• Like
  • Save

Reliability analysis for wireless sensor networks

  • 1,150 views
Uploaded on

Monitoring and control systems that are based on networked wireless sensors have been recognized as an indivisible component for current and future smart systems in many applications such as …

Monitoring and control systems that are based on networked wireless sensors have been recognized as an indivisible component for current and future smart systems in many applications such as healthcare, home security, disaster response, and environmental monitoring. From the viewpoint of researchers, developers and even consumers, reliability analysis is an indispensable step before wireless sensor network systems can be widely deployed for mission-critical applications. In this talk, reliability modeling and analysis for wireless sensor networks will be discussed under two different communication paradigms: infrastructure communication and application communication. Five data delivery models (sink unicast, sink anycast, sink multicast, sink manycast, and sink broadcast) will be presented for the infrastructure communication reliability analysis. Impact of different routing algorithms and network topology characteristics like connectivity, average path length, average degree, diameter, and clustering coefficient on the network reliability will also be discussed.

More in: Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
1,150
On Slideshare
0
From Embeds
0
Number of Embeds
2

Actions

Shares
Downloads
2
Comments
0
Likes
2

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Reliability Analysis for Wireless  Sensor Networks (无线传感器网络可靠性分析) Dr. Liudong Xing   (邢留冬博士) ©2012 ASQ & Presentation Xing ©2012 ASQ & Presentation Xing Presented live on Nov 17th, 2012http://reliabilitycalendar.org/The_Reliability Calendar/Webinars_‐ y_ /_Chinese/Webinars_‐_Chinese.html
  • 2. ASQ Reliability Division  ASQ Reliability Division Chinese Webinar Series Chinese Webinar Series One of the monthly webinars  One of the monthly webinars on topics of interest to  reliability engineers. To view recorded webinar (available to ASQ Reliability  ( y Division members only) visit asq.org/reliability To sign up for the free and available to anyone live webinars  To sign up for the free and available to anyone live webinars visit reliabilitycalendar.org and select English Webinars to  find links to register for upcoming eventshttp://reliabilitycalendar.org/The_Reliability Calendar/Webinars_‐ y_ /_Chinese/Webinars_‐_Chinese.html
  • 3. Reliability Analysis for Wireless Sensor Networks (无线传感器网络可靠性分析) Presented by Dr. Liudong Xing (邢留冬) E-mail: lxing@umassd.edu University of Massachusetts, Dartmouth www.massachusetts.edu ASQ Reliability Division Webinar Series 2012 US National Science Foundation No. 1112947 & 1112935
  • 4. Wireless Sensor Networks (WSN) A network consisting of many spatially-distributed sensor devices for monitoring physical or environmental conditions and cooperatively passing their data through the network to a main location http://en.wikipedia.org/wiki/Wireless_sensor_network 2
  • 5. WSN Communication Infrastructure communication  Relates to delivery of configuration and maintenance messages  From base station (sink node) to sensor nodes Application communication  Relates to transfer of sensed data collected from physical environment  From sensor nodes to base station http://monet.postech.ac.kr/research.html 3
  • 6. WSN Graph Model: G(V, E) V: sensor nodes E: wireless links (i, j) ϵ E iff d(i,j) ≤ tr  d(i,j): Euclidean distance between nodes i and j  tr: transmitting range; a node can communicate with other sensor nodes within a Euclidean distance of tr  sr: sensing range; a node can monitor any point that is within a radius of sr from that sensor 4
  • 7. Agenda WSN Topologies Infrastructure Communication Reliability Application Communication Reliability 5
  • 8. WSN Topologies (1) Star  Organizes peripheral nodes around central hub Hierarchical/Tree  Natural and logical extension of star  Sink at the root and nodes at different layers connected via direct links Mesh  Each node also functions as router  Multi-hop communication  Multiple paths through the network 6
  • 9. WSN Topologies (2) Hierarchical clustering  Sensor nodes form clusters  Cluster heads in a lower layer are arranged into clusters in higher layer  A cluster head is assigned for each layer cluster 7
