spread spectrum cheating

We make a bunch of boxes that go into a semi fab tool. One measures an
optical waveform and shoots it to a bigger box, over three twisted
pairs (clock, data, data) using shielded RJ45 ethernet type stuff.

When we originally did it, they told us we were exempt from ROHS and
EMI standards, but now we aren't. ROHS is no big deal, but the little
box makes a continuous 62 MHz clock, differential at 5 volt swings,
and radiates too much.

We can't lowpass filter the fundamental of course. We can't drop the
amplitude much. A common-mode balun might help some.

So one idea is to spread-spectrum, wobulate the clock frequency or
phase to smear the spectral peak below the CE limits.

Has anyone done this? I wonder how wide a frequency sweep we'd need
but more important is what the equivalent FM modulation frequency
would have to be so the spectrum analyzer never sees the peak spectral
line. Imagine a sawtooth frequency modulation, which turns the
spectral spike into a nice flat plateau. What sort of sawtooth
frequency would work?

My options are to add a modulated phase shifter in the clock path, or
to replace the main XO with a VCO and apply some waveform to the VCO
input to FM the whole FPGA clock and everything. Clock and data would
sweep together, which is kind of nice.

So, how wide and how fast should I sweep?

John Larkin wrote:
> We make a bunch of boxes that go into a semi fab tool. One measures an
> optical waveform and shoots it to a bigger box, over three twisted
> pairs (clock, data, data) using shielded RJ45 ethernet type stuff.
> 
> When we originally did it, they told us we were exempt from ROHS and
> EMI standards, but now we aren't. ROHS is no big deal, but the little
> box makes a continuous 62 MHz clock, differential at 5 volt swings,
> and radiates too much.
> 
> We can't lowpass filter the fundamental of course. We can't drop the
> amplitude much. A common-mode balun might help some.
> 
> So one idea is to spread-spectrum, wobulate the clock frequency or
> phase to smear the spectral peak below the CE limits.
> 
> Has anyone done this? I wonder how wide a frequency sweep we'd need
> but more important is what the equivalent FM modulation frequency
> would have to be so the spectrum analyzer never sees the peak spectral
> line. Imagine a sawtooth frequency modulation, which turns the
> spectral spike into a nice flat plateau. What sort of sawtooth
> frequency would work?
> 
> My options are to add a modulated phase shifter in the clock path, or
> to replace the main XO with a VCO and apply some waveform to the VCO
> input to FM the whole FPGA clock and everything. Clock and data would
> sweep together, which is kind of nice.
> 
> So, how wide and how fast should I sweep?
> 
How badly did you miss the limit?

Cheers

Phil Hobbs

-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

On Wed, 2 Nov 2022 23:03:54 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>John Larkin wrote:
>> We make a bunch of boxes that go into a semi fab tool. One measures an
>> optical waveform and shoots it to a bigger box, over three twisted
>> pairs (clock, data, data) using shielded RJ45 ethernet type stuff.
>> 
>> When we originally did it, they told us we were exempt from ROHS and
>> EMI standards, but now we aren't. ROHS is no big deal, but the little
>> box makes a continuous 62 MHz clock, differential at 5 volt swings,
>> and radiates too much.
>> 
>> We can't lowpass filter the fundamental of course. We can't drop the
>> amplitude much. A common-mode balun might help some.
>> 
>> So one idea is to spread-spectrum, wobulate the clock frequency or
>> phase to smear the spectral peak below the CE limits.
>> 
>> Has anyone done this? I wonder how wide a frequency sweep we'd need
>> but more important is what the equivalent FM modulation frequency
>> would have to be so the spectrum analyzer never sees the peak spectral
>> line. Imagine a sawtooth frequency modulation, which turns the
>> spectral spike into a nice flat plateau. What sort of sawtooth
>> frequency would work?
>> 
>> My options are to add a modulated phase shifter in the clock path, or
>> to replace the main XO with a VCO and apply some waveform to the VCO
>> input to FM the whole FPGA clock and everything. Clock and data would
>> sweep together, which is kind of nice.
>> 
>> So, how wide and how fast should I sweep?
>> 
>How badly did you miss the limit?
>
>Cheers
>
>Phil Hobbs

