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

[802.3ae] 802.3ae PRBSs are upside down

Calling 10GE chip designers:

Pseudo random bit sequences of length 2^31-1, 2^23-1 are each uniquely
defined in ITU-T's O.150 and implemented in test equipment, as is a PRBS7.
Meanwhile 802.3ae D4.1 has defined:

In clause 49, a PRBS31 which differs from O.150 in that for every 1 in
O.150, D4.1 has a 0.  We think the scrambler polynomial is the same.
In clause 50, a PRBS31 identical to the one in clause 49 but independently

In clause 50, a mixed frequency test pattern based on O.172 and G.957 using
a PRBS23.

In clause 48, a PRBS7, X^7 + X^3 + 1 or X^7 + X^6 + 1, which drives the
A,K,R randomizer.  And has no other purpose in the standard?

The 58 bit scrambler in clause 49 is neither used as a PRBS alone, nor does
it compete with any 58 bit scrambler or PRBS in O.150.  So we are free to
stay with our own choices for this.

Are there any other PRBSs to be considered?

Almost no-one knows exactly what the actual bits in these patterns are,
no-one wants to.  But we all expect that pressing e.g. the PRBS31 button on
a test equipment will get us to a working test.  It would be particularly
inconvenient for manufacturers and networks which use both SONET and
10GEthernet to let this discrepancy continue indefinitely.  Therefore, we
should make the O.150 PRBS31 the preferred variant in our standard.  We
should allow its inverse, the one currently in D4.1, so as not to disqualify
implementers who have implemented it in good faith.  We may wish to define
optional inversion bits in clause 45.  This way, over time, we can all
migrate to the consistent set of PRBSs which are in the test equipment, and
not have to train users, customers and testers in this detail.

To avoid snarl-ups like this we should not define things that are already
defined (just as, "we don't do fiber standards".  We should refer
normatively to O.150 where appropriate, yet repeat the information,
informatively, in our standard to assist the reader. 

Some test equipment can invert its PRBSs.  We don't know if all makes can -
input welcome.

For those who want to know more, O.150 is available for about US$12 from
.150 .  Each PRBS has one run of identical bits as long as the generator
polynomial, of one type only.  For example, PRBS31 has one run of 31 zeroes
per pattern, in the O.150 definition, or one run of 31 ones per pattern, in
the D4.1 definition.

Does anyone know where the PRBS7 in the test equipment is defined?  I
believe some, maybe all, test equipment uses !(X^7 + X^6 + 1) (i.e. contains
a run of 7 zeroes) which is O.150 style, by default.  Are all makes of test
equipment the same on this?  Can all invert the pattern (generating and