Re: 8b/10b and EMI
I think you missed Joe's point. He makes reference to a 1-mm hole.
that is the size hole you would have if you routed only the
bare fiber through the hole. This implies usage of a buried
module, or a module with a fiber pigtail. In either case, the
fiber is routed through a metal plate, significantly removed from
the LASER diode and driver. The plate is generally constructed as
a buried wall in the chassis, with a second bulkhead used to mount
> Dear Joe,
> > Joe Gwinn writes:
> > Yes. All it takes is a small enough hole, and even 10 gbaud will be
> > confined. If the transmitter and receiver are correctly designed, the
> > photons will pass through a 1-mm or 2-mm diameter hole in a metal wall,
> > which will cut off everything below a few hundred gigahertz: (3*10^8
> > m/s)/(10^-3 m)= 3*10^11 Hz= 300 GHz.
> I don't think it is so simple. Your analysis might be correct for pure
> wave propagation, but it is common for the voltage waveform on the laser
> diode to capacitively coupling through the dielectrically loaded hole
> onto the fiber ferrule. Many fiber connectors use metal ferrules, which
> make a dandy little antennas.
> It is quite difficult to make a well grounded 2-mm hole. Usually the
> hole in the main panel is much bigger, accommodating the entire module
> dimension, and a metal shim in the module acts as the 2-mm shield. For
> this shield to work, it needs a good seal to the front panel. Many
> vendors try to do this with spring loaded flanges, etc., but some
> customers prefer to keep system-ground separate from chassis ground.
> This causes many module vendors to capacitively couple the occluding
> shim rather than actually connecting it to the system ground. Of course,
> the coupling cap has a series inductance, and really doesn't do much good
> above a 1G or so.
> Even at 1Gb/s, the various ground loops intrinsic to this fiddly game
> make it difficult for most designs to solidly meet EMI regulations
> under a wide range of construction practices.
> Combine this with the fact that radiation from a current loop
> is proportional to (Iloop*Area*frequency^2), and you have a problem
> that is 100x more difficult at 10G than 1G. Anyone who had a mild
> headache at 1G will definitely be doing some tough engineering at 10G.
> Rick Walker