itself is not a major part of the problem.

First thing to do is find how it is getting off the SBC. What happens if you
remove the Audio input and output cables from the SBC ?

Dave

 

[John Woodgate]

In message
<44e14e0e$0$495$61c65585@uq-127creek-reader-03.brisbane.pipenetworks.com.
au>, dated Tue, 15 Aug 2006, Mike Warren

I've tried shielding various cables but nothing makes more than about
1dB of difference. Switching the audio chip off in the CMOS makes all
my problems go away.

Shielding is not appropriate; you need ferrite beads on the cables. But
this is a 'fix', not a proper cure.

I wonder if the audio chip maker 'improved' it and it now has much
faster edges than before. It's happened.

 

[Mike Warren]

In message
<44e14e0e$0$495$61c65585@uq-127creek-reader-03.brisbane.pipenetworks.com.
au>, dated Tue, 15 Aug 2006, Mike Warren


Shielding is not appropriate; you need ferrite beads on the cables.
But this is a 'fix', not a proper cure.

Sorry, I forgot to mention we tried ferrite beads at the pre-compliance
test and they also made very little difference. Even completely removing
all the internal cables except the power and network only dropped the
problem signals by a few dB.

Maybe our case maker has used fake steel.

I wonder if the audio chip maker 'improved' it and it now has much
faster edges than before. It's happened.

It's possible. I would love to be able to see if fitting the recommended
R/C network helps.

I have just been trying some weird things and am amazed by how little
difference anything makes.

First, I made a metal shield to mount over the board without interfering
with the heatsinks. It was grounded at the board mount points. This would
be commercially viable if it worked.

Then I tried putting the board in a bubble wrap bag without grounding it and
finally covered the bag with aluminium foil.

There has to be a clue in how little the signals were affected.

The numbers here are dB relative to the centre of my spectrum
analyser display.

Freq Start Metal Bag Bag+foil
Shield
110.60 +22 +18 +22 +18
122.89 +12 +14 +12 +14
135.17 +8 +12 +10 +10
159.75 +18 +16 +16 +18
172.04 +10 +8 +12 +14
184.33 +16 +14 +16 +16
258.06 +6 +6 +8 +6
282.64 +6 +2 +2 +8
515.99 -12 -12 -12 -12
573.98* -4 -4 -6 -8
749.60 -6 -4 -4 -6
774.18 0 0 +4 0

* This is the only emission not a harmonic of the 12.288MHz
bit clock. It didn't show up on the pre-comp test.

I designed a similar device a couple of years ago using a different
SBC and it passed EMC testing first time.

Looks like I'm going to have to bite the bullet and try another board.
That will however, mean major redesign.

-Mike

 

[Mike Warren]

I'd guess that the xtal waveform itself is not a major part of the
problem.
First thing to do is find how it is getting off the SBC. What happens
if you remove the Audio input and output cables from the SBC ?

Very little change. It seems to me that it is being radiated from the PCB,
not conducted down cables.

There is more info in my reply to John.

At over AUS$325 per hour there wasn't a lot of time to play at the test lab.

-Mike

 

[Mike Warren]

What's the source of the 12.288MHz clock ?

Sorry Graham. Just realised you probably wanted more info.

The ALC655 has a 24.576MHz crystal which I assume is divided
by 2 internally to generate the bit clock. The chip is capable of
using an external bit clock but in this case it is being generated
by the chip.

-Mike

 

[PeteS]

Sorry Graham. Just realised you probably wanted more info.

The ALC655 has a 24.576MHz crystal which I assume is divided
by 2 internally to generate the bit clock. The chip is capable of
using an external bit clock but in this case it is being generated
by the chip.

-Mike

That's an AC97 clock, driven from the ALC device. Assuming you aren't
driving a huge load (it *should* only be the AC97 master), you could
get away with up to 100 ohm series resistance, although it's hard to
know.

I would be tempted to put a 49.9 ohm device in series rather than the
33. If you *did* move the resistor, it would probably do more good near
the destination rather than the source (that varies board to board
though).

If you want a link to the AC97 spec (which has the physical layer) post
again and I'll dig it out (You could simply google, but it's hidden in
the bowels of Intel's site).

Keep in mind that there is data travelling on the SDataout (to sound
device) and SDatain (from sound device) lines and these will
**definitely** give you 12.288MHz harmonics as they will have
(typically) stochastic activity, splattering all over the spectrum.

I would suggest you look closely at those lines too. I had some issues
with my data lines on the AC97 bus I have in my handheld device in the
early stages.

Cheers

PeteS

 

[Mike Warren]

I would be tempted to put a 49.9 ohm device in series rather than the
33. If you *did* move the resistor, it would probably do more good
near the destination rather than the source (that varies board to
board though).

Thanks for the information. I'll give that a go

Keep in mind that there is data travelling on the SDataout (to sound
device) and SDatain (from sound device) lines and these will
**definitely** give you 12.288MHz harmonics as they will have
(typically) stochastic activity, splattering all over the spectrum.

Something else for me to check. Those lines also have 33 ohm
resistors in series. All go to an Intel 6300ESB

-Mike

 

[Eeyore]

The audio chip.


I think that as well but don't know what I can do differently. The board is
mounted straight to the chassis, as is everything else.

The SBC only has 4 mount points, one on each corner. It might be worth
looking at what changes if I don't attach it to the chassis.


Everywhere it can. The worst area is through the LAN cable. I come out
of the RJ45 on the SBC with a short cable which goes to a socket on my
I/O PCB. This is the same technique used by the board manufacturer
in their sample kit.

