Hi,

I work on a project involving an optical link. The transceiver is a
finisar one with a CML interface and AC coupled serial links with
internal 100 ohm differential impedance termination for the receiver of
the optical transceiver. I use one RocketIO in an XC2VP7FF672-6I with a
BREFCLK running at 125 MHZ for a 2.5 Gbit/s serial link. I use the
internal terminations provided with the rocketIO with a 50 ohm value
(see below).
VTTX is 2.5V filtered.

                VTTX
                 |
                 R1
     |\          |                              |\
     | \------------------------------C----|----| \
-----|  >                                  R2   |  >
     | /o-----------------------------C----|----| /
     |/          |                              |/
                 R1
                 |
                 |
                VTTX

R1 is 50 ohms in the Rocketio and R2 is 100 ohms in the transceiver.
All discrete parts are inside either the FPGA or the transceiver. This
is the design I have implemented and the swing seems to be correct out
of the Rocketio (500 mV single ended swing => 1000mv differential
swing). I have a DC level of about 1.8V.
The question is am I doing anything wrong in this implementation ? The
reason for this is that the optical transceiver does not seem to drive
the correct power ie the extinction ratio is very poor (close to 4)
with the low level being around 300uW which is high. If anybody has an
idea or can tell me what I did wrong I would be very thankfull.

Best regards,

JF

Hi,

I have forgotten to mention that the traces between the FPGA and the
transceiver are routed as very loosely coupled traces where each trace
has a 50 ohm characteristic impedance. Each trace is about 5 cm long.

Best regards,

JF

JF,
What sets the DC bias of the CML receiver? I.e. where R2 is.
Syms.
"jfh" <j...@fr.thalesgroup.com> wrote in message 
news:1...@z14g2000cwz.googlegroups.com...
> Hi,
>
> I have forgotten to mention that the traces between the FPGA and the
> transceiver are routed as very loosely coupled traces where each trace
> has a 50 ohm characteristic impedance. Each trace is about 5 cm long.
>
> Best regards,
>
> JF
> 

Hi,

R2 is inside the transceiver so I do not know how to answer this
question. All i know is that normally everything inside the transceiver
is taken care of so I do not have to worry about it (anyway I hope so
!!!).

Best regards,

JF

Symon a =E9crit :

> JF,
> What sets the DC bias of the CML receiver? I.e. where R2 is.
> Syms.
> "jfh" <j...@fr.thalesgroup.com> wrote in message
> news:1...@z14g2000cwz.googlegroups.com...
> > Hi,
> >
> > I have forgotten to mention that the traces between the FPGA and the
> > transceiver are routed as very loosely coupled traces where each trace
> > has a 50 ohm characteristic impedance. Each trace is about 5 cm long.
> >
> > Best regards,
> >
> > JF
> >

jfh wrote:
> [...] the swing seems to be correct out of the Rocketio
> (500 mV single ended swing => 1000mv differential swing).

Howdy JF,

Did you measure this differentially, or just one leg at a time?  If you
don't have a differential probe, get one.  In the mean time, you can
get an approximation by using two probes, placed at the transceiver,
and attempt to verify that the edges of each leg transistion at exactly
the same time.  If the scope has a "Difference mode", where it
subtracts one probe from another, it can come in handy here.

> The question is am I doing anything wrong in this implementation ?

In your other message, you mentioned that the nets "are routed as very
loosely coupled traces where each trace has a 50 ohm characteristic
impedance".  I'd think you'd want a fairly tightly coupled 100 ohm
differential pair.  How loosely are they coupled?  Are the _p and _n
legs close to the same length?

> The reason for this is that the optical transceiver does not seem to drive
> the correct power ie the extinction ratio is very poor (close to 4)
> with the low level being around 300uW which is high. If anybody has an
> idea or can tell me what I did wrong I would be very thankfull

Do you have an optical scope so you can look at the eye diagram?

Good luck,

   Marc

Hi,

I have measured differentially and single ended. The single ended
signal is somehow poorer than the differential one but still the
results are satisfying. I did look at the optical eye diagram and it is
quite closed because of the low ER of the transceiver on my board. BTW,
th _p and _m are closely matched in terms of length but they are quite
far away from each other compared to the height above the reference
planes.

Best regards,

JF


Marc Randolph a =E9crit :

> jfh wrote:
> > [...] the swing seems to be correct out of the Rocketio
> > (500 mV single ended swing =3D> 1000mv differential swing).
>
> Howdy JF,
>
> Did you measure this differentially, or just one leg at a time?  If you
> don't have a differential probe, get one.  In the mean time, you can
> get an approximation by using two probes, placed at the transceiver,
> and attempt to verify that the edges of each leg transistion at exactly
> the same time.  If the scope has a "Difference mode", where it
> subtracts one probe from another, it can come in handy here.
>
> > The question is am I doing anything wrong in this implementation ?
>
> In your other message, you mentioned that the nets "are routed as very
> loosely coupled traces where each trace has a 50 ohm characteristic
> impedance".  I'd think you'd want a fairly tightly coupled 100 ohm
> differential pair.  How loosely are they coupled?  Are the _p and _n
> legs close to the
 same length?
>
> > The reason for this is that the optical transceiver does not seem to dr=
ive
> > the correct power ie the extinction ratio is very poor (close to 4)
> > with the low level being around 300uW which is high. If anybody has an
> > idea or can tell me what I did wrong I would be very thankfull
>
> Do you have an optical scope so you can look at the eye diagram?
>=20
> Good luck,
>=20
>    Marc

"Marc Randolph"
<m...@my-deja.com> wrote in message 
news:1...@z14g2000cwz.googlegroups.com...
>
> In your other message, you mentioned that the nets "are routed as very
> loosely coupled traces where each trace has a 50 ohm characteristic
> impedance".  I'd think you'd want a fairly tightly coupled 100 ohm
> differential pair.  How loosely are they coupled?  Are the _p and _n
> legs close to the same length?
>
Hi Marc,
The coupling between them doesn't matter if the individual lines are 50 ohms 
and the propagation time. How can the electricity know the difference 
between that and a 100 ohm diff pair, or indeed, all the hybrid combinations 
in between?

JF,
You should check the datasheet or call the manufacturer to make sure you 
don't need the put any DC bias on the optics' rx CML port. Normally it's 
something like Vicm, input common mode voltage.

Cheers, Syms. 

"Symon" <s...@hotmail.com> wrote in message 
news:42f0d979$0$18645$1...@news.sunsite.dk...
> The coupling between them doesn't matter if the individual lines are 50 
> ohms and the propagation time is the same for each.

Whoops, typo! 

Symon wrote:
> "Marc Randolph" <m...@my-deja.com> wrote in message
> news:1...@z14g2000cwz.googlegroups.com...
> >
> > In your other message, you mentioned that the nets "are routed as very
> > loosely coupled traces where each trace has a 50 ohm characteristic
> > impedance".  I'd think you'd want a fairly tightly coupled 100 ohm
> > differential pair.  How loosely are they coupled?  Are the _p and _n
> > legs close to the same length?
> >
> Hi Marc,
> The coupling between them doesn't matter if the individual lines are 50 ohms
> and the propagation time. How can the electricity know the difference
> between that and a 100 ohm diff pair, or indeed, all the hybrid combinations
> in between?

Howdy Symon,

The main point of tigher coupling is so that anything that affects one
leg will also affect the other leg - allowing it to be cancelled out by
the differential receiver.  I don't know exactly what his definite of
loosely coupled is, but if it is "loose" enough that an aggressor
signal could affect one leg of the pair much more than the other, it
throws out the main benefit of using differential pairs in the first
place.

> JF,
> You should check the datasheet or call the manufacturer to make sure you
> don't need the put any DC bias on the optics' rx CML port. Normally it's
> something like Vicm, input common mode voltage.

If he's using the more common internally AC-coupled transcievers (which
I got the impression from the first post that he is), any external
DC-biasing is not going to have an affect on his signals.  They do make
DC-coupled transceivers though, so it is worth checking to make sure
that JF didn't end up with some.

I realize the chances of it are very low, but I assume JF has ruled out
the possibility this is just a bad transciever.

Regards,

   Marc

"Marc Randolph"
<m...@my-deja.com> writes:

> Symon wrote:
> > "Marc Randolph" <m...@my-deja.com> wrote in message
> > news:1...@z14g2000cwz.googlegroups.com...
> > >
> > > In your other message, you mentioned that the nets "are routed as
very
> > > loosely coupled traces where each trace has a 50 ohm characteristic
> > > impedance".  I'd think you'd want a fairly tightly coupled 100 ohm
> > > differential pair.  How loosely are they coupled?  Are the _p and _n
> > > legs close to the same length?
> > >
> > Hi Marc,
> > The coupling between them doesn't matter if the individual lines are 50 ohms
> > and the propagation time. How can the electricity know the difference
> > between that and a 100 ohm diff pair, or indeed, all the hybrid combinations
> > in between?
> 
> Howdy Symon,
> 
> The main point of tigher coupling is so that anything that affects one
> leg will also affect the other leg - allowing it to be cancelled out by
> the differential receiver.  I don't know exactly what his definite of
> loosely coupled is, but if it is "loose" enough that an aggressor
> signal could affect one leg of the pair much more than the other, it
> throws out the main benefit of using differential pairs in the first
> place.
> 

But most of the aggressors are likely to be other PCB traces, and they
are *not* going to couple symmetrically onto the pair.  My
understanding is that you just can't couple PCB traces that well.

The main benefit of using differential pairs for signals is off-board
using twisted pair cabling.  So long as your PCB traces match in delay
(not length), you've got what you can out of them.

Cheers,
Martin

-- 
m...@trw.com
TRW Conekt, Solihull, UK
http://www.trw.com/conekt