All,
I will not be able to attend next week's IBI-ATM meeting. We do not care
about making the PAM4 thresholds optional, reasons are explained below.
There is also a discussion below about Redriver flow that is based on a
new reserved parameter that specifies that the input to all AMI_Init
functions is the IR matrix of the cumulative upstream channel.
SiSoft/MathWorks does not care if we change returning PAM4 thresholds to
be optional.
* SerDes Toolbox will continue to output valid PAM4 thresholds
* QCD already calculates PAM4 thresholds from the IR output of Rx
AMI_Init or the waveform output of AMI_GetWave if the model does not
output them
The Redriver flow solutions is quite simple, and requires some realistic
assumptions and a new reserved parameter Tx_IR_Input_Is_Upstream_Channel
True|False
* Default is False
* If False then no change to current flows, although certain
combinations of Init_Only, GetWave_Only and Dual models are problematic
* If True the input to a Tx is the IR of the Upstream Channel
* For a primary Tx, the Impulse Matrix are Unit Impulse Responses
* For Redriver Tx, the Impulse Matrix is the output of the Rx
AMI_Init
* Whether True or False, the input to each Rx is the total upstream
channel
* Redriver flows (all assume Tx_IR_Input_Is_Upstream_Channel=True)
* The flow is straightforward if all of the AMI models are Dual
models
* The flow is straightforward if all of the AMI models are Init only
models
* The flow is straightforward if all of the AMI models are GetWave
only models
* Assume one redriver, there are 81 combinations of Init_Only,
GetWave_Only and Dual models
* As a practical mater, the only buffer that equalizes itself is the
terminal Rx
* With the exception of the terminal Rx, any Init_Only model can be
made into a Dual model by making its input a unit impulse response,
therefore the output IR of the model will be the equalization of the
model. The IR that can be passed to the next stage is the total upstream
IR convolved with the model equalization convolved with the IR of the
downstream channel to the next model. The "GetWave" output of this model
is the total upstra, waveform convolved with the model equalization
convolved with the IR of the downstream channel to the next model. Note
the symmetry on how GetWave and Init work.
* The only real problem is if the downstream Rx is Init_Only and it
optimizes itself based on the total upstream impulse response and any one
of the upstream models is GetWave_Only.
Walter
