Geolocating with WiFi RTT measurements

As it is known, Google/Android is already working on other-than-GNSS technologies for navigation. One of such technologies is WiFi, exploiting the 802.11mc protocol that offers Round Travel Time (RTT) readings. So far, WiFi navigation is largely based on Received Signal Strength (RSS) and fingerprinting, which is very susceptible to environment (walls, windows, doors, ...).  RTT, being the time of flight is less susceptible to environment, thus leading to potential meter-level accuracy indoors (see this article from Google at GPS World).

In the context of the European Space Agency (ESA) Hansel project, Rokubun had the chance to perform an exploratory work of this new technology. The equipment purchased for this activity consisted of a Google Pixel 2 smartphone as well as 3 Google WiFi routers. These are one of the few hardware fully supporting the 802.11mc equipment. As far as the authors know, the Google Pixel 3 and 4 smartphones as well as the Compulab WILD router also support the protocol but they have not been tested.

Indoor positioning with RTT measurements

One of the objectives within the Hansel project was to test and demonstrate the capability of the algorithms developed by Rokubun to navigate with both GNSS and WiFi ranges (RTT) as well as purely indoor positioning using only WiFi RTT measurements (no GNSS data) and known locations of the Wifi Access Points (WAPs).

The data collection was done on the 25th February 2020 at the OpenLab of the UAB campus using a Google Pixel 2 in static mode, placed on a table in the main room of the OpenLabs, as shown in the figure below (white circle).

The figure above shows the results, with the floor-plan of the OpenLabs (yellow lines). WiFi routers (WAPs) are shown as yellow diamonds, blue squares are the positions resulting from the positioning algorithm, white circle is the reference position. Orange and red circle represent 3m and 5m deviation relative to that reference position.

A first quality assessment of the positioning solution could be based on the standard deviation of the East/North/Up (ENU) components relative to the reference position (white circle in the figure above). The time series of the ENU components are shown in the figure below.

It can be seen that, under optimal conditions (static device with no nearby obstructions), precisions of the solution of 30cm/50cm/1.5m in ENU respectively can be achieved. The North component is the one performing the worst due to a similar reason as in GNSS, the Dilution of Precision (DOP): all routers were placed at the same height as the device. If the WAP were located at different heights (providing more geometric diversity in the height component), the precision of the Up component would also be improved.

Quality of the RTT measurements

An important aspect when considering RTT measurements for navigation is to assess its quality. In particular:

  • Variability of the measurements under static conditions (i.e. observable noise)
  • Potential biases, that could be related to the WAP and/or device. Being WiFi RTT a peer-to-peer (bidirectional) technology, there is no clock errors (as in GNSS), but other hardware biases in the range could potentially be non constant over time and degrade the final navigation solution.

To conduct this quality assessment, the Google Pixel 2 smartphone has been placed at an exactly well known (measured) distance relative to the Google Wifi routers. With this setup, data-takes were taken and analysed. The figure below shows the RTT ranges minus the actual range that has been measured with the tape.

In the ideal case, this difference should yield a constant 0 with a certain noise. The plot evidences a certain biases that is router dependent (see the bias values for each router MAC address in the lower left area of the figure) but might as well depend on the smartphone. Note that the biases in this setup can go from 0.4m to 1m. A setup to determine the actual source of this bias would require an additional smartphone used to take data samples at exactly the same distances. Ultimately, a database of the WAP position should also include an estimate of the WAP bias, as proposed in Rokubun's patent EP19382753.2. An operational system to perform the automated computation of such database (based on this patent) using smartphone data is also being implemented by Rokubun.

In addition to the biases, the noise of these ranges might be also derived from this test. Note that the tests have been performed under a benign environment, where no nearby objects or bodies were present and thus the multi-path patterns might be minimised. Under these circumstances, the standard deviation of the RTT measurements amount to ca. 20cm. This can be considered a best case scenario.

References

Other resources that might be of interest have been compiled in this list: