2 1. Introductionlocalization solutions has been induced by the RADAR[2] system developed at Mi-crosoft Research at the year 2000. The system uses the received signals strengthsof a number of APs and an analytical model for the impact of an obstacle on thesignal strength to determine the position of a mobile device with respect to a 2Dfloor map. From the structure of the RADAR system can be concluded that theproblem formulation has two major aspects:What is the distribution of the signal strengths and how is this information processedalgorithmically?1.1 Radio PropagationThe first aspect relates to the nature of the Wi-Fi signals and rises the followingquestions. How are the signals distributed in the localization space? How are theypropagated from the AP source? These questions lead to the concept of radio prop-agation models. These models can be specified at different levels of complexity butthey have in common, that they allow a prediction of the Wi-Fi signal distributionover the targeted areas. This prediction can then be used to drive the decisionprocess that leads to a localization result.Consequently, the generation of an accurate radio propagation model was the firstfocus of this thesis. The primary source for the investigated propagation model isthe so called PHOTON raytracer[23] that was developed recently by Arne Schmitzat the chair of I8 of the RWTH . The performance of the GPU-driven raytracer,with respect to radio signal propagation, was in the first place examined for urbanoutdoor environments, but it is designed for the general application to arbitraryindoor and outdoor environments. A basic evaluation of the model capabilities foran indoor scenario was conducted in an earlier work by Schmitz[24].In order to simulate the propagation of the AP emitted radio signals accurately, theraytracer has to be configured with parameters that relate to the physical propertiesof the involved entities. The first entity is the AP that is basically configured tobe an isotropic radio sender with a scalar antenna gain. The other simulation rele-vant entities compose the structure of the building and can basically be understoodas material annotated scene geometry. Thus, the scene geometry is a mandatoryprerequisite for the raytracer and the material parameters have to be determinedindependently.The target indoor scenario for the evaluation of this thesis is the UMIC office buildingwith four levels and a size of 15 m × 60 m × 9 m. The 3D geometry of the buildingwas modelled by using the software Blender 2. 10 different types of materials weredefined and accordingly attached to the mesh model. Further details on the modelproperties and the materials are described more formally in the evaluation chapter6.1.1.The parameters of the materials, specifically the coefficients controlling the rate ofabsorption and reflection of the given building are assumed to be unknown in order toperform the raytracing simulation. To acquire these parameters, a training techniquebased on evolutionary concepts, more precisely Genetic Algorithms , is devised and2The open source toolkit Blender is freely available athttp://www.blender.org/.