1 / 42

Scalable and reliable wireless sensor network systems

Scalable and reliable wireless sensor network systems. Vinod Kulathumani Dept. of Computer Science and Electrical Engineering West Virginia University CS/EE 796 Graduate seminar series. Embedded systems. Found in variety of devices Aircraft, radar systems, nuclear and chemical plants

miriam
Download Presentation

Scalable and reliable wireless sensor network systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Scalable and reliable wireless sensor network systems Vinod Kulathumani Dept. of Computer Science and Electrical Engineering West Virginia University CS/EE 796 Graduate seminar series

  2. Embedded systems • Found in variety of devices • Aircraft, radar systems, nuclear and chemical plants • Vehicles, TVs, camcorders, elevators • > 90% of CPUs used for embedded devices • Part of a larger system • Application known apriori • Little flexibility in programming

  3. Networked embedded systems • What if embedded processors were connected ? • Not wired but wireless Enter Wireless Sensor Networks - Really a network of embedded systems

  4. Enabling technology • Micro-sensors (MEMS, Materials, Circuits)‏ • acceleration, vibration, gyroscope, tilt, motion • magnetic, heat, pressure, temp, light, moisture, humidity, barometric • chemical (CO, CO2, radon), biological, micro-radar • actuators (mirrors, motors, smart surfaces, micro-robots)‏ • Communication • short range, low bit-rate, CMOS radios

  5. The Vision for WSNs • Combine wireless networks with sensing / actuation  Ubiquitous computing • Fine-grained monitoring and control of environment • Network and interact with billions of embedded computers Reasons • Wireless communication - no need for infrastructure setup • Drop and play • Nodes are built using off-the-shelf cheap components • Feasible to deploy nodes densely

  6. Number Crunching Data Storage Mainframe Minicomputer productivity interactive Workstation PC Laptop PDA A new class of computing log (people per computer)‏ streaming information to/from physical world year Slide courtesy: Murat Demirbas

  7. Application areas Science: oceanography, seismology Engineering: industrial automation, structural monitoring Daily life: health care, disaster recovery

  8. Emerging applications • Combination of sensors with mobile devices • Social networking • Participatory urban sensing • Assisted living – health monitoring • Vehicular networks with variety of sensors • Control systems using sensor networks

  9. Trends • Increasing in scale • Increasing in complexity Intel Developer Forum ExScal Intel Hillsboro Fab Middle America Subduction Experiment

  10. Outline of talk • Research challenges / goals • Summary of contributions • Centralized classification / tracking [SRDS’05, Computer Comm’03] • Distributed vibration control [MSNDC’05] • Sensor network service for object tracking [EWSN’07, IPSN’06] • Distance sensitive snapshot service [OPODIS’07] • Details of a specific contribution • Sensor network service for object tracking • Future research interests

  11. My research focus

  12. Interests • Distributed systems / networking • Fault-tolerance • Self-healing systems • Scalability Sensor networks pose plenty of problems in these areas !

  13. Research challenge Industrial, medical, military Observation based / control based Static / mobile Scales: < 100 to 10000 Application Rising in scale, complexity Performance crucial How to design scalable, reliable WSN applications despite unreliable networks ? Middleware services Network design Network abstraction layer Resource constrained nodes Low bandwidth, fading, interference Harsh, malicious environments Network Unreliable

  14. Scenario – asset protection • Dense deployment; Resource and bandwidth constrained • Goal: classify and observe tracks of objects • Application design • Reliable estimation of influence fields [SRDS ‘05] • Influence field (IF) – region over which an object can be detected • IF estimated using binary detections • Classification – Estimating size of IF • Tracking – Estimating shape of IF • Scenario – asset protection • Dense deployment; Resource and bandwidth constrained • Goal: classify and observe tracks of objects • Application design • Reliable estimation of influence fields [SRDS ‘05] • Network design • Network abstractions for IF separation • Distance insensitivity, contention insensitivity • Network abstractions for IF shape • Routing uniformity • Network parameters (density) • Scenario – asset protection • Dense deployment; Resource and bandwidth constrained • Goal: classify and observe tracks of objects • Requirement : low latency (<2 s), high accuracy (> 99%) Aggregator Soldier and vehicle influence fields wrt magnetometer Classification and tracking (monitoring) • Scenario – asset protection • Dense deployment; Resource and bandwidth constrained • Goal: classify and observe tracks of objects • Application design • Reliable estimation of influence fields [SRDS ‘05] • Network design • Network services for separation • Network services for uniformity • Network parameters (density) • Deployment and testing • Line in the sand [Computer Communications’ 03] • ExScal (RTSS’05)

