Hello John, Javier,

you may get bogus numbers when calculating jitter from phase  noise
measurements with limited bandwidth.

For example what about an asyncronous (deterministic jitter) component
sitting at 103KHz when you only measure with a 100KHz bandwidth? Depending on  the
level of this 103KHz signal the resulting jitter could look horrible on a
scope, but would be completely invisible on a 100KHz BW measurement. Such a
103KHz spur may come from switching power supplies etc, and have a  detrimental
effect on the signal.

For example, I have a Wenzel Ultra low noise OCXO which is extremely low
noise (around -170dBc/Hz floor) but at the same time it has a large amount of
noise at around 620MHz or so. I attribute this noise to the digital CMOS  gates
they seem to use in the buffer stages, followed by low-pass filters that  are
not attenuating properly at those high frequencies. It can be easily seen on
a spectrum analyzer for example.

BTW: I asked Wenzel about this, they said "we see this too, it's probably
caused by mixing artifacts in the spectrum analyzer itself". I don't agree since
the noise goes away if I use a 50MHz low pass filter on the signal..

Will such higher frequency noise/jitter affect your system? That  depends on
your systems' requirements.

Generally, when measuring jitter it always looks better to measure  with a
limited bandwidth such as 10MHz or 20MHz as used on PN equipment such as
E5052B's etc.

To give you an example, I have measured the same 10MHz OCXO source  with both
a limited 10MHz BW and a ~2GHz bandwidth Wavecrest Jitter Analyzer.  The
result:

Jitter as indicated based on PN measurement: ~350fs rms

Jitter as indicated by the Wavecrest: ~8ps rms (noisefloor of  <3ps).

That's a big difference, and depending on your application could result in  a
significant increase in your theoretical Bit Error Rate etc.

Another item to keep in mind: even if your application has limited  bandwidth
(say a receiver with only 5MHz IF bandwidth etc) you may get aliasing  back
into your bandwidth of interest by the mixing of two deterministic noise
sources etc. So if your source has noise at say 600MHz due to the 60th overtone  of
your crystal, and your system happens to have some noise at 603MHz, you may
get a signal down at 3MHz in your IF due to the non linearity of the amps  etc.

In my opinion it is always better to verify PN based jitter measurements
with a wide-bandwidth SA and/or wide-bandwidth jitter analyzer rather than just
rely on over simplistic PN-to-jitter calculations such as are available online
(Wenzel website etc). While they are useful, these may not tell you the
entire  story.

bye,
Said

In a message dated 10/13/2008 12:26:04 Pacific Daylight Time,  jmiles@pop.net
writes:

Normally  the PN reaches a broadband floor determined by the circuit's  own
limitations or its semiconductor process.  This happens between  100 kHz and
10 MHz depending on what's generating the signal.  So you  wouldn't want to
extrapolate the slope indefinitely.

A high-quality  crystal oscillator's broadband floor will be sufficiently
quiet (typically  better than -160 dBc/Hz) that it won't matter much whether
you integrate  out to 100 kHz or to 1 MHz.  The difference will be on the
order of  attoseconds.  When making the measurement, you'd typically place
the  upper integration cursor one decade into the broadband-floor region,  and
call it good.

**************New MapQuest Local shows what's happening at your destination.
Dining, Movies, Events, News & more. Try it out!
(http://local.mapquest.com/?ncid=emlcntnew00000001)

Said, what is the operating principle of the Wavecrest Jitter Analyzer?
Is it a sampling scope like the ones Agilent and Tektronix offer? How
does it compare to them? I've googled it but I only found a Chinese site
with little information in English. Another advantage I see for
time-based measurements is that you can go arbitrarily low in offset
from the carrier. In the 5052B you have a PLL with non-zero bandwidth,
so below some offset you probably get an optimistic estimate because the
PLL is actually following the noise. Another item to bear in mind, if my
understanding is right, is that integrating the phase noise plot gives
you the rms jitter between your noisy waveform and the perfect sinewave,
or rather an approximation of it, which is what comes out of the PLL in
say the 5052B. A sampling scope typically measures the time jitter
between two (noisy) rising edges of a clock waveform. In the simplistic
case of white phase noise this should give you a factor sqrt(2) more
jitter than the loose PLL measurement, right?
Thanks again,
Javier