The problem of full data recovery from a sensor network of thousands of randomly-placed nodes is addressed as it relates to the complexity of the network data. In the applications of interest, the user requires a representation of the full two-dimensional data generated by a field underlying the area of the network. The usefulness of data recovery algorithms is judged by the units of energy consumption to recover sufficient data to reconstruct the field to a specified error. Energy consumption is measured primarily in units of data transmission and reception as these constitute the main limiter of node performance in current and envisioned networks.
A set of novel algorithms broken up into two major classes is simulated on several continuous test fields spanning a broad range of complexity. The first class consists of algorithms that transmit raw data directly to the user with no intra-network processing and differ in the mechanism for which data is selected. The second class consists of algorithms that rely on local data collection and compression prior to sending a full representation of the local field to the user. As the complexity changes, different algorithms become the best performers in terms of the proposed energy measures.
In addition to the continuous field tests, the algorithms are applied to discontinuous fields and images. A lower-bounding algorithm relying on regularly-placed nodes and hierarchical compression is analyzed. The best performing algorithms are simulated in a full system setting using a single energy metric, namely the overall energy consumption in the network in Joules. Depending on the complexity of the data, different strategies have the lowest overall energy cost.
May 31, 2004
Doherty, L. R. (2004). Energy Measures for Sensor Networks. United States: University of California, Berkeley.