Clock Synchronization
in Sensor Networks
Mostafa Nouri
Outline
Need for time synchronization in sensor
networks
 Definition of time synchronization
 Common problems in time synchronization
 Typical schemes and algorithms.

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
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
ADC
Battery
Low-power
CPU/mC
Memory
Radio
Sensors
Base
Station
Why Need for Time
Synchronization

Link to the physical world
 When

does an event take place?
Key basic service of sensor networks
 Fundamental

Crucial to the efficient working of other basic
services
 Localization,

to data fusion
Calibration, In-network processing, …
Several protocols require time synchronization
 Cryptography,
Topology management.
Sensor Networks

Characteristics of SN
 Cheap

No Accurate Oscillator
 Limited


Energy
Together
Data Fusion
Conclusion
 must
?
?
Need to Sleep
 Work

?
and Small
initiate T-Sync in WSN
?
?
Metrics for Synchronization
Protocols
Precision
 Longevity of synchronization
 Time and power budget available for
synchronization
 Geographical span
 Size and network topology

Examples

Energy efficient radio scheduling
Sender
Receiver
Time

Array processing
Radio On
Radio Off
Radio Off
Guard band due to clock skew; receiver can’t
predict exactly when packet will arrive
Computer Clocks

Clocks in computers
C(t)=k∫0tω(τ)d τ + C(t0)
 ω is frequency of oscillator, C(t0).
 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



C(t)
a is a clock drift (rate)
B is an offset of the clock
Perfect clock


Rate = 1
Offset = 0
b
C(t)
a
t
Definition of time synchronization

Let C(t) be a perfect clock
 A clock

If Ci(t)=C(t)
 A clock


Ct(t) is called accurate at time t
If dCi(t)/dt = dC(t)/dt = 1
 Two

Ci(t) is called correct at time t
clocks Ci(t) and Ck(t) are synchronized at time t
if Ci(t)=Ck(t)
Time synchronization
 Requires
knowing both offset and drift
NTP Overview

Most widely used time synchronization
protocol
 Hierarchical:

Perfectly acceptable for most cases.
 Coarse

C/S model
grain synchronization
Inefficient when fine grain synchronization
is required
Why not Use NTP

Link



Topology


Ratio of packet loss is very low in Internet (fiber, cable)
Links are short range and short lived in sensor networks
(wireless)
Static, robust and configurable
Energy aware



Frequent message exchange
Listening to the network is free
Using the CPU in moderation free
Difficulties in Sensor Networks

No periodic message exchange is guaranteed
 There

Transmission delay between two nodes is hard
to estimate
 The

may be no links between two nodes at all
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
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
SR vs. RR
A
A
B
Beacon
B
A

B
A
Sources of error-variation of packet delays and
clock drifts
B
Basic Mechanism

Pair-wise synchronization
A


T2=T1+delay+offset
distance
B
T1
T2
T3
T4=T3+delay-offset
T4


offset=((T2-T1)-(T4-T3))/2
delay=((T2-T1)+(T4-T3))/2
hi
hj
Sources of Errors

Send time

Kernel processing
 Context switches
 Interrupt processing

Access time

Specific to MAC protocol



Transmission time
Propagation time



e.g. in Ethernet, sender must wait for clear channel
Very small in WSNs, can be ommited
Reception time
Receive time
Illustration
Sender
Receiver
Receive time
Send time
NIC
Acess time
Transmission time
Propagation time
NIC
Reception time
Physical Media
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)
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
Level Discovery Phase
2
1
K
B
0
1
A
H
1
2
C
F
1
2
E
3
I
L
2
2
G
D
3
J
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
Time Synchronization Algorithm
K
B
0
1
A
H
1
2
C
F
1
2
E
3
I
L
2
2
G
D
3
J
LTS

Features
 Minimize
synchronization complexity rather than
maximizing accuracy

Authors claim that wireless sensor networks need quite low
synchronization accuracy
 Sender-receiver bidirectional
 Two LTS algorithms
 Centralized


mechanism
Node sends a synchronization request to a closest reference
node by any routing mechanism
Distributed

Requires a spanning tree to be constructed firstly
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
TS/MS

Features


Determining relative offset and drift between two nodes
Sender-receiver bidirectional scheme




Two clocks c1(t) and c2(t) are linearly related as


Node 1 sends a probe message to node 2 time-stamped with t0
Node 2 generates timestamp tb and responds immediately
Node 1 generates timestamp tr
C1(t)=a12C2(t)+b12
The following inequalities are held:

t0<a12tb+b12<tr
TS/MS
C1(t) = a12H2(t) + b12
C1(t)
Sample 3
tr
t0
Sample 1
b12
Node 1
t0
a12
Node 2
tb
tr
tb
C2 (t)
Parameter Estimation

Relative drift a12 and offset b12

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.


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.
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”
Thank you