[ibis-macro] Re: de-convolution in new flow
- From: "C. Kumar" <kumarchi@xxxxxxxxx>
- To: wkatz@xxxxxxxxxx, ibis-macro@xxxxxxxxxxxxx, fangyi_rao@xxxxxxxxxxx
- Date: Fri, 23 Oct 2009 15:05:36 -0700 (PDT)
hello:
I have a suggestion. Instead of complicating the flows , if the eda tool wants
buffer equalization , it can make two calls to the model. One of them with unit
impulse. Then the modification of impulse will return buffer equalization
--- On Thu, 10/22/09, fangyi_rao@xxxxxxxxxxx <fangyi_rao@xxxxxxxxxxx> wrote:
From: fangyi_rao@xxxxxxxxxxx <fangyi_rao@xxxxxxxxxxx>
Subject: [ibis-macro] Re: de-convolution in new flow
To: wkatz@xxxxxxxxxx, ibis-macro@xxxxxxxxxxxxx
Date: Thursday, October 22, 2009, 10:05 PM
Thanks, Walter. So your point is
if a RX returns modified impulse in Init and does not provide GetWave, then you
can safely assume RX is LTI and perform de-convolution.
Fangyi
From: Walter Katz
[mailto:wkatz@xxxxxxxxxx]
Sent: Thursday, October 22, 2009 6:47 PM
To: RAO,FANGYI (A-USA,ex1); ibis-macro@xxxxxxxxxxxxx
Subject: RE: [ibis-macro] de-convolution in new flow
Fangyi,
When
Rx Init returns a modified impulse response it is telling the EDA tool that
there is an LTI approximation for the Rx. This telling the EDA tool that
this output can be convolved with a digital stimulus pattern to get a time
domain waveform, or it can be run through a statistical analysis such as
StatEye. In either of these two cases a deconvolution is not necessary. The
only times that just hREI(t) (or a deconvolution) is required is when Rx
GetWave does not exist, and either Tx GetWave does exist or one is using an
external waveform at the Rx Pad. So if you are generate an Rx model with an LTI
approximation to a non-LTI Rx (that by the nature of your LTI approximations
does not support deconvolution), then you really should supply an Rx GetWave in
addition to the LTI approximation to the Rx model.
As
Kumar suggests in his flows, you can put as input to the Rx Init a channel that
is a Dirac-Delta function, then the out put of the Rx Init function will be an
hREI(t). This technique will work for Rx models that the EDA tool can set the
tap coefficients. But I assume that code in your Rx Init function has some sort
of algorithm which picks an equalization based on the input, and I suspect you
really want the real channel including the real Tx equalization to generate
that input.
The
bottom line is that if you have a non LTI Rx, and also want to supply an LTI
approximation (which we think is a good thing) then you also should include a
Rx GetWave call in your model. With the existence of an Rx GetWave, one never
needs to do a deconvolution.
Walter
Walter Katz
303.449-2308
Mobile
720.333-1107
wkatz@xxxxxxxxxx
www.sisoft.com
-----Original Message-----
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx]On Behalf Of fangyi_rao@xxxxxxxxxxx
Sent: Thursday, October 22, 2009 9:05 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] de-convolution in new flow
Hi,
Walter;
In this
week’s meeting we acknowledged the possibility that some model
can’t provide an equalizer filter and have to return a modified impulse.
This indicates that the modified impulse is not likely to be the convolution of
the input impulse with a unknown filter. In this case de-convolution in the new
flow won’t generate any meaningful result.
Can you
provide some clarification?
Thanks
and regards,
Fangyi
Other related posts:
