RE: 8b/10b and EMI
Brad,
RE:
Does scrambling give better spectral density than 8b/10b? I was under the
impression that scrambling has longer run lengths of 1's or 0's compared to
8b/10b.
Response:
If data and control are scrambled before encoding then the output of the
encoder will appear as a random stream of encoded characters. In this case
the rf spectrum will be that of random encoded data. But in addition the
properties of the of code in terms of run length, disparity, clocking
content etc., are all maintained. This principle is well known and has been
used in IEEE 802.12 for 5b6b encoding and IEEE 1394b for 8b10b encoding. In
both cases the reason for doing this was to reduce the spectral peaks
associated with idle and other repeating control signals to minimize EMI.
Although in theory EMI can be avoided by careful control of differential
signals, differential balance etc., it can never be removed in practice. The
Gigabit Ethernet (GbE) Idle is a common test signal and from experience for
large GbE switches even with good design the idles cause major EMI concerns.
10GbE has the opportunity to avoid this issue by scrambling repeating
patterns like idle. This will greatly ease product EMI qualification and
design.
RE:
Does this not make it harder to perform clock and data recovery
with just a single scrambler, hence the reason SONET uses two scramblers?
Response:
Clock recovery is just the same as for the normal 8b10b case. The only
questions to be answered are what type of scrambler to use, how to
synchronize the Tx and Rx scramblers and what part of data or control
actually requires the scrambling function.
The second scrambler was added to SONET essentially to remove the
possibility of user defined data locking the scrambler or causing clock
recovery failures. The transition density and run length of SONET systems
are not constrained by the scrambling operation other than in a statistical
sense.
Best regards,
David.
-----Original Message-----
From: Booth, Bradley [mailto:bradley.booth@intel.com]
Sent: Wednesday, March 15, 2000 11:23 AM
To: stds-802-3-hssg@ieee.org
Subject: RE: 8b/10b and EMI
Does scrambling give better spectral density than 8b/10b? I was under the
impression that scrambling has longer run lengths of 1's or 0's compared to
8b/10b. Does this not make it harder to perform clock and data recovery
with just a single scrambler, hence the reason SONET uses two scramblers?
Just curious,
Brad
Brad Booth
bradley.booth@intel.com <mailto:bbooth@Level1.com>
Intel Network Interface Division, Austin Design Center
(512) 407-2135 office
(512) 589-4438 cellular
-----Original Message-----
From: Tom Truman [mailto:truman@lucent.com]
Sent: Wednesday, March 15, 2000 10:13 AM
To: Ed Grivna
Cc: stds-802-3-hssg@ieee.org
Subject: Re: 8b/10b and EMI
<< File: Card for Tom Truman >> Ed,
Thanks for the response.
If 8b/10b were to be scrambled, then it would appear
to me that all it is providing at the XAUI interface is
packet delineation
and some error monitoring capability. I imagine that each
lane would need
a separate scrambler/descrambler, initialized to different
states so that
the transitions across the lanes are uncorrelated.
Synchronizing these
scramblers,
and deskewing the lanes would require some thought -- it
isn't difficult,
but it isn't as straightforward as the "alignment column"
proposed for HARI.
At that point, the 25% overhead of the 8b/10b scheme
seems to be a staggering price to pay for delineation and
error monitoring -- why not start with scrambling, at a
lower baud rate, and
make the overall design problems simpler?
Best regards,
Tom Truman
Lucent Technologies