[802.3_NGAUTO] tx_rx_channel adhoc - relation to alien crosstalk and review of 802.3bq board noise measurements
All - following up on the discussion today, note that our link segment alien crosstalk leaves a budget of -150 dBm/Hz in it for noise introduced at the receiver between the MDI and the PHY receiver (this is calculated for the 10GBASE-T1 case). This is about 3 dB greater than the alien crosstalk specified to be coupled in at the link segment. Also note that the link segment alien crosstalk specification is just for the cabling (MDI to MDI), and doesn't include coupling within the MDI on a multi-port ECU itself. That coupling would be part of the board's budget, which is a focus of our ad hoc.
What this means is that if the board noise is better (less than -150 dBm/Hz), we could reduce the burden on the cabling and relax the link segment's alien crosstalk specification; BUT, if the board's noise is expected to be significantly worse (more than 1-2 dB) then we really can't fix it by tightening up on the link segment specification.
For example, the board's budget (-150 dBm/Hz) plus the cabling's budget (-153 dBm/Hz) gives a total noise from the link of -148.2 dBm/Hz, which the PHY needs to tolerate at the receiver input. If the board were -148, then you could have NO coupling in the cabling and still have the same level at the PHY's input (and if the cabling stays at -153, the phy input is only -146.8 dBm/Hz - only 1.4 dB worse than spec, and if the cabling were made 3 dB better, -156 dBm/Hz, you get -147.4 dBm/Hz, not much better) - in either case you'd have to recover most of the difference by eating into the PHY's implementation margin - which is going to cost complexity and power. The only real recourse is to make the insertion loss less. All of these things change the balance of relative costs of parts of the system - there is no free lunch for taking shortcuts on the board - think of it like a sensitive radio receiver, because that is what it is.
We faced these same problems in 802.3bq, and chartered a similar ad hoc, the "Channel Ad Hoc" to go out and measure the board noise on high-volume server and switch systems. Most of this work happened at Intel between March 2014 and September 2014, and can be found in the bq public area, as well as the area for the ad hoc (http://www.ieee802.org/3/bq/public/channelmodeling/index.html )
The final ad hoc report is at http://www.ieee802.org/3/bq/public/jul14/cibula_3bq_01_0714.pdf
The ad hoc was also concerned with impulsive noise. The last set of measurements are at http://www.ieee802.org/3/bq/public/may14/cibula_3bq_02_0514.pdf
The methodology is also described, along with measurements at http://www.ieee802.org/3/bq/public/mar14/cibula_3bq_02a_0314.pdf , with follow up at http://www.ieee802.org/3/bq/public/channelmodeling/cibula_3bq_channel_ad_hoc_supplemental_BG_noise.pdf - you will note that these second measurements are more sensitive and take into account the test system's noise levels, which are significant when you are measuring at these sensitive levels. You'll also see noise from other (non-PHY) electronics.
Hope this helps. If anyone needs to get in contact with Pete Cibula (now retired from Intel), please let me know.
-george
George Zimmerman, Ph.D.
President & Principal
CME Consulting, Inc.
Experts in Advanced PHYsical Communications
george@xxxxxxxxxxxxxxxxxxxx<mailto:george@xxxxxxxxxxxxxxxxxxxx>
310-920-3860
________________________________________________________________________
To unsubscribe from the STDS-802-3-NGAUTO list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-NGAUTO&A=1