I was led to a really useful paper by Ramsey Faragher at CUCL. This is a really good paper about BLE location. There are lots of little gems in it. I might see if I can go talk to him.
His experiments, and theoretical results bear out our empirical results. Fascinating extracts…
These spurious values would be highly detrimental to fingerprinting if not filtered from the data. The issue appears to stem from the use of active Wi-Fi scanning (rather than passive) in the latest version of Android (4.4.4). A conclusion that can be drawn at this point is that it may be useful to switch off all radios that are not being used while RSS fingerprint measurements on a particular radio band are being recorded.
the speed of the user (slower users can exploit lower beaconing rates to gather the same number of samples in unit distance travelled). The beaconing rate affects the battery life of the BLE beacon, and for most iBeacon purposes will typically be set to an advertising rate of a few Hertz. For positioning purposes it is clear that good multipath mitigation is highly dependent on the properties of the receiving device and the beaconing rate. During our experiments we determined that 10 Hz beaconing and 1 second batch processing of median or mean values produced the best (smooth and repeatable) RSS profiles.
The desire for metre-level positioning suggests that the windows should not exceed 1 second, as a pedestrian will walk further than 1 metre in this time. By a similar argument, the window should not fall below 0.1 s since fast fades occur at a spatial separation of around half the signal wavelength (around 12 cm) or longer.
For reference the positioning error for a WiFi tracking scheme was less than 8.5 m 95% of the time, this performance was limited by the poor signal geometry afforded by the existing WiFi infrastructure available in our laboratory. The advantage of BLE lies in the ability to freely locate beacons to provide good signal geometry.
Repeating this analysis with BLE data rather than WiFi achieved an accuracy of less than 2.6 m 95% of the time.
it was determined that there is little benefit in having more than 6-8 beacons available for a given fingerprint
Latapy  has previously proposed that one should assume 1 dB of attenuation per metre of an indoor office/residential environment (accounting for furniture and walls) for broadcasts of similar frequency (cellular) to BLE
We observe that each wall led to an additional loss of around 2 – 3 dB.
The experiment demonstrates that a ~10 dB reduction in RSS caused by this body effect will still result in a reasonable proximity measurement when within 10 cm of the transmitter (the receiver will still be estimated to be within a metre of the transmitter), whereas out at 1 metre the body effect can result in a range estimate of 5-10 m.