  • 10. Which topology is the most reliable one? b a b a c c d d Base Station Base Station Node deployment Mesh b a Level-1 Cluster Head Level-0 Cluster Head c Gateway Node Ordinary Sensor Node d Base Station Hierarchical cluster Tree 8
  • 11. Intuitively Speaking... Star/Tree (least reliable)  when a link is obstructed, there are no alternate paths from affected node to base station Mesh (most reliable)  highly fault tolerant: offers multiple redundant paths through network Hierarchical cluster (intermediate)  maintain multiple redundant paths through network  cluster heads: single-point of failures A. Shrestha and L. Xing, “Quantifying Application Communication Reliability of Wireless Sensor Networks,” International Journal of Performability Engineering, Special Issue on Reliability & Quality in Design, 2008; 4(1): 43-56 9
  • 12. Example Verification c b a d Failure rate (hr-1) for nodes and links Links Base station Cluster head Nodes Base Station 2e-6 1e-7 1e-6 1e-6 b a Level-1 Cluster Head Level-0 Cluster Head c Gateway Node Reliability values for mission time of 10,000 hours Ordinary Sensor Node d Base Clustered Mesh Tree Station  Hierarchical Base Station a 0.96407761 0.93120456 0.91301771 b 0.96404580 0.95712310 0.91301771 c 0.99340368 0.93124373 0.91301771 d 0.96495799 0.95960433 0.91301771 a, b, c, d 0.92097894 0.83314192 0.80977407 10
  • 13. Agenda WSN topologies Infrastructure communication reliability  Data delivery models  Network characteristics Application communication reliability 11
  • 14. Data Delivery Models Unicast  To a single sensor Infrastructure communication Multicast reliability (ICR): probability  To a group of sensors that there exists operational Broadcast path from sink node to ......  To all sensors Anycast  To any one sensor out of a group of qualified sensors Manycast  To a subset of sensors out of a group of qualified sensors 12
  • 15. ICR under Unicast Probability that there exists an operational path from sink node (sink) to destination sensor node (a). Example: tree topology    E 2 sink to PN t ht  E 2 PN t ht to PN t 1ht 1      Pr        ICRunicast 1h to PN 0 h  E PN 0 h to a E 2 PN 1   0 2 0  hk: PN that is hierarchically above a at level-k, 0 ≤ k ≤ t E2: event - there exists an operational path between a given pair of nodes Pr(E2): two-terminal reliability (BDD-based method) L. Xing, “An Efficient Binary Decision Diagrams Based Approach for Network Reliability and Sensitivity Analysis,” IEEE Trans. Systems, Man, and Cybernetics, Part A: Systems and Humans 2008; 38 (1): 105-115. 13
  • 16. ICR under Anycast Probability that there exists an operational path from sink to any one sensor node out of a qualified group (Q) Q = {n1, n2}       E 2 sink to PN t ht  E 2 PN t ht to PN t 1ht 1      ICR anycast  Pr    1 0   aQ E 2 PN h1 to PN h0  E 2 PN h0 to a 0        14
  • 17. Example Results (unicast, anycast) Failure rates: link (2e-6/hr), sink (5e-7/hr), sensor node (1e-6/hr) n2 1 n1 n5 PN1 Q1 PN2 n3 0.8 Sink Q2 n4 Reliability 0.6 Q3 Level-1 PN Level-0 PN Q4 Sensor Node 0.4 Q1 = {n1} (unicast) 0.2 Q2 = {n1, n2} Q3 = {n1, n2, n3} 0 0 1 2 3 4 Q4 = {n1, n2, n3, n4} Mission time: hours 5 x 10 15
  • 18. ICR under Multicast Probability that there exists an operational path from sink to all the sensor nodes in a qualified group (Q) Q = {n1, n2}    iH          E 2 sink to PN t i    iH t E 2 PN t i to PN t 1 j    t   jH t 1   ICR multicast  Pr      1   0     0      ijH1 E 2 PN i to PN j    iaH 0 E 2 PN i to a        H 0    Q  16
  • 19. ICR under Manycast  Probability that there exists an operational path from sink to at least one subset of nodes (Rx) out of a qualified group (Q) Q = {n1, n2, n3, n4} n = 4, m = 2     Cnm  iH t ,x       jH t 1,x     E 2 sink to PN t i  iH t ,x E 2 PN t i to PN t 1 j    ICRmanycast  Pr         x 1     iH1,x E PN 1i to PN 0  j    iH 0 ,x E PN 0 i to a     jH 0 ,x 2  aRx 2          17
  • 20. Example Results (multicast, manycast) 1 n2 n1 Q1 n5 0.8 PN1 PN2 n3 Q2 Sink Reliability n4 0.6 Q3 Level-1 PN Level-0 PN 0.4 Sensor Node 0.2 Q1 = {n1, n2} (multicast) 0 Q2 = {n1, n2, n3} 0 1 2 3 4 Q3 = {n1, n2, n3, n4} Mission time: hours 5 x 10 18
  • 21. ICR under Broadcast Probability that there exists an operational path from sink to all sensor nodes in WSN     i  E 2 sink to PN t i     i ,j    E 2 PN t i to PN t 1 j       ICRbroadcast  Pr          E PN  i ,j 2    1i to PN 0  j      i ,a 2   E PN   0 i to a     19
  • 22. Example Results (all models for treetopology) 1 Unicast 0.8 Anycast Multicast Reliability 0.6 Manycast 0.4 Broadcast 0.2 0 0 1 2 3 4 Mission time: hours 5 x 10 C. Wang, L. Xing, V. M. Vokkarane, and Y. Sun, "Reliability of Wireless Sensor Networks with Tree Topology," 20 International Journal of Performability Engineering 2012; 8 (2): 213-216
  • 23. Example & Results for Clustering Topology n1 n5 n2 Level-1 CH Level-0 CH 1 Level-1 Gateway n3 CH2 Level-0 Gateway CH1 1 Sink n4 Sensor Node 1 1 1 Q1 Broadcast Multicast Q2 Multicast:Q1 Manycast 0.8 0.8 0.8 Manycast:Q2 Anycast Q3 Unicast Q4 Reliability 0.6 ReliabilityReliability 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0 0 0 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 Mission time: hours 5 Mission time: hours 5 Mission time: hours 5 x 10 x 10 x 10 Q1 = {n1} (unicast) Q2 = {n1, n2} Q1 = {n1, n2} Q = {n1, n2, n3} Q3 = {n1, n2, n3} Q4 = {n1, n2, n3, n4} Q2 = {n1, n2, n3} C. Wang, L. Xing, V. M. Vokkarane, and Y. Sun, "Manycast and Anycast-Based Infrastructure Communication Reliability for Wireless Sensor Networks," The 18th ISSAT Intl Conf. on Reliability and Quality in Design, Boston, MA, July 2012 21
  • 24. Summary WSN with anycast is the most reliable WSN with broadcast is the least reliable WSN with manycast is more reliable than WSN with multicast for a given qualified group WSN reliability increases as number of sensor nodes in the qualified group increases for anycast and manycast models 22
  • 25. Agenda WSN topologies Infrastructure communication reliability  Data delivery models  Network characteristics Connectivity, average path length, average nodal degree, network diameter, clustering coefficient Application communication reliability 23
  • 26. Connectivity A graph is connected if every pair of vertices is connected via a path A graph is k-connected if the graph remains connected when fewer than k vertices are deleted from the graphAverage Path Length Average number of hops along the shortest path for all possible pairs of network nodes Indicates the efficiency of information transfer over the network 24
  • 27. Average Nodal Degree Nodal degree: number of edges connected to the node Average nodal degree for entire network is the average number of edges connected per node Indicates the density of a networkNetwork Diameter The longest path length of all the shortest paths for all possible pairs of network nodes Indicates the linear size of a network 25
  • 28. Clustering Coefficient For a node  ratio of existing links connecting a node’s neighbors to each other to the maximum possible number of such links  average fraction of pairs of neighbors of a node which are also neighbors of each other For the entire network  average of clustering coefficient of all the nodes of the network 26
  • 29. Five-Node Connected Graphs 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 failure rates: links (2e-6/hr), nodes (1e-6/hr) 27
  • 30. Number k-connected Average Path Length Average Nodal Degree Network Diameter Average Clustering Coefficient 01 1 1.6 1.6 2 0 02 1 1.8 1.6 3 0 03 1 1.5 2 2 0.43 04 1 1.6 2 3 0.33 05 1 1.6 2 3 0 06 1 1.4 2.4 2 0.6 07 2 1.4 2.4 2 0 08 2 1.3 2.8 2 0.8 09 1 2 1.6 4 0 10 1 1.7 2 3 0.47 11 1 1.4 2.4 2 0.87 12 1 1.5 2.4 3 0.53 13 1 1.3 2.8 2 0.7 14 2 1.5 2 2 0 15 2 1.4 2.4 2 0.33 16 2 1.3 2.8 2 0.77 17 2 1.2 3.2 2 0.87 18 2 1.3 2.8 2 0.4 19 3 1.2 3.2 2 0.67 20 3 1.1 3.6 2 0.9 28
  • 31. Connectivity 1 0.9 1-connected 2-connected 0.8 3-connected Fully-connected 0.7 0.6 Reliability 0.5 0.4 0.3 0.2 0.1 0 Time: Hours 29
  • 32. Average Path Length 0.7 t=100000hours t=200000hours t=300000hours 0.6 0.5 0.4 Reliability 0.3 0.2 0.1 0 Average Path Length (Graph Number) 30
  • 33. Average Degrees 0.7 t=100000hours t=200000hours t=300000hours 0.6 0.5 Reliability 0.4 0.3 0.2 0.1 0 Average Degree (Graph Number) 31
  • 34. Network Diameter 0.7 t=100000hours t=200000hours t=300000hours 0.6 0.5 Reliability 0.4 0.3 0.2 0.1 0 Diameter (Graph Number) 32
  • 35. Average Clustering Coefficients 0.7 t=100000hours t=200000hours t=300000hours 0.6 0.5 Reliability 0.4 0.3 0.2 0.1 0 Average Clustering Coefficient (Graph Number) 33
  • 36. Summary In general, higher connectivity, shorter average path length, larger average nodal degree, and shorter network diameter lead to higher network reliability Clustering coefficient property is not a good indicator of the network reliability C. Wang, L. Xing, V. M. Vokkarane, and Y. Sun, "Reliability Analysis of Wireless Sensor Networks using Different Network Topology Characteristics," Proc. of Intl Conf. on Quality, Reliability, Risk, Maintenance, and Safety Engineering (QR2MSE2012), 34 Chengdu, China, June 2012.
