RE: [SI-LIST] : Re: (2) Positive ECL (PECL) voltage swings,

L D Miller (ldmiller@nortelnetworks.com)
Thu, 29 Jul 1999 16:29:15 -0700

We have encouraged/demanded that our fiber optics transceiver vendors put
as much of the transceiver-SERDES coupling network inside their package
(talking about 1 X 9 packages here) as possible.

What that means is that the TX+ and TX- (transmitter) input signals have
the .01 uF capacitors, and any terminating [100 ohms differential] and
biasing circuitry needed, in the module (and inside the module electrical
shield). Thus, the circuit from the SERDES output is simply the pulldown
resistors (we use 200 ohms or so) right at the SERDES TX pins and 50 ohm
traces arranged to have very small area--i.e, parallel traces, going right
into the fiber optics transceiver TX pins.

Similarly, the fiber optics transceivers contain the RX+ and RX- output
components: pulldowns, and the .01 uF coupling capacitors. Thus, the
circuit to connect to the SERDES is just 50-ohm parallel traces running
from the transceiver RX+ and RX- pins over to the SERDES RX inputs, with a
pair of series-connected 50 ohm resistors across the inputs at the SERDES
[100 ohms differential], with the centertap formed by the two resistors AC
grounded with .01 uF or so.

With is arrangement we have the absolute minimum number of parts and traces
to radiate at 1250 MHz and harmonics, and we can also manage to avoid any
extra vias-- above and beyond what is required for automated testing-- we
put TX on one side of the PCB, and RX on the other side. Also, we are very
careful to keep the RX and TX pairs as close to the same lengths as=
possible.

Meeting FCC/IEC emission specs is actually fairly difficult with these
units, and we arrived at these requirements through bitter, expensive
experience. Meanwhile, over the last year+, the transceiver manufacturers
have quieted down their units significantly, so we now usually have 8-10 dB
margin over the FCC "A" limits.

We are using the same arrangement with Small form Factor (SFF) Gb
transceivers; as I said, there seems to be some concensus that this is a
preferred arrangement.

Hope this helps,

Larry Miller
At 01:51 PM 7/29/99 -0700, you wrote:
>Larry,
>
>Please explain your "these coupling caps be located in the fiber optics
>transceiver package for reduced EMI" comment.
>
>Jim Knighten
>________________________________________________________
>Dr. Jim Knighten e-mail: Jim.Knighten@SanDiegoCA.NCR.com
><mailto:Jim.Knighten@SanDiego.NCR.com> =20
>Senior Consulting Engineer
>NCR
>17095 Via del Campo
>San Diego, CA 92127 http://www.ncr.com <http://www.ncr.com> =20
>Tel: 858-485-2537
>Fax: 858-485-3788
>
>***** Notice the Area Code change from 619 *****
>
>
> -----Original Message-----
> From: ldmiller@nortelnetworks.com
>[mailto:ldmiller@nortelnetworks.com]
> Sent: Thursday, July 29, 1999 11:50 AM
> To: si-list@silab.eng.sun.com
> Subject: [SI-LIST] : Re: (2) Positive ECL (PECL)
>voltage swings, but designed to be AC coupled
>
> (Sorry, I hit the SEND button prematurely)
>
> I believe I can answer some of that, at least for 1 Gb
>Ethernet.
>
> (A) The input circuits for fiber optics transceivers and
>SERDES inputs need
> to be common-mode DC biased to set the input stage operating
>point. In
> "classical PECL" this would normally be Vcc-2.0 VDC.
>However, the various
> transceiver manufacturers and SERDES manufacturers do not
>agree on what
> this voltage should be.
>
> Therefore, if you use a "classical" 50-ohm
>thevenin-equivalent DC circuit
> (usually 82 ohms and 130 ohms) many combinations of fiber
>optics
> transceiver and SERDES are not happy, ranging from poor
>noise margins to
> total non-operation.
>
> Since virtually all of these ICs have internal bias
>generators we have
> found that the best thing is to simply AC couple the ciruits
>and let each
> component go to its own DC "sweet spot".
>
> (B) We have always insisted (and the manufacturers seem to
>slowly be coming
> around to this as a standard) that these coupling caps be
>located in the
> fiber optics transceiver package for reduced EMI. Also, the
>SERDES
> manufacturers all seem happy with using a pair of 50 ohm
>resistors on their
> inputs, with the center of the resistors tied to a 0.1 uF
>capacitor to
> produce a 50-ohm AC termination and let the IC set its own
>DC operating point.
>
> Hope this helps,
>
> Larry Miller
> Nortel Networks
>
> At 11:28 AM 7/29/99 -0500, you wrote:
> >
> >
> >
> >All,
> >
> >Will some one shed some light on why differential lines (to
>and from) a ECL
> >device are AC coupled? I've been unsuccessful at finding
>an answer from my
> >internet searches.
> >
> >I am speaking specifically to the AC coupling that is found
>in fibre-channel
> >ser-des transmitters.
> >
> >(P)ECL C1
> >|\-------------o-----||---
> >| \ |
> >| / | C2
> >|/o------o-----------||---
> > | |
> > | |
> > < R1 >R2
> > > <
> > | |
> > | |
> > _|_ _|_
> > GND GND
> >
> >Where C1=3DC1=3D0.01uF
> >
> >Why is the capacitor needed? What is the theory behind
>this implementation?
> >
> >Thanks,
> >
> >--------------------------------------------------------
> > _ __ Salvador Aguinaga Jr.
> >/_/ )/ _| Carrier Systems Group=20
> >=B7 R&D
> >__ |( (_ _ _ Mt. Prospect, Illinois
> >\ \_)\__|(_)|)) (847) 222=B72833
> >
> >
>
> **** To unsubscribe from si-list: send e-mail to
>majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
>si-list, for more help, put HELP. si-list archives are accessible at
>http://www.qsl.net/wb6tpu/si-list ****
>
>**** To unsubscribe from si-list: send e-mail to
majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
si-list, for more help, put HELP. si-list archives are accessible at
http://www.qsl.net/wb6tpu/si-list ****
>
>
>

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****