Wireless Sensor Networks: Minimum-energy communication Mario Čagalj supervised by prof. Jean-Pierre Hubaux (EPFL-DSC-ICA) and prof. Christian Enz (EPFL-DE-LEG, CSEM) mario.cagalj@epfl.ch Wireless Sensor Networks Large number of heterogeneous sensor devices Ad Hoc Network Sophisticated sensor devices communication, processing, memory capabilities Wireless Sensor Networks: Minimum-energy communication 2 Project Goals Devise a set communication mechanisms s.t. they Minimize energy consumption Maximize network nodes’ lifetimes Distribute energy load evenly throughout a network Are scalable (distributed) Wireless Sensor Networks: Minimum-energy communication 3 4 Minimum-energy unicast Wireless Sensor Networks: Minimum-energy communication 5 Unicast communication model Link-based model each link weighed how to chose a weight? B Power-Aware Metric [Chang00] 1 C 1 1 A 1 E 1 D Maximize nodes’ lifetimes include remaining battery energy (Ei) B x x2 c (e r ) 1 E i ij ij 0 E i e energy spent in transmitting ij r energy spent in receiving 0 cBC C cAB A Wireless Sensor Networks: Minimum-energy communication cCD cAE E cED D 6 Unicast problem description Definitions undirected graph G = (N, L) links are weighed by costs the path A-B-C-D is a minimum cost path from node A to node D, which is the onehop neighbour of the sink node minimum costs at node A are total costs aggregated along minimum cost paths D C Minimum cost topology Minimum Energy Networks [Rodoplu99] optimal spanning tree rooted at one-hop neighbors of the sink node each node considers only its closest neighbors - minimum neighborhood Wireless Sensor Networks: Minimum-energy communication B A 7 Building minimum cost topology Minimum neighborhood notation: N i - minimum neighborhood of node i N P1: minimum number of nodes enough to ensure connectivity P2: no node N falls into the relay space of any other node N i i Finding a minimum neighborhood nodes maintain a matrix of mutual link costs among neighboring nodes (cost matrix) the cost matrix defines a subgraph H on the network graph G 1 c21 c 31 c41 c 51 c12 c13 c14 1 c23 c24 c32 1 c34 c42 c43 1 c52 c53 c54 c15 c25 c35 c45 1 C A Wireless Sensor Networks: Minimum-energy communication B Finding minimum neighborhood 8 We apply shortest path algorithm to find optimal spanning tree rooted at the given node subgraph H Theorem 1: The nodes that immediately follow the root node constitute the minimum neighborhood of the root node Theorem 2: The minimum cost routes are contained in the minimum neighborhood Each node considers just its min. neighborhood Wireless Sensor Networks: Minimum-energy communication Distributed algorithm Each node maintains forwarding table E.g. [originator ¦ next hop ¦ cost ¦ distance] Phase 1: find minimum neighborhood Phase 2: each node sends its minimum cost to it neighbors upon receiving min. cost update forwarding table Eventually the minimum cost topology is built Wireless Sensor Networks: Minimum-energy communication 9 An example of data routing Different routing policies different packet priorities Properties energy efficiency nuglets [Butt01] scalability packets flow toward nodes with increased fault-tolerance lower costs Wireless Sensor Networks: Minimum-energy communication 10 11 Minimum-energy broadcast Wireless Sensor Networks: Minimum-energy communication Broadcast communication model b Eab Eac a c Omnidirectional antennas By transmitting at the power level max{Eab,Eac} node a can reach both node b and node c by a single transmission Wireless Multicast Advantage (WMA) [Wieselthier et al.] Trade-off between the spent energy and the number of newly reached nodes Power-aware metric Ebc 12 include remaining battery energy (Ei) embed WMA (ej/Nj) b Every node j is assigned a broadcast cost c j X2 X1 E j e j Ej b cj X3 U ( j) N j node j ' s neighbourh ood Oij overlappin g set of nodes i and j U j node j ' s uncovered set Wireless Sensor Networks: Minimum-energy communication 13 Broadcast cover problem (BCP) Set cover problem F {S ,..., S m}, S j N 1 Covering C F s.t. N j : S Example: j C S S1 j S3 j C1={S1, S2, S3} cost (S j ) associated with S j C cost (C ) j : S S2 S4 cost (S ) C j S5 C2={S3, S4, S5} C1 , cost (C1 ) cost (C2 ) C2 , cost (C1 ) cost (C2 ) C*= Find cover C * arg min {cost (Ci )} BCP C i Greedy algorithm: Sj Nj at each iteration add the set Sj that minimizes ratio cost(Sj)/(#newly covered nodes) cost ( S j ) e j cost (C ) broadcast cover cost Find cover that minimizes broadcast cover cost The set of forwarding nodes belong to a tree rooted at originator Wireless Sensor Networks: Minimum-energy communication X2 Ej e Xj 1 E j c bj X3 U ( j) Distributed algorithm for BCP Phase 1: learn neighborhoods (overlapping sets) Phase 2: (upon receiving a bcast msg) 1: if neighbors covered HALT 2: recalculate the broadcast cost 3: wait for a random time before re-broadcast 4: if receive duplicate msg in the mean time goto 1: Random time calculation cib random number distributed uniformly between 0 and b c0 Wireless Sensor Networks: Minimum-energy communication 14 15 Simulations GloMoSim [UCLA] scalable simulation environment for wireless and wired networks average node degree ~ 6 average node degree ~ 12 Wireless Sensor Networks: Minimum-energy communication Simulation results (1/2) Wireless Sensor Networks: Minimum-energy communication 16 Simulation results (2/2) Wireless Sensor Networks: Minimum-energy communication 17 Conclusion and future work Power-Aware Metrics trade-off between residual battery capacity and transmission power are necessary Scalability each node executes a simple localized algorithm Unicast communication link based model Broadcast communication node based model Can we do better by exploiting WMA properly? Wireless Sensor Networks: Minimum-energy communication 18 19 Minimum-energy broadcast b Pab a c if (Pac – Pab < Pbc) then transmit at Pac As the number of destination increases the complexity of this formulation increases rapidly. Requirement for distributed algorithm. - forwarding nodes What are good criteria for selecting forwarding nodes? Pac Propagation model: Pab kdab , [2..6] Omnidirectional antennas Wireless Multicast Advantage (WMA) [Wieselthier et al.] Minimum-energy broadcast: Challenges: Pbc Broadcast Incremental Power (BIP) [Wieselthier et al.] Add a node at minimum additional cost Centralized Cost (BIP) <= Cost (MST) 2 8 4 9 1 10 5 Improvements? 4 2 1 Take MST as a reference Branch exchange heuristic… … to embed WMA in MST 2 6 8 5 5 4 7 3 Wireless Sensor Networks: Minimum-energy communication 5