80 6. EvaluationDevice NallallNasusasusNedueduNwrtwrtNpisapisaIconia 924 3.3 367 3.5 271 2.6 72 4.3 214 3.2Nexus 919 3.8 327 4.3 282 3.2 90 4.0 220 3.7Galaxyp 496 2.1 166 2.2 152 2.1 58 1.8 120 2.0Nexus1 515 2.0 177 2.1 156 1.7 62 2.2 120 2.0Messbook 492 2.6 247 2.6 206 2.1 12 7.7 27 4.6Total/Avg 3346 3.0 1284 3.3 1067 2.5 294 3.2 701 2.9Table 6.2 The training corpus for the radio propagation model is composed ofmeasurements from different devices.to assume that more elaborate measurement techniques lead to more precise readingswith less variance.Most readings were taken by using the OS-level Wi-Fi-APIs of the Android andLinux devices. These readings were measured in RSSI with minimal readings ofaround- 10 dbm and a maximum of- 100 dbm6.1.2 Training of free ParametersThe training of the free parameters for the PHOTON propagation model was con-ducted by using the process described in 4.2.2. It is reasonable to assume, that anoptimal trained radio propagation model, a high quality representation of the realworld, leads to lower error rates for localization algorithms relying on this infor-mation source. Therefore, the first part of the evaluation investigates how manyprerequisites in form of a 3D-geometry and how much training data in form of man-ual measurements is needed to derive a good propagation result. The optimizationtarget of the Genetic Algorithm is given by the RPE. The quality of the propagationmodel is assumed to be proportional to the reached RPE in the optimization processwhich is therefore used for evaluation as well.For the following evaluations, a default optimization parameter set is used that isdefined in the following table:PHOTON ParametersName ValueNumphotonsDensity in mAP Power in WResolutionBBox in m5000000.310- 5(320, 99, 58)(-1.0, 60.0),(-1.0, 18.0),(-4.0, 7.1)Genetic ParametersName ValueStartpop 80Childcount 80ChildcullMutprobMutamt300.150.15Table 6.3 The default optimization parameter set.This parameter set contains the PHOTON configuration and the initialization val-ues of the genetic algorithm. Their effects are briefly explained starting with thePHOTON Parameters: