Motion Sensing without Sensors: Information Harvesting from Signal Strength Measurements D. Puccinelli and M. Haenggi Department of Electrical Engineering University of Notre Dame Notre Dame, Indiana, USA {dpuccine, mhaenggi}@nd.edu Abstract This letter shows how physical phenomena affecting radio communication can be exploited to turn any wireless network into a wireless sensor network for motion detection without any actual sensing hardware. Motion of individuals or objects in the network area produces shadowing and multipath fading effects altering the received signal strength. The ability to measure signal strength, which wireless terminals normally have, is sufficient to enable motion detection. This idea is particularly suitable for 802.11-based networks, but its simplicity and the lightweight nature of its possible implementations also allow its use as a virtually free add-on feature for lower-end wireless sensor networks. Introduction: In the wireless realm, the term fading refers to deviations of the received signal strength (RSS) from its expected value [1]. Localized fluctuations are due to small-scale fading (a.k.a. multipath fading), whereas stronger variations of the signal are brought about by shadowing 1
and the large-scale path loss. Fading is a spatial phenomenon; temporal variations occur either because of the motion of the terminals or because of changes in the environment where the nodes are deployed. Shadowing falls somewhere in between large-scale path loss and small-scale fading and accounts for large-scale variations in the signal strength due to the interposition of obstacles between two terminals and the consequent interruption of the line-of-sight (LOS). Large obstacles create shadow zones that cause deep fades if a receiver happens to enter them. These physical phenomena modulate the RSS and thus enrich it with information that can be harvested for the detection of changing conditions in the surroundings of the network. In particular, since any such variation is due to some form of motion, the constructive exploitation of fading and shadowing for motion detection is a particularly interesting example of information harvesting and constitutes the focus of this paper. We illustrate our idea with examples obtained with IEEE 802.11b-compliant hardware and two different platforms from the Berkeley mote family: MICA2, equipped with a 433MHz narrowband radio, and MICAz, built around an IEEE 802.15.4- compliant radio operating at 2.4GHz. Principles of RSS-based Motion Detection: The key concept is that the motion of individuals or objects between or near wireless transceivers leaves a characteristic footprint on the RSS. Given a pointto-point wireless link, the normalized receiver gain may be modeled as 2 MX S i G = exp ( j2πd i), (1) d i i=0 where M is the number of paths, d i is the length of the i th path (note that d 0 is the transmitter-receiver distance), and S i is the reflection/penetration coefficient of the i-th path. If individuals or objects move across the LOS path or a strong reflected path, the receiver gain G 2
is subject to significant, abrupt variations whose amplitude can be easily estimated given a particular geometry. For the following numerical example, we focus on the wavelength of MICA2 (λ = 69cm). Let us consider the scenario in Figure 1: in a room, a person or an object moves between two wireless terminals (a transmitter T and a receiver R). When the body is static in its original location away from the terminals, the signal from T propagates to R through 3 different paths: a LOS path and 2 reflected paths (M = 2). We assume a penetration coefficient S 0 = 0.2 in case the LOS is shadowed and S 0 = 1 otherwise. Further, for i > 0 we assume that S i = 0 if the i the path is interrupted, and S i = 0.8 otherwise, due to the reflection off the obstacle. When the body is in position A, it shadows path 1, and the gain at R with respect to the static conditions is about 2.4dB. At position B, the body shadows the LOS, and the gain is -9.7dB. In C, path 2 is blocked off and the gain at R is -1.8dB. This is just an example and is contingent on the particular geometry and propagation conditions described above. However, our experimental experience confirms that an RSS drop of the order of -10dB can be expected when the LOS is shadowed, as shown by Figures 2 and 3. In particular, Figure 2 shows an experiment with MICA2 in which the motes are placed by the sides of a door, and a person exits and reenters the room through that very door. Figure 3 shows a similar motion detection experiment with MICAz; here several people go back and forth in and out of the room. Motion can be easily detected from the clear footprint that it leaves on the RSS; in essence, we are exploiting the shadowing of the LOS. When other paths are blocked off, smaller fluctuations can be expected, which are indicative of motion near the nodes. Note that the shadowing effect is dependent on the distance d 0 between the transmitter and the receiver. As a rule of thumb, d 0 must be smaller than the average distance between the transceivers and the obstacles. In rich scattering environments, d 0 should be as small as possible, as long as the receiver lies in the far field 3
of the transmitting antenna. Application in 802.11-based networks: An RSS-based motion detection scheme is particularly appealing if implemented in the form of an overlay to regular network operation. In networks where wireless terminals routinely exchange packets, motion detection capabilities may be added as long as the radios are able to measure RSS. This enables motion detection at no additional cost: there is no need for sensors or additional hardware of any kind. For instance, any 802.11-based wireless network can be used to log the signal strength in a particular environment. Users can simply leave their wireless network on and later refer to the signal strength log to extract motion-related information betraying a particular kind of activity. As shown in Figure 4, it is sufficient to look at the variations in the RSS to conclude whether or not people are active inside the room where a point-to-point motion detection system is located. A very simple signal processing algorithm can automatically recognize RSS variations. One possibility is comparing a moving average over time windows of different sizes; a difference between ±2dB and ±5dB would be indicative of activity near the terminals, whereas a variation of about 10dB would most likely signal a shadowing effect due to motion between them. The choice of the window sizes obviously depends on the characteristics of the motion events one wishes to detect. For ordinary motion patterns (i.e., people walking into a room), it makes sense to compare a moving average over a window of 10s to a moving average over a window of 1s. The links between an 802.11-compliant terminal and wireless access points of known location can also be exploited. Conclusions: We have introduced the exploitation of signal strength variations for sensorless motion detection in wireless networks. We interpret the information contained in the fluctuations of the RF signal strength in order to detect activity and motion in point-to-point settings, 4
and we illustrate this with examples obtained with simple experimental setups. This form of motion detection is particularly appealing as an added feature to existing wireless networks. It provides a minimal overhead (due to its simplicity), and has a large number of interesting applications, especially in surveillance and monitoring. This idea is not necessarily an alternative, but rather a complement to conventional sensing: our sensorless approach can be used in conjunction with traditional motion sensors for increased robustness and reduction of false alarm rates. Last but not least, another benefit of RSS-based activity detection lies in its educational value. In a classroom setting, RSS-based motion detection can be easily used to impressively demonstrate the causes and the effects of fading. References 1. GOLDSMITH, A. : Wireless Communications, Cambridge University Press, New York, NY, USA, 2005 Authors affiliations: D. Puccinelli and M. Haenggi (Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN 46556, United States) dpuccine@nd.edu Figure captions: Fig. 1 Layout for our motion detection experiment. Fig. 2 Motion detection experiment: two MICA2 transceivers are placed by the sides of a door, and a person walks out and back in the room through the door, leaving a clear footprint on the RSS. 5
Fig. 3 A motion detection experiment with MICAz hardware: people walk in and out of the room, each time leaving an unmistakable footprint on signal strength. Fig.4 Motion detection with an 802.11b link. 6
path 2 3m C Moving Body T B R path 1 A 2m 2.5m Figure 1: 56 Received Signal Strength [dbm] 58 60 62 64 66 68 70 72 0 5 10 15 Time [s] 20 25 Figure 2: 7
50 55 60 RSS [dbm] 65 70 75 80 85 90 0 20 40 60 80 100 120 Time [s] Figure 3: 8
75 80 RSS [dbm] 85 90 Motion near the nodes Motion between nodes 95 0 10 20 30 40 50 60 70 80 Time [s] Figure 4: 9