  • 37. Agenda WSN Topologies Infrastructure Communication Reliability Application Communication Reliability 35
  • 38. Application Communication Acquisition of sensed data from a specific area by senor nodes  Related to sensing coverage: ability to monitor every point in the region by at least one node  More generally, K-coverage requires every point to be covered by at least K sensors Having a reliable communication from sensor nodes which observe data to the sink node.  Related to network connectivity and routing procotols WSN ACR = Pr {(every point in the sensed field is observed by at least K nodes) AND (there exists an operational path from each of these nodes to sink node)} 36
  • 39. K-Coverage Every point in the area that is covered by at least K sensors 1-covered 2-covered 3-covered K Unit-disk Coverage (K-UC)  each sensor has the same sensing range K Non-unit-disk Coverage (K-NC)  each sensor may have a different sensing range 37
  • 40. K-Coverage Analysis The distance between each point in the monitored area and each sensor is calculated to check which points are in the sensing range of which senor. Each sensor is associated with a matrix modeling all points within the monitored area; an element in matrix is 1 if the corresponding point is within sr of sensor. Adding all sensors matrixes to obtain an overall coverage matrix: ratio between # of elements not less than K and total # of elements in the matrix  K-coverage analysis. 38
  • 41. Example 98.29% of the whole area is covered by at least one sensor node corresponding to 1- coverage 65.08%: 2-coverage 44.6%: 3-coverage 30 sensors (sr=1) randomly distributed a 5 by 5 area (density of 1.2 sensors /sq.) 39
  • 42. Effect of Density on Coverage Percentage of the whole area that is K-covered (K=1, 2 and 3) increases as density increases. 40
  • 43. K-coverage Set A minimal set of sensors such that each point in the specific area is covered by at least K different sensors.  Identify all sensors that cover the whole or part of the specific area  For all possible combinations of those sensors.  Check summation of the corresponding matrixes for all sensors in the combination  If each element of the summation matrix is greater than or equal to K, then that combination supports K- coverage.  Remove combinations that have redundancy. 41
  • 44. K-coverage Reliability Probability that all points in the specific area are covered by at least K different sensors RK  Pr{ SN1,1  SN1,2 ... SN1, M1  SN 2,1 SN 2,2 ... SN 2, M1  ...  NK  Mi    SN N K ,1 SN N K ,2 ... SN N K , M N   i 1  j 1  }  Pr   SNi, j  ,     K  NK : number of K-coverage sets  Mj : number of sensors in ith K-coverage set  SNi,j : jth sensor node in ith K-coverage set Evaluation methods  Inclusion/exclusion (I/E), Sum of Disjoint Products (SDP), or BDD A. E. Zonouz, L. Xing, V. M.Vokkarane, and Y. Sun, “K-coverage Reliability Evaluation for Wireless Sensor Networks,” The 18th ISSAT International Conference on Reliability and Quality in Design, Boston, MA, July 2012 42
  • 45. Example: Randomly Deployed WSN 50 sensors (sr=1.5m, tr=2m, λ=5e-5/hr ) are randomly distributed in an 8m by 8m area. {1}, {10}: 1-coverage {1, 10}:2-coverage 43
  • 46. Example: Predefined Deployed WSN λ=5e-5 K K-coverage sets for area (0.5 ~ 1, 0 ~ 0.5), 1 {{2}, {3}, {6}, {8}, {5, 10}} 2 {{2, 3},{2, 6},{2, 8},{3, 6},{3, 8},{6, 8},{2, 5, 10}, {3, 5, 10},{5, 6, 10},{5, 8, 10}} 3 {{2, 3, 6},{2, 3, 8},{2, 6, 8},{3, 6, 8},{2, 3, 5, K-coverage reliability decreases as 10},{2, 5, 6, 10},{2, 5, 8, 10},{3, 5, 6, 10},{3, 5, 8, 10},{5, 6, 8, 10}} K value increases for the same 4 {{2, 3, 6, 8},{2, 3, 5, 6, 10},{2, 3, 5, 8, 10}, deployment {2, 5, 6, 8, 10}, {3, 5, 6, 8, 10}} 5 {{2, 3, 5, 6, 8, 10}} 44