  15. Scenario • Control vibrations during payload launch • Sensors / actuators distributed across surface • Low computational resource, fault-prone • Experimental study on Boeing fairing simulator [MSNDC’05] • Faults impact – potentially severe • Hard to detect in real time • Requirement – mission critical stability Distributed vibration control • Scenario • Control vibrations during payload launch • Sensors / actuators distributed across surface • Application design • Use on-off control scheme • Model plant as linear system; vibration modes assumed • Model unreliability as Byzantine behavior of actuators • Worst input to plant at all times • Network design • Determine actuator placement for plant to be stable despite Byzantine actuators [MSNDC’ 05] • Scenario • Control vibrations during payload launch • Sensors / actuators distributed across surface • Application design • Use on-off control scheme • Model plant as linear system; vibration modes assumed • Model unreliability as Byzantine behavior of actuators • Worst input to plant at all times

  16. Scenario • WSN laid to protect asset • Evader’s goal: minimize distance to asset • Pursuer’s goal: intercept evaders at maximum distance • Pursuers query sensor network for mobile evader locations Distributed tracking – optimal interception • Scenario • WSN laid to protect asset • Pursuers query sensor network for mobile evader locations • Application design • Model as zero-sum game • Formulation of optimal pursuit control strategies [IPSN’06] • Presence of delay • Under discrete sampling rate • Nash equilibrium conditions for successful pursuit  information of nearer objects required at faster rate  information of nearer objects required with lower delay • Scenario • WSN laid to protect asset • Pursuers query sensor network for mobile evader locations • Application design • Model as zero-sum game • Formulation of optimal pursuit control strategies [IPSN’06] • Network design • Trail – a distance sensitive network service • O(d) find time, cost for object distance d away • O(d*log(d)) update time, cost for distance d moved • Fault-tolerant, energy-efficient, family of tunable protocols • Scenario • WSN laid to protect asset • Pursuers query sensor network for mobile evader locations • Application design • Model as zero-sum game • Formulation of optimal pursuit control strategies [IPSN’06] • Network design • Trail – a distance sensitive network service • Deployed and tested in Catch Me If You Can • Demonstrated at Richmond Field Station, Berkeley, August 05

  17. Distance sensitive snapshots in WSN • Scenario • Distributed object tracking using WSN • Goal: Pursuers should eventually catch all evaders • Application design • Perfect information not necessary • State of evaders distance sensitive in error, latency and rate  eventual catch • Network design • Network service for distance sensitive snapshots [OPODIS 07] • Exploit alternate forms of compression to gain efficiency • State of nearby nodes is fresher • State of nearby nodes more precise • State of nearby nodes refreshed more often

  18. Systems built • ExScal (Extreme Scaling Experiment) • Goal: classify between person, soldier, SUV and ATV and track • Deployment area: 1,260m x 288m • 1000+ sensor nodes, 200+ Stargates • Technology transferred to Northrup Grumman • 10,000 node experiment using ExScal software • Roles • Classification / tracking subsystem • Integrating communication chain • Yield studies [ICNP’05] • Identify and study impact of faults ExScal field

  19. Other systems built • Kansei • WSN testbed at Ohio State • 432 TelosB, 150 Stargates, 150 XSM, 100 i-mote2 • Software services for data injection, data collection • Mobile network PeopleNET • Cellphones integrated with psi-mote • Buddy messaging, elevator status • Vehicle classification • Los Alamos National Labs [2007] • Seismic + Acoustic sensors

  20. Trail: network service for tracking

  21. Motivating scenario • Mobile Objects tracked by network of static sensors over a large area • Network runs a tracking service • Application (residing on mobile objects) issues query of the form “Find object X” to the tracking service

  22. Motivation for Trail • Queries answered by one (or more) central nodes not scalable • Depletes energy • Increases latency • One way to make queries local • Publish object state everywhere • But upon every move, global update needed • Global update for every object move not scalable • We need to publish object information systematically

  23. Requirement 1: Find distance sensitivity Network tracking service returns query results in time and work proportional to distance from object Requirement 2: Update distance sensitivity When an object moves, tracking protocol updates the track in time and work proportional to distance moved Informal problem statement

  24. Trail tracking structure • Trail protocol based on geometric ideas • Properties proved on continuous 2-d plane • Then implemented on discrete plane • Model • 2-d real bounded plane, C denotes center of this plane • Cost measured in Euclidean distance • One track maintained for each object • Let P be object being tracked located at point p • Tracking data structure for P denoted as trailP • Pointers that lead to current location of P • All tracks rooted at C

  25. Trail intuition • If trailP restricted to be a straight line, each move will involve update from C p’ C p • Instead, trailP marked with vertices on-the-fly • Vertices serve as anchor points for update • Distance between vertices increases exponentially moving towards C • Anchor updated depending on distance moved • After sufficiently large distance, update from C

  26. C C C C C C N3 N3 N3 N3 N3 N3 N2 N2 N2 N2 N2 N2 N1 N1 N1 N1 N1 N1 c2 c2 p p p p c2 c2 c3 c3 c3 c1 c3 c1 c2 c2 c3 c1 p c1 p c3 c1 c1 Examples of trailP

