Return on Investment (roi) for Multi-Technology son juan Ramiro, Mark Austin and Khalid Hamied

A.5.2. Event Status Report

As mentioned in Section A.3.3, there are events with distorted OTD measurements due to propagation conditions, and hence, they should not be geo-located. Apart from those filtering criteria, an event may be discarded during the geo-location process for other two reasons: the synchronization reference is not properly recovered, and the RLS method does not converge. Events are classified according to their status and the corresponding PDF is shown in Figure A.3, from which it can be drawn that:

• 
  Most of the events reporting at least three different sites are geo-located, that is between 65–80%, which indicates that using OTD measurements for geo-location in a real network is a feasible approach.
  
• 
  The synchronization recovery algorithm seems to be robust since only around 5% of the events in both scenarios are discarded due to lack of synchronization reference.
  
• 
  The US scenario shows worse numbers in multipath propagation and wrong SC assignment as compared to the selected European scenario. The largest difference occurs for the divergence criterion, which can be explained by the fact that an event is more robust when it reports more different sites. And events from the US scenario include noticeably fewer reported sites than in the European case.
  

A.5.3. Geo-Location Accuracy

Accuracy numbers are shown in Figure A.4 for both scenarios in Europe and US. As expected, results are worse in the US scenario due to the higher number of distorted OTD measurements
Appendix A: Geo-Location Technology, UMTS 269

0 50 100 150 200 250 300
Error [m]
Figure A.4 CDF of the geo-location error for scenarios in US and Europe, including only events reporting at least three different sites.
(see Figure A.3). However, the median error for the scenario in US is less than 150 m, which is very promising considering that OTD measurements have an intrinsic uncertainty of a chip.
When the event filtering is disabled, firstly the upper tail of the Cumulative Distribution Function (CDF) gets worse and, secondly, the number of events that diverge increases, which proves that bad-conditioned events are more likely to be erroneously geo-located. Figure A.5 shows the CDF of the geo-location error (a) and the event status summary (b) for the European scenario comparing results when enabling and disabling the event filtering.