# [10GBT] PMA Training signal

```Hi all,

I apologize. I discovered an error with my PMA training signal proposal in slide 8-10
of my startup presentation entitled "Startup protocol for 10GBASE-T", thanks to comments from Brett at SolarFlare.

In my original proposal, training seaquense is generated
from side-stream scrambler as follows

Syn[0] = Scrn[0]
Syn[1] = g(Syn[0])
Syn[2] = g(Syn[1])
Syn[3] = g(Syn[2])        when loc_rcvr_status = NG
Syn[3] = g(Syn[2])^Syn[0] when loc_rcvr_status = OK
g(x)=x^3+x^8

These 4 bit seaquense are mapped into 2PAM symbol.

I misunderstood that we can distinguished between pair A,B,C and D through try and error
even when channel skew and pair swaps are present.
Because I misunderstood that there is only one pair combination wicth satisfy following relationship
Ry(n)[1]+Ry(n-3)[0]+Ry(n-8)[0]=0
Ry(n)[2]+Ry(n-3)[1]+Ry(n-8)[1]=0
Ry(n)[3]+Ry(n-3)[2]+Ry(n-8)[2]=0 when loc_rcvr_status = NG
Ry(n)[3]+Ry(n-3)[2]+Ry(n-8)[2]+Ry(n)[0]=0 when loc_rcvr_status = NG

Ry(n)[x]: PAM2 demapping deta of lane x

However, Brett found the relationship between pair D and pair A when loc_rcvr_status = NG.
That relationship is Sy(n+1)[0]+Sy(n-3)[3]+Sy(n-8)[3]=0.
So, I realize that my original proposal can't distinguish ABCD(normal) from BCDA, CDAB, and DABC
when channel skew is present.

On the other hand, 1000BASE-T had a mechanism to easily identify pair A.
This was done by transmitting the inverse of Syn[1],Syn[2] and Syn[3] during time n+1.

Below are a detail explanation by Brett,
"In 1000BASE-T 3PAM was used for training.
The 4 scrambler sequences were used to generate 2-level sequences, but the 2 levels were (0,-2).
An additional polarity scrambler randomly flipped the polarity of the -2 to get a 3-level signal.

The important thing to note is that although the scrambler sequences were used to directly generate
the 2-level sequences for pairs A, pair B,C & D was different.

At even time periods (n = 2*k, for integers k) pair A is the same as the scrambler sequence).
At odd time periods (n = 2*k+1, for integers k) pair A is the inverse of the scrambler sequence from time n-1.
From this construction, pair B,C & D will always have a 1,0 or 0,1 at times n=2*k and 2*k+1.
From this pair A can be identified."

I didn't apply 1000BASE-T approach, because pair B,C and D are the inverse of the scrambler sequence
from time n-1 at odd time periods. In this case, these signals has higher transmitting PSD at high freaquency.
So, these signal should be scrambled by another scrambler such as polarity scrambler in 1000BASE-T.
This might bring us increasing complexity in polarity correction, deskew and pair swapping by 2PAM training signal.
On the other hand, 1000BASE-T can correct polarity due to its 3PAM training signal.

So, I would like to modify my proposal so that pair D is distinct from pairs A,B,C

Original:
Syn[3] = g(Syn[2])        when loc_rcvr_status = NG
Syn[3] = g(Syn[2])^Syn[0] when loc_rcvr_status = OK

Modified:
Syn[3] = g(Syn[2])^Syn[0] when loc_rcvr_status = NG
Syn[3] = g(Syn[2])^Syn[1] when loc_rcvr_status = OK

From this, my modified proposal can determine the order of A, B, C & D.

Anyway, my proposal is just an idea to start discussion about PMA training signal.
I am welcome refinement to my proposal and other more efficient proposal.

Could you take a vote on providing corrected presentations on the website
during Friday's closing session?

Regards,
Seki

Katsutoshi SEKI
k.seki@necel.com

-----Original Message-----
Sent: Wed, 15 Sep 2004 16:59:30 -0700
Subject: Re: [10GBT] Reminder: Presentation submittal deadline
This is a multi-part message in MIME format.

A clarification so that there is hopefully less confusion on this:
Addition of supporter names is permitted after the presentation