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

finally, the right SNR table

Hello 10G'ers,

And finally, the really correct complete SNR table that
should replace the one I presented in Idaho:

    The Complete SNR Table
    Architecture vs Optical SNR

1) 1000BASE-X                                    - 1 dB
2) 8b/10b + 4-WDM                             - 5 dB
3) 10G-BASE-T- EVEN coding        - 5.5 dB
4) 10G-BASE-T- Trellis coding        - 4 dB

The main differences between the two 10 Gbps
solutions appear to be:

a) speed of the electronics and active optics
(1.25 GHz vs 3.125 GHz) (technology, cost)

b) optical fiber (installed fiber  vs higher bandwidth
fiber to allow a reasonable timing budget margin for
the 0.32 nsec bit period solution). Waveshaping is
still an additional option in 10G-BASE-T to reduce
dispersion in the optical fiber and increase the link

The full explanation for the last change (line # 2 in
the Table) follows below. I sent it for double-checking
to the same HP researcher and it looks correct to
him too:

" ... it occurred to me that I might be penalyzing twice
the 8b/10b + 4-WDM approach.

In my presentation, I wrote that this approach incurres
in a signal penalty due to the smaller bit period of 0.32
nsec (compared to the 1000BASE-X reference of 0.8 nsec).
This signal penalty (or difference in bit energies) is 4 dB.

    10 * log (1 * 0.32) = - 5 dB

(compared to the 1000BASE-X reference of
10 * log(1 * 0.8) = - 1 dB).

Then, when I calculted the thermal noise of the
8b/10b + 4-WDM approach I said that we get here
another 4 dB penalty in noise because the bandwidth
is 3.125 GHz, instead of 1.25 GHz:

        10 *log(3.125/1.25) = 4 dB

Therefore, I concluded, the total penalty of 8b/10b + 4-WDM
is 8 dB compared to the reference 1000BASE-X.

Shouldn't I have calculated the  optical SNR of
8b/10b + 4-WDM compared to the 1000BASE-X as
follows ? :

    Signal Power = the same as in 1000BASE-X = - 1 dB

(the power levels of 1000BASE-X and 8b/10b+4-WDM
are the same, only the bit energies are different).

    Noise Power = (4kT/R) * B = (4kT/R) * 3.125 =

             = Noise power of 1000BASE-X  +  10*log(3.125/1.25)

            = Noise power of 1000BASE-X    +  4 dB

Hence, the optical SNR penalty  of 8b/10b + 4-WDM compared
with the reference of 1000BASE-X should be only 4 dB.

Therefore, in the optical SNR table, at the end of the presentation,
the first two entries should be:

1) 1000BASE-X                - 1 dB
2) 8b/10b + 4-WDM        - 5 dB

(and not -1 dB and - 9 dB, respectively, as I wrote)"


Jaime E. Kardontchik
Micro Linear
San Jose, CA 95131
email: kardontchik.jaime@xxxxxxxxxxx