|Title||Space-partitioning with cascade-connected ANN structures for positioning in mobile communication systems|
The world around us is getting more connected with each day passing by – new portable
devices employing wireless connections to various networks wherever one might be. Locationaware
computing has become an important bit of telecommunication services and industry. For
this reason, the research efforts on new and improved localisation algorithms are constantly
being performed. Thus far, the satellite positioning systems have achieved highest popularity
and penetration regarding the global position estimation. In spite the numerous investigations
aimed at enabling these systems to equally procure the position in both indoor and outdoor
environments, this is still a task to be completed.
This research work presented herein aimed at improving the state-of-the-art positioning
techniques through the use of two highly popular mobile communication systems: WLAN and
public land mobile networks. These systems already have widely deployed network structures
(coverage) and a vast number of (inexpensive) mobile clients, so using them for additional,
positioning purposes is rational and logical.
First, the positioning in WLAN systems was analysed and elaborated. The indoor test-bed,
used for verifying the models’ performances, covered almost 10,000m2 area. It has been chosen
carefully so that the positioning could be thoroughly explored. The measurement campaigns
performed therein covered the whole of test-bed environment and gave insight into location
dependent parameters available in WLAN networks. Further analysis of the data lead to
developing of positioning models based on ANNs.
The best single ANN model obtained 9.26m average distance error and 7.75m median distance
error. The novel positioning model structure, consisting of cascade-connected ANNs, improved
those results to 8.14m and 4.57m, respectively. To adequately compare the proposed
techniques with other, well-known research techniques, the environment positioning error
parameter was introduced. This parameter enables to take the size of the test environment into
account when comparing the accuracy of the indoor positioning techniques.
Concerning the PLMN positioning, in-depth analysis of available system parameters and
signalling protocols produced a positioning algorithm, capable of fusing the system received
signal strength parameters received from multiple systems and multiple operators. Knowing
that most of the areas are covered by signals from more than one network operator and even
more than one system from one operator, it becomes easy to note the great practical value of
this novel algorithm. On the other hand, an extensive drive-test measurement campaign,
covering more than 600km in the central areas of Belgrade, was performed. Using this algorithm and applying the single ANN models to the recorded measurements, a 59m average
distance error and 50m median distance error were obtained. Moreover, the positioning in
indoor environment was verified and the degradation of performances, due to the crossenvironment
model use, was reported: 105m average distance error and 101m median distance
When applying the new, cascade-connected ANN structure model, distance errors were
reduced to 26m and 2m, for the average and median distance errors, respectively.
The obtained positioning accuracy was shown to be good enough for the implementation of a
broad scope of location based services by using the existing and deployed, commonly