3.2 Measuring Behavior
3.2.2 Proximity Measurement
Recognizing co-location between mobile nodes is an application of lo- calization technologies (see Section 2.2.2) to mobile nodes in a sensor network (see Section 2.2.3). Consequently, co-location was initially seen and measured as a by-product of localization [104, 105]. The NearMe Wireless proximity server [104] computed proximity between mobile users by analyzing the exact location of each mobile node. However, as precise indoor location is still a challenge, a variety of technologies were developed to measure proximity directly without knowledge of the location of both nodes [16, 106, 107, 108, 109, 12]. Because they do not need reference nodes, these systems can operate without the localization infrastructure. Therefore, they can operate independently of the environment and loca- tion. The developed systems can be split into systems that use dedicated hardware [16, 106, 107] and those that reuse the interfaces of the mobile phone [108, 109, 12].
Systems that directly measure proximity calculate the distance by using the TOF or the RSSI and, based on these values, decide if a node is within proximity. The decision is often based on a threshold value that defines the proximity range. However, RSSI values vary because of reflections and interference of multiple or reflected signals. The multipath propagation of the signal can also distort TOF measurements. Repeated measures can minimize these problems, but will increase the required communication overhead. In summary, a higher precision requires more power and reduces the run time of a sensor. Hence, the determined proximity is only a probability that depends on the trade-off between the precision and power consumption that guided the system design.
The mobile phone as a wearable device provides the required capabilities to act as a proximity sensor and has been used as such [108, 110, 12]. The
mobile phone provides a variety of radio interfaces that can be used for TOF or RSSI measurements. Today, nearly everyone carries a smartphone on their bodies. The Virtual Compass system [108] uses WiFi and Bluetooth signals to estimate the distance between mobile phones and laptops. The Bluetooth interface was also used by Eagle and Pentland [109] to measure contacts between students. They built a dataset to analyze student behavior by their contacts. Matic et al. [12] used the WiFi radio to estimate proximity. The benefits of the mobile phone approach are that they are available in large quantities and that they come with rather large batteries. The main problem is that the solutions depend on the actual phone. For example, the used radio chip and the orientation and type of the antenna influence the measurements. Hence, two phones with WiFi may measure different values and the proximity estimation must be device dependent [111]. Furthermore, the implementations are not energy efficient because, the hardware is not designed for proximity measurement and is also used for other purposes. For instance, Matic et al. [12] periodically switch the phone’s WiFi module between client and access point mode.
Dedicated devices can deliver more precise results and can be developed with a smaller form factor. An overview of available devices for proximity measurement is shown in Figure 3.3.
The Sociometric badge [115] combines a number of sensors to measure social contact. A Bluetooth module and a 2.4 GHz receiver are combined
(a) Sociometric Badge [112]
(b) OpenBea- con [113]
(c) Wren Mote [114]
Figure 3.3: Available proximity sensors in chronological order of devel- opment from left to right
to measure proximity. In addition, an infrared sensor recognizes when two persons face each other and voice detection recognizes conversations. The Sociometric badge was successfully used in a variety of domains [112, 115]. The badge is worn around the neck so that the direction of the infrared sensor conforms to the line of sight. The system captures rich data on social interaction, but the additional sensors (e.g. the microphone) may compromise the privacy of third parties. Furthermore, the system’s size is similar to that of a mobile phone.
The OpenBeacon system [113] is an open source initiative to create low-cost proximity sensors for social network research. The system is based on active radio-frequency identification (RFID) and uses a single coin cell to operate over several months. If proximity is recognized, the event is directly transmitted to an access point. Hence, OpenBeacon differs from other dedicated proximity sensors because it requires an infrastructure. The system was used in various settings, including a large conference with more than 500 users [106]. In the conference venue, several access points had to be deployed to measure contacts. Moreover, the access points allow a rough location estimation. While the OpenBeacon system allows large-scale deployments at low costs, infrastructure has to be deployed and managed. OpenBeacon combines mobile devices and stationary access points in a similar fashion as classic localization systems. The access points are powered by cable, whereas mobile devices have to rely on their batteries. Hence, the architecture aims to reduce the load for wearable devices by implementing functionality that consumes more power into the access point. Power-relevant functionality involves, for instance data storage and management of media access (e.g. by synchronizing communication into time slots).
The WREN mote [114] was designed for rapid deployments in large-scale studies. The system uses an 802.15.4-compliant radio to measure proximity and has a rechargeable lithium-polymer battery. The rapid deployment is supported by management racks that can charge and download data from 100 motes at the same time. The system was used with school-age children to research how diseases spread through the social networks [107]. The system should be able to operate for five days at a sampling rate of 20 Hz.
The application domains of proximity sensing are manifold. Envisioned applications include dating services [104], community applications [116],
and location-based services [105]. The majority of these sensor systems target human contact research to analyze businesses [112, 115], social contacts on conferences [106], the impact of social interaction on psychol- ogy [12], and epidemiology research [117, 107]. In all of these cases, the correctness of a single contact is not decisive because a huge amount of data is analyzed and minor errors do not have an observable impact on the results. The data are anonymized and abstracted from the specific situation. Social network analysis (SNA) algorithms [118] have been used to analyze the social network automatically and deduce the underlying patterns. This research, until now, underestimated the value of the detailed data to measure and understand work processes.