4.

Time Synchronization with GPS

Wireless Sensor Networks

1. Introduction to clock synchronization
1. GPS example
2. Wired clock synchronization overview
3. WSN characteristics and design space
4. WSN clock synchronization overview
5. WSN protocol evaluation and trade-offs
1. Quantitative and qualitative comparison

Wireless Sensor Networks

 Application data are often fused and often
time stamped to give a global view of the
physical system
Environmental monitoring, target tracking, security
systems

Time-based systems or protocols need

accurate timing
Time Division Multiplexed Access (TDMA)
RT-Link

Wireless Sensor Networks

 A computer clock is an electronic device that
counts oscillations at a certain frequency
An accurately-machined quartz crystal
Use a counter register and a holding register
Each cycle decrements the counter until it is 0.
The (zeroed) counter is replaced with the holding register
value when a timing interrupt is generated

Clock skew
Quartz imperfections and environmental influences
such as temperature, pressure, and power voltages
lead to clock drift
Clocks run at different frequencies and will diverge
from each other in a multi-node system
A frequency deviation of 0.001% will cause a clock to
drift of ~1sec/day
Wireless Sensor Networks
 Physical Clocks
Adhere to real-world time
Universal Coordinated Time (UTC)

Logical Clocks
Relative synchrony with respect to other nodes in
the distributed system

Wireless Sensor Networks

 Coordinated Universal Time (UTC)
International time standard
High precision atomic time, ~300 atomic clocks world-wide

Radio Clock
Receives time-coded RF signal and transfers to a digital value

Global Positioning System (GPS)

24 or more solar-powered satellites
As of 2005, a fleet comprised of Boeing Block II and IIA
satellites, and Lockheed Martian Block IIR satellites
Each satellite contains one primary atomic clock, and backups
Block II and IIA satellite two cesium and two rubidium atomic clocks
Block IIR has three rubidium atomic clocks

Reference: http://gps.losangeles.af.mil/jpo/gpsoverview.htm

Wireless Sensor Networks

 QuickTime and a
QuickTime and a TIFF (Uncompressed) decompressor
TIFF (Uncompressed) decompressor are needed to see this picture.
are needed to see this picture.

QuickTime and a QuickTime and a

TIFF (Uncompressed) decompressor
TIFF (Uncompressed) decompressor are needed to see this picture.
are needed to see this picture.

Reference: http://gps.losangeles.af.mil/jpo/gpsoverview.htm
Wireless Sensor Networks
 Satellites are tightly time-synchronized
Primary (master) clock onboard is within 1s of the

U.S. Naval Observatory's Master Clock

Atomic precision is ~3ns, drift 1s per 300,000 years
Satellites are always within 250ns of each other

Receivers
Broadcast time of flight is 65-85ms @ 10,900
nautical miles
Receivers use quartz crystals
Synchronizes time during initialization, and later
perform triangulation

Wireless Sensor Networks

 Protocol should be resilient to unbounded
message latencies

Nodes should be able to estimate local clocks

on remote nodes
Applies to bi-directional clock synchronization

Time must never run backwards

Do not want to repeat events in time
Do not want to violate causality

Synchronization overhead should not degrade

system performance
Wireless Sensor Networks
 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Medium access
Network latency
Packet loss
Context switches
Protocol overhead, if any

Wireless Sensor Networks

 Highly dynamic, mobile, and sparsely distributed

Communication links are short range and may

have short lifecycles
Store and forward techniques are needed for transient
nodes

Traditional assumptions do not hold

Message delay can be estimated in wired, but ad hoc
networks may have arbitrarily long communication
delays
Periodic message exchange common in wired networks,
but energy constraints in ad hoc networks make this
very expensive. Use on demand techniques?

Wireless Sensor Networks

 Remote Clock Reading
Time-Transmission Protocol (TTP)
Offset Delay Estimation

Network Time Protocol (NTP)

Set-value Estimation

Wireless Sensor Networks

 Query remote process for time

Several queries are performed

Round-trip time is an average or the lowest bound
Latency is defined as the round-trip time

Round-trip time is not deterministic

Inconsistencies and variance in routing, network load,
system load and scheduling can vary results

Reference: F. Cristian, Probabilistic Clock

Synchronization, Distributed Computing, 3:146-158,
Springer-Verlag, 1989

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

1. Client sends request at T0

2. Server replies with timestamp S-time (local
time)
3. Client receives message at time T1
4. Client sets time to [S-time + (T1-T0)/2]
This is usually achieved by averaging several rounds
or using the shortest round-trip time (T 1-T0)

