GSM signal based localization

In today’s world, the need for mobile communication systems and the high increase in the number of users have also made the development of new generation mobile applications indispensable. Obtaining location information has been one of the most interesting and significant areas of improvement. The purpose of the services used to determine the location is generally to obtain the information of the users such as approximate location, speed and time. The GPS system is the most preferred and globally accurate positioning system among global positioning systems. However, in addition to requiring a high installation cost of this system, it is one of the biggest constraints that galactic and meteorological factors, high buildings and other physical obstacles, and especially closed areas can lead to serious signal weaknesses and losses which may cause the system to be out of service. Considering these issues, it is seen that there is an urgent need for positioning systems that will be alternative and complementary to global positioning systems. At this point, it is an extremely important alternative to make location estimation by making use of the infrastructure of the GSM network, which is widely used by almost everyone and whose coverage area is increasing day by day. It is thought that cellular networks will be more advantageous than global positioning systems when compared to the signal levels that can be obtained in closed areas and in bad weather conditions. Within the scope of this study we have carried out, data sets were created by recording the GSM signal strength, GSM base station and user location information measured in indoor and outdoor areas through a mobile application we developed in the Android Studio environment for mobile phones. The network to be created in the Matlab simulation environment will be trained with some part of this data, using some artificial neural networks (ANN) methods, namely extreme machine learning (ELM), generalized regression neural network (GRNN) and k-nearest neighborhood algorithm (k-NN), and by testing with the other part of the data it is aimed to perform approximate location estimation. In the tests conducted with indoor, outdoor and blended data sets, it has been observed that the proposed GSM signal-based positioning system can obtain real location information with error rates below a meter at the minimum, and between 76-216 meters on average.