This is a collaborative research proposal that involves the Net-Centric Software Systems I/UCRC sites at Arizona State University and the University of North Texas. Advances in sensor design coupled with improvements in data processing and communication are creating new possibilities for developing sensor networks that can monitor borders, events and people. These advances include continuing development of sensor modalities, integration of collaborative signal processing algorithms, and developments in wireless ad hoc networks. These algorithms and methods can be used in several net-centric applications - such as border control - where microphones and accelerometers collect acoustic data and data from events such as footsteps and use signal processing algorithms to detect, localize, and extract appropriate information to describe an acoustic scene. Furthermore, it is possible to manufacture simple and low-cost sensors, which when deployed in large numbers, are capable of monitoring with high precision, and pass on this information via wireless links. The individual sensors or a dedicated fusion center performs inference from the aggregated data. It is imperative for the distributed inference systems to be robust to a wide variety of sensing, and channel impairments. One such impairment is network interference which corrupts the sensing and communication and is typically non-Gaussian with an unknown distribution. This necessitates the joint design of the sensing apparatus and the communication system while respecting low-power constraints. A robust statistic of the sensed data as well as one which satisfies peak-power and computational requirements are required. The marriage between robust statistics which traditionally works with nonlinear operations of ranks and signs of the data, and transmit power constraints for communications, sparks a transformative connection between statistics and communication theory. We propose to design a suite of robust, distributed detection and estimation algorithms for topologies with or without fusion centers with the following approach: 1. Design of nonlinearities at the sensor side which create robust statistics in the presence of unknown noise distributions and low peak power constraints and quantify the robustness versus statistical efficiency tradeoffs in the presence of sensing and channel impairments. 2. Consider spatial diversity by multiple antenna reception at the fusion center, and frequency diversity through ultra wideband communications for distributed inference. 3. In network topologies without a fusion center, we propose to extend robust processing at the sensors to networks where each sensor shares information with its neighbors for the purpose of reaching consensus on a statistic for robust detection and estimation. 4. Optimize the algorithms to implement them on low-power systems. 5. Experiments that use the theories developed, and at the same time serve to verify the theories, especially while implemented on a real-time sensor platform.
|Effective start/end date||7/1/12 → 6/30/15|
- National Science Foundation (NSF): $198,300.00