## Section: New Results

### Energy-aware Geographical Routing in Wireless Networks

Participants : Essia Hamouda, Nathalie Mitton, David Simplot-Ryl.

Routing in wireless sensor networks is a challenging task. A promising approach is position-based (or geographical) routing approaches. Indeed, such an approaches assume that a node only needs local information (its position, the ones of its neighbors and of the destination) to perform a routing decision). These approaches are memory-less, distributed and scalable, which makes this kind of routing very suitable for wireless sensor networks. Then, several communication primitives may be designed based on these principles. To date, several georouting algorithms have been proposed in the literature but the POPS team has proposed the first protocols for unicast and anycast routing which are both energy-aware and guaranteed delivery.

In [1] , we propose EtE an end-to-end energy-aware routing protocol for wireless sensor networks. EtE is localized and guarantees delivery. To forward a packet, a node s in graph G computes the cost of the energy weighted shortest path between s and each of its neighbors in the forward direction towards the destination which minimizes the ratio of the cost of the shortest path to the progress (reduction in distance towards the destination). It then sends the message to the first node on the shortest path from s to x : say node x^{'} . Node x^{'} restarts the same greedy routing process until the destination is reached or an obstacle is encountered and the routing fails. To recover from the latter scenario,
local minima trap, our algorithm invokes an energy-aware Face routing that guarantees delivery. Our work is the first to optimize energy consumption of Face routing. It works as follows. First, it builds a connected dominating set from graph G, second it computes its Gabriel graph to obtain the planar graph G^{'} . Face routing is invoked and applied to G^{'} only to determine which edges to follow in the recovery process. On each edge, greedy routing is applied. This two-phase (greedy-Face) End-to-End routing process (EtE) reiterates until the final destination is reached.
Simulation results show that EtE outperforms several existing geographical routing on energy consumption metric and delivery rate. Moreover, we prove that the computed path length and the total energy of the path
are constant factors of the optimal for dense networks.

In the anycasting problem, a sensor wants to report event information to one of sinks or actors. In mitton-infocom-09, we describe the first localized anycasting algorithms that guarantee delivery for connected multi-sink sensor-actor networks. Let S(x) be the closest actor/sink to sensor x , and |xS(x)| be distance between them. In greedy phase, a node s forwards the packet to its neighbor v that minimizes the ratio of cost cost(|sv|) of sending packet to v (here we specifically apply hop-count and power consumption metrics) over the reduction in distance (|sS(s)|-|vS(v)| ) to the closest actor/sink. A variant is to forward to the first neighbor on the shortest weighted path toward v . If none of neighbors reduces that distance then recovery mode is invoked. It is done by face traversal toward the nearest connected actor/sink, where edges are replaced by paths optimizing given cost. A hop count based and two variants of localized power aware anycasting algorithms are described. We prove guaranteed delivery property analytically and experimentally.

Nevertheless, despite all the results we have achieved so far in geographical routing, existing theoretical and simulation studies on georouting appear detached from experimental studies in real environments. Therefore, in [14] , we set up our test environment by using WSN430 wireless sensor nodes. To overcome the need for significant number of wireless nodes required to perform a realistic experiment in high density network, we introduce a novel approach - emulation by using relatively small number of nodes in 1-hop experimental setup. Source node is a fixed sensor, all available sensors are candidate forwarding neighbors with virtual destination. Source node makes one forwarding step, destination position is adjusted, and the same source again searches for best forwarder. We compare three georouting algorithms. We introduce here Greedy geographical routing Algorithms in a REal environment (GARE) which builds a RNG by using as edge weight (ETX(uv) counts all transmissions and possibly acknowledgments between two nodes until message is received), and selects RNG neighbor with greatest progress toward destination (if none of RNG neighbors has progress, all neighbors are considered). Our experiments show that GARE is significantly more efficient than existing XTC algorithm (applying RNG on ETX(uv)) in energy consumption. COP GARE selects neighbor with progress that minimizes , and outperforms both algorithms.