Wireless Indoor Location Estimation Based on Neural Network RSS Signature Recognition (LENSR) Page: 3 of 7
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
techniques follow. These systems and techniques typically
have an offline training phase and an online location
determination phase . Some location estimation techniques
build a radio map in the offline phase that represents the RSS
values to each reachable AP from every location in the area of
interest. A location estimation algorithm run on a mobile
device requires a copy of the radio map. The advantage of
running the algorithm on the mobile device is the preservation
of user privacy and improved scalability.
RADAR [12, 13] uses RSS measurements obtained from
multiple locations to triangulate a user's position in an area. An
experimental radio map is built offline by taking measurements
in all possible grid locations of the area of interest. The system
performs both location estimation and user tracking to within 2
to 3 meters of the actual location.
The Clustering and Decision Tree-based  method
(CADET) selects the set S of wireless access points that give
the best performance for each location of a wireless area in the
offline phase. The grid space is then partitioned into clusters
and a decision tree is built for each cluster. In the online mode
the RSS values from selected access points are used to
determine which cluster and coarse location is associated with
the device. Next, the decision tree for the identified cluster is
evaluated resulting in a specific grid location. The best
accuracy of CADET is 83.4% within 1.5 meters.
The Joint Clustering (JC)  technique uses clustering of
location maps and probability distributions. A cluster, which
represents a set of locations that shares a common set of access
points, is calculated offline, as well as the joint probability
distributions of the signal strength of different access points.
During the online location estimation phase RSS values are
acquired from some set of APs , which are used to determine
the cluster to search for the probable location. The radio map
and Baye's theorem are used to determine the most probable
user location within a cluster. The accuracy of this system is
90% to within 7 feet.
Uncertainty in RSS signal measurements can be modeled as
fuzzy sets.  divides the area of interest into zones. A radio
map is developed offline and is used to train the fuzzy
inference system. There are six fuzzy sets for RSS: Excellent,
Very Good, Good, Low, Very Low, and None. The degree of
membership of a mobile device to a specific area is used to
determine the location estimate, providing an accuracy of near
A generalized regression neural network (GRNN) is used for
a pattern matching algorithm in . A measured RSS value for
each AP and the corresponding grid location in the radio map
are inputs to the neural network during the training phase. The
GRNN has one hidden layer, and an output layer
corresponding to two neurons representing the x and y
locations in the radio map. During the online phase a set of
RSS values are input to the GRNN and the output is the
estimated user's location in x and y grid map coordinates. The
maximum error between estimated and true positions for the
test data was 43.2 meters. The location accuracy for the test
data is 45% to within 5 meters.
III. MOBILE DEVICE LOCATION ESTIMATION
The approach to mobile device location estimation presented
in this paper is based on a comparison of RSS signal vectors
recorded by a mobile device and RSS vectors from a radio grid
map. The radio grid map can be created offline in two ways.
One way is either to have a person manually collect and record
RSS signal strength values for each grid location, or to do this
task automatically with a robot. With this approach, both
precision (actual signal is recorded) and computational time
savings are achieved (no analytic determination of attenuation
and reflection is needed). Another way is the creation of a
theoretical propagation model representing the RSS signal
levels that are calculated for every location of the radio map
based on propagation equations. For the sake of simplicity, the
latter approach is taken in this paper. The effectiveness of the
presented algorithm is the same regardless of the way in which
the radio grid is created. The advantage of a recorded map is
that complex, analytic modeling  of signal attenuation and
reflection in an indoor environment can be effectively avoided,
resulting in more correct, actual radio map. The use of a
theoretical model allows the algorithm to be deployed in a new
environment without having to physically acquire signal
strength readings for building a radio map. Alternatively, the
radio map may be constructed by 1) manually recording RSS
readings from access points in each designated grid location in
the area of interest, or 2) using an autonomous vehicle/robot to
collect signal strength information. The CPN with k-NN AP
approach to location estimation is valid regardless of how the
radio map is built. Locations and WiFi transmitters and
receivers power are all that is needed to build the radio map.
The APs are expected to be homogeneous; i.e., same
transmitter power. Cable and connector losses are omitted to
simplify the model. The theoretical model is based on the
following equations for received signal strength RX and path
loss Lp :
where RX is received signal strength value. TX is AP power
in dB, and LP is path loss, and
P 33dB + N * logio(D) + 20 * logio(t) (2)
where f is a frequency in gigahertz. N is a path loss
exponent, and D is a distance in meters.
Based on equations (1) and (2), for factory environment path
loss and access point power (N - 5.5, TX -- +20dBm), and AP
frequency f= 2.4, received signal strength RX is:
RX - 20 - 33 - 5.5 logio(D) - 20 * logio(2.4) (3)
Equation (3) shows how the signal strength decreases
exponentially with distance from a WiFi access point
regardless of transmission power and antenna gain.
The theoretic model is developed by using equation (3) for
calculating the received signal strength for each access point in
every location of the radio map. Some test values are
calculated by adding noise to the theoretical model. The
theoretical map of RSS values for a 20 meter square area (n x
Here’s what’s next.
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Reference the current page of this Article.
Derr, Kurt & Manic, Milos. Wireless Indoor Location Estimation Based on Neural Network RSS Signature Recognition (LENSR), article, June 1, 2008; [Idaho Falls, Idaho]. (https://digital.library.unt.edu/ark:/67531/metadc894963/m1/3/: accessed April 26, 2019), University of North Texas Libraries, Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.