Neven,
I've worked with the EMC Engineers at Lexmark who invented the
spread-spectrum clock generator (SSCG), since before they started its
development. I used SSCG in a number of products that I designed at
Lexmark. Below is my understanding of how SSCG works, based on numerous
discussions with the *real* experts.
Consider an ordinary clock generator, that creates a continuous
near-trapezoidal waveform switching between Alow and Ahigh volts at
frequency f1 Hertz. Depending on the duty cycle, rise time, fall time,
shape of the rising edge, and shape of the falling edge, the frequency
spectrum of this clock will have:
* A DC bias, A0 volts, somewhere between Alow and Ahigh volts.
* A fundamental of A1 volts at f1 (and -f1) Hertz.
* A second harmonic of A2 volts at f2 = 2*f1 (and -f2) Hertz.
* A third harmonic of A3 volts at f3 = 3*f1 (and -f3) Hertz.
...
* An Nth harmonic of AN volts at fN =N*f1 (and -fN) Hertz.
...
If we tune over frequency with a peak/quasipeak/average detector having
a bandpass filter with bandwidth BW < f1 Hertz, we will see steady
spikes at frequencies f1, f2, f3, etc. with amplitudes A1, A2, A3, etc.
volts, each having a width BW Hertz and a shape reflecting the filter's
s21 shape factor. (If BW > f1 Hertz, at frequency f Hertz we will see a
composite of all of the spikes between f-BW/2 and f+BW/2 Hertz,
attenuated by the filter's shape factor, which is much messier to deal
with conceptually.)
A given detector has a finite risetime for a signal that appears within
its filter's bandwidth, and an associated falltime when *no* signal is
seen within its filter's bandwidth. For a peak detector, this risetime
is very short, and the falltime essentially infinite. For a quasipeak
detector, the risetime is somewhat longer, and both the risetime and the
falltime are finite, as specified by some standard-- or by the parts
chosen and how the detector is built. The quasipeak detector imitates
the response of human ears and eyes, where a single brief stimulus
doesn't affect us much, but an on-going stimulus-- even at a much lower
level-- bugs the dickens out of us (Chinese water torture). An average
detector has a risetime and a falltime that extend over several (many)
cycles of our clock, so it measures the average value.
With a continuous clock at frequency f1, a peak, quasipeak, and average
detector will all measure close to:
* A1 volts at f1 Hertz.
* A2 volts at f2 = 2*f1 Hertz.
* A3 volts at f3 = 3*f1 Hertz.
...
* AN volts at fN =N*f1 Hertz.
...
Now let us vary (dither) the clock frequency between f1- and f1+ Hertz,
at some reasonably-fast rate, without changing the rising and falling
edges of the clock (i.e. spread-spectrum clock generator (SSCG)):
* The DC bias, A0 volts, stays about the same.
* The fundamental is still about A1 volts, but it varies between
f1- and f1+ Hertz.
* The second harmonic is still about A2 volts, but it varies between
f2- = 2*f1- and f2+ = 2*f1+ Hertz.
* The third harmonic is still about A3 volts, but it varies bwteen
f3- = 3*f1- and f3+ = 3*f1+ Hertz.
...
* The Nth harmonic is still about AN volts, but it varies between
fN- = N*f1- and fN+ = N*f1+ Hertz.
...
Essentially, over time, the fundamental and its harmonics each occupy a
band of frequencies. At high-enough harmonics
(f1+ + f1-)/2
N >= -------------
f1+ - f1-
these bands will overlap one another, but hopefully the amplitudes of
the harmonics have dropped enough that these overlaps won't cause us a
problem.
Instantaneously, if the clock is at frequency f1x, f1- <= f1x <= f1+,
we still have spikes at f1x, 2*f1x, 3*f1x, ..., N*f1x, ...
But if we look at the SSCG clock with a detector with a bandpass filter
set at frequency f Hertz, within one of these frequency bands, either
* The fundamental or harmonic remains within the bandwidth (BW Hertz)
of the filter-- and we measure essentially the same amplitude as for
a constant clock.
OR
* The harmonic flits in and out of the filter's bandwidth, thus we see
only part of the signal.
If we flit through the filter's bandwidth fast enough-- and infrequently
enough-- the detector doesn't have time to respond fully, and it *looks
like* the fundamental/harmonic is much lower than it really is!
Since most of the Radiated and Conducted Emission standards specify
limits based on quasipeak and average detectors, the spread, modulation
frequency, and modulation waveform are all controlled to "fool" the
detector as much as possible, without affecting the functionality of the
product.
In some of the products that my department and I designed at Lexmark,
SSCG lowered our *measured* emissions by up to 15dB over ordinary clock
generators.
To see how well SSCG works on your product, you really need to test it
with an appropriate quasipeak detector.
The frequency of your modulating waveform sounds way too slow for me.
You want to get through the filter's bandwidth as fast as you can, to
keep the quasipeak detector from seeing your signal. You might try to
increase the size of your steps, so that at a frequency that is causing
you grief, no two adjacent steps of the clock frequency are within the
120kHz bandwidth of the quasipeak detector.
John Barnes KS4GL, PE, NCE, NCT, ESDC Eng, ESDC Tech, PSE, Master EMC
Design Eng, SM IEEE
dBi Corporation
http://www.dbicorporation.com/
-
----------------------------------------------------------------
This message is from the IEEE Product Safety Engineering Society emc-pstc
discussion list. To post a message to the list, send your e-mail to
<emc-p...@ieee.org>
All emc-pstc postings are archived and searchable on the web at:
http://www.ieee-pses.org/emc-pstc.html
Attachments are not permitted but the IEEE PSES Online Communities site at
http://product-compliance.oc.ieee.org/ can be used for graphics (in well-used
formats), large files, etc.
Website: http://www.ieee-pses.org/
Instructions: http://listserv.ieee.org/request/user-guide.html
List rules: http://www.ieee-pses.org/listrules.html
For help, send mail to the list administrators:
Scott Douglas <emcp...@radiusnorth.net>
Mike Cantwell <mcantw...@ieee.org>
For policy questions, send mail to:
Jim Bacher: <j.bac...@ieee.org>
David Heald: <dhe...@gmail.com>
