[time-nuts] Comparison of Logic Standards for Clock Distribution
stephan at rrsg.ee.uct.ac.za
Wed Oct 25 07:06:37 EDT 2006
Thank you for the answer and also for the attached measurements. The
measurements I find particularly interesting and note worthy.
It seems that digital signals provide high noise immunity in the sense that
one relies only on one dimension to convey information: the cross-over
point. The signal is amplified and amplitude limited to such an extent that
amplitude noise should have little bearing on the cross-over point. All in
all, this works very well for conveying information in digital "word" form
as long as you obey the setup and hold times of the particular hardware.
However, as you move higher up in frequency it is exactly this hard
switching that turns around to bite you: in the form of all sorts of
parasitic effects. And, constantly we are finding more ways to cure and make
amends to these parasitics because it is so convenient to transmit data in
bit form. For example: As long as "101010" is received at the other end
there has been no "signal degradation". As long as the amount of jitter,
skew, etc. is still within the setup and hold times of the logic.
Now, I realise that I state the obvious in the above. But, my point is that
clock distribution is actually an analog problem. Fundamentally, one is not
interested in setup and hold times. Instead, in a very idealistic way, one
is very concerned with getting the clock to its destination with no
degradation in that precious cross-over point. The down-side of this is that
with an analog signal information is conveyed (the cross-over changes)
through both amplitude and frequency modulation. (Not that these effects are
completely independent of one another in any how)
It didn't come as a surprise that the transmissions of sine waves are
superior to digital signals. I guess for best results that sine wave is
converted to digital (squared off) only at the final stage before entering
the digital system.
Am I correct in my view of clock signals or am I just living blissfully
ignorant? Can you recommend any reading on this subject?
> -----Original Message-----
> From: time-nuts-bounces at febo.com [mailto:time-nuts-bounces at febo.com] On
> Behalf Of SAIDJACK at aol.com
> Sent: 24 October 2006 07:40 PM
> To: time-nuts at febo.com
> Subject: Re: [time-nuts] Comparison of Logic Standards for Clock
> Hello Stephen,
> to help with your question, here is my 2-cents from empirical experience:
> Fundamental Sine waves cause the lowest amount of noise since they do
> (should) not include any harmonics, and mixing products that can arise
> from these
> harmonics. Thus sine waves are used extensively in T&M measurement
> to distribute 5 or 10MHz references over cable.
> They also have the benefit of allowing the longest cable runs, since
> signals are comprised of harmonics (particularly 3rd and 5th harmonic)
> are affected much more by cable dispersion, skin effect, and other
> than a fundamental Sine wave is.
> That said, LVDS is a much more modern standard than PECL or ECL. It was
> specifically designed for low-power (laptop and HDTV) signal transmission
> really cheap cables between a graphics chip and an LCD panel such as an
> HDTV. You
> may get cheaper, and better chips when using LVDS. PECL and ECL were not
> specifically designed to drive long cables I believe, rather they are PCB
> backplane specific.
> Attached are two phase-noise measurements of our new Jackson-Labs Fury
> which has both digital and analog 10MHz outputs. Both outputs drive about
> 4-feet of Belden 316 cable, then go into a TSC5120A analyzer. You can see
> a much
> more significant spur spectrum on the digital signal, there are
> no spurs at all on the Sine wave cable!
> I think this is not only caused in the digital driver chip, but also in
> ADC circuits on the receiving side due to harmonics on the square wave all
> way up to 1GHz...
> BTW: our FireFox GPSDO referenced signal generator has LVDS, PECL, CMOS,
> Sine wave outputs all driven in paralell.
> The cable used is also very important, since cheap RG58 has very poor
> shielding, lot's of dispersion and a high loss-tangent etc...
> hope this helps,
More information about the time-nuts