From: Larry Miller (firstname.lastname@example.org)
Date: Mon Feb 26 2001 - 13:56:46 PST
I have used this termination extensively; it is the "standard" one for
gigabit Ethernet fiber optics transceiver receiver to SERDES interfaces.
Many transceivers now have the coupling capacitors built in, so you do not
need them if your transceiver has them (check the transceiver data sheets).
Originally the transceivers were all DC-coupled (standard PECL outputs) and
you had to supply a Thevenin equivalent load and
DC bias network. This was SUPPOSED to be the usual Vcc-2V circuit that you
can get with voltage dividers.
The problem was that the SERDES inputs from different manufacturers (and
sometimes different models or production runs from the same manufacturer)
required differing DC bias voltages, none of which were the standard Vcc-2V.
This rapidly turned into a manufacturing nightmare.
A meeting of SERDES vendors was held by Dr Johnson (he was technical editor
on the Gigabit Ethernet fiber IEEE 802.3z committee) and at that meeting it
came out that virtually all of the standard SERDES chips had their own
receiver input bias generators built in. So the easiest termination to use
was an AC one: 50 ohm resistors from each differential input line to a
common AC ground (the center tap cap). This lets each and every SERDES
vendor and part's internal DC bias generator set the DC level to the "sweet
spot" for that receiver. The AC coupling in each line lets each and every
transceiver vendor have ITS own DC levels on the other side of the caps.
So much for history. Please note the following:
1. The center-tap capacitor value is not critical, since it sees only the
unbalance (common-mode) signal, which the SERDES is supposed to ignore. This
signal is typically comprised of mostly glitches during signal transitions.
What is important is that it be a low-inductance circuit. This is typically
satisfied by using C0603 surface mount caps with short traces to the ground
plane. The package inductance (low) is the main thing.
2. This is a termination scheme cooked up ONLY for the receive end of a
differential line that is going into a differential PECL receiver. It may
not be suitable for the transmitter of a fiber optics transceiver, nor for a
SERDES technology backplane transmitter. This is not to say that it cannot
work, but that it would be uncommon. The main reason is that it provides no
DC bias (which was the idea in the first place); this was desirable in a
receiver but probably not for a transmitter.
3. This circuit was intended for PECL edge-coupled (loosely-coupled) traces.
That is, it primarily provides the 50-ohm termination to ground and
secondarily provides the 100 ohm line-to-line termination through the AC
ground path. This means that LVDS circuits may not like it, nor, perhaps,
broadside-coupled (closely-coupled) differential pairs.
4. The AC coupling capacitors can be a small value because of the 8B/10B
encoding used in GbE; it has a run length of 3 to 5 bit times (maximum
number of consecutive 1's or 0's) so you do not need to worry about code DC
component nor baseline wander. This would not apply to uncoded data or to
codes that do not closely control the DC component like 8B/10B.
[mailto:email@example.com]On Behalf Of Stockalis, Anthony
Sent: Friday, February 23, 2001 2:36 PM
Subject: RE: [SI-LIST] : Differential End Termination
The suggested termination scheme is a .01uF series capacitor on each line
into the load along with a 50ohm .01uF AC terminator to gnd. The other
termination scheme is to have .01uF series capacitors into the load and a
100ohm resistor between the two pairs to match with the 50 ohm impedence of
the trace. I have contacted the manufacturer, but I am waiting to get a
response. I wanted to see if anyone had tried using either of these
BXR-48000 Hardware Team
From: Fred Balistreri [mailto:firstname.lastname@example.org]
Sent: Friday, February 23, 2001 4:12 PM
To: Stockalis, Anthony
Subject: Re: [SI-LIST] : Differential End Termination
Hello Anthony, the two resistors with a common capacitor method will
work. However I would get better clarification with what the IC chip
manufacturer told you. The 50 ohms with a .01uf capacitor acting as AC load
work. The .01uf they are talking about may be in series with the driving
device to block any DC currents. This is common on some Transeiver chips
don't like DC currents on the input. In this case the terminating resistor
across the differential and should be choosen for the differential impedance
connecting transmission lines not necessarily 50 ohms. I don't believe that
manufacturer would make such a mistake. The problem may be in communication.
sketch/schematic of what they are saying for clarification.
"Stockalis, Anthony" wrote:
> I am investigating the ac end termination of a differential data signal.
> have reviewed section 6.4 of Howard Johnson's "Handbook of Black Magic" on
> ac end terminators, and it seems that two resistors to a common capacitor
> ground is the way to go. The recommendation of the chip manufacturer is
> have a 50 ohm to .01uF capactor end termination on each line of the
> differential pair. The chip is a Gigabit Ethernet Chip running at
> This capacitance value seems rather high to me, and I believe that the
> common capacitor would be the correct solution. I was wondering if anyone
> else has any valuable experience doing ac end terminations on differential
> data lines. I have used the Mentor Graphics Interconnectix simulator, but
> have not gotten very conclusive results because a random data signal
> be used in the tool. Any help would be appreciated.
> Anthony Stockalis
> Marconi Communications
> BXR-48000 Hardware Team
> email: email@example.com
> Phone 724-742-6124
> **** To unsubscribe from si-list or si-list-digest: send e-mail to
> firstname.lastname@example.org. In the BODY of message put: UNSUBSCRIBE
> si-list or UNSUBSCRIBE si-list-digest, for more help, put HELP.
> si-list archives are accessible at http://www.qsl.net/wb6tpu
-- Fred Balistreri email@example.com
**** To unsubscribe from si-list or si-list-digest: send e-mail to firstname.lastname@example.org. In the BODY of message put: UNSUBSCRIBE si-list or UNSUBSCRIBE si-list-digest, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu ****
This archive was generated by hypermail 2b29 : Tue May 08 2001 - 14:30:56 PDT