Fangyi,
I think we have to revisit the straw poll. We will discuss in today’s ATM
meeting.
Walter
Walter Katz
Work 508.647-7633
Cell 720.417-3762
From: ibis-macro-bounce@xxxxxxxxxxxxx <ibis-macro-bounce@xxxxxxxxxxxxx> On
Behalf Of Fangyi Rao (Redacted sender "fangyi_rao" for DMARC)
Sent: Tuesday, March 9, 2021 2:13 PM
To: Walter Katz <wkatz@xxxxxxxxxxxxx>; ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: Redriver Flow - I totally agree that the flow in
Fangyi's BIRD 210 is correct, it just does not need two additional columns
in the Impulse Matrix or a new reserved parameter
Walter,
Variant 3 is a superset of Variants 1 and 2 and has the benefits of both, as
summarized in the ATM link. In Variant 3 the EDA tool can still execute
Variant 2 if it only uses information in the two additional columns.
Below is the straw poll result on Jan 26.
Straw poll:
Ambrish - Variant 1
Michael - Variant 3
Radek - Variant 3
Todd - Abstain
Arpad - Variant 3
Randy - Abstain
Justin - Abstain
Walter - Variant 3
Bob - Abstain
Curtis (by mail) - Variant 3
The results of the straw poll were:
Variant 1 - 1
Variant 2 - 0
Variant 3 - 5
Abstain - 4
Fangyi
From: Walter Katz <wkatz@xxxxxxxxxxxxx <mailto:wkatz@xxxxxxxxxxxxx> >
Sent: Tuesday, March 09, 2021 4:17 AM
To: Fangyi Rao <fangyi_rao@xxxxxxxxxxxx <mailto:fangyi_rao@xxxxxxxxxxxx> >;
ibis-macro@xxxxxxxxxxxxx <mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: Redriver Flow - I totally agree that the flow in Fangyi's BIRD
210 is correct, it just does not need two additional columns in the Impulse
Matrix or a new reserved parameter
Fangyi,
I think we need to revisit that decision. Can you please explain any
advantage of Variant 3 over Variant 2, with the clarifications to Variant 2
I have given below?
Walter
Walter Katz
Work 508.647-7633
Cell 720.417-3762
From: Fangyi Rao <fangyi_rao@xxxxxxxxxxxx <mailto:fangyi_rao@xxxxxxxxxxxx> >
Sent: Tuesday, March 9, 2021 3:46 AM
To: Walter Katz <wkatz@xxxxxxxxxxxxx <mailto:wkatz@xxxxxxxxxxxxx> >;
ibis-macro@xxxxxxxxxxxxx <mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: Redriver Flow - I totally agree that the flow in Fangyi's BIRD
210 is correct, it just does not need two additional columns in the Impulse
Matrix or a new reserved parameter
Walter,
What you proposed is basically Variant 2 we discussed at ATM meetings back
in Jan. Back then the ATM group compared Variants 1 and 2 and decided to
combine these two into Variant 3, based on which BIRD 210 was written.
Comparison between Variants 1 and 2 was documented in the following ATM
link.
http://ibis.org/atm_wip/archive/20210120/fangyirao/AMI%20Redriver%20Flow/AMI_redriver_flow_2021_v0.pdf
<https://urldefense.com/v3/__http:/ibis.org/atm_wip/archive/20210120/fangyirao/AMI*20Redriver*20Flow/AMI_redriver_flow_2021_v0.pdf__;JSU!!I5pVk4LIGAfnvw!xbzGPNDJLYGBJDYrzl_0PjcRswi2SsL_iGvWgv9hnrvbaL7qzqA2A4KQrp7Piy-ojJc$>
Fangyi
From: ibis-macro-bounce@xxxxxxxxxxxxx
<mailto:ibis-macro-bounce@xxxxxxxxxxxxx> <ibis-macro-bounce@xxxxxxxxxxxxx
<mailto:ibis-macro-bounce@xxxxxxxxxxxxx> > On Behalf Of Walter Katz
Sent: Monday, March 08, 2021 6:06 AM
To: ibis-macro@xxxxxxxxxxxxx <mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-macro] Redriver Flow - I totally agree that the flow in
Fangyi's BIRD 210 is correct, it just does not need two additional columns
in the Impulse Matrix or a new reserved parameter
CAUTION: This message originates from an external sender.
All,
The statistical redriver flow in BIRD 210 specifies that the impulse
response input to each AMI_Init is the cumulative impulse response of
everything upstream, and that the output contains the models equalization
applied to that input impulse response. This is correct. There is no need to
add a column to the impulse response to do this, just use column 1.
The handling of crosstalk needs to be done by the EDA tool. It needs to find
the impulse response of all paths of the connection from all Tx to it’s Rx,
and to the Rx’s it aggressors. All the EDA tool needs to do this is the
equalization of all of the models between the primary Tx (included) and the
terminal Rx (excluded). The equalization of all models (except the terminal
Rx) is determined by initialing the second column in the Impulse Matrix to a
Unit Impulse Response. The output of the second column of the Impulse Matrix
is the impulse response of the equalization of that model. The “aggressors”
argument of these calls to AMI_Init will be 1. The Impulse Matrix input to
the terminal Rx will have “aggressors” equal to the actual number of Tx
aggressors to this Rx. The terminal Rx will apply the “LTI” equalization
(generally the equalization before the Rx DFE) to these aggressor impulse
response columns.
All legacy models support this flow. Models that are not terminal Rx already
apply their equalization to all of the columns of the Impulse Matrix, and
terminal Rx apply just the LTI part of their equalization to the aggressor
columns. There is no need for a reserved AMI parameter that says this, and
certainly no need to add additional columns beyond the number of
“aggressors”.
We have seen one Tx model in the past 12 years that optimized its FFE taps
based on the impulse response of the downstream channel. This feature was
never used because it over-equalized the channel and gave overall poorer
results. Thus the need for back channel which is addressing this.
I am OK if IBIS decides it needs a new parameter to support this flow, and
document this flow by replacing the flow in column 1 and indicating that
“aggressors” can include columns that the EDA adds to extract the impulse
response of model. But lets keep the number of columns in the impulse matrix
= aggressors+1.
Walter
Walter Katz
Work 508.647-7633
Cell 720.417-3762