Chapter 11: Time Synchronization

1
Chapter 11Time Synchronization

2
Time Synchronization
What is time synchronization in sensor networks?

HHMMSS
120110
120110
120110
Objective Allow all sensor nodes to maintain the
same time frame
3
Time Synchronization

• Temporal relations play an important role in
  sensor fusion ? Which event happened first?
• Physical time is itself part of information
• Estimation of target position Direction, Speed
• Providing fire breaking time
• Due to the clock drift, the local clock needs to
  be periodically synchronized to maintain an
  accurate global time

4
Time Synchronization Challenges

• Why is it so difficult to synchronize the sensor
  nodes?
• Low-end clock crystals are used
• Clock drifting may be significant and clock
  jitters may occur often
• Communication links are noisy
• Some sensor nodes may become unsynchronized
• Node failures occur often
• Cannot depend on a single sensor node to be the
  master clock

5
Factors Influencing Time Synchronization

• Temperature
• Temperature variations during day may cause the
  clock speed up or down (a few ?sec/day).
• Phase Noise
• Access fluctuation at the hardware interface,
  response variation of the operating system to
  interrupts, jitter in delay, etc.
• Frequency Noise
• The frequency spectrum of a crystal has large
  sidebands on adjacent frequencies.

6
Factors Influencing Time Synchronization

• Asymmetric Delay
• The delay of a path may be different for each
  direction
• Clock Glitches
• Hardware or software anomalies may cause sudden
  jumps in time

7
Sources of Time Synchronization Error

• Sending Time (Time spent at the sender to
  construct the message)
• Kernel protocol processing
• Variable delays caused by OS
• Transfer within the host to NIC
• Access Time (Delay caused to wait for access to
  the channel)
• Specific to MAC protocol

8
Sources of Time Synchronization Error

• Propagation Time
• Can be neglected for air
• Important for underground, underwater
• Receiving Time
• Processing required for the antenna to receive
  the message from the channel and notify the
  processor of its arrival (A/D conversion)
• Common Denominator non-deterministic!!!!

9
Further Difficulties

• Periodic message exchange is not guaranteed to
  occur among nodes
• Transmission delay between two nodes is hard to
  estimate

10
TERMINOLOGY

• FREQUENCY
• The rate at which a clock progresses.
• CLOCK OFFSET
• Difference between the time reported by a clock
  and the real time.
• SKEW OF A CLOCK
• Difference in the frequencies of the clock and
  the perfect clock.
• DRIFT (RATE) of a CLOCK
• Second derivative of the clock value with respect
  to time.

11
Traditional Synchronization
Problem for SYNCH Many sources of unknown,
nondeterministic delays (send time, access time,
propagation time and receive time) between sender
and receiver
Sender
Receiver
Send Time
Receive Time
At t1
Radio
Radio
Access Time
Physical Media
Propagation Time
12
Reference Broadcast Synchronization (RBS)J.
Elson, et. al., Fine-Grained Network Time
Synchronization using Reference Broadcasts,
Proc. of the Fifth Symp. on Operating Systems
Design and Implementation (OSDI 2002), Boston,
December 2002.
Does not try to sync Sender Receiver Tries to
sync receivers
Sender
Receiver
Receiver
Receive Time
Radio
Radio
Radio
I saw it at t4
I saw it at t5
Physical Media
Propagation Time
13
TRADITIONAL SYNCH vs RBS
Radio
Radio
Sender
Sender
Receiver

Receiver 1

Receiver 2

Time
Critical Path
Critical Path
Traditional Critical Path From the time the
sender reads its clock, to when the receiver
reads its clock Nondeterministic delay Send
time and Access time. Receiver time small.
RBS Only sensitive to the differences in receive
time and propagation delay Critical path length
is shortened to include only the time from the
injection of the packet into the channel to the
last clock read. (Send time and Access time
are eliminated!)
14
Reference-Broadcast Synchronization (RBS)

• Algorithm to Estimate the Phase Offset between
  the Clocks of Two Receivers
• A transmitter broadcasts a reference packet to
  two (or more) receivers
• Each receiver records the time at which the
  packet was received according to its local clock

15
Reference-Broadcast Synchronization (RBS)

• 3. The receivers exchange the observed times at
• which they received the packet.
• 4. The clock offset between two receivers is
• computed as the difference of the local times
  at
• which the receivers received the same message.

