Wireless Local Area Network Positioning - Ambient Intelligence for Scientific Discovery - page 251

The WAF factor is measured empirically by determining how much the signal
drops as the number of walls increase. Researchers have tested this by measuring
the signal strength when the wireless device and access point were in the same
line of site. The signal strength was then measured with a various number of
walls between the transmitter and receiver. The average of the difference between
these signal strength values produced the Wall Attenuation Factor.
Accuracy of the Propagation Model. The propagation model offers accu-
racy as good as 4.3m at the 50 th percentile and 1.86m at the 25 th percentile [1],
provided users are within a reasonable proximity to the access points. However,
as the distance between the mobile device and access point increases, accuracy
decreases [15]. In addition, radio signals in an indoor environment are victim to a
barrage of obstacles causing reflection, diffraction and scattering, thereby signif-
icantly affecting signal propagation [1]. Environments with frequently changing
interferences or significant obstacles can decrease accuracy. In general, the com-
plexity of the model implemented is proportional to the accuracy achieved.
2.4 Empirical Method
The empirical method entails creating a radio frequency map by taking many
measurements within a wireless network and recording Signal-To-Noise ratio
(SNR) values from all access points within the range of each location. The SNR
values at each location, also known as calibration nodes, are stored in a database
with their associated location coordinates [2]. Subsequently, to find the position
of a given mobile device its SNR values from all nearby access points are sent
to the server which determines the closest match in the database by interpo-
lating the probability distributions using the calibration nodes [18]. Two em-
pirical model implementations, location fingerprinting and a k-nearest neighbor
approach are discussed below. Algorithms used in the empirical model include
Na ıve Bayes, K-Nearest Neighbor, Neural Networks, Fuzzy Logic Subspace Tech-
niques, Viterbi-like algorithms, and Hidden Markov Model based techniques.
Location Fingerprinting. Location fingerprinting, a positioning method re-
searched at the University of Pittsburgh, is one method for implementing the
empirical model [11]. This approach, like all empirical methods, requires a col-
lection phase and a calculation phase. During the collection phase, a rectangular
grid of points is collected by recording the received signal strength (RSS) from
the access points on the site. Multiple measurements are taken from each point
creating a vector of RSS values for a given coordinate (x,y). This vector of val-
ues is referred to as the location fingerprint of that point. During the calculation
phase the Euclidean distance between the observed RSS vector and each fin-
gerprint in the database is calculated. The fingerprint that returns the smallest
Euclidean distance is returned as the user's position. The mathematical model
for this method is represented as follows:
A sample =[ ρ 1 ,ρ 2 ,ρ 3 ,...,ρ N ]
A fingerprint =[ r 1 ,r 2 ,r 3 ,...,r N ]