When we did the box, the customer insisted that all cable shields be
grounded on one end only "to prevent ground loops" ! So we didn't
ground the cat5 shield at the ESM, the little box that transmits the
clock and data. That fails CE test by about 30 dB. The crazies are
gone so we can ground the shield now and fail by maybe 10.

Grounding, and some baluns, might just fix it, but spread spectrum
would help a lot too.

Strangly, we fail at 62 MHz, the clock frequency, and we can't lowpass
filter that.

On a sunny day (Wed, 02 Nov 2022 20:00:00 -0700) it happened John Larkin
<jlarkin@highlandSNIPMEtechnology.com> wrote in
<sha6mhlu5664pfrguubvalkr56i845tan6@4ax.com>:

>We make a bunch of boxes that go into a semi fab tool. One measures an
>optical waveform and shoots it to a bigger box, over three twisted
>pairs (clock, data, data) using shielded RJ45 ethernet type stuff.
>
>When we originally did it, they told us we were exempt from ROHS and
>EMI standards, but now we aren't. ROHS is no big deal, but the little
>box makes a continuous 62 MHz clock, differential at 5 volt swings,
>and radiates too much.

Google double shielded RJ45 cable?

On Thursday, 3 November 2022 at 05:28:19 UTC, Jan Panteltje wrote:
> On a sunny day (Wed, 02 Nov 2022 20:00:00 -0700) it happened John Larkin 
> <jla...@highlandSNIPMEtechnology.com> wrote in 
> <sha6mhlu5664pfrgu...@4ax.com>:
> >We make a bunch of boxes that go into a semi fab tool. One measures an 
> >optical waveform and shoots it to a bigger box, over three twisted 
> >pairs (clock, data, data) using shielded RJ45 ethernet type stuff. 
> > 
> >When we originally did it, they told us we were exempt from ROHS and 
> >EMI standards, but now we aren't. ROHS is no big deal, but the little 
> >box makes a continuous 62 MHz clock, differential at 5 volt swings, 
> >and radiates too much.
> Google double shielded RJ45 cable?

Are you testing to class A or B?  As this is an industrial application you should be
able to go for class A which gives you an extra 9.6dB.
The measurement receiver will almost certainly have a bandwidth of 120kHz plus
a bit extra for the finite slope of the filters, so if you want a 3dB reduction in the
received signal you will need the clock to be uniformly spread over at least a 250kHz
range.  That way, the unwanted signal is within the measurement bandwidth for
only half the time.  This may of course be too much for your system to cope with.
Spread spectrum clocks are much more effective at dealing with high harmonics
than the clock fundamental as the deviation gets higher for the harmonics but the
receiver bandwidth stays the same (up to 1GHz).
How is the box powered?  If there is a separate power supply, then adding some
common mode inductance such as a ferrite ring or clamp to the power cable
should help a lot.
If the box only connects to one other device plus a power supply,  then I don't
understand how grounding the shield at one end only can make any sense.  It
should be solidly grounded at the main equipment end and be connected to the
internal ground reference in the small box.  Any break in the ground structure
to avoid "ground loops" should be obtained by having a floating power supply
if possible.  A simple double insulated wall wart would achieve this.
An alternative way of breaking round loops if the customer really insists is to do
what ethernet often does - put a small capacitor in series with the shield so that
there is minimal 60Hz coupling but the impedance is low at 62MHz.  A ceramic
capacitor selected to be self-resonant at close to 62MHz would be ideal.  Murata
and some others are good about providing such data.
If there is a separate power supply, then keep its cable as close as possible to the
data cable during testing to make the dipole antenna created by the two cables as
small as possible.

