so it probably could only work (more or less) in cities because there you have the most chance of having at least 3 gateways in reach of that node … but then also you have more reflection from buildings which corrupts the measurement.
Well here is a small (?) problem. Especially in cities you have a lot of multipath fading that will disturb the time of reception heavily. As a result of that the probability of location will decrease.
what I said before 
Currently from the LoRaWAN document I have, the precision of time stamp in GW is only about 1ms, which means TDOA based positioning accuracy cannot be better than hundreds of metres.
Would following approach work to improve localisation accuracy of TdoA
- Put a reference node (for which you known the exact loacation) close to your unknown mobile location node
- Let TdoA calculate the location of your reference node using TdoA.
3)Calculate the error vector (real vs measured location)
4)Subtract the error vector from you aproximated mobile location node to get the real location of the mobile node more accurately
Thoughts ?
so its not an unknown location , you would need many many reference nodes imho.
if you need accurate position go gps (a good one)
It seems that the docs on LoRa localisation aren’t publicly available.
LoRaLoc modules provide fine time stamp with 1 nanosec precision (and 50-100 ns, accuracy 
). Also, timestamps are encrypted, so you have to pay for decryption keys.
First, you have to have good coverage. It seems that most of gateways installed by KPN don’t support localisation.
At all, from practical point of view, LoRa-based localisation doesn’t work well enough for the most cases.50 meters accuracy in ideal conditions like tens of gateways installed on oil platforms in Nothern Sea (almost flat surface, yea…), and you have to send some hundreds messages first…
That would assume all nodes in that area have the same error due to reflection/multipath. I doubt that’s a reasonable assumption.
1 Like
I have the same questions with you, have you solved these problems now?
That’s amazing. But it’s my ignorance, I have difficulties in acquiring the time of signal arrival. How to use a GPS to achieve such accuracy? Could you provide some details? Many thanks!
Your feedback is very welcome!
The GPS 1PPS signal is (should be) connected to the baseband chip in the gateway and that gives a sub micro-second accuracy of the timing.
This means you need a special gateways and all gateways need to do this in the same way. In theory this is possible but not on TheThingsNetwork.
How about the development of a smart antenna which estimates the angle of the incoming signal, that way updating the gateways with these antennas would allow localization ?
And how would this solve the multipath problem exactly? The signal might be striking the antenna on a totally different angle and direction of where the gateway is.
This technology (beam forming) exists, it’s what those fancy 802.11ac routers with their three antenna’s do, but it only works under LOS or minimal multipath (e.g. high carrier frequency) conditions.
2 Likes
I recently attended a meetup in Zurich. One of the topics was iot geolocation of nodes (without gps chip in the nodes). The technology is based on the time of arrival of the same message in 3 to 4 gateways with a gps chip. The problem is to determine the exact time of arrival of messages which may take up to almost 2 seconds to arrive (e.g. sent with SF12). This is a very complex problem and the algorithms to determine the TOA are owned by Semtech.
A document from the CSEM website https://www.csem.ch/ has some more detailsDoc.pdf (567.6 KB)
Just realised that the CSEM solution was mentioned earlier in the Application development thread
February 1 - The Things Conference
Hello, I don’t know if they are working on the same issue but at https://www.loracloud.com/
discuss similar issues with geolocation and the use of LORA
LoRa Cloud - as owned by Semtech aka the mothership, if they don’t know how to do it, then it’s probably not really commercially viable.
Whilst they provide an estimate of 1000m to 2000m accuracy for RSSI based lookups, it’s really at $2,000 gateway that provides fine-timestamp data that will give you any form of accuracy but still not quite in the realm of GPS.
Formerly Collos - with a TTN integration… 