Wireless Control Systems - from theory to a tool chain, Mikael Björkbom, Aalto University
1. Mikael Björkbom
Wireless Sensor and Actuator Networks for Measurement and Control
Phase II
Wireless Control Systems
- from theory to a tool chain
Aalto University
Department of Communications and Networking
Control Engineering Group
KTH
Radio Communication Systems Group
Automatic Control Group
Royal Institute of Technology KTH
3. Nordite WISA Project
Quality of service
Requirements for
current control algorithms
Data fusion
Increase jitter margin
PID Controller tuning
and tolerance to errors
New control algorithms
Wireless automation systems
Increase robustness Coexistence protocols
Performance of Decrease jitter Multi-path routing (mesh)
current wireless networks Synchronization
Royal Institute of Technology KTH
4. Workpackages
• WP1: Reliable and secure communication protocols for
wireless automation
• WP2: Communication constrained reliable control
• WP3: Implementation of WiSA toolchain
• WP4: Project management
Aalto KTH
Royal Institute of Technology KTH
5. WISA Phase I & II
WISA Phase I WISA Phase II
Control, data fusion and networking algorithms,
testbeds and simulation tools Control and Wireless Design
data fusion networking tools
Cross-layer design
Tool chain
Royal Institute of Technology KTH
6. Results: Toolchains
• PiccSIM – Simulation of wireless control systems
• WirelessTools – Planning of wireless network schedule
• PROSE – Node and simulated network
Royal Institute of Technology KTH
7. WP 1: Reliable and secure communication
• T1.1. Interference avoidance and dynamic spectrum
management
– Time and frequency domain methods
– Adaptive frequency hopping
• T1.2. Reliable networking
– SIRP, Antenna switching
– Tools for scheduling
• T1.3. Sensor and network monitoring, fault detection,
and fault recovery
– Fault detection part is partly missing
– Fault recovery: Code dissemination tool
Royal Institute of Technology KTH
8. WP2: Communication constrained reliable control
• T2.1. Communication-aware data fusion and control
– New data fusion schemes
– Network jitter aware PID tuning rules
• T2.2. Control structures, architectures and scalability
– Impact of MAC on control and data fusion were analyzed
– Tuning of PID controllers for distributed MIMO systems
• T2.3. Adaptive and robust control
– Delay adaptive Internal Model Control based tuning
– Network performance adaptive controller
Royal Institute of Technology KTH
9. WP3: Implementation of WiSA Tool Chain
• T3.1. Automated implementation of routing protocols
– This was not accomplished! There is no automation in the
development of routing protocols
– PROSE tool for hardware in the loop simulation
• T3.2. Automated control algorithm implementation
– Part of PiccSIM
• T3.3. Design tools and interfaces for the WiSA tool chain
– Part of PiccSIM
• T3.4. Demonstrator development
– Several demo sessions were arrange (including NORDITE
workshop)
Royal Institute of Technology KTH
10. WP 1/T1.1-T1.2: Results
• Objective: wireless sensor nodes should be able to
communicate in a reliable fashion despite bad channel
conditions (interference, fading).
• We aim at improving reliability by means of:
– Interference Avoidance through Dynamic Spectrum Access
– Frequency Hopping
– Channel Coding
Dynamic Spectrum Access
Frequency Hopping Channel Coding
Antenna Switching RELIABILITY Hybrid ARQ
Receiver diversity
Royal Institute of Technology KTH
11. WP 1/T1.1: Dynamic Spectrum Access
• An Example: Experimental Comparison of DSA schemes:
Spectrum Holes in the Time domain Performance of DSA in the time domain depends heavily on
channel conditions:
Energy increased
of up to 5 times for
high interference!
