GIN Phase 1 – Preparations and initial development

GIN project has had it’s virtual kick-off. Not the real kick start we wanted with everybody keeping social distance and mostly working from home. But we have started and the project is moving and we have already made 3 months and everybody is now looking forward to summer vacations. Here are the main accomplishments we have made:

  1. Test environment. We chose Combain Office as first test set up. Easy access, limited space and control of the set up. Combain Office has a total area of 400m2 and we bought and installed 15(!) Google WiFis in this limited area. 1 WiFi per 27m2, very dense and very good coverage. This makes our office ideal for doing round trip time (RTT) measurements. We also acquired different phones supporting RTT: Google Pixel1, Pixel2, Pixel4. Why does it seem to still be only Google supporting RTT? Almost all WiFi chip sets support RTT and it is a 802.11 standard.

2. RTT measurements.

We made lot’s of Line Of Sight (LOS) RTT measurements to understand the performance. Generally these measurements are consistent with the early prototype hardware we used some years ago. We achieve around +/-1m with RTT. About 5-10x better than with RSSI! RTT measurements are consistent over time, which is great. We can repeat measurements and get same performance.

RTT offset. There is an error/offset in the RTT distance measurements that often are in the order of a meter that is not just noise but an offset. According to sources on the internet it is WiFi AP dependent. But our measurements indicated that it is not that much WiFi AP dependent but more mobile dependent. Seems like every phone behaves slightly (but consistently) different compared to each other. Largest difference is between models, newer models like Pixel 2 and Pixel 4 seem to be better than the old Pixel 1. We believe the offset is due to that the mobile reports it’s delay from receiving the message to transmitting the ACK back wrongly to the measuring AP and that causes this offset. With newer and better hardware, this offset should be reduced in the future.

Doing RTT positioning by measurements during a track walking around the office and using an optimal trilateration for creating a position gave us a median error around 1.8m, a little worse than expected and needs to be investigated further.

3. Test App.

We need to make lot of tracks with ground truth in our project. So we developed a very good Android app for creating tracks that can be downloaded for algorithm development. We made it as part of the Traxmate solution. Download and use the app if you want to try some indoor positioning and indoor model creation:

4. Algorithm development. We read a lot of articles about indoor floor map generation and started to do some basic tests. Our initial simulations show that we probably need a indoor accuracy of 1-2m to be able to create the paths etc. Thus better than what we achieve currently with our indoor 3D slam (5-10m). RTT is definitely going to work, but since the deployment still is very low, we are evaluating if we can increase accuracy further using either a fingerprinting method and/or creating a denser beacon environment. Will continue to investigate after summer vacations.

Have a great summer and keep social distance!