[ibis-macro] Re: RX Optimization

  • From: Gregory R Edlund <gedlund@xxxxxxxxxx>
  • To: Ambrish Varma <ambrishv@xxxxxxxxxxx>
  • Date: Fri, 8 Mar 2013 13:14:00 -0600

Ambrish,

The original question from my colleagues was simply whether or not the RX
algorithmic model considered TX equalization during optimization (link
training).  I answered "yes" to them but wanted to understand the details
of where optimization occurs in the flow.

I believe you're telling me RX optimization can occur in one of two places,
depending on how the RX algorithmic model is constructed:

Step 3:  RX AMI_Init

Step 7:  RX AMI_GetWave

Did I understand your note?

Greg Edlund
Senior Engineer
Signal Integrity and System Timing
IBM Systems & Technology Group
3605 Hwy. 52 N  Bldg 050-3
Rochester, MN 55901





From:   Ambrish Varma <ambrishv@xxxxxxxxxxx>
To:     Gregory R Edlund/Rochester/IBM@IBMUS
Cc:     "ibis-macro@xxxxxxxxxxxxx" <ibis-macro@xxxxxxxxxxxxx>
Date:   03/08/2013 12:47 PM
Subject:        RE: [ibis-macro] Re: RX Optimization



Greg,
When I said that the simulators should not be expected to do anything
special, I meant that they should not be treating the Tx in any special
manner just because the Rx is in training mode. They should still be
providing a valid (or as best as it can) impulse response and/or analog
waveform to the Rx for it to do its thing.

So to go back to the statement in your original email – generally, training
in the Rx occurs in the Getwave – and if the Tx AMI model also includes
Getwave, in time domain simulations, any Tx equalization in the Init is
ignored. However, if the Rx AMI model only has an Init function, it could
include an optimization algorithm and the simulator must provide the
impulse response *with* any Tx equalization to the Rx Init function.
I consider that ‘normal’ simulator behavior – and one that tries to mimic
real life.
Of course, there is a good chance that the Tx Init does not perform any
equalization as there is a Getwave in that model.

Sorry – I think I misunderstood your original question.
Thanks,
Ambrish.

From: Gregory R Edlund [mailto:gedlund@xxxxxxxxxx]
Sent: Friday, March 08, 2013 12:51 PM
To: Ambrish Varma
Cc: ibis-macro@xxxxxxxxxxxxx
Subject: Re: [ibis-macro] Re: RX Optimization



Amrish,

Thanks for your response.  Do I understand correctly that you're
disagreeing with IBIS 5.1, which states:

"Under certain circumstances, for example when the Rx AMI_Init function
includes an optimization algorithm, the impulse response presented to the
Rx AMI_Init function must include the Tx equalization effects for the
optimization to work correctly."

Or are you saying that the simulator and the hardware do not behave the
same way?

Greg Edlund
Senior Engineer
Signal Integrity and System Timing
IBM Systems & Technology Group
3605 Hwy. 52 N  Bldg 050-3
Rochester, MN 55901



Inactive hide details for Ambrish Varma ---03/08/2013 11:38:00 AM---Hi
Greg, Rx training, when you consider real devices, don'tAmbrish Varma
---03/08/2013 11:38:00 AM---Hi Greg, Rx training, when you consider real
devices, don't have any idea how Tx's are behaving. In

From: Ambrish Varma <ambrishv@xxxxxxxxxxx>
To: "ibis-macro@xxxxxxxxxxxxx" <ibis-macro@xxxxxxxxxxxxx>
Date: 03/08/2013 11:38 AM
Subject: [ibis-macro] Re: RX Optimization
Sent by: ibis-macro-bounce@xxxxxxxxxxxxx




Hi Greg,
Rx training, when you consider real devices, don’t have any idea how Tx’s
are behaving. In fact they don’t even know if there is any Tx equalization
exist. As such, there cannot and should not be any expectation that the
simulators should handle Tx equalization any differently during Rx
training. The only thing that the simulator has knowledge of is how many
bits to ignore before it starts gathering data for analysis purposes.

Glad you consider section 10 (or at least a part of it) clearly written J.

Thanks,
Ambrish.

From: ibis-macro-bounce@xxxxxxxxxxxxx [
mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Gregory R Edlund
Sent: Friday, March 08, 2013 12:04 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] RX Optimization



Hello IBIS Comrades,

My co-workers and I were having a discussion today about how simulators
should handle TX equalization during RX training.  I took an action item to
run this past the Open Forum and make sure I was speaking correctly.

"RX optimization (adaptive equalization), if it exists, occurs during Step
3, i.e. the call to RX AMI_Init.  During optimization the RX algorithmic
model must consider TX equalization, if it exists, and there is more than
one way to accomplish this.  Two are documented in the IBIS 5.1."

By the way, Section 10.2.3.2 was clearly written.  Most of our questions
were answered simply by reading two pages of the spec.

Greg Edlund
Senior Engineer
Signal Integrity and System Timing
IBM Systems & Technology Group
3605 Hwy. 52 N  Bldg 050-3
Rochester, MN 55901

GIF image

Other related posts: