[ibis-macro] Re: "AMI Impulse Response"
- From: "Dmitriev-Zdorov, Vladimir" <vladimir_dmitriev-zdorov@xxxxxxxxxx>
- To: "wkatz@xxxxxxxxxx" <wkatz@xxxxxxxxxx>, IBIS-ATM <ibis-macro@xxxxxxxxxxxxx>
- Date: Sun, 2 Jun 2013 22:22:06 +0000
Walter,
I agree that a definition, different from conventional step/impulse response is
needed. In some literature, they call it 'edge response' (from where we can
find time derivative) but I'm not sure they have a definition of this as well.
I also feel that we won't be able to give a definition without providing some
clue of how one can compute of measure this response. Unambiguous definition is
possible for the linear non-autonomous SISO system, here we need more details.
1. It makes sense to assume that prior to transition, the buffers and the
channel remain in a certain logical state sufficiently long, so that the
transition starts from the steady state.
2. After transition starts, we continue simulation or measurement long enough
to reach another steady state (with a reasonable accuracy) that corresponds to
the logic opposite to the initial stat
3. The resulted waveform becomes the DC offset edge response taken with factor
2. From this waveform, we subtract the initial value (to make it start from
zero) and divide the result by 2. If the end value is negative, change the
polarity.
4. Find time derivative, use it as 'AMI impulse response'.
I can see at least two problems in this definition, for which we may want to
add some explanations.
- Since electrically, the buffers are not LTI, the result may depend on the
switching polarity. This is true for both primary and the x-talk responses. We
need however only one 'impulse response'. Even if EDA tool can support two
responses, AMI Init function does not allow more than one for the same Tx/Rx
pair. A reasonable note we can add to this definition: use the average between
the two responses found when switching in opposite direction.
- Sometimes, DC is not a proper initial state for the transition since it does
not correspond to the way the channel operates most of the time. Therefore we
should allow other methods as well. Often it would make sense to consider
transition after running several edges as follows. For example, one can find
the waveform at Rx with the following stimuli:
(a) DC(0)101010101000000000000...
(b) DC(0)101010101111111111111...
Then, find the difference, eliminate unnecessary initial delay for the first 9
transitions, divide the result by 2.
Evidently, there could be many other methods of response generation. These
should be allowed as possible computation techniques.
Vladimir
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx [ibis-macro-bounce@xxxxxxxxxxxxx] on
behalf of Walter Katz [wkatz@xxxxxxxxxx]
Sent: Saturday, June 01, 2013 3:21 PM
To: IBIS-ATM
Subject: [ibis-macro] "AMI Impulse Response"
All,
We will be going around in linguistic circles unless we careful define how we
should be using Step Response and Impulse Response. First we can no longer call
the input to Tx AMI_Init an “Impulse Response”. We need to start calling this
an “AMI Impulse Response” understanding that the “AMI Impulse Response” is
the “Unit Impulse Response” scaled by a non-Unit voltage swing. In BIRD 158 we
should in addition only reference the “AMI Step Response”. Thus eliminating any
confusion between “AMI Step Response” and “Unit Step Response” and “Step
Response”, and between “AMI Impulse Response” and “Unit Impulse Response” and
“Impulse Response”.
Walter
Walter Katz
wkatz@xxxxxxxxxx<mailto:wkatz@xxxxxxxxxx>
Phone 303.449-2308
Mobile 303.335-6156
Other related posts:
