Hi Vladmir,
I thank you for the reply in sharing your thoughts on sampling when in the
AMI_Init.
Below is my reply (orange) to the points brought up by you.
Vladmir: If EDA tool and AMI model use different approaches we may expect this
little timing mismatch. For typical 32 samples per UI we expect about 0.03UI
mismatch, which might be acceptable, however it could be more with coarser
granulation.
Hansel:
* Good point! We are proposing the model report to the EDA tool a delay
between the stimulus and decision point.
* Let me setup an offline meeting with you in the week after Good
Friday in going over a few slides in gathering feedback.
Vladmir: I think that sampling point should be reported by the model in time
units (not the sample number),...
Hansel:
* Agreed! The unit should be in time (seconds) and not index.
Glad to hear no one is proposing to have a cursor sampling point with regards
to an impulse response given impulse response does not hold any UI information.
:)
Wish you a happy weekend ahead! Stay safe and healthy given the current events
in the world.
Regards,
Hansel Dsilva
From: Dmitriev-Zdorov, Vladimir <vladimir_dmitriev-zdorov@xxxxxxxxxx>
Sent: 01 April 2020 02:40 AM
To: Hansel Dsilva <hanseldsilva@xxxxxxxxxxxx>; Walter Katz
<wkatz@xxxxxxxxxxxxx>; ibis-macro@xxxxxxxxxxxxx
Subject: RE: Gathering feedback. In the AMI_Init flow, the RX IBIS-AMI model to
output the cursor sampling point with regards to a pulse response or step
response?
Hansel,
We all know that from the impulse response (IR, a response to Dirac pulse) we
can restore step response by integrating IR over time. Then the pulse response
(or unit symbol response) can be found as a difference between the step
response and its copy delayed by single UI. AMI Init function knows symbol
interval and time resolution therefore has enough data to translate found
sampling on the equalized pulse response into sample number or sampling time.
One little caveat is related to the fact that we deal with discretized, not
continuous functions of time. For this reason we find step response from
impulse response by numerical integration or summation. Numerical integration
can be done with trapezoidal, Simpson, and higher polynomial formulas, or, with
zero order approximation of the function with rectangular bins (e.g. by finding
the samples of the step response as Sk = dt*(h1+h2+...hk)). In worst case
scenario the resulted step response (and hence pulse response) found by
different techniques may have a time shift of about the time step specified by
EDA tool when calling Init function. If EDA tool and AMI model use different
approaches we may expect this little timing mismatch. For typical 32 samples
per UI we expect about 0.03UI mismatch, which might be acceptable, however it
could be more with coarser granulation.
I think that sampling point should be reported by the model in time units (not
the sample number), and EDA tool and AMI should both assume it a sample time
for the "true" continuous pulse or step response, and make some efforts to
minimize timing error when restoring pulse or step response on its own for the
purpose of statistical simulation.
Vladimir
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Hansel Dsilva
Sent: Tuesday, March 31, 2020 1:59 PM
To: Walter Katz <wkatz@xxxxxxxxxxxxx<mailto:wkatz@xxxxxxxxxxxxx>>;
ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-macro] Re: Gathering feedback. In the AMI_Init flow, the RX
IBIS-AMI model to output the cursor sampling point with regards to a pulse
response or step response?
Hi Walter,
I thank you for the prompt reply.
Yes, that is correct the number is the same but the jargon that needs to go
into the specification may require us to mention either a pulse or step
response.
I too have been scratching my head over the last week and hence thought it
would be best to enable a discussion if a BIRD is to be drafted.
Thanks,
Hansel Dsilva
From: Walter Katz <wkatz@xxxxxxxxxxxxx<mailto:wkatz@xxxxxxxxxxxxx>>
Sent: 01 April 2020 01:25 AM
To: Hansel Dsilva
<hanseldsilva@xxxxxxxxxxxx<mailto:hanseldsilva@xxxxxxxxxxxx>>;
ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: Gathering feedback. In the AMI_Init flow, the RX IBIS-AMI model to
output the cursor sampling point with regards to a pulse response or step
response?
Hansel,
I think the number reported is the same whether doing a Pulse response or a
Step response analysis.
Walter
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
<ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>> On
Behalf Of Hansel Dsilva
Sent: Tuesday, March 31, 2020 3:23 PM
To: ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-macro] Gathering feedback. In the AMI_Init flow, the RX IBIS-AMI
model to output the cursor sampling point with regards to a pulse response or
step response?
Hi,
Subject, "Gap in IBIS for sampling with statistical mode AMI models",
presentation at the DesignCon 2020 IBIS Summit.
We are working on a BIRD in enabling when in the AMI_Init flow for an RX
IBIS-AMI model to tell the EDA tool the cursor sampling point time (seconds).
In IBISv7.0, the AMI_Init flow calls out an impulse response which involves a
stimulus of 1/time step and does not have any information on the unit interval
(UI).
The question to ourselves is whether it would it be preferable to make the RX
IBIS-AMI model when in the AMI_Init flow to output the cursor sampling point
with regards to a pulse response or step response?
Below is a overview on impulse, pulse and step response in enabling a
discussion. Feel free to correct me if I happen to be going in the wrong
direction.
[cid:image001.jpg@01D60A25.4FBB48F0]
Thanks,
Hansel Dsilva