Our research objective is to route data to mobile sinks under hard latency constraints. One way to address latency constraints it to decrease the distance that data has to travel along the low-bandwidth 805.15.4 links. We address this problem in our two-tier architecture by routing data from sensors (data sources) to their cluster-heads first. Cluster-heads then cooperate to find a cluster where the user (data sink) is located and deliver data to this cluster over high bandwidth 802.11 links. Data packets again travel along 805.15.4 links in the final part of the route.

From the networking point of view, routing data from a data source to its cluster-head is straightforward: any collection tree protocol works fine. However, a number of alternatives exists for transmitting data from a cluster-head to a sink. We are currently exploring a cluster-wide flooding, point-to-point routing, route-to-sink along collection tree and sink-centered collection tree approaches.

Cluster-wide flood routing

Two main problems need to be addressed when data sink changes its location: 1) routing data-structure needs to be updated, and 2) data packets en route to the sink need to be re-routed to the new location. Clearly, cluster-wide flood requires no change of the data structure, but introduces larger communication overhead. Other two routing approaches can greatly benefit from timely prediction of the sink's next communication neighbor, as the sink moves around. We are currently exploring RSSI based neighbor prediction techniques, in particular, locally weighted linear regression, Gaussian process regression, and discrete path selection based on training sample set.