John

On Thursday, November 3, 2022 at 2:00:11 PM UTC+11, John Larkin wrote:
> We make a bunch of boxes that go into a semi fab tool. One measures an 
> optical waveform and shoots it to a bigger box, over three twisted 
> pairs (clock, data, data) using shielded RJ45 ethernet type stuff. 
> 
> When we originally did it, they told us we were exempt from ROHS and 
> EMI standards, but now we aren't. ROHS is no big deal, but the little 
> box makes a continuous 62 MHz clock, differential at 5 volt swings, 
> and radiates too much. 
> 
> We can't lowpass filter the fundamental of course. We can't drop the 
> amplitude much. A common-mode balun might help some. 
> 
> So one idea is to spread-spectrum, wobulate the clock frequency or 
> phase to smear the spectral peak below the CE limits. 
> 
> Has anyone done this? I wonder how wide a frequency sweep we'd need 
> but more important is what the equivalent FM modulation frequency 
> would have to be so the spectrum analyzer never sees the peak spectral 
> line. Imagine a sawtooth frequency modulation, which turns the 
> spectral spike into a nice flat plateau. What sort of sawtooth 
> frequency would work? 
> 
> My options are to add a modulated phase shifter in the clock path, or 
> to replace the main XO with a VCO and apply some waveform to the VCO 
> input to FM the whole FPGA clock and everything. Clock and data would 
> sweep together, which is kind of nice. 
> 
> So, how wide and how fast should I sweep?

You probably shouldn't sweep at all, but make the box and it's cables radiate less.

A balanced clock and balanced data shouldn't radiate much at all, and shielded twisted pair ribbon cable with a solid foil shield might well radiate less than RJ45.

Forcing balance with a clamp-on ferrite common mode choke - balun - would help too, as you say.

A 5V swing is probably more than you need. ECL and LVDS both swing through only a volt or so and put a lot less hash on your power supply rail and through your ground returns.

-- 
Bill Sloman, Sydney

On Wed, 02 Nov 2022 20:00:00 -0700, John Larkin
<jlarkin@highlandSNIPMEtechnology.com> wrote:

>We make a bunch of boxes that go into a semi fab tool. One measures an
>optical waveform and shoots it to a bigger box, over three twisted
>pairs (clock, data, data) using shielded RJ45 ethernet type stuff.
>
>When we originally did it, they told us we were exempt from ROHS and
>EMI standards, but now we aren't. ROHS is no big deal, but the little
>box makes a continuous 62 MHz clock, differential at 5 volt swings,
>and radiates too much.
>
>We can't lowpass filter the fundamental of course. We can't drop the
>amplitude much. A common-mode balun might help some.
>
>So one idea is to spread-spectrum, wobulate the clock frequency or
>phase to smear the spectral peak below the CE limits.
>
>Has anyone done this? I wonder how wide a frequency sweep we'd need
>but more important is what the equivalent FM modulation frequency
>would have to be so the spectrum analyzer never sees the peak spectral
>line. Imagine a sawtooth frequency modulation, which turns the
>spectral spike into a nice flat plateau. What sort of sawtooth
>frequency would work?
>
>My options are to add a modulated phase shifter in the clock path, or
>to replace the main XO with a VCO and apply some waveform to the VCO
>input to FM the whole FPGA clock and everything. Clock and data would
>sweep together, which is kind of nice.
>
>So, how wide and how fast should I sweep?

Wobbling the clock frequency to reduce EMI is in fact a standard trick
going back decades, with commodity chips to do just that.

