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

Re: [10GMMF] LRM Jitter Tolerance



Title:
David and Nick

The original comment number from draft 2.2 is #67.

Thanks,

Ali

Ali Ghiasi wrote:
David and Nick

Just want to provide you my feedback as I will not be attending the Interim in Corning
and a proposed resolution on the unresolved TR, comment #49 on draft 2.2 regarding
receiver jitter tolerance test (Table 68-6).

First: we need to examine the purpose of jitter tolerance test as exist in FC, 802.3z and .ae.
There are numerous jitter sources in a operating system such as; power supply noise, SSO,
ground bounce, oscillator, magnetic noise, and etc., which often not present during component test.
Jitter tolerance test provide some margin for these elements and to make sure the CDR does not
have a detrimental frequency sensitivity.

Second: It is very difficult and expensive to filter low frequency noise in the transmitter
and limits CDR implementation,  so standards such as FC, 802.3z, 802.3ae, and 802.3aq
specify a golden PLL to tracks / filter some of the low frequency jitter during transmitter
compliance test.  Effectively you are given some credit for the low frequency jitter generated,
but the current LRM receiver may not tolerate the credited jitter or the jitter coming through
the XFI channel!

The 10GBASE-LRM jitter tolerance test is inadequate and leaves the end use shorted!!!
Current draft only specifies two test frequency at 40KHz and 200 KHz with a B2B channel,
since when power supply noise only show up during the B2B test!  The implications are
that the LRM transmitter are tested with 4 MHz CRU which will track and credit the module
for low frequency (<4MHz) jitter, but on the other hand the receivers are never verified with
the same low frequency jitter.

Without a comprehensive stress receiver jitter tolerance you can not build an XFP
implementation of LRM and X2/Xenpak more complex, because the current LRM
receiver has zero margin for low frequency jitter.

My recommendation is to make LRM as robust as other 10Gig E variants such as SR, ER,
LX4, and LR by adopting 802.3ae jitter tolerance mask of Figure 52-4 with high frequency
range > 4MHz at 0.05 UI.  Just as the cases of the above standards the jitter tolerance must
be applied on top of the channel stressor, the sole propose of applying SJ stressor at these
levels to a B2B channel just increases the document size by about 4 lines.

Thanks,
Ali
                                                                                                               



begin:vcard
fn:Ali Ghiasi
n:Ghiasi;Ali
org:Broadcom
adr;dom:;;3151 Zanker Road;San Jose;CA;95134
email;internet:aghiasi@broadcom.com
title:Chief Architect
tel;work:(408)922-7423
tel;cell:(949)290-8103
url:http://www.broadcom.com
version:2.1
end:vcard