Space estimates
===============

Mode is probably going to be slower than mean.  After a person moves, it
takes a while to collect enough samples to find the mode.  

Mean may be erratic if a person is standing "in between" two beacons.
Mode may have a bit more hysterisis (e.g. if the values land in the same
bin) but it's not guarenteed.

Space estimates do not work well when there is a large number of
beacons.  In order to find the correct space estimate, the averaging
window size must be at least as big as the expected cycle time.  The
expected cycle time is the cycle that a beacon will be received
successfully from every beacon within range, with high probability.  One
key difference between space vs position inference is that for space
inference, the nearest beacon MUST BE HEARD.  For position inferencing,
any most recently received beacons from 3 to 4 different beacons will
do.

Hence, after a person moves, the device should wait the expected cycle
time until the correct space can be reported.  Otherwise, there will be
a sequence of "jumps", depending how close you were wrt to the last
associated beacon.

This might be a consequence that cannot be resolved unless there is a
positioning model (e.g. Kalman filter).  We need a model to handle the
case that when the listening receives a distance estimate from a
particular beacon B, it finds that the distance has changed.  From this
"distance change", we need a model that allows us to model that all the
distances relative to other beacons have also changed.  Unfortunately,
some beacons will become closer, while others become further.  This
depends on the user's position and trajectory, etc.  Accounting for all
this, the model we get will be something close to a Kalman filter.

Position Estimates
==================

Pick which beacons?

Four with smallest stddev.

While the position accuracy may be good, the delay increases.  The
beacons with the lowest stddev tends to be beacons that do not have a
recent update after a move.

Four closest beacons.

Probably suffers the same problems as above but to a lesser degree.  If,
after a move, a beacon becomes closer, then an update is possible.  But
results may be erratic.

Four most recent.

This should work best if 