.<https://www.eetimes.com/isscc-spread-spectrum-clocks-mitigate-emi/>

Joe Gwinn

On Thu, 3 Nov 2022 03:45:02 -0700 (PDT), John Walliker
<jrwalliker@gmail.com> wrote:

>On Thursday, 3 November 2022 at 05:28:19 UTC, Jan Panteltje wrote:
>> On a sunny day (Wed, 02 Nov 2022 20:00:00 -0700) it happened John Larkin 
>> <jla...@highlandSNIPMEtechnology.com> wrote in 
>> <sha6mhlu5664pfrgu...@4ax.com>:
>> >We make a bunch of boxes that go into a semi fab tool. One measures an 
>> >optical waveform and shoots it to a bigger box, over three twisted 
>> >pairs (clock, data, data) using shielded RJ45 ethernet type stuff. 
>> > 
>> >When we originally did it, they told us we were exempt from ROHS and 
>> >EMI standards, but now we aren't. ROHS is no big deal, but the little 
>> >box makes a continuous 62 MHz clock, differential at 5 volt swings, 
>> >and radiates too much.
>> Google double shielded RJ45 cable?
>
>Are you testing to class A or B?  As this is an industrial application you should be
>able to go for class A which gives you an extra 9.6dB.
>The measurement receiver will almost certainly have a bandwidth of 120kHz plus
>a bit extra for the finite slope of the filters, so if you want a 3dB reduction in the
>received signal you will need the clock to be uniformly spread over at least a 250kHz
>range.  That way, the unwanted signal is within the measurement bandwidth for
>only half the time.  This may of course be too much for your system to cope with.
>Spread spectrum clocks are much more effective at dealing with high harmonics
>than the clock fundamental as the deviation gets higher for the harmonics but the
>receiver bandwidth stays the same (up to 1GHz).

250K is well inside the sweep range (some do an octave) but outside
the modulation range of most little commercial VCOs, but maybe I can
find one, to replace the 125 MHz XO in the box.

The two knobs to turn, to fool the spectrum analyzer, is how wide a
range to sweep over, and what is the sweep rate? An SA has both a
front-end bandwidth and "video bandwidth", essentially lowpass
filtering after the detector.

Another option is a circuit that phase modulates the 62.5 MHz clock
after the FPGA generates it. Shades of the old "phasing method" of FM
generation. 


>How is the box powered?  If there is a separate power supply, then adding some
>common mode inductance such as a ferrite ring or clamp to the power cable
>should help a lot.

The customer provides 24 volts DC, and they insisted I not ground the
low side so I have an isolating dc/dc converter. The claim is that
they are not seeing EMI from the power cable.

>If the box only connects to one other device plus a power supply,  then I don't
>understand how grounding the shield at one end only can make any sense.

It never made sense. One fix is to change the RG45 connector and the
box end plate to get a good shield ground, but for some reason the
customer says that's not quite good enough.

Of course, customers have been known to be wrong before.


>It
>should be solidly grounded at the main equipment end and be connected to the
>internal ground reference in the small box.  Any break in the ground structure
>to avoid "ground loops" should be obtained by having a floating power supply
>if possible.


The PCBs in both boxes are hard grounded to our grounded metal
enclosures. The "ground loop" concept was stupid. There are giant
metal pipes between the relevant parts of this machine and they make
"ground loops" too. Not to mention all the electrical power conduits
and the building structure steel.

>  A simple double insulated wall wart would achieve this.
>An alternative way of breaking round loops if the customer really insists is to do
>what ethernet often does - put a small capacitor in series with the shield so that
>there is minimal 60Hz coupling but the impedance is low at 62MHz.  A ceramic
>capacitor selected to be self-resonant at close to 62MHz would be ideal.  Murata
>and some others are good about providing such data.
>If there is a separate power supply, then keep its cable as close as possible to the
>data cable during testing to make the dipole antenna created by the two cables as
>small as possible.
>
>John

My real question remains, what sweep width and modulation do I need to
spread the EMI spectrum enough to fool a CE-test-class spectrum
analyzer?

If the SA RF bandwidth is 125K, I might sweep 1 MHz p-p to help. The
digital stuff might tolerate that.

