From: Mike Monett
X-Mailer: Mozilla 2.02 (Win16; I)
Subject: Re: Twisted Pair High Data Rate Design
References: <3DAB9294.5753@Spam.Bots> <3DAD69CA.firstname.lastname@example.org>
Date: Thu, 17 Oct 2002 00:07:03 -0400
NNTP-Posting-Date: Thu, 17 Oct 2002 00:07:01 EDT
Organization: Bell Sympatico
> The scope photos in the data sheet are pretty funny! The scope
> photos of "before equalization" show that the eye-pattern is gone
> due to the high-freq attenuation, in this case very severe. Look
> on page 5 of the data sheet. You can see the progression of the
> eye pattern degradation at various data rates. 100 meters of CAT5
> cable has around 22 dB attenuation at 100 MHz. The data sheet is
> showing a data rate of 311 Mb/s (equivalent to 155.5 MHz). Thus, I
> would expect the data sheet scope photos to be correct.
> To understand why the eye pattern looks so bad, you must draw out
> signals on paper and superimpose them on each other. Draw 1x,
> 0.5x, 0.333x, .... data rates plus the inverted signals. Be sure
> to include the rise time of the cable which will effectively give
> you attenuation at the higher freqs. You will easily see why the
> eye closes up with the high freq attenuation of the cable. You
> will see a very small amplitude 1x data rate signal, larger 0.5x
> data rate signal, even larger 0.333x data rate signal, .... At
> really low data rates, you get resistive losses only.
> The tests we ran were over 10 to 20 meter cables at 400 to 500
> Mb/s. In our tests we could still make out the eye in the eye
> pattern. The CLC014 was pretty magical in beautifying the eye
> I suggest getting the evaluation board that National puts out.
> Connect a pseudo-random signal generator (you can make one out of
> some shift registers) at one end of the cable and the evaluation
> board at the other end.
> Mark Chun
> Santa Barbara, CA
Thanks for the reply. Yes, the photos are puzzling. For example, I'm
looking at the top photo on page 5 of
It shows "Before Equalization: 100m of Belden 8281 Coaxial Cable"
The waveform is very noisy, but the transitions are clearly visible.
This shows the scope is triggered properly.
It suddenly dawned on me why the transitions look so noisy. The
pattern is pseudo-random, which contains various length runs of ones
and zeros as well as short patterns of alternating bits.
The losses due to skin effect in the coax causes high frequency
rolloff, which changes the amplitude of the waveform according to
the data pattern. Short runs of alternating bits have low amplitude,
long runs of ones and zeros have higher amplitude.
So the entry point on the left side of the eye pattern depends on
the pattern that preceded each trace.
If the scope were real-time, like a 7104, there would be many
individual traces with different amplitudes that make up the eye
However, the sampling scope changes the appearance to look like
random dots, similar to Gaussian noise.
I'm used to looking at the read signals of a hard disk drive. If
they were as bad as most of the pictures, no amount of equalization
would recover the data. In fact, the high frequency boost shown
later in the datasheet would just increase the noise.
But the signals on the coax start out in the hundreds of millivolts
instead of hundreds of microvolts, so equalization will work fine.
OK, I'm happy now:)