DSA in the frequency domain (channel selection) requires larger
energy for spectrum sensing but allows to avoid interference:
By selecting the
communication channel
effects of interference
Spectrum Holes in the Frequency domain can be mitigated
Royal Institute of Technology KTH
12. WP 1/T1.2: Spatial diversity
TABLE I
CHANNEL PARAMETERS
Tap CHANNEL 1 CHANNEL 2
Relative Relative tap Relative tap Relative tap
tap delay amplitude delay [ns] amplitude
[ns] [ns] [ns]
1 0 0 0 -0.1
2 20 -0.9 20 -0.6
3 30 -2.6 50 -2.9
4 40 -3.5 100 -5.8
5 100 -6.7 150 -8.7
6 300 -17.9 200 -11.6
Time Diversity Approaches for 0.1km/h and 1km/h
1
0.95
0.9
Packet Delivery Ratio
0.85
Pure Time Diversity (0.1km/h)
Elektrobit’s: Channel Emulator PropSIM-c2 0.8
Piggybacking (0.1km/h)
Switch if No Acknowledgement (0.1km/h)
0.75
Piggybacking (1km/h)
Pure Time Diversity (1km/h)
0.7
Switch if No Acknowlodgement (1km/h)
Multiple receiving antennas: 0.65
26% increase in packet delivery ratio 0.6
-90 -85 -80 -75 -70 -65
Mean RSSI (dBm)
12
Royal Institute of Technology KTH
14. Performance of Multi-Channel MAC Protocols
• Performance of G-McMAC analyzed and compared to other existing
protocols
• G-McMAC outperforms other protocols with respect to delay
regardless of the used parameters
• G-McMAC achieves the highest throughput in many cases.
Royal Institute of Technology KTH
15. Time-Synchronization in Multi-Channel WSN
• Multi-Channel Time-Synchronization (MCTS) protocol
• Time-synchronization
– Critical for many WSN applications, e.g. control
– Enables efficient communications and deterministic operation
– Multiple channels can be used simultaneously in order to
minimize convergence time
Royal Institute of Technology KTH
16. WP 2/T2.1 communication aware data fusion and control
Control over WirelessHART networks
P
Data stream characteristics:
WirelessHART network • Slotted time
• Minimum transmission delay
• Time-varying latency, loss
C
Many analysis tools and control design techniques, but no perfect match
– theory most complete for linear-quadratic control
Here: explore sampling interval as ”interface parameter” in co-design.
Royal Institute of Technology KTH
17. Realiable real-time challenge
Meeting hard deadlines on unreliable multi-hop network
Maximize deadline-constrained reliability (the “timely
throughput”)
i
i
Royal Institute of Technology KTH
18. WISA-II solutions
Focusing on WirelessHART-compliant solutions
New theory, algorithms and software for network
scheduling
– minimize multi-source data collection delay
– maximize deadline-constrained reliability for unicast
joint routing and transmission scheduling
Limits of performance, rules of thumb, and optimal
algorithms
Royal Institute of Technology KTH
19. WP 1/T1.2 : Convergecast
Given: sensors with single packet to send at time zero
Find: schedule that delivers all packets to sink (in an optimal fashion)
Key operation WirelessHART’s
sensing-compution-actuation cycle:
Royal Institute of Technology KTH
20. WP 1/T1.2 : Convergecast
Optimal convergecast on trees
Proposition. The minimum evacuation time for a wireless HART
network with tree topology is max(N, 2Nmax-1) timeslots, where
Nmax is the number of nodes in the largest subtree.
Also here, we can characterize the channel-latency tradeoff.
Efficient (O(N2)) time-optimal policies, channel-limited case slightly
harder.
Royal Institute of Technology KTH
21. WP 1/T1.2 : Convergecast
If links are unreliable, then the complete operation might fail.
Observation. If links fail with probability pl, convergecast fails with
probability (1-pl)S where S=# transmissions in the schedule.
For line with N nodes, S=N(N+1)/2Schedule quickly becomes unreliable!
Several simple ways of improving reliability of a given schedule
– duplicating each slot, repeating schedule, …
Need methods for quantifying the resulting latency distributions.
