[ibis-macro] Re: ignore_bits and TX vs. RX

  • From: Mike LaBonte <mike@xxxxxxxxxxx>
  • To: michael.mirmak@xxxxxxxxx
  • Date: Sat, 30 Jun 2012 03:07:47 -0400

For both TX and RX, Ignore_Bits is usually based on the idea that output is
affected by prior inputs that were absent, or did not reflect typical
steady state operation. If a model does reflect actual buffer operation
from the very first bit, it seems Ignore_Bits could be zero. After all, at
startup a typical TX initially will have it's FFE registers preloaded with
fixed values, and the model pretty much starts the same way. The model
could easily be spot on.

But when an RX calls for thousands of bits of it's output to be ignored,
that is usually a matter of allowing slow control cycles with long
effective memories of prior inputs to reach a state where it is believed
the current state reflects nothing but "real" input values. If the RX
initially received and used for adjustment a few bits from the TX that the
TX said should be ignored, well it's hard to believe RX convergence
thousands of bits later will be significantly or even measurably impacted.
An experiment could be performed to prove or disprove that however.

So on one hand maybe the counter starts on RX Ignore_Bits only after it
begins receiving trustworthy input, ie. after TX Ignore_Bits. On the other
hand I agree that RX Ignore_Bits usually has a degree of arbitrariness
built in that exceeds all of the TX Ignore_Bits. Until Arpad's more
symmetric case comes along we have little to fear.


On Wed, Jun 27, 2012 at 4:31 PM, Mirmak, Michael

>  A deceptively simple question…****
> ** **
> Ignore_bits as a parameter can apply to the TX or the RX.  The language of
> the specification states that it determines the number of bits to be
> ignored by the simulation tool.****
> ** **
> In the case where I have a link where only an isolated TX or an isolated
> RX, this is easy to understand.  But what happens when I combine a TX with
> an RX on the same link where both use ignore_bits?****
> ** **
> I would assume that the TX (using GetWave) would provide the tool with a
> waveform that the tool would then shorten by the stated number of bits
> (say, 21.  But would the RX GetWave then use the shortened waveform in its
> own analysis, before the tool reports back an even shorter waveform?****
> ** **
> Specifically, if ignore_bits is 21 in the TX, and ignore_bits is 35 in the
> RX, does that mean that the final waveform reported out by the tool would
> be shorter than the original pattern by 56 bits?****
> ** **
> **-          **MM****

Other related posts: