| Thread Previous • Date Previous • Date Next • Thread Next |
Hola Yibin,That is a good point, although I am reluctant to try to compensate dynamic adjustment in the engine because we would need a model that would be really hard to tweak and almost certainly would be wrong most times. I think we just have to live with the limitations of WiFi, however, these kinds of issues can be more easily (although not cheaply) overcome with the addition of custom Bluetooth beacons, which we are planning to put together and test after the summer.
In terms of your suggestion for gradient descent, that would work best on the 'rank distance', which is a measurement of difference between fingerprints (which in turn are the collections of APs visible at particular locations). The rank distance is normalized, so it should even be a bit more consistent than raw RSSI measurements.
cheers! Jorge On 11-04-20 02:22 PM, yibin wrote:
For the idea of room finder application, one thing to notice is that some APs actually have energy preservation feature (and I think it will become more and more popular) which allows AP dynamically adjust RSSI by load. Some more sophisticated machine learning technique is needed to counter this time variant fingerprint value. The room finder is the kind of application requires most accuracy and needs not just positioning but also direction guidance (which maybe could be done using gradient descend of RSSI of the landmark APs ?). Bests, Yibin
| Thread Previous • Date Previous • Date Next • Thread Next |