Royal Institute of Technology KTH
22. Optimal co-design
Understanding what controllers need, and what network can provide
Key result: optimal co-design is modular, can be computed efficiently
deadline-constrained maximum reliability and control under loss
optimal parameters found by sweeping over sampling interval
Royal Institute of Technology KTH
23. WirelessHART tools
Key features:
• Powerful network editor
• Interactive scheduler
• Integrated schedule optimizer
• Reliability analysis
• Matlab/Simulink integration
• Multiple superframe support
• Sensors, actuators, relays
Royal Institute of Technology KTH
24. WP 2/T2.1 Communication aware data fusion and control
• Tune the controller s.t. stable even with varying delay
• One proposed method: Extended plant PID tuning
Step experiment Filtering Extended plant response
1
G f (s)
G(s)
1 sT
n
f
Filter design
0 (t) max Tf f max , n AMIGO design
on extended plant,
tuning rules
• Measurement filter design based on the network delays
1
3 max ,
n 1
Tf (1 n )/ 2
1 n
max , n 1.
3 n n 1
Royal Institute of Technology KTH
25. WP 2/T2.3. Adaptive and robust control
• Network congestion causes packet drops
• Adjust control speed and required communication rate
• Maintain good network QoS
– Keep packet drop at 3 %
0.08 40
0.07
20
0.06
0.05 0
0 200 400 600 800 1000 1200
QoS
0.04 Time [s]
0.03 6
0.02 4
h [s]
0.01 2
0 0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Time [s] Time [s]
Royal Institute of Technology KTH
26. WP 3: WISA Toolchains, PiccSIM
• PiccSIM
– Control simulation in Simulink
– Network simulation in ns-2
– Graphical user interfaces for
network design
– Data-based modeling tools,
controller design and tuning
GUI
– Automatic code generation, and
code reusability
• All in one tool
• Released as open-source to
researchers
• wsn.tkk.fi/en/software/piccsim
Control design
Royal Institute of Technology KTH
28. WP3: Using Field data in Simulations
Light Machinery Heavy Machinery Simulation: Crane Control
60 100 90
90 80
50
80
70
Packet Delvery Ratio (%)
70
40 60
FreeSpace
Packet drop [%]
Packet drop [%]
60 Light Machinery
50
30 50 Medium Machinery
40
Heavy Machinery
40
20 30
30
20 20
10
10 10
0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 0
1 2 3
Link [#] Link [#] Mobility Models
10000 1 10000 1 400 1 2000 0.5
5000 0.5 5000 0.5 200 0.5 1000
0 0 0 0 0 0 0 0
Good Bad Good Bad Good Bad Good Bad
400 1 200 1 1000 1 200 1
SINK
200 0.5 100 0.5 500 0.5 100 0.5
NODES
0 0 0 0 0 0 0 0
Good Bad Good Bad Good Bad Good Bad 20M
2000 1 200 1
4000 1 200 1 20M
1000 0.5 100 0.5
2000 0.5 100 0.5
0 0 0 0
0 0 0 0 Good Bad Good Bad
Good Bad Good Bad
4000 1 200 1
100 1 4000 1
2000 0.5 100 0.5 10M
50 0.5 2000 0.5
0 0 0 0
0 0 0 0 Good Bad Good Bad
Good Bad Good Bad 40M
• A Gilbert-Elliot packet drop model is implemented on PiccSIM
– Each link model consists of the state residence times and the packet
drop probabilities for each state
Residence time
Packet drop probability Royal Institute of Technology KTH
29. ID = 1
Data N
Data N 0 T 1
WP3: Automatic Code Generation Node _ KF
Process
• Simulation -> Implementation and testing on real Process
hardware AD 0 DA 6
• Generic node block in PiccSIM library DA 7
Radio timestamp 1
– Make implementation in block
Radio recv 1 Radio send 1
• Simulink blocks, Matlab code... Process_interface
do { ... } while
• Radio blocks for communication between nodes
