1. Clock Synchronization in Sensor Networks Mostafa Nouri

2. Outline Need for time synchronization in sensor networks Definition of time synchronization Common problems in time synchronization Typical schemes and algorithms.

3. Sensor Networks Sensor networks and applications  Deeply embedded into the environment  Sense, monitor, and control environments for a long period of time without human intervention Vast collection of miniature, lightweight, inexpensive, energy-efficient sensor nodes

4. Sensor Networks Applications  Biological & Environmental: habitat monitoring, wildlife, pollution, natural catastrophes  Civil: infrastructure, machine health, human health, traffic monitoring  Military: surveillance, tracking, detection Network Architecture / Sensor Platforms Low-power CPU/  C ADC Radio Memory Sensors Battery Base Station

5. Why Need for Time Synchronization Link to the physical world  When does an event take place? Key basic service of sensor networks  Fundamental to data fusion Crucial to the efficient working of other basic services  Localization, Calibration, In-network processing, … Several protocols require time synchronization  Cryptography, Topology management.

6. Sensor Networks Characteristics of SN  Cheap and Small No Accurate Oscillator  Limited Energy Need to Sleep  Work Together Data Fusion Conclusion  must initiate T-Sync in WSN ? ? ? ? ?

7. Metrics for Synchronization Protocols Precision Longevity of synchronization Time and power budget available for synchronization Geographical span Size and network topology

8. Examples Energy efficient radio scheduling Array processing Radio Off Radio On Time Sender Radio Off Receiver Guard band due to clock skew; receiver can’t predict exactly when packet will arrive

9. Computer Clocks Clocks in computers  C(t)=k∫ 0 t ω( τ)d τ + C(t 0 )  ω is frequency of oscillator, C(t 0 ).  Time of the computer click implemented based on a hardware oscillator Computer clock is an approximation of a real time t  C(t)=a*t+b a is a clock drift (rate) B is an offset of the clock Perfect clock  Rate = 1  Offset = 0 C(t) t a b

10. Definition of time synchronization Let C(t) be a perfect clock  A clock Ci(t) is called correct at time t If Ci(t)=C(t)  A clock Ct(t) is called accurate at time t If dCi(t)/dt = dC(t)/dt = 1  Two clocks Ci(t) and Ck(t) are synchronized at time t if Ci(t)=Ck(t) Time synchronization  Requires knowing both offset and drift

11. NTP Overview Most widely used time synchronization protocol  Hierarchical: C/S model Perfectly acceptable for most cases.  Coarse grain synchronization Inefficient when fine grain synchronization is required

12. Why not Use NTP Link  Ratio of packet loss is very low in Internet (fiber, cable)  Links are short range and short lived in sensor networks (wireless) Topology  Static, robust and configurable Energy aware  Frequent message exchange  Listening to the network is free  Using the CPU in moderation free

13. Difficulties in Sensor Networks No periodic message exchange is guaranteed  There may be no links between two nodes at all Transmission delay between two nodes is hard to estimate  The link distance changes all the time Energy is very limited  Nodes sleep most of the time to conserve power Node need to be small and cheap  No expensive clock circuitry

14. Basic Approach Collaboration among sensor nodes  Establish pair-wise relationship between nodes.  Extend this to network level. Two approaches for collaboration  Receiver-Receiver  Sender-Receiver

15. SR vs. RR Sources of error-variation of packet delays and clock drifts AB BA A A B B Beacon

16. Basic Mechanism Pair-wise synchronization T2=T1+delay+offset T4=T3+delay-offset offset=((T2-T1)-(T4-T3))/2 delay=((T2-T1)+(T4-T3))/2 A B h i h j distance T1 T4 T2 T3

17. Sources of Errors Send time  Kernel processing  Context switches  Interrupt processing Access time  Specific to MAC protocol e.g. in Ethernet, sender must wait for clear channel Transmission time Propagation time  Very small in WSNs, can be ommited Reception time Receive time

