[ibis-macro] Re: Comments on Redriver Flow discussion.

From: "Fangyi Rao" <dmarc-noreply@xxxxxxxxxxxxx> (Redacted sender "fangyi_rao" for DMARC)
To: "wkatz@xxxxxxxxxxxxx" <wkatz@xxxxxxxxxxxxx>, "ibis-macro@xxxxxxxxxxxxx" <ibis-macro@xxxxxxxxxxxxx>
Date: Tue, 16 Mar 2021 10:56:16 +0000

Walter,

There are many problems in the modified Variant 2 flow you proposed. For
example, if Tx2 adapts, Tx2 Init only gets the combined impulse of upstream and
downstream. As a result, Tx2 Init cannot estimate the input signal to Tx2 due
to the lack of the upstream impulse. Therefore,

  *   Tx2 Init cannot adjust input gain properly to minimize compression
  *   Tx2 Init cannot estimate S/N ratio to balance signal and internal noise
  *   Tx2 Init cannot properly optimize output gain to compensate compression
  *   Tx2 Init cannot account for waveform distortion caused by compression
when optimizing EQ

Also, this flow doesn’t work automatically on legacy models (you claimed it
does) because the EDA tool (and most likely model user) doesn’t know whether Tx
adapts or Rx adapts using crosstalk in legacy models.

Fangyi

From: ibis-macro-bounce@xxxxxxxxxxxxx <ibis-macro-bounce@xxxxxxxxxxxxx> On
Behalf Of Walter Katz
Sent: Tuesday, March 09, 2021 4:33 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Comments on Redriver Flow discussion.

CAUTION: This message originates from an external sender.
All,

There was a question today, to paraphrase: “Is an AMI_Init function required to
apply it’s filtering to crosstalk impulse responses?”  The answer to this
question is clearly yes, see pages 201/202 and 225 of the IBIS 7.0 specification

“If present, any filtering in the transmitter and/or receiver AMI_Init
function(s) must also be applied to the crosstalk impulse responses to properly
account for the crosstalk effects.”

“When the Boolean_literal (Init_Returns_Impulse) value is set to “True”, the
model returns the convolution of the impulse response with the impulse response
of the equalization.”

I would interpret “properly account for the crosstalk effects” to include the
LTI filtering in a Terminal Rx, and exclude the DFE filtering of a Terminal Rx,
while the DFE filtering would not be excluded from the impulse response column
one.

I would offer the folowing imporvements to this section of the documentation.

“In order to assure consistant AMI simulation results, all models in a channel
should have Init_Returns_Impulse=True, and have Max_Init_Aggressors>=1”

“An EDA tool may choose to use an a Unit Impulse Response for any one of the
aggressor columns. A Unit Impulse Response is zero for all rows except one row
shall be 1.0/sample_interval. The output impulse response of this column will
be the filtering preformed by the channel used to account for crosstalk
effects.”  (Note, the EDA tool can use this information to eliminate the need
for deconvolution when doing time domain simulation with Init Only models).

“Rx models may adjust their filtering based on the impulse responses in the
crosstalk columns in addition to the first column. Rx models should ignore
crosstalk columns that contain a Unit Impulse Response.”

A new Reserved Parameter (Tx_Adapts (Value True|False) (Usage Info)(Type
Boolean)
(Description “If True then the Tx filtering can adapt to improve the waveform
eye at its downstream Rx”)

A new Reserved Parameter (Rx_Adapts_Using_Crosstalk (Value True|False) (Usage
Info)(Type Boolean)
(Description “If True then the Rx filtering can adapt to improve the waveform
eye at its latch”)

Note: Essentially all existing AMI models are Tx_Adapts=False,
Rx_Adapts_Using_Crosstalk=False.

Change the current Redriver Flow:

[cid:image002.jpg@01D71A15.A6CD5970]
To optionally, either of the following two

[cid:image003.jpg@01D71A15.A6CD5970]
[cid:image004.jpg@01D71A15.A6CD5970]
Note that the first of these is required if Tx_Adapts=True.

The EDA tool is responsible for determing how crosstalk propagates from the
primary Tx models to the terminal Rx models. There are many ways that an EDA
tool can accomplish this task. It is important that Rx models have crosstalk
columns for all aggressors if an Rx model adjusts its equalization based on the
amount of crosstalk (e.g., Max_Init_Aggressors>3).

With this modified Varient 2, then all all the columns (except Hide EQ info?)
under Varient 2 are Yes. Hide EQ info? Is No for Variant 1 since Variant 1
allows using Unit Impulse Response in an aggressor column.

[cid:image005.png@01D71A15.A6CD5970]

Most importantly, the following should be compelling:

  1.  This flow will work with legacy models that have
Init_Returns_Impulse=True.
  2.  Existing EDA flows will continue to work.
  3.  Existing EDA flows will work on models built to support these flows.
  4.  EDA tools will properly support Redriver flows without requiring new
models to support these flows.

Walter

Walter Katz
Work  508.647-7633
Cell      720.417-3762
[Description: Description: Visit MathWorks.com]

References:
- [ibis-macro] Comments on Redriver Flow discussion.
  - From: Walter Katz

[ibis-macro] Re: Comments on Redriver Flow discussion.

Other related posts: