Fingerprinting

A type of Scene Analysis. Perhaps the dominant method of using Wi-Fi to determine location. To contrast with the next most dominant method, triangulation.

Location Fingerprinting
Using readings about a mobile unit's environment (some characteristic of a signal) to determine the MU's location. Could be readings from a Wi-Fi antenna, a Bluetooth antenna, or any type or combination of sensors. A type of scene analysis. A scene is analogous to a fingerprint, which is made up of features. Also a type of pattern recognition.

Wi-Fi Fingerprinting
Also known as WLAN Location Fingerprinting. Uses RSS readings (almost exclusively) to make up the fingerprint. Implementations using Signal-Noise Ratio (SNR) have also been created. Also known as RF-based Scene Analysis. One of many signal attenuation based methods.

Location Fingerprint
A set of scan results (beacons in the case of Wi-Fi), either together, or summarized in some way (like mean) that define a location. When made up of actual readings stored for later comparison, called a calibration point (CP), defined in a training phase.

Calibration Point
Also called a Reference Point or a trained position. One spot from which your train the radio map.

Training Stage
Also known as the calibration phase or offline stage. During this step, reference (or calibration) points are recorded by populationg the radio map with scan result vectors.

Positioning Phase
Also known as the online stage or the live phase or the location estimation phase or the real-time phase or the localization stage. Matches the online measurements to the closest offline fingerprint.

Algorithms
Algorithms can be either static or filtered. Algorithms can also be either deterministic or probabilistic.

Deterministic Algorithms
Deterministic Algorithms apply a distance norm (Euclidean, Manhattan, Mahalanobis, etc) to each calibration point. Implies that the method will come up with the same answer each time it is run against the same data.
 * Nearest Neighbor in Signal Space (NNSS)
 * k-Nearest Neighbor (KNN)
 * Weighted k-Nearest Neighbor (WKNN)

Probabilistic Methods
In a probabilsitic (statistical) method, a probability density function is used. Can compute a likely state (location) and a likelihood. Probabilstic approaches break down into parametric and non-parametric approaches. Often uses Gaussian approximation.

Other Methods

 * Neural Networks
 * Multilayer Perceptron (MLP)
 * support vector machine (SVM) for support vector classification (SVC) and support vector regression (SVR) also known as machine learning or supervised learning
 * smallest M-vertex polygon (SMP) such as Multiloc