Towards Reliable Wireless Sensor and Actuator Networks

1
Towards Reliable Wireless Sensor and Actuator
Networks
Presentation at 8th Scandinavian Workshop on
Wireless Adhoc Networks May 7-8, 2008
Johannesberg Estate

• Prof. Riku Jäntti
• Department of Communications and Networking
  (Comnet)

2
Contents

• Intelligent monitoring systems
• System requirements
• Challenges
• Wireless sensor networking solutions
• State of art
• Open issues current research
• Tools for overall system design
• PiccSim
• Examples
• Building automation Identification of
  bottlenecks
• Robotics Routing
• Conclusions

3
Wireless automation today A journey towards
Reliable Wireless automation

• Wireless Networked control systems are real-time
  computing and control systems over wireless
  networks.
• Embedded systems where the different devices
  (sensors, controllers and actuators) communicate
  seamlessly using wireless technology
• Connection of field devices through a field bus
  requires a lot of network planning, wiring and
  troubleshooting as a result, for many automation
  systems the cost is in all in the wires
• Wireless vision autonomic communications and
  computing gets rid of the human-in-the-loop by
  making the systems self-configuring,
  self-healing, self-optimizing and self-protecting

4
Requirements

• We aim at developing an intelligent system
• The events in the environment need to be detected
  and acted upon in real-time (sensing,
  timeliness),
• The information must be seamlessly delivered from
  the environment to the decision making units and
  users (communications),
• The decisions need to be taken in real-time
  (information fusion and control),
• Actions are to be made in order to affect the
  environment (actuation),
• Reconfiguration of the used devices must be easy
  and fast (tasks, networking, algorithms),
• The system should be able to monitor its
  performance and adapt if necessary (fault
  detection, self-healing), and
• Security, safety and application management need
  to be inbuilt features of such systems, not
  add-ons.
• The system should be energy efficienti such that
  battery operation time is maximized
• The system should be based on open standards
  (IEEE802.15.4, 6LOWPAN,ROLL,ISA SP100,)

5
Physical layer challenges

• Statistical fades can cover multiple IEEE802.15.4
  channels
• In large networks, there might not be a single
  channel that provides full connectivity
• Air interface is vulnerable to interference and
  jamming.
• E.g. WiFi is efficient jammer for IEEE802.15.4
• Network capacity is limited by interference

Ward et al. Improved quality of service in
802.15.4 wireless mesh networks, In
Internationalworkshop on wireless and industrial
automation, San Francisco, 2005.
M. Petrova et al, Performance study of IEEE
802.15.4 using measurements and simulations,
IEEE WCNC 2006.
6
State of Art in WSANs

• ISA SP100 and WirelessHART

7
State of Art in WSANs

• ISA SP100 and WirelessHART
• IEEE802.15.4 radio on 2.4 GHz ISM band
• Frequency hopping over all 16 channels
• Conflict free FDMA/TDMA MAC
• End-to-end security
• End-to-end reliable transport
• Centralized control
• Network Manager NM
• Security Manager
• WirelessHART
• All data routed through Gateway/access point
• Supports HART protocol for communication with
  field devices
• ISA SP100
• Direct paths betwen nodes
• 6LoWPAN IPv6 over sensor networks (mesh under)
• Supports multiple fieldbus protocols

8
State of Art in WSANs

• Both ISA SP100 and WirelessHART use Time
  Synchronized Mesh Protocol (TSMP)
• Frequency hoppin over 16 channels
• Conflic-free MAC based on FDMA/TDMA
• Multi-path transmission
• (Graph routing with two link disjoint paths)
• Centralized control of network
• Dust Network has reported carrier class
  reliability 99.9995 over 26 days
• 44 nodes
• 33 packets/15 min
• 1kbit/s access point

9
State of Art in WSANs

• Limitations of the state of art solutions
• Centralized network and security managers
• Single point of failure (although redundancy
  possible)
• Does not support autonomous applications
• Network / frequency planning
• Does not scale well (network manager is very
  complex entity)
• Setting up network is slow. This can be problem
  if the network needs to re-establish itself after
  crash. (Self-healing)

• Frequency hopping (with blacklisting)
• Diversity technique
• Interference is not avoided automatically
• There can still be co-existence problems (e.g.
  deployment in a city area / office buildings
  where high density WLAN occupy most of the
  channels.)
• Network topology is inflexible
• Sensor data is often correlated and redundant.
  Unfortunately, the end-to-end security does not
  allow data aggregation and data fusion in the
  intermediate nodes.
• No support for multicasting

10
Nordite WISA Project

Quality of service
Increase robustness Degrease jitter
Requirements for current control algorithms
Data fusion PID Controller tuning New control
algorithms
Increase jitter margin and tolerance to errors
Wireless automation systems
Coexistence protocols Multi-path routing
(mesh) Synchronization
Performance of current wireless networks
11
Research problems

• Cognitive radio sensor network
• Adapt to the interference situation rather than
  average it
• Find white spaces in the spectrum
• Decentralized operation with self-organization
  and self-healing would enable autonomous systems
• Flexible topologies
• Store-process-and-forward to allow data
  aggregation and data fusion which reduces the
  need to transmit data.
• This would require new security solutions (packet
  level authentication?)
• Co-design of communications and data fusion and
  control
• Intelligent data fusion and control methods can
  tolerate certain packet loss and jitter and thus
  relax the requirements for communications
• Jitter and asynchronous sampling are challenges
  for control engineers, but tools for handling
  them are under development.

• Communication protocols, data fusion and control
  schemes can have complex interactions not always
  foreseen by the designers. This calls for tools
  for system design and testing.
• The system should support reconfiguration
• Software updates without shutting down the
  operation
• Adaptation to changes in the environment e.g. by
  changing the sampling intervals or data fusion
  techniques

12
WISA Phase I II
Cross-layer design
13
Project organization

• Strategic framework (Shankar Sastry, University
  of California Berkeley, 2006).

