WiFi case study

Measured indoor signal strength follows the exponentiated Ornstein–Uhlenbeck model, and fast reversion decides which search policies lead.

The data

The Feng, Nguyen and Luo (2024) dataset covers a 92 by 15 metre floor on a 0.6 metre grid. It has 642 reference points, 120 samples each, and 13 access points.

The model holds

After detrending by a path-loss model, the residual signal strength along traces is close to Gaussian. Its spatial correlation is exponential with a length of about 11 grid units, 6.6 to 7.1 metres, plus a short-scale component carrying roughly half the variance.

Two independent implementations, with different detrending, agree. This matches the classical shadowing model of Gudmundson (1991).

Replaying the policies

The model was fitted on 25 training traces, and 125 search episodes were replayed on 25 held-out traces. The payoff is the exponentiated signal, the objective the policies optimize.

policymean payoffgain over the incumbent
incumbent2.596
two-shot rule3.184+22.6%
three-shot rule3.442+32.5%
knowledge gradient4.224+62.7%
equispaced probes4.409+69.8%

Why far probes lead here

The correlation length of about 11 grid units is short against traces of 15 to 92, so far probes are nearly fresh draws. For small correlation $\rho$ the exact formulas give a revisiting window

$$\big(b_-,\ e^{2\Theta}\big) \;\to\; \big(0.586\sqrt\rho,\ 1 + 2\rho\big),$$

and an interior premium of order $\rho(1 - b^2)$. The interior option's edge is of order $\rho$, while the variance of a fresh draw is of order one.

Fast reversion therefore makes measuring far away and taking the best optimal to leading order. Equispaced probing and the knowledge gradient lead because the regime says they should.

Refinements

Code: casestudy/wifi/.