Synchronize with Ns -2
• Matlab Real-Time Workshop
Timestamps
Node
Send to N 1 T 1
– Target Language Compiler (TLC) u
Data N 2 T 3
ID = 0
– Generate code from Simulink block Interface node
• Wrapper main file for Sensinode node hardware
– Other wrappers can easily be implemented
Royal Institute of Technology KTH
30. WP3: Automatic Code Generation
1 u -2 .5 /2 90 /2 .5
AD 0 Bias 2
3 u +0 .4 2 .5 / 0 . 8 1 Gain 1
Radio recv 1 L D: 3 3
Saturation Bias1 DA 6
Gain 2 Signal Specification Radio send 1
Constant
Rate Transition
4
2 U ~ = U /z double Delays 2 .5 2
Radio timestamp 1 DA 7
Detect Data Type Conversion Tapped Delay Add Gain 3
Change
Royal Institute of Technology KTH
31. PROSE – Hardware in network simulation
• Test hardware with simulated
network
• Wireless protocol
– Testing, debugging
– Logging all activities
– Controllable channel conditions
Royal Institute of Technology KTH
33. Collaboration
between research groups
• Researcher visits: Lasse Eriksson @ KTH, 5/2006 and 6/2007
• Researcher visits: Mikael Björkbom @ KTH, 5/2009
• One day visits from KTH to Aalto
• Joint publications
between research groups and industry
• Active participation of the industry in the steering board meetings
(e.g. simulation testbed demo attracted Åkerströms (Sweden) to
travel to Helsinki)
• Tomorrow PiccSIM demo at ABB, Sweden
• Joint workshop on ”Standards and research challenges for industrial
wireless control” with industrial partners in Stockholm, Sweden 4th
of March 2008
Royal Institute of Technology KTH
34. Information dissemination
• Results summary (2008-2010)
– 5 (+1) Ph.D theses
– 3 Masters theses
– 5 Bachelor theses
– 8 Journals papers
– 38 Conference papers
• Seminar presentations and invited talks:
– DoD/TEKES workshop in Washington 11 - 12 March 2008
– Rutgers/HIIT Workshop on Spontaneous Networks in Rutgers 5-9 May, 2008
– Third International Summer School on Applications of WSN and Wireless
Sensing in the Future Internet (SenZations) in Slovenia 1 - 5 September 2008
– 8th Scandinavian Workshop on Wireless Adhoc Networks (Adhoc' 08) May 7-8,
2008 Johannesberg Estate
– Sensinode research seminar, Vuokatti, Finland, 16th of September 2008
– Lecture on Reliable WSNs at Prairie View Texas A&M, 15th of October 2009
– Presentation at Scandinavian Electronics Event, 14.4.2010
Royal Institute of Technology KTH
35. Wireless Sensor Systems group at Aalto
• Started in 2008 to collaborate in the field of WSS
– Made possible by WiSA project
– Aalto University Workshop on Wireless Sensor Systems 2010
• Currently 4 projects, multiple departments, about 15
researchers
• Research fields
– Network Management
– Wireless Automation (Gensen, RELA, RIWA)
– Indoor Situation Awarenes (WISM II)
– Structural Health Monitoring (ISMO)
Royal Institute of Technology KTH
36. Final thoughts
• Nordic cooperation
– Closeby, initial visits
– Still videconference more convenient
• NORDITE program
– Nordic cooperation good
– Basic research oriented
• Industry involvement
– Only interest group
– Less feedback than in industrially financed projects
Royal Institute of Technology KTH
37. Contact information:
Mikael Björkbom
Aalto University
School of Electrical Engineering
Dept. of Automation and Systems
Technology
P.O.Box 15500
FI-00076 AALTO
Finland
Tel. +358 9 470 25213
Email: mikael.bjorkbom@tkk.fi
http://wsn.tkk.fi
Royal Institute of Technology KTH