Distributed m-D State Space Models

This project focuses on in-situ distributed processing and sensing of signals in a grid sensor network. The outlined method provides optimal load balancing, fast processing, minimum communication overhead, and in-situ marking of an event.

Wireless sensor networks with regularly placed nodes in a grid configuration are useful in certain applications including structural monitoring, agriculture, surveillance, and target tracking. Such applications use nodes that are spatially distributed in 1-D, 2-D or 3-D arrays over the area of observation, each generating sensor data over time. In fact, grid-based, periodic sensor deployment becomes essential when regular spatial sampling is required. Previously, researchers have theoretical analyzed the capacity bounds and performance limits of lattice sensor networks, and the robustness of grid-based deployment in wireless sensor networks.

Nodes in these networks are often equipped with more than one sensor type. A single sensing modality might not be sufficient in certain detection tasks. Moreover, multi-modality sensing offers increased robustness and accuracy in decision making, the rationale being that individual modalities provide complementary information. Given the large numbers of nodes employed in certain applications, this approach also has the potential to significantly decrease the overall system deployment and maintenance cost. This is due to the fact that a few, expensive, high accuracy sensors can be replaced by many inexpensive sensors that are sensing multiple modalities.

Distributed information processing methods are attractive in grid-based wireless sensor networks due to limitations in energy, radio range and data throughput. Network lifetime can be significantly extended and the system robustness can be improved by using distributed algorithms. Such algorithms minimize unnecessary transmission of information over the network and allow the network resources to deplete evenly across the network. Furthermore, applications requiring local actuation in response to a local detection are best supported by such distributed algorithms, yielding a minimum response delay as compared to centralized schemes.

In this research, we propose a novel method for distributed multi-modality information processing in grid sensor networks. It is an amalgamation of two fundamental theories: (a) Fornasini-Marchesini (FM) multidimensional (m-D) distributed state space model and (b) Dempster-Shafer (DS) evidence theory. The advantages of this method includes its general model applicable to implementation of any linear system, extremely high scalability, ease of reconfiguration, minimized communication costs and extended system lifetime, integration of multiple sensing modalities for robust detection, and its support for local actuation based on local processing results. Distributed evidence filter implementations based on this approach can successfully exploit spatio-temporal correlations in grid sensor networks using data derived from multiple sensor modalities.

To see the resume of the expert associated with this case study, see the link below.