36 3. Related Workthe Uniform Theory of Diffraction(UTD). AROMA is the only presented systemthat models the human body shadow. Detailed antenna modelling is also possible.The different properties of the antenna model that can be configured are given bytransmitting power, frequency, maximum gain and radiation pattern. Figure 3.1 il-lustrates these model assumptions. Configuration values for isotropic and half-wavedipole antenna patterns are predefined. The input for the scene geometry can begiven either in Blender or Google SketchUp format. Different material parametersare assigned to objects of the scene. It is unclear how the material parameters are de-rived, but the system exposes a preconfigured database for commonly used buildingcomponents. The simulation of the human body shadow relies on the UTD modelwith the basic assumption that a human body can be represented by the shape of acircular cylinder with radius 0. 15 m and height 2 m.As testbed for the evaluation, a 80 m2apartment room was chosen. 2 APs were placedand measurements at 11 locations were collected. At each location, 60 measurementsamples were taken and subsequently averaged. A RPE of 3. 2 dBm and RMS-RPEof 4 dBm is reported on this environment.Another advanced 3D-Raytracer model with a 2D fallback mode is presented inthesis of Martin Klepal[15]. The engine simulates the optical effects of reflection,refraction and diffraction. Non-isotropic radiation patterns for the antenna modelare also supported. The scene geometry is represented by a voxel space with avoxel edge size of 30 cm for the 2 GHz wave frequency. Multiple classes of materialsare supported by assigning them to the corresponding voxels. As in the presentedlocalization framework(i4lf), the material parameters are trained by optimizationwith Genetic Algorithms . The optimization criterion, also called the fitness value,is given by a minimization of a variance value which is also used during evaluation.The author reports that the Conjugate Gradient and Hill Simplex optimizationalgorithms have shown worse convergence behaviour with respect to the reachedoptimum. The impact of the number of rays on quality of the simulation is alsoanalysed. The author concludes that using 0.3- 2 Mio rays leads to a good APcoverage estimation without artefacts due to undersaturated areas. This relatesnicely to the experience with the PHOTON raytracer where 0.3 Mio rays lead tousable and 2 Mio to excellent results with respect to the amount of artefacts inundersaturated areas.An extensive performance evaluation was conducted on a single floor of 4 differentoffice style buildings. The floor area of all sites combined is around 5000 m2. Thefour different materials, that are used in all scenes, are given by: light wall, heavywall, windows, floor. Special care has been taken by collecting the measurements.A wooden stick, that holds the device around 1 m in front of the operator, wasused to minimize the body shadow effect. The number of measurement locationsand the corresponding number of received values that were probably averaged areunfortunately not disclosed. The RMS-RPE is reported to be between 3. 5 and4. 5 dBm over the four scenes2and corresponds to the PHOTON results as well.2On page 84 of the thesis the author reports a computed mean and variance over the measure-ment deltas without applying the abs() function. This variance is a lower bound for the RMS-RPE.Therefore, the RMS-RPE is assumed to be around 0. 5 dBm higher.