1. Localized Techniques for Power Minimization and Information Gathering in Sensor Networks EE249 Final Presentation David Tong Nguyen Abhijit Davare Mentor: Farinaz Koushanfar

2. Outline Introduction Assumptions Project Goals Problem Formulations Related Work 1. Node coordination for power minimization 2. Network traversal algorithm 3. Generation of optimal solution Experimental Results Conclusions Future Work

3. Introduction Ad-Hoc wireless sensor networks Unattended autonomous operation Limited energy sources Idle power consumption Not just point to point routing, but gathering information using only local information  Uncertainty about node status (active, standby)

4. Assumptions for the sensor network Unit Disk Communication Model  Nodes can communicate iff (Euclidean distance  R c ), where R c is fixed communication range E Communication >> E Computation E idle ~ E Communication (While radio is on) The algorithms run above the MAC layer protocol Node Information includes its ID, geographic position and status (active/standby) Each node has information about its neighbors

5. Project Goals Efficient localized node coordination for extending the network lifetime Power efficient information gathering method  Gathers the queries from all of the nodes within a predefined area in the deployment field  Attempts to visit as few nodes as possible, minimizing communication energy consumption

6. Problem Formulations 1 – Localized power efficient coordination:  Objective: Maximize the number of nodes in standby mode using only local information.  Constraints: Global network connectivity should be preserved, i.e. A node cannot go into standby if it disconnects the network. 2 – Localized efficient information gathering:  Objective: Minimize the number of communications required for gathering complete information from a network, where some nodes are in standby. 3 – Generation of optimal solution for network traversal  Objective: Find a network and an optimal traversal path through that network that minimizes the number of nodes visited while gathering data from each node

7. Related Work Power aware MAC layer  PAMAS [Kravets et al. 2000], [Woo et al. 2001], S-MAC [Ye et. al. 2002] Coordination power saving strategies  Span [Chen et al. 2001], GAF [Xu et al. 2001] Ascent [Cerpa et. al., 2002] They do not state necessary and sufficient conditions for putting a node in standby & have less power savings. Network discovery  Birthday protocols [McGlynn et al. 2001], TopDisk [Deb et. al. 2002], ad-hoc routing survey [Stojmenovic et al. 2002] We also consider the network shape & regions of low density. Perimeter routing  Guaranteed delivery [Bose et al. 2001], GSPR [Karp et al. 2001] They did not consider perimeter routing for studying the shape of the network and has just used it for coming out of local minima in routing.

8. 1 - Efficient Node Coordination for Power Minimization We guarantee that enough nodes stay active to maintain network connectivity  Necessary and sufficient condition for putting a node into standby is to ensure an alternate path exists between any two of its neighbors Fair power saving method Only local information used

9. 1 - Initial Phase: Token Assignments Token defines the current active node that has the control of the flow of procedure Distributed local computation  multiple tokens required Handshaking between tokens is done through a semaphore-like mechanism During the initial phase, tokens are assigned to the nodes  Such that every node has a token  Tokens act in a localized area

10. 1 - Node Selection for Standby Mode Each token uses updated information from its local area to make a decision. Token considers itself and its neighbors. Each token “locks” the nodes it is considering. Token chooses node whose neighbors will be able to communicate for the longest time if the node stays in standby mode. Each node sleeps for T s interval, dependent on the energy in its local area Token is then passed to node which gone the largest amount of time without obtaining the token (“miss me?”)

11. 1 - Parameter Tuning Flexibility in choosing: Ts vs. density Number of tokens Ts vs. number of tokens

12. 2 - Information Gathering: What is new? While there exist many point-to-point routing algorithms, no major contribution for complete area traversal. Guarantee complete information gathering Graph theoretic and geometric abstraction of the network area:  Perimeter (shape) of the network  Ranking w.r.t connectivity Completely localized traversal procedure

13. Find the perimeter of the network using method similar to Right-Hand Rule. 2 - Perimeter Routing Starting Node Problem: If edges cross in the network, right-hand rule fails

14. 2 - Perimeter Routing - Planarizing Solution: Planarize the Graph uv w To include the edge (u, v) in the graph, the shaded circle must not contain any node w. (Gabriel graph planarization)

15. 2 - Partitioning the graph While traversing the perimeter, find partitioning points of the planarized graph. Starting Node Partitioning Point

16. 2 - Traversal Method Network traversal begins at a perimeter node Next node is determined locally according to:  Rank – Distance from perimeter  Parity – Even or odd rank  Section – Prefer unvisited nodes in same section  Novelty – # of unvisited neighbors

17. 3 - Generating Instances with Known Solutions To accurately evaluate the quality of network traversal heuristic, must know the optimal solution. However, given a network, generation of optimal network traversal is NP-complete. Alternative: Generate an optimal solution first, then generate network around it.  A path through the network is the optimal if each node on the path has at least one unique neighbor, and no other nodes have unique neighbors

18. 3 - Generating Instances with Known Solutions (cont’d) 1. Place the initial node randomly. 2. Choose a unique neighbor node within r c. 3. Choose a second node that is in range of the first node, but not of the unique node 4. Iterate until path is required length 5. Filler nodes can be added that are in range of the path nodes

19. Experimental Results: Coordination As number of nodes in the network increases, standby strategy becomes more effective

20. Experimental Results: Network Traversal Algorithm a b

21. Conclusions Tremendous energy savings using a localized standby strategy Necessary and sufficient conditions to maintain the network connectivity Energy efficient information gathering, which uses both geometric and graph theoretic information

22. Future Work Find efficient information dissemination methods Integrate other power saving strategies into the simulations