  27. N3 N2 N1 c2 p’ p c3 c1 Cost for update and find Theorem Cost of updating trailP over a move of distance d is O(d*log(d)) worst case structure: log spiral

  28. Algorithm for find • Draw successive circles of radii 20, 21, 22 .. 2(log dist(C,q)) • Until trailP is intersected • Or reach C • Follow trailP to reach current location of P C Theorem N3 Cost of finding P from object Q at point q is O(d) where d is dist(p,q) m N2 q Cost includes reaching trailP, following trailP, returning to q N1 c3 p c2

  29. Fault-tolerance and adaptivity of Trail • Fault-tolerance • Nodes may fail after creating trail or old trails may not be deleted • Self-stabilizing actions using heartbeats along trail structure • Tolerating failures during update and find • Route around failures using a method such as left hand rule in GPSR • As size of holes increases, update and find cost proportionally increase • Trail supports graceful degradation • Adaptivity (Trail yields family of protocols) • Can be tuned based on update and query frequency • When query frequency higher, publish structure increases and find increasingly straight • Extreme case – find is a straight line to C and publish in circles

  30. Performance evaluation • Experimental evaluation (Kansei testbed at OSU) • Used to demonstrate PE tracking application for NEST DARPA project • Intruder tracks collected from Richmond Field Station [140m X 60m] • Tracks injected into Kansei testbed nodes to emulate motion of evaders • 15 X 7 node network, 3 ft spacing • 3 pursuer 3 evader scenario • Study effect of interference on scaling in • Objects [2 - 10] • Query frequency [0.25 – 1 Hz] • Simulations [JProwler] • 8100 nodes (90 by 90) • Up to 50 objects (uniformly separated and collocated) Garcia Robots as Pursuers

  31. Summary of Trail features • Trail – a distance sensitive network service • Assumes no hierarchical partitioning of network • O(d) find time, cost for object distance d away • O(d*log(d)) update time, cost for distance d moved • Fault-tolerant • Self-stabilizing, graceful degradation • When many objects come close together, network interference can cause delay • Synchronized push version? • Distance sensitive snapshot service

  32. Distance sensitive snapshot service A brief overview

  33. Informal problem statement Given • N nodes, with bounded memory, in f dimensions • each can sense m-bit information at any time • each can communicate at W bits per second Deliver a global snapshot • at each node (can be relaxed to a subset) • that uniformly has distance sensitive latency (and distance sensitive resolution, and distance sensitive rate) • State of nearby nodes is fresher • State of nearby nodes more precise • State of nearby nodes refreshed more often • periodically, as fast as possible(can be relaxed to lower rate)

  34. Illustration

  35. Illustration

  36. Results Maximum staleness in state of a node i received by a snapshot at node j is O(log(n) * m * d) where d = dist(i, j) Resolution of state of a node i in a snapshot received at node j is Ω(1 / d2) where d = dist(i, j) Communication cost to deliver a snapshot of one sample from each node to all nodes is on average O(N * log(n) * m)

  37. Conclusions • Research focus • Reliable network services for WSN applications • Applications for classification, tracking, distributed control • Network services tested in actual field deployments • Key role in integrating complete WSN systems • ExScal, Line in the Sand, Kansei, Catch Me If You Can • Facility monitoring at Los Alamos National Labs • Provided deep insight into real problems in wireless and sensor networks

  38. Future research interests WSNs combined with mobility, actuation

  39. Mobile heterogeneous wireless networks • Convergence of mobile devices with sensors • Urban surveillance, online health monitoring, disaster relief, mobile gaming, vehicular networks • Realization of ubiquitous systems • Research questions • Low power self – localization of mobile units • Scenarios: low cost indoor tracking, security, identity management • Reliable, secure information management • Protect against eavesdropping, jamming • Provide reliable content delivery • Architecture • Composing applications across heterogeneous networks [MODUS 2008] • Convergence / inter-operability with Internet, cellular networks

  40. Wireless sensor networks for control • WSNs suited for control applications • Wireless feature: industrial control and process control applications • Large scale feature: control of distributed parameter systems, power grids • Challenges / research questions • Performance • How to guarantee reliability / low latency and meet wire-line standards? • How to secure the network against jamming? • Architecture • Underlying network independent of control system / application ? • Theory • Joint stabilization of control application and network layer

  41. Information processing Control systems Wireless communication technology Database systems Data Mining Computer vision (urban surveillance) MEMS / sensor fabrication Cross cutting research • Network protocols • Network architecture • Reliable • Secure

  42. Thank you Contact Information Vinod Kulathumani Vinod.kulathumani@mail.wvu.edu http://www.csee.wvu.edu/~vkkulathumani

More Related