We present imaging sensor nets that provide scalable architectures by moving the complexity from the sensor, which plays the role of pixels, to the collectors. This analogy to imaging addresses the key issues of data collection and localization in large-scale randomly deployed sensor networks; in such networks, conventional multihop relay to extract the data is often inapplicable and sensor geolocation capability is expensive. The sensors can be made extremely "dumb" and low-cost with minimum functionality (no geolocation and networking). In one instance of an imaging sensor net, a sophisticated airborne collector node queries the sensor field with a radio frequency beacon. Sensors electronically reflect the beacon adding low-rate data, thus, creating a virtual radar geometry, which provides fine-grained localization, and conveying data to the collector. We show that these reflecting sensors can be located using a Synthetic Aperture Radar-like two dimensional matched filtering algorithm and its decision-directed extensions.

This concept is translated into a millimeter-wave prototype, where a stationary collector sweeps a sensor field with a spread-spectrum beacon that is reflected by an RFID-like sensor after imposing data on it. The collector is built from off-the-shelf components and the semi-passive sensors are implemented on low-cost printed circuit boards. We develop low-complexity algorithms for collector baseband processing to demodulate the data and locate the sensor. Preliminary experimental results demonstrate the feasibility of our approach and ability to support data rates up to 100 kbps, while providing a few centimeters location accuracy.

We investigate source localization based on angle of arrival (AoA) measurements at a geographically dispersed network of collectors in real-world propagation environments. Accordingly, a sequential localization algorithm capable of suppressing the outlying AoA measurements due to multipath is presented. This algorithm achieves close to optimal performance at a complexity that is linear in the number of measurements. A possible application of this algorithm is in another instance of an imaging sensor net, where the sensor transmits its data without prior coordination with a network of collectors that are responsible for locating the sensor and decoding the data.