Architectures for Wireless Sensor Networks

swarmtellingMobile - Wireless

Nov 21, 2013 (3 years and 11 months ago)

82 views

Architectures for
Wireless Sensor Networks

Luca Necchi
April 16th, 2007
Politecnico Di Torino
Dipartimento di Elettronica
In Collaboration with:
ST Microelectronics
University of California
Berkeley
Advisor: Luciano Lavagno
2
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Wireless Sensor Networks (WSN)
Ad-hoc wireless networks
Sensing (monitoring)
Computation (decision)
Communication
Actuation (control)
Ad-Hoc Network topology:
No network infrastructure
Hierarchical or homogeneous
Distributed or centralized
Static or dynamic
3
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Car manufacturing
Monitoring
Agriculture
Environmental monitoring
Animal tracking
Infrastructure monitoring
Inventory handling
Security
Emergency response
Health – medical care
Control (Automation)
Building automation
Industrial automation
Ambient intelligence
Automotive
Application areas
Wine growing
4
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
WSN Architecture Performance Metrics
Crucial metrics in WSN
Lifetime
Power, energy efficiency, energy storage and/or scavenging
Cost
For design, deployment, management, maintenance
Size
Miniaturization, pervasive sensing/networking, smart dust
Robustness
Against: device failures, interference, voltage supply variability
Reactivity
For event-driven real time applications and power efficiency.
Other metrics:
Ease of use/deployment/management
Safety
Privacy
Distributed system behavior
5
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Ph.D objectives
Architectures for Wireless Sensor Networks (WSN)
HW and SW platform definition and co-optimization
Consider all layers in the HW and SW stack
Major improvements with respect to key metrics:
Energy efficiency
longevity, energy scavenging, while keeping flexibility
Reaction latency
fast wake-up, fast adjustment to new performance or power requirements
Ease of use
backward compatibility with current platforms
Cost
integrated hardware/software
Robustness
6
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Work carried-out (1/2)
HW platform domain:
Design of an asynchronous microcontroller (MCU) for WSN
Obtained by de-synchronization methodology
Optimized for Dynamic Voltage and Frequency Scaling
Compatible with actual WSN software/toolchain (same ISA)
Implementation using 130 nm technology (STM)
Remarkable improvements for key metrics
Extreme energy efficiency: 2,7 - 14 pJ/instr (vs ~1nJ/Instr of TI MSP430)
Better Lifetime and Cost
Fast wake-up
Better Reactivity
Wider voltage supply range
Better Robustness and Lifetime
Presented to International Symposium on Asynchronous Circuits
and Systems 2006 (ASYNC ’06)
7
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Work carried-out (2/2)
SW platform domain:
Design of EERINA: an Energy Efficient and Reliable In Network
Aggregation algorithm for WSN
Efficient network service to collect data on a “converge-cast” WSN
Tolerant to HW faults and radio interference
Modeled on a network simulator (Omnet++)
Remarkable improvements over key metrics
Increased Energy efficiency in any multi-hop converge-cast collection
Need for Lifetime, Cost
No need for a unique Cluster Head
Need for Robustness and Distributed system behavior
Tolerant to faults, allows on-line maintenance
Need for Robustness - Reliability
Presented at IEEE Wireless Communications and Networking
Conference 2007 (WCNC ‘07)
8
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Metrics for data aggregation
Energy saved
Sensor networks have a lot of built-in redundancy that saps
the limited energy available
Accuracy
Difference between data obtained by aggregation and actual
values
Completeness
Percentage of all readings included in final results
Network congestion reduction
Latency
Message overhead
9
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
Routing with Aggregation
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
10
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Data Aggregation
Consist in performing some distributed, in network activity to
reduce the overall amount of data flowing over multi-hop paths
Less traffic
Less energy consumption
Better network scalability
Depending on network topology, aggregation can be useful or
pointless
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
11
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Application:

Real-life Manufacturing Plants
20 – 50 mt
Homogeneous
sensing from
many sensors
in narrow
space
Single
sink
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
12
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Clustered topology:
The sensing areas are clearly identified and separated
Cluster
Sink
All nodes
are within
radio range
Forwarding
path
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
13
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Strategy behind E
2
RINA
Leverage Node
density
For reliability
Leverage inherent
broadcast
nature in communication channel
To minimize the number of transmitted messages
To create a high level of data redundancy.
Snooping suppression
/ Idle listening
Used to control the algorithm behavior at cluster level.
Randomized
approach
Algorithm never depend from a unique node or event but each choice
will just “tune” the behavior .
Every error or malfunction will just bring a sub-optimal behavior.
Clear
temporal division
between “data” and “control” tasks
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
14
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
How do E
2
RINA works?
Cluster behavior (distributed)
Cluster Leader
is elected
or re-affirmed
(Control)
Checks if any
exit
conditions
is met.
Fully
randomized
broadcasting
(Data)
Cluster-Leader
driven broadcasting

(Data)
Schedule
next
aggregation
Timed
schedule
looping
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
15
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Initialization
and
exchange
phases:
nodes’ behavior
Random wake-up to transmit
data
(Broadcast)
Everyone tries to send during
Initialization
Just who is needed tries to send during
Exchange
Wake-up to listen for
data
Random during
Initialization
Just the Cluster Leader in
Exchange
No Acknowledgement (not now!)
CSMA MAC, to avoid destructive collision
The volume of
traffic
being generated is fine-tuned
By knowing how many nodes are still trying to transmit
Traffic in the cluster will always be
Poisson distributed
@every time, each node stores a random number of
data
This natural asymmetry guides the
Cluster Leader
selection.
Data
-oriented
phases
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
16
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Contention
phase
node’s behavior
Cluster leader is elected by a “Contention-backoff”
Each node starts a back-off timer
Nodes with more data wait for less:
The node whose timer expires first is the “Cluster Leader”
The “Cluster Leader” sends a special packet (CP)
Just a single “Contention packet” (CP) loads ACK and orders.
The transmission of CP prevents the others from transmitting
In short time, by CSMA MAC collision avoidance
In mean time, by revoking timer after correct CP reception
The “Contention Phase” leverages snooping suppression
to achieve the required control functionality effectively
Control
-oriented
phases
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
17
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Why am I so confident it works?
To prove the effectiveness of the idea we develop both a
Simulation infrastructure (in Omnet++) in order to:
Test
the correctness and efficacy of the ideas
Stress
the algorithm in various working conditions
Verify
robustness with respect to:
Clock drift, synchronization errors
Node’s death and/or malfunction
Interference on radio channels
Mathematical model: in order to
Predict
the performance of the algorithm with respect to the free
variables of the system.
Optimize
Energy consumption or Performance metrics
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
18
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Energy minimization analysis
In a cluster with 20 nodes (Simulation results)
Optimal working point
(energy consumption)
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
19
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Energy consumption and convergence time
Comparison between different models
Simulated
Simulated
Mathematical
Mode
l
Mathematical
Model
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
20
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Simulation results:
Time needed to aggregate and its variance
Optimal working point
(performance)
Politecnico Di Tori
no


Uni
ver
sity
of
Californi
a B
erkeley
21
16/04/2007
Architecture for Wireless Sensor Networks Luca Necchi
Thanks
Questions?
Comments?
luca.necchi@polito.it