where xk|0 is a RSS value at a constant reference distance
0 from APk. This xk|0 is unique to the device, which
explains why different devices receive different RSS values
at the same location. ψk is a Gaussian noise with zero mean.
β is a constant denoting the path loss exponent. According
to Eq.(2), if a device S can observe a larger RSS value
from an AP1 than that from an AP2, then S is generally
closer to AP1 than AP2
7. Hence, another device T in the
same location will (by expectation) also observe a larger
RSS value from AP1 than that from AP2. Hence, if we use
this boolean pairwise RSS comparison between AP1 and
AP2 as a new feature, it is robust to both S and T.
Secondly, the pairwise feature is unfortunately not discriminative,
and we want to extend it to high-order comparison
feature8. The pairwise features are not discriminative
to differentiate locations. For example, we can find two
locations y1 and y2, which are on the same side of an