18. Illustration SenderReceiver NIC Send time Acess time Transmission time Propagation time Reception time Receive time Physical Media

19. Typical Schemes and Algorithms RBS (Reference Broadcast Synchronization) TPSN (Timing-Sync Protocol for Sensor Networs) DTMS (Delay Measurement Time Synchronization) LTS (Lightweight Time Synchronization) FTSP (Flooding Time Synchronization Protocol) TS/MS (Tiny-Sync and Miny-Sync) TSync AD (Asynchonous Diffusion)

20. TPSN Features  Sender-receiver bidirectional mechanism  Two phases of operation: Level discovery and synchronization  Level discovery phase Root node initiates level discovery A node on receiving its level broadcasts it Any node receiving multiple level packets takes the first one and ignores the others Any node not receiving a level packet times out and sends a request for level packet

21. Level Discovery Phase A B H E C F K D G L J I 0 1 2 3 1 1 1 2 2 22 3

22. TPSN Synchronization phase  Root node sends a start synchronization packet  All nodes of level 1 synchronize themselves to the root node  For every level i, the nodes of that level synchronize to the nodes of level i-1  If a node in level i-1 is not synchronized, then it does not respond to synchronization requests from level i

23. Time Synchronization Algorithm A B H E C F K D G L J I 0 1 3 1 1 2 2 22 3

24. LTS Features  Minimize synchronization complexity rather than maximizing accuracy Authors claim that wireless sensor networks need quite low synchronization accuracy  Sender-receiver bidirectional mechanism  Two LTS algorithms Centralized  Node sends a synchronization request to a closest reference node by any routing mechanism Distributed  Requires a spanning tree to be constructed firstly

25. LTS LTS optimizes synchronization frequency with required precision  The synchronization frequency is calculated from the requested precision, from the depth of the spanning tree, from the drift bound Simulation results  500 nodes (120 x 120)  Target precision: 0.5  Duration: 10 hrs  Centralized: 65% of all nodes request synchronization 4-5 synchronization operation on average per node  Distributed: Average 26 pair-wise synchronization per node

26. TS/MS Features  Determining relative offset and drift between two nodes  Sender-receiver bidirectional scheme Node 1 sends a probe message to node 2 time-stamped with t 0 Node 2 generates timestamp t b and responds immediately Node 1 generates timestamp t r  Two clocks c 1 (t) and c 2 (t) are linearly related as C 1 (t)=a 12 C 2 (t)+b 12  The following inequalities are held: t 0 <a 12 t b +b 12 <t r

27. TS/MS C 2 (t) Sample 1 Sample 3 t0t0 trtr C 1 (t) = a 12 H 2 (t) + b 12 C 1 (t) Node 1Node 2 t0t0 trtr tbtb tbtb a 12 b 12

28. Parameter Estimation Relative drift a 12 and offset b 12 The tighter the bound, the higher the synchronization precision Requires high amount of data points and quite complex Algorithms precision increases with the increased number of data points.

29. Difference Between TS and MS Different methods in selecting useful data points  Tiny-Sync Keep only four constrains of all data points Does nor always give the best solution for the bounds  Mini-Sync is an extension of Tiny-Sync More optimal solution with increased complexity Keeps also the data points which may be useful by some future data points to give tighter bounds A data point is discarded only if it is definitely useless Quite complex selection criterion.

30. Reference Elson, Girod, Estrin, “Fine-Grained Network Time Synchronization using Reference Broadcasts” Sichitiu, Veerarittiphan, “Simple, Accurate Time Synchronization for Wireless Sensor Networks” Ganeriwal, Kumar, Srivastava, “Timing-Sync Protocol for Sensor Networks” Genunen, Rabaey, “Lightweight Time Synchronization for Wireless Sensor Networks” Li, Rus, “Global Clock Synchronization in Sensor Networks” Dai, Han, “TSync: A Lightweight Bidirectional Time Synchronization Service for Wireless Sensor Networks”

31. Thank you