Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: XAUI AC coupling

Hi Paul,

Its actually both simpler AND more complicated than what you list here.
Just specifying a cap without also knowing its parasitics will get
you into no end of trouble.

Lets start at the beginning.  The fastest base frequency that you 
can get from an 8B/10B coded signal is when sending a D21.5 character
(which generates a continous alernating 01 bit sequence).  With
a serial signaling rate of 3.125 GBd, this is equivalent to a 1.56 GHz
square wave.

Now the first thing I usually get thrown back at me when I say this is
that "but you need to take into account the odd harmonics of this at
4.7GHz and 7.8GHz.  Bunk.  This is only true if you have a true square
wave with extreemly crisp edges.  In the case of these links, most
can barely get a signal to switch at the 3.125GBd signaling rate--
which is the BEST thing you can have.  Those higher frequency spectral
components CAUSE much of the dispersion in the interconnect.  Any
components present that are faster than 1.56 GHz only serve to degrade 
the link performance.

OK, so now you know where the upper part of the passband needs to be.  The
lower part is much more on the "iffy" side, and you will get some strong
discussions as to how low you need to go.  If you look back at the 
8B/10B characters, the longest run length that you can get is 5 bits.
The K28.7 character, when repeated, actually provides a square wave
consiting of 5-bits low followed by 5-bits high.  If you convert this to
an equivalent frequency, you get 312.5 MHz.  No one will argue that this
this must also be in the passband, however some will state that you also
need to go lower.

Mr. Widemer (one of the creators of the 8B/10B code) made a presentation
to a T11 ad hoc on this a few years back, and at that time was of the 
opinion that you needed to go substantially lower than this.  The reason
was that you can find some sequences of characters, which when strung
together, will actually contain a DC or near-DC offset.  It isn't much,
but it is real.  The problem with this analysis is that it assumes a
transmission interconnect that is lossless and does not degrade the signal.

Unfortunalely, a copper interconnect (which is what a XAUI interface
is specified for) is not lossless.  Even at the short distances targeted
by XAUI, there will be significant phase dispersion, and upper frequency
attenuation (primarily skin-effect loss, with some minor dielectric losses).
In our attemts to correct for these effects, we try to add equalization
to the link, either in the form of precompensation at the transmitter,
or by attenuating the lower frequencies at the receiver (using a high-pass

Both (either?) of these forms of equalization result in reducing the 
amount of low frequency content in the link.  In many of the equalized 
cables (especially the longer ones) the low-frequency attenuation
that must be added to properly compensate for high-frequency losses is
on the order of 7-to-10 dB.

One of the simpler ways to get some of this low-frequency attenuation
is by using a smaller coupling cap.  Since the Xc increases as you
lower the frequency (at least it does when you're below resonance), 
you can actually perform some link compensation just by proper selection 
of a coupling cap.

In the spectral range that XAUI is targeting, you are fortunate that 
you do not need to deal with the high-K titanate dielectrics.  Instead
you get to use the much more stable C0G or NP0 dielectrics.  Within these
you want to find a part with a series resonant frequency (when mounted
on your circuit board) preferably above 1.56 GHz.  You also want to 
limit the attenuation at the low-end (312. 5 MHz).  I'll assume for
now that you are using 50 Ohm lines on your board (100 Ohm coupled).
For myself I generally try to keep the reactance below an Ohm.  For this
operating frequency, the cap would be around 500pF.  But depending on the
amount of EQ present in your link, you may want to make this smaller.

For example, a non-equalized 1GBd link made using a 0.01uF coupling cap, 
will work about 20% farther if you change the cap to 1000pF.  The reason 
is it is performing an equalization function by removing low-frequency
signal components.

There are also manufactures of capacitors that are designed specifically
for DC blocking.  I don't know if its appropriate to name companies here
(so I won't).  Please contact me directly if you want additional information.


Ed Grivna
Cypress Semiconductor
(952) 851-5046

> To all XAUI users vendors,
> My appologies if this question has already been answered.
> AC coupling is part of the XAUI specification but no minimum capacitor 
> size has been included. At some point all XAUI users have to answer 
> the question of what capacitor size to use so I propose that it is 
> included in the specification. The format could be something like:
> Maximum low frequency cut-off caused by AC coupling of XAUI data: ____ Hz.
> Minimum recommended AC coupling capacitor size in XAUI channels: ____ Farads.
> To fill in the blanks above requires some spectral information of the 
> 8B10B data going over the XAUI link. Can anyone help here?
> Paul Wilson
> Nortel Networks