6.2. Localization 83Patheg-room-changeog1-classicog1-egog1-eg-rightog1-long-roomsog1-long-straightog1-room-changestairs-upwardLength40. 0 m36. 3 m59. 8 m69. 2 m81. 2 m41. 9 m14. 3 m47. 6 mTurns1281113167911avg. Signals30.540.665.462.393.340.643.266.7avg. APs8.912.16.610.47.38.49.94.4FigureA.1A.2A.3A.4A.5A.6A.7A.8Table 6.4 Physical properties of the defined paths in the first two columns and APcoverage extracted from the evaluation corpus.The avg. Signals column of the table 6.4 represents the number of these readings.The avg. APs column indicates how many APs are received in average for a singlereading. From the avg. Time column of table 6.5 can be concluded, that successiveRSSI reading are received in approximately 1 Hz which was empirically determinedto represent an appropriate pulse frequency.Pathog1-long-straightog1-eg-rightstairs-upwardog1-classiceg-room-changeog1-egog1-room-changeog1-long-roomsavg. Time38 s69 s51 s39 s52 s60 s32 s86 savg. Speed1. 1 m/s1. 0 m/s0. 9 m/s0. 9 m/s0. 8 m/s1. 0 m/s0. 4 m/s0. 9 m/sIconia(F/B)11/1113/1317/1413/1010/1011/1112/1211/10Nexus(F/B)12/1311/1110/1016/1510/1010/1010/1012/12Table 6.5 Time and speed related properties of the defined paths of the evaluationcorpus. And the size of the evaluation corpus with respect to the employed devices.As can be seen in the columns Iconia(F/B) and Nexus(F/B) of table 6.5, at least 10evaluation samples were collected for the forward(F) and the backward(B) variantof each defined path. Such an evaluation sample is a sequence of location annotatedRSSI readings. The measurements were recorded at different sessions mostly duringoffice hours and no special care was taken to ensure perfectly constant environmentconditions. This means, that sometimes doors, which probably effect the radiopropagation, were either open or closed. In some path samples, people are eithercrossing the way or occupying radio propagation relevant places. Therefore, theevaluation corpus can be assumed to represent an appropriate representation of liveconditions which lead to random noise artefacts in the RSSI readings.The synthetic and the real world evaluation use radio propagation models that aretrained and optimized according to 6.1.2. For the evaluation of synthetic data, amodel that was only trained on Iconia measurements is used. The same model isalso employed in the real world evaluation. Additionally, the real world evaluation