Machine Learning Model for Localization in an Urban Environment

Abstract: The indoor localization problem has seen great improvements during the last tenyears and today it is possible to determine the location of electronic devices, such assmartphones, with centimeter precision. The aim of this project is to apply similar localizationmethods on multichannel antenna data gathered outdoors in an urban environment andanalyze the viability of this approach. Specifically, machine learning models are created andtested. We also answer the question if this method can be an alternative to the GPS. Themodels utilize the received gain, channel impulse response and power delay profiles and putthis data through different kNN-regression models. To measure the results, both accuracyand computational time are taken into consideration. Through this method, we producemodels that are highly accurate with longer computational times as well as less accuratemodels that are more computationally efficient. This method is viable but may becomplicated to implement on a bigger scale, for example as a GPS alternative.