• WISA-II
• Workpackages

Algorithms Protocols Tools Theory
Schemes Testbed
14
Development tools for co-design

• There is a lack of design tools that are able to
  deal with integrated communication and control
  systems
• The performance of data fusion and control
  schemes depend on the network and the network
  performance depend on the traffic generated by
  these schemes gtCo-design is needed

Distributed systems Communication Systems
Control Systems Components composition
Layer composition Plant
controller model
PiccSIM
15
Platform for integrated control and
communications simulation PiccSIM

• Platform for (wireless) communications and
  applications co-simulation
• Integrates MATLAB computation environment
• algorithm development,
• control design and analysis tools,
• simulation of dynamical systems etc.
• with Ns-2 network simulator
• de-facto standard network simulator in research
• Implementation of e.g. data fusion and control
  algorithms and routing protocols on the same
  simulation tool -gt simulation of networked
  control systems (NCS)
• Automatic code generation capabilities provide
  ease of use
• Allows for testing various algorithms (e.g.
  speaker identification, coordination of robots)
  with a realistic communications model

16
PiccSIM Architecture
17
PiccSIM ExampleBuilding Automation
Identification of bottlenecks

• Testing of wireless temperature and ventilation
  control
• 40 ZigBee nodes
• Shadowing from blueprint

18
PiccSIM ExampleBuilding Automation
Identification of bottlenecks

• Physical Models
• Heat balance in rooms (PID control)
• CO2 concentration in rooms (relay control)
• Event driven signals, lighting (on/off)
• Communication models
• IEEE802.15.4 radio
• Indoor propagation modeling

19
PiccSIM ExampleRobotics - Multipath routing

• Sensor Motes equipped with Ultra sound receivers
  and a radio module forms a Grid network
• A mobile Node (Trolley/Robot) emits Periodic
  Ultrasound pulse
• Sensor Motes estimate the distance to the Mobile
  using
• Distance information is forwarded to the
  Controller, where Position estimation is done
• Controller estimates the position using 3-D
  Position Sensing scheme, where the Differences in
  the Time-of-Flights from a Wave Source to Various
  Receivers Ajay.
• Finally controller sends Control (Action) Message
  to the Mobile nodes.

Sensors-gtController Controller-gtMobile Node
20
PiccSIM ExampleRobotics - Multipath routing
AODV
AOMDV

• LMNR (Localized Multiple next hop routing)
• Set up multiple routes
• Next hop is locally decided based on load,
  interference, and link availability
• gt Increase robustness against link faults
  (decrease the need for rerouting in case of
  failures)

LMNR
S. Nethi, C. Gao and  R Jäntti, Localized
Multiple Next-hop Routing Protocol, in Proc. 7th
international conference on ITS telecommunication
(ITST 2007), Paris, France, June 5-8, 2007
21
PiccSIM ExampleRobotics - Multipath routing

• LMNR (Localized Multiple next hop routing)
• Intermediate nodes are given the liberty to
  choose from multiple Next-hop to the destination,
• Basically, intermediate nodes choose from
  possible next hop node depending upon cost
  function of the respective node.
• Finally, Update cost, using Periodic hello
  messages.
• In order to achieve load balancing, the cost
  function should somehow reflect the traffic
  volumes carried by the nodes on a route

Cn Cost associated with node n Rn Number of
entries in the routing table of node n Dtr
Life-time of routing table entry r ter Expire
time of route entry r t Current time
22
PiccSIM ExampleRobotics - Multipath routing

• NS2 Simulation results
• Simulation time 250 s
• Number of nodes 50
• Number of traffic flows 10
• Packet size 512 bytes

Performance improves, but what would this mean
from the application point of view?
23
PiccSIM ExampleRobotics - Multipath routing

• Control of mobile node over wireless network
• Singlepath vs. multipath

24
PiccSIM ExampleRobotics - Multipath routing
Packet delivery fraction and Avg. end-to-end delay
Outage time and Error estimate
25
PiccSIM ExampleRobotics Communications
Control

• Robot squad
• Leader controls positions of other nodes
• Multihop scenario
• Routing is very important
• Communication constrained control
• Dropped packets degrade control performance
• Halt when several packets dropped
• Rerouting when link breaks

26
PiccSIM ExampleRobotics Communications
Control

• Control options
• PID controller
• Tuned stable for specified jitter
• Slightly conservative
• Kalman filter PID
• KF estimates current process state
• No jitter margin needed
• Tight tuning
• Networking options
• AODV Single path routing
• LMNR Multi path routing
• Priorization of traffic based on number of hops

27
PiccSIM ExampleRobotics Communications
Control

• Control cost
• Average distance from reference trajectory, rk
• Only when communication
• No cost if break (large dominating cost)
• Costs

28
PiccSIM ExampleRobotics Communications
Control

• Packet delay for singlepath routing as a function
  of hops.

2. Average delay as function of number of total
hops, with different priorizations.
Multipath Vs Single path
29
Concluding remarks

• Current WSAN solutions can achieve very high
  reliability, but they are
• Inflexible
• Complex to implement and thus do not scale well
• The use of WSAN in autonomous systems would
  require
• decentralized communication and data fusion
  solutions
• re-configurability
• This calls for cognitive networking
• Co-design of communications, data fusion and
  control can help relax the reliability
  requirements and thus achieve
• better scalability
• energy efficiency
• Store-process-and-forward data transmission could
  be utilized to
• reduce the network load
• improve estimator/controller performance
• but they would require new transport and data
  security solutions

30
Any Questions ?