  • 47. Effect of Density on K-coverage ReliabilityK-UC reliability (sr=1, λ=5e-5) for K-NC reliability (avg sr=0.6, λ=5e-5) forA) 2.5 sensors/sq. A) 2 sensors/sq. B) 3 sensors/sq.B) 5 sensors/sq. C) 4 sensors/sq. Larger K-coverage can be supported as density becomes higher For specific K, WSN with higher density provides higher K-coverage reliability 45
  • 48. Application CommunicationReliability (ACR) Communication reliability of delivering the observed data from sensor nodes within the identified K-coverage sets to sink node  N K th   ACR  Pr  i K - coverage set  sink    i 1  Two single-path routing algorithms:  Shortest-path distance algorithm (D): Dijkstra’s algorithm  Shortest-path hop algorithm (H): Breadth-first search (BFS)A. E. Zonouz, L. Xing, V. M.Vokkarane, and Y. Sun, “Application Communication Reliability of Wireless Sensor NetworksSupporting K-coverage in the Presence of Shadowing,” IEEE International Conference on Communications, 2013 (under review) 46
  • 49. Link Unreliability Lognormal shadowing radio propagation model    r   10 log     1   tr    ,   ; iff r  1, PLink (r )  1  erf  2  2 log(10)    tr        tr: transmitting range of sensor node  ψ: ratio between standard deviation of shadowing (σ) and pathloss exponent (η) 47
  • 50. Example: Predefined Deployed WSN 20 sensors (sr=1.5m, tr=2m, λ=5e-5/hr) in a 5m by 5m area Monitored area: (0.5 ~ 1, 0.5 ~ 1) {2}, {9}: 1-coverage {2, 9}: 2-coverage 48
  • 51. ACR Results D algorithm is more reliable than H algorithm. Both algorithms generate paths with 3 hops, but links on paths generated by D are shorter and thus more reliable. Single-paths from sensor #2, # 9 to sink D algorithm: {212821} {911621} H algorithm: {21521} {911621}. 49
  • 52. Example: Randomly Deployed WSNParameter Value# of Sensors 20 (density of 0.8)Sensing range (sr) 1.5mTransmitting range (tr) 2mFailure rate (λ) 5e-5Deployment area 5m by 5mSpecific area (0.5~1,0.5~1)mChannel condition (Ψ) 2Sensor node failure rate 5.0 e-5 (fph) D algorithm is more reliable than H algorithm at the beginning (sensor node has high reliability, effect of link reliability on ACR is relatively more significant) H algorithm can be more reliable than D algorithm as time passes (reliability of sensor node decreases greatly, its effect on ACR would become more significant; D algorithm may involves more hops/nodes) 50
  • 53. Different Channel Conditions ψ=2 ψ=5 Similar trend can be observed ACR results with larger ψ are smaller because link failure probabilities increase with increasing value of ψ (worse channel condition) 51
  • 54. Different Network Densities density=0.8 density=1Density 0.8 1 Increasing density leads to shorter pathsDiameter 6.435447 5.926134 and fewer hops involved in sending theAvg. node degree 6.34919 8.667 sensed data to the sink node  better ACRClustering coefficient 0.70534 0.72497 resultsAvg. distance using D 3.1885 2.9408Avg. no. of hops using H 3.556 3.337 52
  • 55. Conclusion WSN reliability under infrastructure communication and application communication were discussed  Different network topologies: start, tree, mesh, hierarchical clustering  Different data delivery models: unicast, anycast, multicast, manycast, and broadcast  Different network characteristics: connectivity, average path length, average nodal degree, network diameter, clustering coefficient  Different routing algorithms: shortest-path distance (D) and shortest-path hop (H)  Different K-coverage requirements and densities 53
  • 56. Thank You!谢谢! Dr. Liudong Xing (邢留冬) E-mail: lxing@umassd.edu Phone: +1-508-9998883http://www.ece.umassd.edu/faculty/lxing/home.html 54