[Ns-developers] Overhaul of IEEE 802.11 modeling and simulation architecture in NS-2

[Mathieu Lacage]

hi nicola,

I don't know why but your email escaped me for a very long time, hence
this very late reply/comment. I think that we talked in person in Nantes
about some of the items within this email but I thought it might be
interesting to reply here.

On Tue, 2007-10-16 at 11:49 +0200, Nicola Baldo wrote:

> > For your interest I attached our upcoming publication on our approach of 
> > overhauling the MAC/PHY stack. One main advantage here is that we took a 
> > very modular and structured approach. 
> 
> Again, I definitely agree: a modular architecture which allows to
> separate PHY, MAC and Channel functionality, and to reuse these
> components across different wireless technologies and implementations,
> is the way to go for NS.

This is a pretty cool idea and I have to confess that I did share this
goal at one point but it is pretty hard to actually implement. Here are
some considerations to ponder:

   - it is impossible to design an abstract PHY layer API which works
the same on a wide variety of real PHYs. For example, a GSM PHY is so
damn far from a 802.11 PHY that there is no chance on earth you can
design a useful API which provides access to some sort of least common
denominator which is not the empty set. 

   - the only practical solution is to do what you attempted to do in
your MPhy library, that is, attempt to provide a bunch of building
blocks which can be used to implement a PHY. 

   - Over the past 2 or 3 years, I have talked to a lot of people who
work on PHY-layer modeling: what I understood from these discussions is
that they all have very different perspectives on the fundamental
characteristics of a PHY layer model. A lot of them make the assumption
that the signal strength is constant over the reception of a signal,
that is, the signal strength is calculated when the first (or last) bit
of the signal is received. However, a lot of the more digital-signal
people scream if you mention this: this approximation makes very little
sense except in a few very specific PHY+channel conditions none of which
are verified in practice for most PHYs.

I could go on and on about the many "little" fundamental modeling
details which seem to have a large impact on simulation results but my
feeling is that providing an MPhy-like library is a very nice idea but
the most important thing you want to do is write a small document which
_describes_ the fundamental model characteristics you can achieve with
such a library. For example, that document would say:
  - if you make the assumption that the received energy of a signal is
constant over the reception window, you can use the models X, Y, Z to
build a PHY and a channel model.
  - if you make another assumption, here are the models we provide, etc.

In my experience, it is the clear statement of all these implicit
assumptions, which is the hardest part of building, using, and
explaining a PHY model.

my 2 cents,
Mathieu