Hi Hongming and Conrad, thanks for this contributions, I was actually working on a similar solution myself. I think an even better approach (thanks Steffen Graber for sharing ideas) would be to send the scrambler seed after the JJJK start. So, the preamble would look like this: JJ, JK, SL, SH, 55*, 55*, 55*, 5D*, ...
Where: SL = low byte of 16-bit scrambler status (5B encoded) SH = high byte of 16-bit scrambler status (5B encoded) * = scrambled preamble (5B encoded)
NOTE: the very first preamble nibble (5B symbol) after the JJ,JK,SL,SH sequence shall be descrambled with the given scrambler status.
This could also be easily extended to cover the PLCA "COMMIT" signal by modifying the PCS RX/TX State Diagrams accordingly. This approach further enhances the scrambler effect, because sending the same packet repeatedly will give different outputs. It would also solve the killer packet problem (which, OK, we don't really have with DME but still...). Note also that we cannot replace any more preamble bytes since this would kill the ethernet preemption support for the point-to-point objective. I think this is yet another reason by which we cannot use the proposed golay preamble as it is defined now anyway.
Please find also my comments to previous discussion marked as [PB] inline, below.
Kind Regards, Piergiorgio
------ Original Message ------
From: "Hongming An" <Hongming.An@xxxxxxxxxxxxx>
To: STDS-802-3-10SPE@xxxxxxxxxxxxxxxxx
Sent: 28/2/2018 17:43:35
Subject: Re: [802.3_10SPE] Scrambler proposal Hongming An
Thanks Conrad!
1) - the first four 4b5b symbols need to be robustly detected to start the scrambler in the receiver
Yes this is valid argument. First for DME theoretically if you see a "0" to"1" transition you will determine the clock edge and data edge therefore recover the data and clock, JJJK has 32bits which should be sufficient for the scrambler after last "K". In fact the PLCA is based on this assumption otherwise it will not work;
[PB] exactly, this is the actual reason we proposed the starting 'J' as it gives three consecutive '0' (clock with no data transition) to synchronize the DME decoder (BTW, the JJJK sequence is 20 bit long).
2) short control frames with the first four 4b5b symbols should also be free of spikes in the emissions PSD
Yes these JJJK is a repeated pattern but just 4bytes which is a very small portion of the rest of preamble and payload. In our sim, the effect of these K codes are take into account.
[PB] We made some EMC/EMI measurements (which will be presented in Rosemont) which also evaluate the effect of the non-scrambled part of the preamble and I can confirm it has negligible effect on the resulting conducted/radiated emissions.
Best regards,
Hongming
-----Original Message-----
From: Zerna, Conrad [mailto:conrad.zerna@xxxxxxxxxxxxxxxxx]
Sent: Wednesday, February 28, 2018 8:30 AM
To: STDS-802-3-10SPE@xxxxxxxxxxxxxxxxx
Subject: [802.3_10SPE] Scrambler proposal Hongming An
Hi Hongming,
your proposal to apply the scrambler earlier in the frame than payload data, is interesting. And I agree, that any preamble should have a good DC balance.
However, you have to take care of two things:
- the first four 4b5b symbols need to be robustly detected to start the scrambler in the receiver
- short control frames with the first four 4b5b symbols should also be free of spikes in the emissions PSD
Regards, Conrad
________________________________________________________________________
To unsubscribe from the STDS-802-3-10SPE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-10SPE&A=1
________________________________________________________________________
To unsubscribe from the STDS-802-3-10SPE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-10SPE&A=1
To unsubscribe from the STDS-802-3-10SPE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-10SPE&A=1 |