Processing-aware compression for sensor networks

In sensor networks, where power and bandwidth are at a premium, there is a clear need to use compression to limit the amount of information exchanged by the sensors. We study the signal compression problem in situations where signals are being processed for the purpose of source localization. In these scenarios, compression should be optimized for the accuracy of source localization, rather than to provide a reproduction of the signals with some desired fidelity. We show how this leads to novel design techniques that have clear advantages over standard quantizer design approaches.