Section: Scientific Foundations
Sensor network, distributed measure and distributed processing
- Sensor network for distributed embedded measure.
In order to gather information on the dynamic of a specific physical phenomena, a distributed embedded measure must be performed. The quality of the measure is crucial and largely impacts the analysis. Moreover, by conducing and controlling the measure and its bias during the experiment, one may adapt and optimize the analysis.
Sensors networks offer an efficient way to measure physical phenomena at various space and time scales. The important challenge is to take advantage of a communicating sensor node that can be associated to a physical object in order to design a large scale distributed measurement system that can monitor and sense the physical world. Given a target application, the goal is to design adequate sensor nodes and to set up the way they communicate, cooperate and collect their data in order to fulfill the application constraints. Fulfilling this goal requires the development of theoretical and practical techniques to help the dimensioning and deployment of such distributed sensing tools, to manage the distributed measures and to perform efficient and reliable distributed computing on top of the network. With these main tasks in mind, we define the following objectives:
- Design of a global sensing tool.
Based on the deployment context, we should propose a methodology to design the most appropriate and accurate measurement architecture that matches the application constraints. Heterogeneity is a fundamental, beneficial quality of distributed sensing tool, not just a problem to overcome. Heterogeneous sensing systems are more immune to the weaknesses of sensing modalities and more robust against defective, missing, or malicious data sources than even carefully designed homogeneous systems. However, data heterogeneity also presents main challenges when trying to integrate data from many different sensors.
- Measure characterization and dimensioning.
Measure characterization and dimensioning must take into account the different correlations in space and time that exist between all sensors. It must also handle the various time scales that may exist in the measures. The challenge is due to the heterogeneous data resolution. Moreover, data is generally multi modal and multi scale with possible irregularities and offer much correlation in time and/or space. Data sets collected by various sensors may be characterized by the lack of most common simple statistical properties such as stationary, linearity, or Gaussianity. Relevant time scales may be difficult to identify, or may even not exist. Observed properties have non-trivial relations and even the choice of the time scale granularity that should be used in the measure and the analysis is a complex problem as it may bias the analysis in an uncontrolled way.
- In-network distributed processing.
As some computation can be delocalized closer to the sensed phenomena, in the sensor nodes, space/time correlations can be exploited in order to optimize the amount of data sent through the network.