Multi-Session Data Gathering with Compressive Sensing for Large-Scale Wireless Sensor Networks

This paper studies the scaling law of the data gathering capacity of large-scale wireless sensor networks. Many previous researches on data gathering capacity focus on a many-to-one scheme, but we study the capacity in a multi-session data gathering paradigm, where some of the nodes in the network act as sinks and each sink has a set of source nodes to collect data. The analysis of this paradigm is meaningful in that it may be more commonplace in wireless sensor networks, because in real world, we often hope different sinks to get different kinds of data from sensors deployed in the same region. In the multicast scenario, a source node just sends the same data to all of its destinations, while in multi-session data gathering, the sink node has to receive different data from all its sensor nodes, which makes the last hop to the sink become a capacity bottleneck. We use compressive sensing (CS), a newly introduced sampling theory, to simplify the analysis of data gathering capacity into a similar way as the situation of multicast. Meanwhile, compressive sensing can achieve a capacity gain of $k/M$ for each data gathering session.