1. Acoustic localization for real-life wireless sensor network applications Michael Allen Cogent Computing ARC in collaboration with: Centre for Embedded Networked Sensing, UCLA WaveScope project, CSAIL, MIT

2. Wireless networked sensing Wirelessly networked, embedded, battery powered, sensor enabled computers Wirelessly networked, embedded, battery powered, sensor enabled computers Sample and process data about a physical phenomena Sample and process data about a physical phenomena Temperature, light, sound, image Temperature, light, sound, image Aims/advantages Aims/advantages Cheap, pervasive, collaborative Cheap, pervasive, collaborative Distributed computation Distributed computation

3. My Research Physical phenomena is sound - Acoustic localization: Physical phenomena is sound - Acoustic localization: For self/node-localization (locate nodes using acoustics) For self/node-localization (locate nodes using acoustics) For source localization (locate acoustic event of interest) For source localization (locate acoustic event of interest) Real-life aspect Real-life aspect Real problems/questions, real environments Real problems/questions, real environments Systems research (reliability, robust behaviour) Systems research (reliability, robust behaviour) Field-usable tools Field-usable tools Theoretical aspect Theoretical aspect Design principles, algorithms Design principles, algorithms Scalability Scalability Data fusion Data fusion

4. Motivating applications Primary motivation: bioacoustics Primary motivation: bioacoustics Acoustic source localization of animals/bird calls Acoustic source localization of animals/bird calls Position estimation is helpful for behaviour analysis Position estimation is helpful for behaviour analysis Problems Problems Exploratory systems development is often required Exploratory systems development is often required Currently available platforms are not suited to this Currently available platforms are not suited to this

5. Current work - VoxNet An Interactive platform for bioacoustics research An Interactive platform for bioacoustics research Hardware and software Hardware and software Forms real-life, systems aspect of thesis research Forms real-life, systems aspect of thesis research Allow on-line and off-line operation Allow on-line and off-line operation React to events in-field React to events in-field Full data set gathered at node Full data set gathered at node Network consists of x nodes and 1 sink Network consists of x nodes and 1 sink Sink is endpoint for programs Sink is endpoint for programs Nodes talk over multi-hop IP to sink Nodes talk over multi-hop IP to sink Sink/control

6. Hardware – Acoustic ENSBox More capable than current WSN research platforms: More capable than current WSN research platforms: 32-bit ARM CPU, 64MB RAM 32-bit ARM CPU, 64MB RAM Four channel 48KHz audio Four channel 48KHz audio wi-fi/802.11b wi-fi/802.11b internal battery (5-10hr) internal battery (5-10hr) Rapidly deployable: Rapidly deployable: Attended, short-lived deployments Attended, short-lived deployments Self-localization and time synchronisation: Self-localization and time synchronisation: cm accuracy acoustic based localization (up to 100m range) cm accuracy acoustic based localization (up to 100m range) 10us time synchronisation across network 10us time synchronisation across network V2 (2007) L. Girod, M. Lukac, V. Trifa, and D. Estrin. "The Design and Implementation of a Self-calibrating Acoustic Sensing Platform." in Proc. of SenSys 2006

7. Deployment in Colorado Acoustic localization application running on platform Acoustic localization application running on platform In-situ, on-line operation (detecting marmots) In-situ, on-line operation (detecting marmots) Nodes run adaptive event detectors Nodes run adaptive event detectors Signal energy in frequency bands of interest Signal energy in frequency bands of interest On detection, data is passed to sink (4 channels/node) On detection, data is passed to sink (4 channels/node) Sink clusters together related events Sink clusters together related events Makes DoA estimates based on each node’s detection Makes DoA estimates based on each node’s detection Estimates position from crossing of DoAs Estimates position from crossing of DoAs Allen, M., Girod, L., Newton, R., Madden, S., Blumstein, D., Estrin, D., “VoxNet: An Interactive, Rapidly-Deployable Acoustic Monitoring Platform”, International Conference on Information Processing in Sensor Networks (IPSN 2008)

8. Problems/Observations Latency problems Latency problems Uncoordinated, interfering network traffic Uncoordinated, interfering network traffic Event grouping at sink Event grouping at sink Grouped by arrival time – BAD Grouped by arrival time – BAD Events arrive out of order, late Events arrive out of order, late Overall position estimate took far too long Overall position estimate took far too long Link quality Link quality Multi-hop data transfer latency Multi-hop data transfer latency

9. Improvements On-line clustering algorithm On-line clustering algorithm Group events based on detection time Group events based on detection time Smart event grouping Smart event grouping Nodes only send notification of detection Nodes only send notification of detection Sink requests data Sink requests data Adaptive behaviour trade-off Adaptive behaviour trade-off Nodes monitor network links Nodes monitor network links Decide to process locally or pass raw data Decide to process locally or pass raw data

10. Future work Scalability of acoustic localization networks Scalability of acoustic localization networks Coverage, density – they make sense? Coverage, density – they make sense? Bounds on performance Bounds on performance Data fusion for position estimate Data fusion for position estimate Quickest way to get data and fuse it Quickest way to get data and fuse it Information theory/Bayesian approaches to data fusion Information theory/Bayesian approaches to data fusion