Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Thank you Sudeep. What strikes me, is that we are trying to create a set of specifications that are sufficient to guarantee interoperability. In radio land, the standards are “necessary” but not sufficient to guarantee a working system/interop. The results below and in the referenced presentation indicate that the link met BER requirements yet some failed TDECQ. What I didn’t see is a bench mark of the actual receiver. If the actual receiver is much better than the oscilloscope (better equalizer, lower noise floor, better bandwidth – merely speculating here) that makes the TDECQ measurement, then I might expect to see similar results, i.e. better receiver performance makes up for degraded transmitter performance. Why not have a TDECQ requirement that is “necessary” but not sufficient to guarantee performance. (wait a minute, am I a Tx vendor or Rx vendor…..) From: Sudeep Bhoja [mailto:sbhoja@xxxxxxxxx]
Bill, We would then start discarding good optical transmitters for failing TDECQ specs due to specifying a 5 taps T/2 equalizer.
See
http://www.ieee802.org/3/bs/public/17_05/way_3bs_01a_0517.pdf. Clearly 7-9 T spaced taps are required. Sudeep
From:
Bill Kirkland <wkirkland@xxxxxxxxxxx> Ali, in reference to 5 taps, I think it is in part about creating a level playing field for folks using different technologies and processes. (I still claim ignorance , but I believe not everyone is using CMOS/Bi-CMOS and some folks might not even be trying to use an FFE in the Rx). I would hope that the reference receiver architecture would not “force” certain RX technology decisions unnecessarily. From: Ali Ghiasi
[mailto:aghiasi@xxxxxxxxx] Gary The fundamental problem we have based on Marco’s results is that even for 5 T spaced FFE the TDECQ is about 3.4 dB over the current limit of 2.5 dB, effectively we are short of 0.9 dB. Winston presentation indicates changing the reference EQ form 5 tap to 7 tap T spaced EQ improves TDECQ significantly. These results all were taken with 38.9 GHz filter, I expect if the results were taken with 26.55 GHz filter the amount of improvement from 5 taps to 7 taps would improve even more. We have a BW problem here and we should be solving the BW issue using an equalizer with sufficient length! Having more data will help how to better move forward, but going back to the 5 T/2 FFE will put us even in worse predicament. We need to consider 7 T-spaced FFE, increase TDECQ to 3.0 dB, and improve unstressed receiver sensitivity by 0.5 dB. Forcing TDECQ to 2.5 dB will impact yield and cost of the the transmitter as it was illustrated in Winston presentation. The worst part is that you end up trowing away good transmitters that exceed the required BER when tested with link partner as the commercial receivers have more capabilities than the reference equalizer. It has been mentioned several times that we should limit the TDECQ filter to 5 taps because analog implementation are limited
to 5 taps, here are some references that show longer analog FFE can be implemented 9 tap 32 Gb/s FFE filter in Bi-CMOS 7 tap 40 Gb/s FFE in 65 nm CMOS Thanks,
|