16
Observations about RBS

• RBS removes SEND and ACCESS TIME errors
• Broadcast packet is used as a relative time
  reference

17
Observations about RBS

• Each receiver synchronizing to a reference packet
• Ref. packet is injected into the channel at the
  same instant for all receivers
• Broadcast packet does not contain timestamp
• Almost any broadcast packet can be used, e.g ARP,
  RTS/CTS, route discovery packets, etc

18
Phase Offset Estimation

• RBS can produce highly accurate results if
• Message reception by each receiver is high
• Each receiver can record its local clock reading
  as soon as the message is received.

19
Phase Offset Estimation

• Messages can be corrupted.
• Receiving node may not be able to record the time
  of the message arrival promptly (maybe busy with
  other computations)
• Solution Use a sequence of reference messages
  from the same sender rather than a single message

20
Phase Offset Estimation

• Receiver i will compute its offset relative to
  any other receiver j as the average of clock
  differences for each packet received by nodes i
  and j
• Result For all i,j ? n
• m
• Offseti,j 1/m ? (Tj,k - Ti,k)
• k1

n Number of receivers m Number of reference
broadcasts Ti,k Node is clock when it received
the broadcast k
21
RBS Phase Offset Estimation Performance
Evaluation

• In each trial, n nodes were given random clock
  offsets, m message transmission times were
  selected at random.
• Each synchronization message was delivered to
  every receiver and time-stamped using the
  receivers clock.

22
RBS Phase Offset Estimation Performance
Evaluation

• Since every receiver computes its offset with
  every other receiver, O(n2) offsets were computed
  and then compared with the real/actual offsets.
• The maximum difference between computed and
  actual offsets was considered to be the Group
  Dispersion.

23
RBS Phase Offset Estimation Numerical Analysis
Results

• 2-receiver case
• 30 ref broadcasts improve precision from 11 ?sec
  to 1.6 ?sec
• 20-receiver case
• Dispersion reduced down to 5.6 ?sec

24

RBS - Multi-hop Communication
Features

• Use nodes that receive two or more reference
• broadcasts from different transmitters as
• translation nodes

Receivers
Transmitter
Translation Nodes
25
RBS - Multi-hop Communication

• Multiple hops could introduce a high degree of
  variability in message transmission time between
  multiple receivers of the same message ? RBS
  would lose its accuracy
• To avoid loss of precision
• Two nodes located in different neighborhoods are
  synchronized using a third node lying in the
  intersection of the two neighborhoods

26
Time Routing

• Compute a time route -gt dynamically determine a
  series of nodes through which time can be
  converted to reach a desired final time-base
• The physical topology can be easily converted to
  a logical topology links represent possible
  clock conversions

27
Time Routing
1
2
5
1
2
5
A
B
6
6
3
4
7
3
4
7
C
8
9
8
9
D
10
11
10
11
Use shortest path search to find a time
route Edges can be weighted by error estimates
28
TIME ROUTING

• The nodes A,B,C,D send synchronization packets.
• Other nodes are receivers.
• The transmission radii are shown by circles

29
TIME ROUTING

• Logical Topology e.g., E1(R1)-gtE1(R4)-gtE1(R8)-gtE1
  (R10)
• Edges are drawn between nodes that have received
  a common broadcast and therefore have enough
  information to relate their clocks to each other.
• Receivers 8 and 9 share two logical links because
  they have two receptions in common from C and D.

30
ADVANTAGES OF THE RBS SCHEME

• Largest sources of error (send and access times)
  are removed from the critical path by decoupling
  the sender from the receivers.
• Clock offset and skew are estimated independently
  of each other
• In addition, clock correction does not interfere
  with either estimation because local clocks are
  never modified

31
ADVANTAGES OF THE RBS SCHEME

• Post-facto synchronization prevents energy from
  being wasted on expensive clock updates
• Multi-hop support is provided by using nodes
  belonging to multiple neighborhoods (i.e.,
  broadcasting domains) as gateways

32
DISADVANTAGES of the RBS SCHEME

• For a single-hop network of n nodes, it requires
  O(n2) message exchanges (computationally
  expensive for dense networks)
• Convergence time can be high due to large number
  of message exchanges.

