6.1. Radio Propagation Model 7941 special measurements over three APs of the asus AP class, have been taken withthe WISPY spectrum analyzer. The spectrum analyzer measures signals from theAPs at a lower layer that the packet-aware Wi-Fi-chipsets. It reports the energylevel of the electromagnetic field for a frequency vector with a fine grained time-sampling. But this means also, that the AP source of the signal is not known. Theprimary use-case of the device consists of finding the Wi-Fi channels that are lessused by the different consumers of the electromagnetic spectrum. Switching APs tounused channels can increase the service quality of the Wi-Fi infrastructure.The problem of AP identification with the WISPY was solved by customizing thefirmware of three asus-APs. The modification allows the AP to send a burst signalsequentially over all 14 Wi-Fi channels. The underlying burst signal can be identifiedeasily on the Wi-Fi data stream. The main drawback of this technique has beenmaterialized in the efforts that are needed to measure a single AP RSSI value. Thevalues of the different APs need to be recorded serially instead of synchronously asin the case of the Android devices. For a single measurement, around 3 minutes wereneeded with all preparations. To reach the 924 AP readings of the Iconia device, 50hours of measurements would have been required. Therefore, the simpler strategy,to rely only on the reported readings from the Wi-Fi-APIs of the different operatingsystems, was chosen.6.1.1.4 Training CorpusThe training corpus for the raytracer simulated radio propagation models was buildby measuring the RSSI readings for the 22 APs at 100 different locations. This relatesto one location for 81 m3or a cube with edge size of around 4 m. The locations areevenly distributed over the three modelled floors and the measurements were takenabout 1 m above each floor level.The averaged standard deviationallfor measured RSSI values over Nalldifferentlocation/AP tuples is in the range of 2- 4 dBm. The wispy device was excluded asthere were not enough measurements to lead to significant estimates. The class ofthe 9 EDUROAM APs has a slightly better variance withedu= 2. 5 dBm over theirNedu= 1067 readings, than the other three AP classes. These APs consist of CISCOmodels with a larger antenna array for MIMO support, and they are not targeted atthe consumer market. Another possible explanation for the lower variance of thesemodels can originate in a slightly adapted measurement technique. The CISCOAPs emit RSSI values for multiple software simulated SSIDs. By introducing SSIDaliases for these virtual APs all of them have been grouped together. This groupingcan probably also lead to the recognized lower variance. The details for the differentdevice/AP-class combinations are listed in table 6.2.For each location, around 670 readings were taken for each AP over the Androiddevices. This relates to a measurement duration of 17 minutes with a resolutionof 1. 5 seconds per RSSI reading. The effect of the body shadow was minimizedby choosing a neutral position with respect to the placements of the APs. Themeasurements were mostly taken during the office hours, so there was also somesporadic traffic of moving people in the surroundings. It can be concluded, thatunder the trade-off: effort-versus-precision, the effort was reduced. So it is sensible