4.2. Radio Propagation 45to use the raytracer PHOTON for generating the SSMs. Although the raytracer en-ables dynamical simulation of the radio propagation for freely placeable APs, thereare still some required prerequisites. Initially, a 3D model of the building is neededthat should preferably distinguish between materials with different physical proper-ties. For each material, the optical parameters can then be either looked-up or theyneed to be trained. In the context of this thesis, a method to train these parameterswas designed. The main motivation for these efforts was given by the circumstancesthat although material parameters were found in[30], they were not directly ap-plicable to the presented setup. It is also reasonable to assume that the propertiesof the used materials in the UMIC building differ from the ones used in other ex-periments. Furtermore, there exists no unified input format for different raytracersin general and radio propagation simulators in particular. Although probably thishurdle could be overcome by devising some adaptation scheme between differentmaterial parameter representations.4.2.1 ModelThe raytracer model consists of a Blender generated 3D-Scene with multiple meshesrepresenting the walls, doors, windows and furniture of the building. Each meshhas a defined material. For each material, there is a reflection parameter and atransparency parameter . Additionally, a power parameter is defined that controlsthe initial strength of the emitted signal for each AP class. APs of the same modelare grouped into such an AP class. With and as material parameters, theraytracer uses the following BRDF that is described in 2.1.1 for simulating theintersection between rays and materials. Refraction, diffraction and interference arenot simulated by the basic BRDF and there is no change of the direction of theray that passes through the material. For simplicity, the emission of the signal ofan AP is assumed to be isotropic. It is possible to employ more complex antennapatters that are based on spherical harmonics. But that would lead to more free,and therefore trainable, parameters instead of the chosen representation by a singlescalar power parameter for a class of APs.The raytracer was successfully employed to simulate the radio propagation of GSM-stations in outdoor scenarios. So it is reasonable to assume, that indoor scenarios canprobably be simulated successfully as well. The raytracer is GPU-accelerated andsimulates a scene of 3000 m2over three Floors in a resolution of two megavoxels andwith 106Rays in around 30 seconds on a NVIDIA QUADRO 6000. Therefore, theraytracer produces SSMs with a voxel size of(20 cm, 20 cm, 20 cm). For each AP witha given position and power class such a signal strength map is generated. Duringevaluation, these maps have to be generated only once given appropriate materialparameters for the scene and power/location parameters for the APs. Hence, it isvery cheap to adapt the propagation model to a new AP configuration.4.2.2 Parameter EstimationFor training the free parameters of the raytracer model, the implemented systemuses measurements that are collected with Wi-Fi capable devices at predefined po-sitions in the building. After this initialization phase, an evolutionary optimization