An Adaptive DelayMinimized Route Design for Wireless SensorActuator

An Adaptive Delay-Minimized Route Design for Wireless Sensor-Actuator Networks Edith C. -H. Ngai 1, Jiangchuan Liu 2, and Michael R. Lyu 1 1 Department of Computer Science and Engineering, Chinese University of Hong Kong, Hong Kong 2 School of Computing Science, Simon Fraser University, BC, Canada IEEE MASS 2007, Pisa, Italy 9 Oct 2007 1

Outline • • Introduction Related Work Route Design Problem (RDP) Probabilistic Route Design (PROUD) Algorithm Distributed Implementation Enhancements Simulations Conclusion 2

Introduction WSN • Distributed and large-scale network like the Internet • A group of static sensors • resource constrained • wireless communications 3

Introduction WSAN • Collection of sensors and actuators • Sensors • numerous resource-limited and static devices • monitor the physical world • Actuators • resource-rich devices equipped with more energy, stronger computation power, longer transmission range, and usually mobile • make decisions and actuate adaptively in response to the sensor measurements 4

Related Work Mobile Elements • Architecture using moving entities (Data Mules) to collect sensor data [Shah et. al. SNPA’ 03] • Mobile sinks with predictable and controllable moving pattern [Chakrabarti et al. IPSN’ 03, Kansal et al. Mobisys’ 04] • Mobile sinks can find the optimal time schedule to stay at appropriate sojourn points [Wang et al. HICC’ 05] • Message ferry (MF) approach to address the network partition problem in sparse ad hoc network [Zhao et al. Mobihoc’ 04] 5

Related Work Mobile Elements (cont. ) • Joint mobility and routing algorithm with mobile relays to prolong the network lifetime [Luo et al. Infocom’ 05] • Partitioning-based algorithm to schedule the movement of mobile element (ME) to avoid buffer overflow and reduce minimum required ME speed [Gu et al. Secon’ 05] • Using of mobile sensors and controlled mobility to provide quality of coverage on the fraction of events captured [Bisnik et al. Mobicom’ 06] • Vehicle routing problem (VRP) • Considers scheduling vehicles stationed at a central facility to support customers with known demands • Minimize the total distance traveled 6

Route Design Problem (RDP) Network Model • We consider a WSAN with M mobile actuators and N static sensors • Each of the sensors and actuators are equipped with a wireless • • • transceiver Actuators move in the sensing field along independent routes, at constant or variable speeds Each static sensor has a limited buffer to accommodate locally sensed data When an actuator approaches, the sensor upload the data to actuator and free its buffer Sensors have different weights, according to their data generation rates or event frequencies Sensors with higher weights expect shorter average actuator interarrival times 7

Route Design Problem (RDP) 8

Adaptive Delay-Minimized Route Design (PROUD) Adaptive Delay-Minimized Route Design • To minimize the data collection time in a stochastic and • • dynamically changing sensing environment We propose a probabilistic route design algorithm (PROUD) for wireless sensor-actuator networks This is a departure from the previous static and deterministic mobile element scheduling problems PROUD offers delay-minimized routes for actuators and adapts well to network dynamics and sensors with nonuniform weights This is achieved through a probabilistic visiting scheme along pre-calculated routes 9

Adaptive Delay-Minimized Route Design (PROUD) Small-scale Network • STEP 1: Form a Priori Route • A priori route is formed by constructing a TSP path which contains all locations to be visited • Many polynomial-time approximation algorithm has been proposed for the NP-hard TSP problem • We adopt the well-known Approx-TSP-Tour algorithm here for its low cost and bounded performance 10

Adaptive Delay-Minimized Route Design (PROUD) Small-scale Network • STEP 2: Visit sensors probabilistically • Actuators visit the sensors on the priori route in sequence probabilistically • We set the visiting probability pi of a location i to be wi, where wi is the (normalized) weight of the sensors • Sensors with higher weights should be assigned with a higher probability 11

