New Communication Paradigms for Very Large-Scale Sensor Networks B. Ananthasubramaniam, R. Mudumbai, U. Madhow, J. Hespanha and M. Rodwell Department of Electrical and Computer Engineering University of California, Santa Barbara Santa Barbara, CA 93106 Email: {bharath, raghu, madhow, hespanha, rodwell}@ece.ucsb.edu I. I NTRODUCTION In this poster, we will present work in progress on two novel approaches for efficient, scalable communication in truly large scale sensor networks. This work is a follow-up on our papers in IPSN 2004 [1], [2]. Our focus is on applications in which networks of tens of thousands of sensor nodes are required, such as policing activity along a border, monitoring the presence of biological or chemical agents over large urban areas, or exploration of the surface of a planet. Not only is it difficult to scale standard multihop wireless networking protocols to such large networks, but in many such settings, one long hop is required from the cluster of sensor nodes to a remote collector node even if multihop networking among the sensor nodes is feasible. Another key difficulty in large scale networks is localization: such networks would often be randomly deployed, e.g., dropped from an aircraft or spacecraft, so that there is no a priori mapping between sensor node ID and location. Furthermore, geolocation may be unavailable due to either cost (of integrating GPS receivers into sensor nodes) or physical considerations (occlusion of GPS signals in many scenarios, and unavailability in settings such as interplanetary exploration or monitoring of wooded areas). With the preceding context in mind, we introduce two novel concepts. II. I MAGING SENSOR NETS To solve the problem of scale, we draw our inspiration from conventional imaging (ranging from passive optics to active radar), interpreting the sensor nodes as pixels being imaged by a sophis- ticated collector node. While conventional imaging only applies to phenomena with strong enough electromagnetic signatures, imaging sensor nets utilize sensor nodes to translate arbitrary phenomena into data that can be recovered using radio frequency (RF) imaging techniques, while still allowing scaling to tens of thousands of “pixels” as in conventional imaging. Some examples are shown in Figure 1. In IPSN 2004 [1], we had introduced a version of Imaging Sensor Nets that we termed “Virtual Radar”: a moving collector flies by a sensor field, illuminating sections of it with a beacon. Sensor nodes with activity to report respond in a precisely timed fashion to the beacon when they are illuminated by it, thus creating a radar-like geometry. Modifications of synthetic aperture radar processing are then used to localize the “active” sensors. Such drastic simplification of sensor node functionality has compelling cost implications: we can use “dumb” sensor nodes without geolocation or networking 0 This work was supported by the National Science Foundation under grants CCF-0431205, ANI-0220118 and EIA-0080134, and by the Office of Naval Research under grant N00014-03-1-0090. Fig. 1. Imaging Sensor Nets: three possible realizations capabilities, and obtain an adequate link budget with very little energy expenditure on the part of the active sensor nodes. While our presentation last year illustrated the promise of imaging techniques for sensor net data collection, a number of practical issues must be considered before our ideas can be prototyped. For example, it is difficult for low-cost sensors to process collector’s beacon and respond to it in a precisely timed fashion. Also, it is essential to allow sensors to send back more information than just 0 or 1, as in [2]. Finally, stationary collectors are easier to prototype, and are useful for many applications (e.g., border monitoring from remote sites). With this in mind, we propose the following architecture shown in Figure 2: (a) a stationary collector scans the sensor field with a mechanically or electronically steered beam (the carrier frequency is high enough that a relatively small beamwidth can be obtained); (b) the collector sends a beacon with a spread spectrum “location” code; (c) sensor nodes (ultimately to be implemented as ultra low-cost CMOS ICs) electronically reflect the beacon (with or without am- plification) without processing it, except for modulating it at low rate by the data they wish to send; (d) the collector receiver (to be ultimately implemented in software after an RF front end) uses spread spectrum reception techniques to estimate the delays of the reflected components that it receives, and to demodulate the data; (e) the outputs from (d) are fed to an imaging algorithm that processes the information from different scan angles to obtain an image