Mobile networks: how can our movements be tracked while preserving our anonymity?

Signals that have nothing to do with our communications

Unless they are switched off, our smartphones are always “talking”. Not just to send text messages, surf the Internet or receive phone calls: They emit regular signals to detect the possible presence of Bluetooth or Wi-Fi terminals. 

“In technical terms, these signals are known as signalling packets. They can be used to locate mobile devices, explains Aline Carneiro Viana, TRiBE team leader at the Inria Saclay Centre and ANR Mitik project coordinator. They are invaluable sources of information because they are independent of the user’s activity, do not concern the content of any of their communication and are produced automatically. ”

An open source toolkit made available for researchers

It would be easier to access the GPS coordinates stored on smartphones. But that would involve installing an application and actively collecting data, which in turn means recruiting volunteers, with the risk of forming groups that are not entirely representative of the local population. Mobile operators have access to pseudo locations of mobile phones (from relay masts), which are less accurate than GPS, but do not communicate them. 

This goal was reached: at the end of the five-year period (2020 - 2025), the project participants (including two Inria researchers, Aline Carniero Viana and Nadjib Achir) provided the scientific community with an open source toolkit to collect routes within a limited area, render them anonymous, process them, detect the presence of individuals and then retrace their movements, either individually or as a group. When taking measurements over several hours, scientists can pinpoint the number of occupants in a given zone with excellent accuracy, and produce reliable routes for multiple smartphones in an area, including specific places, durations of any stops, etc.

“These movements are like gold dust”

“These movements are like gold dust” enthuses Nadjib Achir, another member of the TRiBE team. 

A public transport operator could use this data to improve the routes offered and the location of stops, a tourist area can use it to adapt the signage system and better channel the flow of visitors, etc.”

Likewise, a region can reallocate or optimise existing transport resources (bus, bike, car sharing), depending on the occupation of different zones and the times of day. Another example is industrial logistics. The Mitik team has entered into a partnership with La Poste so that they can monitor the movements of their mailbags from their sorting centres using a wireless communication tag attached to each mailbag. 

Finally, individual trajectories provide data on whether two people come into contact in a given place, for example in the case of a pandemic. There are other possible uses, such as device to device, whereby smartphones communicate directly with one another, instead of using a base terminal and wasting network resources. 

“Sniffer” devices to collect data

In order to achieve these promising results, the researchers had to overcome several obstacles. The first was to develop a so-called “sniffer” device to collect the signalling packets. “We had to develop this device from scratch, then assess its possible scope to find out how many of them needed to be set up in a given space and to decide on their layout for full and relevant coverage, explains Nadjib Achir. The device we developed has several built-in microcomputers, configurable Wi-Fi cards, a GPS sensor, external antennas and a battery with six to seven hours' battery life. ”

Anonymity has to be ensured from this phase: every smartphone’s unique identifier or MAC (Media Access Control) address, is rendered anonymous on the fly using a hashing technique. The result is then altered to limit re-identification of the original MAC addresses. 

A tough job: extracting reliable data

The second challenge is to “clean up” and process the packets collected. Between a smartphone and the sniffer devices, there are open spaces, walls that act as screens, surfaces that reflect the signals, and so on. What's more, the smartphone emits signals in a 360-degree range, meaning a packet can be collected by several sniffer devices. Finally, other electronic devices generate signals, resulting in interferences and background noise. 

"e put a lot of work into extracting reliable and accurate information, identifying each smartphone, locating it using a multilateration technique by estimating the distance separating it from several sniffer devices, and tracking it over time.

says Aline Carneiro Viana, "As a bonus, we encountered an unexpected obstacle: MAC address randomisation.” 

These addresses make it a little too easy to identify individual smartphones. As such, from 2014 onwards, manufacturers gradually introduced automatic protocols that randomly change the addresses (known as “randomisation”) every fifteen minutes. “Over two hours of measurements, we recorded up to eight addresses for the same mobile, which meant our results were unusable. Fortunately, we developed a software solution to associate these multiple addresses with the right device, and it’s fairly accurate. ”

Accurately predicting contact between people

The ultimate challenge is to retrace individual movements. The researchers knew from the outset that it would be very difficult to obtain the exact trajectory, not least because of interferences and errors in the measurement environment. This was one of Mitik's challenges 

Consequently, in order to estimate the passage area of each person, they recreated each trajectory in the form of a “mobility envelope”. 

These envelopes are more or less wide-ranging, depending on the accuracy achieved, and include a 5 to 10% margin for errors in estimated distances. “This range is small enough to detect contact between people,” adds Nadjib Achir.

Mitik therefore ends on a very positive note, offering researchers specialising in the exploration of human movements a high-level toolkit. This toolkit is already being used in a research project launched in 2023, Mob-Sci-Dat Factory, itself part of a PEPR (Priority Research Programme and Equipment) on the digitisation and decarbonisation of mobility, MOBIDEC.