Adaptive Delay-Minimized Route Design (PROUD) Small-scale Network • STEP 3: Allocate the actuators • The expected route length with probabilistic visiting can be calculated as • For a sensor i with a visiting probability pi, its average actuator inter-arrival time Ai is thus 12

Adaptive Delay-Minimized Route Design (PROUD) Analysis • Time complexity analysis: • Bound analysis: 13

Adaptive Delay-Minimized Route Design (PROUD) Large-scale Network • STEP 1: Form clusters • It divides an MST into two sub-trees by removing its longest edge e 14

Adaptive Delay-Minimized Route Design (PROUD) Large-scale Network • STEP 2: Form priori routes and visit sensors probabilistically • STEP 3: Allocate actuators • Routes with longer expected lengths should be allocated with more actuators 15

Distributed Implementation • STEP 1: Forming R-clusters • Sensors construct MSTs locally by communicating with neighbors • Weight of each edge e in MST is smaller than communication range Rs • We refer such an MST as an R-Cluster, RC(V, E), which includes all the sensors that are within Rs to some sensors in RC(V, E) • The cost of the R-cluster is denoted by Cost(RC), which is the sum of w(e) in RC(V, E) 16

Distributed Implementation • STEP 2: Connecting R-clusters • We divide the network into M subareas, each of which is explored by one actuator • Each actuator looks for the R-clusters in its area and connects them if they are within a certain distance, say |RC 1, RC 2| <= ξ • The new cluster is formed with cost Cost(RC 1) + Cost(RC 2) + |RC 1, RC 2| • Similarly, the actuators also connect their R-clusters/clusters with those in their neighboring areas: 17

Distributed Implementation • STEP 3: Allocating Actuators • Actuators are then allocated to the clusters, such that each cluster is assigned to at least one actuator and no clusters are unassigned 18

Enhancements • Actuators with Variables Speeds • By adjusting the speeds of the actuators, we can ensure sensors with the same visiting probability can achieve similar inter-arrival times • Assume that node i on Rj has a probability pi of being visited by actuator j every cycle. Its average actuator inter-arrival time Ai can be calculated as 19

Enhancements Load Balancing • Multi-Route Improvement Algorithm • Actuators on routes with similar expected length have more balanced energy consumption • Consider two routes R 1 and R 2 involved in multi-route improvement • Their new expected route lengths become ideal if E[R’ 1] = E[R’ 2] = (E[R 1] + E[R 2])/2 • We provide an approximation method to transfer a proportion of sensor locations ζ from MST 1 to MST 2 20

Enhancements Load Balancing • Task Exchange Algorithm • In certain scenarios, it is more energy efficient for one actuator to take up more load than another in the overall energy consumption point of view • It may happen in a network that involves clusters with different sizes or weights • Load balancing among the actuators can be achieved by exchanging their routes 21

Simulations • Simulation parameters • Algorithms: • Partitioning based scheduling algorithm (PBS) [Secon’ 05] • Bounded event loss probability (BELP-2 D) algorithm in 2 D case [Mobi. Com’ 06] • PROUD (Our proposed algorithm) 22

Simulation Results Actuator Inter-arrival Time N = 100, M =5 Uniform random Cluster-based uniform 23

Simulation Results Actuator Inter-arrival Time Cluster-based non-uniform Eye Topology 24

Simulation Results Minimum Actuator Speed N = 100, M =5 Uniform random Cluster-based uniform Cluster-based non-uniform 25

Simulation Results Load Balancing Multi-route improvement Task exchange 26

Conclusion • We focused on WSN with multiple actuators and their route • • • design We studied the route design problem and proposed delayminimized and cooperative solution to coordinate actuators and collect data efficiently We proposed an adaptive route design (PROUD) algorithm It differentiates the visiting frequency to sensor locations with different weights and balances the workload of actuators The proposed algorithm adapts well to the dynamic change of the network and balances the energy consumption of actuators effectively Simulation results suggested that the algorithm remarkably reduces the average inter-arrival time 27

Q&A Thank You! 28