Wireless Sensor Networks

 Network Time Protocol (NTP)
Predominant time-sync service running on switched IP
networks
Huge deployment of 10s of millions of clients and servers
Nominal accuracy of 10ms on WANS, ~200s on LANS
Hierarchical design: root node synchronizes with UTC
Second-tier nodes act as backup to the root
On request, eight iterations a node will send back-to-
back bidirectional messages, calculate and store offset
and delay pairs, pick the lowest delay and set time to the
related offset
Delays are partly compensated
Through the use of a second timing packet
High synchronization overheads
Messaging complexity and storage

References
1. D.L. Mills, Internet Time Synchronization: the Network Time
Protocol, IEEE Trans. Communications, Vol 39, no 10, pp.
1484-1493, Oct. 1991.
2. http://www.eecis.udel.edu/~mills/ntp.html
Wireless Sensor Networks
 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 WSN differs from wired fixed-infrastructure
Limited communication range
Less than 100 meters in most cases
With high message loss
Scalability
1000s of nodes: but may not have hierarchy
Energy conservation
Largest overhead is found in communication
Self configuration and dynamic topologies
Mobile nodes
Not addressed in wired paradigms
May have intermittent connectivity

Wireless Sensor Networks

 Energy, bandwidth and hardware are
constrained
Communication is very costly
Transmission of 1 bit over 100m costs 3 joules, or
equivalently the execution of 3 million instructions.
Communication is limited
Range and high percentage of message loss
Mobility, albeit an implicit benefit of wireless
nodes, comes at the cost of complexity
Intermittent connectivity
Shifting topology changes requires dynamic
reconfiguration

Wireless Sensor Networks

 Reference Broadcast Synchronization (RBS)
Romers protocol
Mocks protocol
Network-wide Time Synchronization
Delay Measurement Time Synchronization
Protocol
Probabilistic Clock Synchronization
Sichitiu and Veerittiphans protocol
Time-Diffusion Protocol (TDP)
Asynchronous Diffusion

Wireless Sensor Networks

 Field is lively, lots of recent research
Tulones Clock Reading protocol, 2004
Meier et als protocol, 2004
Lightweight Tree-based Synchronization, 2003
TSync protocol, 2004
Hu and Servettos protocol, 2003

No generic one size fits all for any specific

situation

Wireless Sensor Networks

 Synchronization Issues
Master-slave or peer-to-peer
Clock correction or un-tethered clocks
Internal or external synchronization
Probabilistic or deterministic algorithms
Sender-to-receiver or receiver-to-receiver

Application-Dependent Features
Single-hop or multi-hop
Mobile or stationary network
MAC-layer or standard layer

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 Reference Broadcast Synchronization
Untethered local clocks
Some non-determinism in synchronization is removed
Exploits broadcast medium of WSN by using receiver-to-receiver
synchronization
Multiple broadcasts to eliminate residual errors, implies fine-grained timing
Determine clock offset by sharing adjacent time-received
timestamps
Estimate clock skew of adjacent neighbors
Use least squares to find best line fit when event of interest occurs

Achieves better accuracy than NTP

1.85 (+-) 1.28 sec 1-hop, 3.68 (+-) 2.57 sec 4-hop

Messaging complexity is very high

Every receiver exchanges messages with every other receiver 0(n)

Convergence time can be high due to large number of messages

Reference: J. Elson, L. Girod, and D. Estrin, Fine-Grained Network Time Synchronization using
Reference Broadcasts, Proc. Fifth Symposium on Operating System Design and Implementation
(ODSI 2002), Vol 36, pp. 147-163, 2002.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 Romers protocol was designed specifically for
ad-hoc networks
Master-to-receiver protocol, using untethered clocks
Uses symmetric bi-directional link to exchange one
additional timing message after a unique event
occurs
Transfer the time of an event on a sending node to the
time on a receiving node
Establishes an upper-bound on the absolute value of
clock error received by a node
e.g. determine lifespan of a timestamp
Very low message overhead, but not very accurate
Does not fare well in multi-hop systems
Synchronization errors will accumulate over time

Wireless Sensor Networks

 Determines upper and lower bounds for real-
time elapsed from timestamp generation at
source to arrival time at destination node

Transform bounds to local time at the

destination node

Subtract resulting values from time of arrival

in the destination node

Reference: K. Romer, Time Synchronization in Ad Hoc Networks,

ACM Symposium on Mobile Ad Hoc Networking and Computing
(MobiHoc 01), pp. 173-182, Oct. 2001.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Synchronization across multiple nodes

requires round-trip time and idle time
(storage time) to be accumulated
Any node n involved in the communication
must account for storage and round-trip
times across hops
Wireless Sensor Networks
 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 QuickTime and a
TIFF (Uncompressed) decompressor
are needed to see this picture.

Wireless Sensor Networks

 Wired clock synchronization overview

WSN design space

Wireless clock synchronization overview

Tough design space with lots of constraints

Wireless protocol comparison

Active area of research, but no single solution
serves all needs

Wireless Sensor Networks