[net2-wg] link estimator interface

[Philip Levis]

On Mon, 2006-03-06 at 11:07 -0800, Joe Polastre wrote:
> On 3/6/06, Philip Levis <pal at cs.stanford.edu> wrote:
> > On Mon, 2006-03-06 at 10:19 -0800, Joe Polastre wrote:
> > > Its ratiometric.  Ie, if node A has a value of 20, then any node (say
> > > node B) with a value < (20 - (20 >> 2)) = < 15 is at least 25% better
> > > (less transmissions required).  That's relative, node B is estimated
> > > to require 25% less transmissions than node A.
> > Some data a student here has been collecting suggests that LQI doesn't
> > behave like this at all. I've attached a plot of LQI vs PRR.
> 
> You're plotting LQI, not the function of LQI.  Please plot the
> function of LQI and resend, otherwise the point is fairly moot.

My point wasn't about the correlation of mean values and PRR/ETX (which
is all the function would change); the point was the observed noise and
outliers, which a stateless function does not change in any way.

E.g., compare the data point with a mean LQI of 100 and a PRR of 70%
with the data point that has an LQI of 75 and a PRR of 90%. Also, note
the data point with a PRR of 20% that has LQI values just under 90.

Phil