On 03/11/2022 03:00, John Larkin wrote:
> We make a bunch of boxes that go into a semi fab tool. One measures an
> optical waveform and shoots it to a bigger box, over three twisted
> pairs (clock, data, data) using shielded RJ45 ethernet type stuff.
> 
> When we originally did it, they told us we were exempt from ROHS and
> EMI standards, but now we aren't. ROHS is no big deal, but the little
> box makes a continuous 62 MHz clock, differential at 5 volt swings,
> and radiates too much.
> 
> We can't lowpass filter the fundamental of course. We can't drop the
> amplitude much. A common-mode balun might help some.
> 
> So one idea is to spread-spectrum, wobulate the clock frequency or
> phase to smear the spectral peak below the CE limits.
> 
> Has anyone done this? I wonder how wide a frequency sweep we'd need
> but more important is what the equivalent FM modulation frequency
> would have to be so the spectrum analyzer never sees the peak spectral
> line. Imagine a sawtooth frequency modulation, which turns the
> spectral spike into a nice flat plateau. What sort of sawtooth
> frequency would work?
> 
> My options are to add a modulated phase shifter in the clock path, or
> to replace the main XO with a VCO and apply some waveform to the VCO
> input to FM the whole FPGA clock and everything. Clock and data would
> sweep together, which is kind of nice.
> 
> So, how wide and how fast should I sweep?

Mock it up on the bench with a signal generator and a spectrum analyser.
(any old frequency will do you can scale the results)

Try a triangle wave at say 1% of f0 to get a feel for it. That should 
give nicer behaviour than a sawtooth. Discontinuities always hurt.

I seriously doubt if you can modulate it in such a way that will be a 
recognisable clock signal that will get the fundamental down by 30dB.

Board redesign so the RF hot traces are between the PSU planes might 
help. Is something pulling serious current at that clock frequency?

-- 
Regards,
Martin Brown

On a sunny day (Thu, 03 Nov 2022 07:51:04 -0700) it happened John Larkin
<jlarkin@highlandSNIPMEtechnology.com> wrote in
<q5k7mh9tchu4r7vtm7pnk1drcbtfvs5ikt@4ax.com>:

>On Thu, 3 Nov 2022 03:45:02 -0700 (PDT), John Walliker
><jrwalliker@gmail.com> wrote:
>
>>On Thursday, 3 November 2022 at 05:28:19 UTC, Jan Panteltje wrote:
>>> On a sunny day (Wed, 02 Nov 2022 20:00:00 -0700) it happened John Larkin 
>>> <jla...@highlandSNIPMEtechnology.com> wrote in 
>>> <sha6mhlu5664pfrgu...@4ax.com>:
>>> >We make a bunch of boxes that go into a semi fab tool. One measures an 
>>> >optical waveform and shoots it to a bigger box, over three twisted 
>>> >pairs (clock, data, data) using shielded RJ45 ethernet type stuff. 
>>> > 
>>> >When we originally did it, they told us we were exempt from ROHS and 
>>> >EMI standards, but now we aren't. ROHS is no big deal, but the little 
>>> >box makes a continuous 62 MHz clock, differential at 5 volt swings, 
>>> >and radiates too much.
>>> Google double shielded RJ45 cable?

You mentioned 62 MHz,
the 74HC4046A at 5 V goes that high. 
Modulate F with VCO pin 9
So modudate with a gaussian ;-)

But I do not like the idea of frequency changing clocks, screening is possibly better.


>My real question remains, what sweep width and modulation do I need to
>spread the EMI spectrum enough to fool a CE-test-class spectrum
>analyzer?

62 MHz? RTL_SDR stick and xpsa:
 http://panteltje.com/panteltje/xpsa/index.html


>If the SA RF bandwidth is 125K, I might sweep 1 MHz p-p to help. The
>digital stuff might tolerate that.

How do they make the real spread spectrum stuff?
Noise generator, low pass, VCO?