33
DISADVANTAGES of the RBS SCHEME

• Reference sender is left unsynchronized in this
  method.
• If reference sender needs to be synchronized, it
  will lead to a significant waste of energy

34
DISADVANTAGES of the RBS SCHEME

• Limited to only one broadcast domain
• Translation errors and forwarding delays make
  RBS impractical and time difference may occur
• Different time-of-occurrences may occur at
  different sinks

35
Timing-Sync Protocol for Sensor Networks (TPSN)
S. Ganeriwal, et.al., Timing-Sync Protocol for
Sensor Networks, ACM SenSys, November 2003.
Features

• Have many root nodes
• Organize into multiple level hierarchy
• Synchronize level-by-level

Root Nodes
A
B synchronizes to A C synchronizes to B
B
C
36
TPSN Basic Concept

• First create a hierarchical topology in the level
  discovery phase
• As a result, each node is assigned a unique
  level 0 (root node)?1?2??N

37
TPSN Basic Concept

• A time synchronization phase is initialized by
  the root node
• Each node synchs to the node which is one level
  lower than itself

38
TPSN Algorithms

• Level Discovery
• Time Synchronization
• New Node Level Request
• Root Node Dies Local Leader Election

39
TPSN Level Discovery

• Root node assigned level 0
• Root node initiates the phase by broadcasting a
  level-discovery packet
• Upon receiving the packet, each node assigns
  itself a 1 level and broadcasts its own new
  level-discovery packet

40
TPSN Level Discovery

• Eventually each node in the network is assigned a
  unique level
• Once a nodes level has been decided, it ignores
• all such packets in the future.

41
TPSN Time Synchronization

• Root node starts this phase by broadcasting a
  time_synch packet
• Nodes in level one wait for some random time and
  start a two way message exchange to synch their
  local clocks
• This is carried out throughout the whole network

42
TPSN Time Synchronization
T2
T3

• At time T1, A sends a sync pulse packet to B,
  which contains the level number of A and T1.
• Node B receives this packet at T2
• T2 T1 D d
• D clock drift between the two nodes
• d propagation delay

Node B
Node A
T1
T4

• At time T3, B sends back an acknowledgement
  packet to
• A with values of T1, T2, T3 and level of B.

43
TPSN Time Synchronization
A can calculate the phase (clock) offset D (clock
drift) and delay d as
(T2-T1) (T4-T3)
(T2-T1) - (T4-T3)
d
D
2
2
Node A corrects its clock to synchronize with
node B, based on the computed offset.
44
TPSN

• When a new node joins a network or when it does
  not get a level discovery packet
• The node waits for some time to be assigned a
  level (listens for a level discovery packet).
• If it is not assigned a level within that period,
  it times out and broadcasts a level-request
  packet.
• The neighbors reply to this request by sending
  their own level and the new node defines its
  level to be one greater than the level it
  received.

45
TPSN When a node i loses all itsneighbors with
level i1

• Protocol is vulnerable to node failures
• When this happens, it is possible for a node at
  level i not to have a neighbor (i.e., a
  synchronization server) at level i-1
• In such cases, the node at level i would not
  receive an ACK to its synchronization message.

46
TPSN

• Node retransmits a message for a fixed number of
  times before assuming that it has lost all its
  neighbors
• If the node does not receive any responses to its
  synchronization messages, it broadcasts a
  level-request packet with a new level

47
TPSN

• Maximum number of retransmissions of a
  synchronization message is subject to two
  constraints.
• If too large, it will increase the time taken for
  synchronization (i.e., the convergence time)
• If too small, a node may erroneously conclude
  that its server has died
• Unnecessary message flooding in the network
  (optimal number of retransmissions is four!)

48
TPSN When the root node dies

• Level 1 nodes will time-out on ACK packets to
  root
• Instead of Level Request query, they run an
  election algorithm and select a new root node
• A brand new level, Discovery Phase is started
  until all nodes get reassigned their new level.

49
TPSN vs RBS
TPSN RBS
Average error (µs) 16.9 29.13
Worst case error (µs) 44 93
Best case error (µs) 0 0
Percentage of time error is less than or equal to average error 64 53
50
ADVANTAGES TPSN

• Scalable
• Synchronization accuracy does not degrade
  significantly as the size of the network is
  increased
• Network-wide synchronization is effectively
  achieved
• Computationally less expensive

51
DRAWBACKSTPSN

• Energy conservation is not very effective
• Requires a physical clock correction to be
  performed on local clocks of sensors while
  achieving synchronization
• Requires a hierarchical infrastructure
• Unsuitable for applications with highly mobile
  nodes
• Support for multi-hop communication is not
  provided

52
DRAWBACKS TPSN

• Forms islands of time
• Whole network does not have the same time frame
• Nodes become unsynchronized when mobile
• All nodes are predefined in a hierarchy.
• No end-to-end common time frame
• Assumed the end users will be able to interpret
  the time gathered from the network
• Requires a pre-defined reliable hierarchy of
  nodes.

53
Time-Diffusion Synchronization Protocol W. Su
and I.F. Akyildiz, Time-Diffusion
Synchronization Protocol for Sensor Networks,
IEEE/ACM Transactions on Networking, April 2005

• Principle problem with other protocols
• Rely on particular nodes to be time servers or
  masters
• Need for a robust solution that
• automatically self-configures
• sensitive to energy requirements

54
Time-Diffusion Synchronization Protocol

• CASE 1
• Precise time servers (sinks act) are present
• Sinks broadcast a reference time to all master
  nodes which are randomly elected to synchronize
  their neighbors
• Master nodes in turn use the received reference
  time to synchronize their neighbor nodes.
• Goal Equilibrium time that the sensor network
  reaches is the reference time broadcast by the
  sinks

55
Time-Diffusion Synchronization Protocol

• CASE 2
• No precise time servers are present
• Sinks cannot be used as time servers
• Line-of-sight or connection to all master nodes
  from sinks may not be possible
• The sensor network should still synchronize on a
  consistent time.

56
Time-Diffusion Synchronization Protocol
Translates the TDP time to the time used by the
users.
Focus on TDP
57
Time-Diffusion Synchronization Protocol

• All sensors have a local time that is within a
  small bounded time deviation from the
  network-wide equilibrium time.
• Protocol operates in alternating active and
  inactive phases.
• Within each active phase there are multiple
  cycles (iterations), each cycle lasting a
  duration t.

58
TDP SCHEDULING

• Protocol operates in alternating active and
  inactive phases
• Active period depends on the convergence time
• Inactive period depends on the tolerance range
  due to local clock drifting
• Performs several iterations (cycles t) frequently
  re-electing master nodes in active periods.
• Does nothing in inactive period.

59
Time-Diffusion Synchronization Protocol

• Every iteration is made of several subparts
• Election/Re-election of Leaders
• False Ticker Isolation
• Load Distribution
• Peer Evaluation
• Time Diffusion
• Time Adjustment

60
Flow Chart of TDP
61
Entire Protocol Operation During One Cycle

• Peer evaluation procedure (PEP) in each cycle is
  performed
• To determine the master node eligibility for the
  next cycle
• Diffused leader responsibility for the remainder
  of the current cycle

62
Entire Protocol Operation During One Cycle

• Step 1 A master (level 1) and its neighbors
• Master sends a large number timestamped scan
  messages to its neighbors
• Neighbors send back ACKs containing the 2-sample
  Allen variance of the local clock from the
  masters clock

63
Entire Protocol Operation During One Cycle

• Based on the received samples, master calculates
• i) an outlier ratio gyz for itself y and each
  neighbor z.
• ii) average of Allen variances
• iii) average of Allen deviations.
• All of these are sent to each neighbor z in a
  RESULT message.
• STEP j (j2,3,,n)
• Repeat the above between each level j diffused
  leader node and its neighbors

64
Entire Protocol Operation During One Cycle

• RESULT All sensors will get their outlier ratios
  and average Allen deviation which are used to
  evaluate the quality of their clocks with respect
  to their neighbors.
• If outlier ratio is gt 1, its local clock deviates
  from the clocks of its neighbors by more than
  twice the Allen variance
• Node does not become a diffused leader during the
  Time Diffusion Procedure of the current cycle or
  a master in the next cycle.

65
Entire Protocol Operation During One Cycle

• Nodes eligible for becoming masters in the next
  cycle, will be decided based on their energy
  availability (threshold)
• Same with diffused leaders in the current cycle
  depends on their energy level

66
Entire Protocol Operation During One Cycle

• STEP 3
• TDP performs the main function of diffusing the
  time from each master in a tree-like manner for n
  hops where n is some pre-determined parameter

67
Entire Protocol Operation During One Cycle

• Message M (tM,I,n,bM,k)
• tM,I is the diffused time of the master M to
  which the nodes synchronize in round I
• n is the number of levels (i.e., depth) to which
  the timing information is to be diffused.
• bM,k is the deviation of the corresponding tM,I
  at a node k hops from the master M (used by time
  adjustment algorithm (TAA) the weight for the
  diffused time tM,I at level k.)

68
Entire Protocol Operation During One Cycle

• Do broadcast for each round i
• 1. Master (level 1) and its neighbors
• i) Send a timing message M (tM,I,n,bM,k) to
  neighboring nodes
• ii) Elected diffused leaders at the next level
  respond with a time-stamped ACK message

69
Entire Protocol Operation During One Cycle

• iii) Master computes D average (Dj) where Dj is
  the round trip time between master and diffused
  leader j.
• Diffused time from master node
• tM,i tM,i D/2 d
• d is the amount of time that the nodes wait
  (relative to tM,i) before adjusting their clocks
  at the end of the round.

70
Entire Protocol Operation During One Cycle

• Standard deviation a of the rtt Dj gives an
  estimate of the quality of diffused time tM,i,
  and is accumulated in bM,k at each hop from the
  master.
• The accumulated deviation is
• bM,k bM,k-1 a
• where kltn is the number of hops from the master

71
Entire Protocol Operation During One Cycle

• b) On receiving a timing message M (tM,I,n,bM,k)
  for j2,3,..,n
• Repeat the above between the elected diffused
  leaders at level j and their neighbors.
• In each round, each node builds a TABLE with
  rows
• ltmaster_ID M, bM,k, tM,igt
• and populates it with the information received
  within that round.

72
Entire Protocol Operation During One Cycle

• After each round i within a cycle, each node
  resets its time to ti, the weighted sum of the
  times tM,i in its table,
• ti is based on the values of all the messages
  received in this round from different master
  initiators
• ti is recorded in the local TABLE
• The table is cleared at the end of each round i

73
Entire Protocol Operation During One Cycle

?
ti
( ßM,k , tM,i ) in Table
? ßM,k ßM,k
ßM,k in Table
tM,i

? ßM,k ßM,k
?
( ßM,k , tM,i ) in Table
ßM,k in Table
74
Entire Protocol Operation During One Cycle

• For any tM,I
• i) The numerator of the weight is the sum of all
  the deviations of all the diffusion messages,
  less the deviation for this particular message
• ii) The denominator of the weight is the sum of
  all such numerators for all the timing entries in
  the table.

75
Entire Protocol Operation During One Cycle

• Thus, the value of each clock is set to the
  weighted average of the clock values of the
  different master nodes of that round
• After averaging the data collected from multiple
  messages received within that round.
• Due to the weighted averaging all the nodes tend
  towards a common equilibrium time

76
ADVANTAGESTime-Diffusion Synchronization
Protocol

• The protocol achieves a system-wide equilibrium
  time across all nodes,
• Computed using an iterative weighted averaging
  technique
• Involves all the nodes in the synchronization
  process

77
ADVANTAGESTime-Diffusion Synchronization
Protocol

• Although there is a hierarchical structure, that
  is neutralized by having multiple master nodes
  distributed across the network
• Designed to be robust to node failures

78
ADVANTAGESTime-Diffusion Synchronization
Protocol

• Network-wide time converges for both static and
  mobile environment
• Small variation in network-wide time
• Small energy variation throughout the network
• Better performance than TPSN

79
DRAWBACKSTime-Diffusion Synchronization Protocol

• High complexity
• Each active period has multiple cycles, and each
  cycle has multiple rounds, in each of which a
  diffusion broadcast is initiated by multiple
  masters

80
DRAWBACKSTime-Diffusion Synchronization Protocol

• The convergence time tends to be high when no
  external precise time servers are used
• However, if the servers are used, the convergence
  time is comparable to a server-based technique