Performance

KPN is continuously improving the performance of Geolocation. The accuracy might therefore vary from time to time. In short, when using the Geolocation functionality, you can expect the following:

• Static devices will have a higher accuracy than moving devices.

• Slow-moving devices will have a higher accuracy than fast-moving devices.

• Rural areas should provide higher accuracy than suburban areas.

• Suburban areas should provide higher accuracy than dense urban areas .

• In urban areas, high rise buildings will negatively influence the accuracy .

• When a static device is turned on, the accuracy will improve during the first subsequent messages.

• The accuracy will be less near the borders of the Netherlands (since less gateways will be available).

Another influence on the accuracy of Geolocation is reflection of the radio signal. Due to the reflection, you might see larger and more unstable deviations than average. This is especially relevant in areas with a lot of water, in densely-built environments and in areas with a lot of machinery or metal constructions.

KPN has done multiple tests and the following is an indication of the performance of Geolocation with different settings and in different environments. The results of these test are provided in the table below. As KPN works on improving this service continuously, the parameters are subject to change. For the results below, the number of different messages used to calculate the position lies between 3 and 8.

KPI	Value for TDoA
Success rate	90% on average
Accuracy	60 meter on average

LoRa Geolocation is designed for outdoor applications. Please note that LoRa Geolocation is not designed to be used at indoor locations or for indoor use cases. While localization will quite frequently work for indoor use cases, KPN will only support he Geolocation functionality for outdoor usage. The main influence for indoor usage is the building material, which may prevent the LoRa message from being received by the minimum number of required gateways.

Accuracy of TDoA

Static device performance in The Netherlands

As stated earlier, several factors impact the accuracy of the Geolocation service. In general, the graph in the figure below is representative of the average accuracy in the Netherlands (all locations and all environments for a Static device profile). Depending on your location and environment, you can experience better or worse performance:

• 95% of the measurements are under 100m

• 50% of the measurements are under 60m

General accuracy of Geolocation in the Netherlands for Static devices

Accuracy in high building environments

To give an example of the impact of different environments, KPN has executed specific testing in a environment with tall buildings (more than 6 floors) in the Hague and Rotterdam with Static devices. The 95-percent interval could increase up to 250 meters in case of reflections (Figure 9). The main influence is the reflection of the uplinks. KPN expects to improve this specific case in the near future.

Specific accuracy in the Netherlands for Static devices in environments with tall buildings

Accuracy of slow-moving devices (max. 25 km/h)

KPN has already performed tests with slow-moving devices. We believe that the majority of use cases should use the Static Device Profile. The main focus of our improvements is the stability and accuracy of Geolocation for static devices. To give an indication, Figure 11 shows the average performance for end devices on slow-moving device profiles.

Average Geolocation accuracy of slow moving devices

Accuracy over time

The relevance of the number of uplinks that can be used by the locsolver is shown in Figure 10. While the accuracy with three messages is around 200 meters, it improves by 30% to around 130 meters after 15 messages. This information is relevant for the design of your solution.

Impact of the number of messages used for Geolocation calculations

Accuracy of RSSI and Both (compared to TDoA)

On introducing the RSSI and Both algorithms, we have done some tests to compare performance in success rate and accuracy.

These tests were performed only to compare the different algorithms. They do not portray the average performance you can expect from our network! For instance, all test devices were place indoor in order to keep our testers dry in the springtime.

These are the results:

Algorithm	Success rate	Median accuracy	Average accuracy
TDoA	78%	181 m	78 m
RSSI	98%	1,475 m	4,173 m
Both	99%	373 m	1,002 m

So RSSI and Both have a higher success rate than TDoA, but the accuracy of TDoA is better than Both, and significantly better than RSSI.

Now let us compare the cumulative distribution function (CDF) of TDoA and Both. In this CDF you can see, for a given accuracy of Geolocation in meters, the percentage of messages that has at least that accuracy. We can see that up to around 80% of the messages, the accuracy of TDoA and Both are about the same. Then, the graph for TDoA becomes a flat line, indicating that there are no more locsolves. The graph for Both still keeps rising, indicating additional locsolves with a higher accuracy.

So from this we can conclude that the Both algorithm is a valid extension of TDoA, leaving the original locsolves intact and filling in missing locsolves with RSSI locsolves.