You put a ferrite lump over that specific cable for sure.

Here is a picture of the prototype.
http://web.aanet.com.au/pics/2600-d.jpg

I've tried shielding various cables but nothing makes more than about
1dB of difference. Switching the audio chip off in the CMOS makes all
my problems go away.

Are all your connector shells firmly bonded to chassis ? That's a classic issue.

Is that a Class D audio amp inside btw ? Whose bits are you using ?

Graham

 

[Eeyore]

Sorry, I forgot to mention we tried ferrite beads at the pre-compliance
test and they also made very little difference. Even completely removing
all the internal cables except the power and network only dropped the
problem signals by a few dB.

Maybe our case maker has used fake steel.


It's possible. I would love to be able to see if fitting the recommended
R/C network helps.

I have just been trying some weird things and am amazed by how little
difference anything makes.

First, I made a metal shield to mount over the board without interfering
with the heatsinks. It was grounded at the board mount points. This would
be commercially viable if it worked.

Then I tried putting the board in a bubble wrap bag without grounding it and
finally covered the bag with aluminium foil.

There has to be a clue in how little the signals were affected.

The numbers here are dB relative to the centre of my spectrum
analyser display.

Freq Start Metal Bag Bag+foil
Shield
110.60 +22 +18 +22 +18
122.89 +12 +14 +12 +14
135.17 +8 +12 +10 +10
159.75 +18 +16 +16 +18
172.04 +10 +8 +12 +14
184.33 +16 +14 +16 +16
258.06 +6 +6 +8 +6
282.64 +6 +2 +2 +8
515.99 -12 -12 -12 -12
573.98* -4 -4 -6 -8
749.60 -6 -4 -4 -6
774.18 0 0 +4 0

* This is the only emission not a harmonic of the 12.288MHz
bit clock. It didn't show up on the pre-comp test.

I designed a similar device a couple of years ago using a different
SBC and it passed EMC testing first time.

Looks like I'm going to have to bite the bullet and try another board.
That will however, mean major redesign.

It's sounds to me as if you're trying to use audio screening style techniques.

Provided that your enclosure is effective, the emissions will be comoing out on
the leads. I know very well that it's popular in audio not to directly ground
the connector shell ( signal low/cold) to chassis but you really have to do this
here.

Does that make sense ?

Graham

 

[John Woodgate]

In message
<44e1b4a6$0$497$61c65585@uq-127creek-reader-03.brisbane.pipenetworks.com.
au>, dated Tue, 15 Aug 2006, Mike Warren

There has to be a clue in how little the signals were affected.

Yes. But without sight of the board, it's difficult to say what it's a
clue to! The 33 ohm being that far from the audio chip suggests a less
than skilled board layout. Maybe a look at the supply line decoupling
of the audio chip, especially at the ground end, might be helpful. Even
check that the ground end really finds ground! No open-circuit via!

 

[Eeyore]

Here is a picture of the prototype.
http://web.aanet.com.au/pics/2600-d.jpg

What material is that ? I'm guessing it's electrogalvanised steel.

If it has a phosphated 'anti-fingerprint' finish, it can be a pig to ground
stuff to it adequately. Just try it with a meter !

Graham

 

[Mike Warren]

You put a ferrite lump over that specific cable for sure.

Putting a ferrite on the external cable at the test lab lowered the
emission slightly but fitting one to the internal cable actually made
the one frequency we were looking at worse. We tried both ends of
the cable. There was no change whether the cable was connected to
the SBC or not. Disconnecting it at the I/O PCB end did make it better.

Are all your connector shells firmly bonded to chassis ? That's a
classic issue.

The sockets are insulated but I have 10nF capacitors from the socket
ground to chassis ground on the PCB next to the sockets. I did try shorting
across the capacitors but there wasn't any difference. I'll try putting foil
between the sockets and the case today.

Is that a Class D audio amp inside btw ? Whose bits are you using ?

Yes. They are made by Hypex in the Netherlands.

-Mike

 

[Mike Warren]

What material is that ? I'm guessing it's electrogalvanised steel.

If it has a phosphated 'anti-fingerprint' finish, it can be a pig to
ground stuff to it adequately. Just try it with a meter !

Zincanneal. Its definitely not 'anti-fingerprint' It's supposed to
have a good conductive surface but I haven't actually checked it.
I'll do that today.

-Mike

 

[Mike Warren]

In message
<44e1b4a6$0$497$61c65585@uq-127creek-reader-03.brisbane.pipenetworks.com.
au>, dated Tue, 15 Aug 2006, Mike Warren


Yes. But without sight of the board, it's difficult to say what it's a
clue to! The 33 ohm being that far from the audio chip suggests a less
than skilled board layout. Maybe a look at the supply line decoupling
of the audio chip, especially at the ground end, might be helpful.
Even check that the ground end really finds ground! No open-circuit
via!

I have tried 2 boards so it's unlikely. One other puzzling thing is the
schematic shows a SM ferrite between the audio chips A_Gnd and
power ground. When I remove that there is still 0 ohms between the
2 grounds. It may be in the chip or there is another ground path.

I'm awaiting a response from the manufacturer on that.

-Mike

 

[Eeyore]

Zincanneal. Its definitely not 'anti-fingerprint' It's supposed to
have a good conductive surface but I haven't actually checked it.
I'll do that today.

Is that a tradename ? Over here in the UK we used 'Zintec' from British Steel
but the Asians don't seem to have any direct equivalent. Either that or the
purchasing guys don't know about it. The phosphating can be an utter PITA.

Graham