wireless control systems - from theory to a tool chain, mikael björkbom, aalto university
TRANSCRIPT
Royal Institute of Technology KTH
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
Mikael Björkbom
Wireless Sensor and Actuator Networks for Measurement and Control
Phase II
Royal Institute of Technology KTH
Wireless Automation: Control
Communication affects control performance
-> Control should be robust to problems in the network
Royal Institute of Technology KTH
Nordite WISA Project
Quality of service
Increase robustness
Decrease jitter
Requirements for
current control algorithms
Performance of
current wireless networks
Increase jitter margin
and tolerance to errors
Data fusion
PID Controller tuning
New control algorithms
Coexistence protocols
Multi-path routing (mesh)
Synchronization
Wireless automation systems
Royal Institute of Technology KTH
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
WISA Phase I & II
WISA Phase I WISA Phase II
Tool chain
Control, data fusion and networking algorithms,
testbeds and simulation tools Control and
data fusion
Wireless
networking
Design
tools
WISA Phase I WISA Phase II
Tool chain
Control, data fusion and networking algorithms,
testbeds and simulation tools Control and
data fusion
Wireless
networking
Design
tools
Cro
ss-layer
desig
n
Royal Institute of Technology KTH
Results: Toolchains
• PiccSIM – Simulation of wireless control systems
• WirelessTools – Planning of wireless network schedule
• PROSE – Node and simulated network
Royal Institute of Technology KTH
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
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
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
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
RELIABILITY
Dynamic Spectrum Access
Frequency Hopping
Antenna Switching
Receiver diversity
Channel Coding
Hybrid ARQ
Royal Institute of Technology KTH
WP 1/T1.1: Dynamic Spectrum Access
• An Example: Experimental Comparison of DSA schemes:
Spectrum Holes in the Time domain
Spectrum Holes in the Frequency 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
can be mitigated
Royal Institute of Technology KTH
WP 1/T1.2: Spatial diversity
-90 -85 -80 -75 -70 -650.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Mean RSSI (dBm)
Packet
Deliv
ery
Ratio
Time Diversity Approaches for 0.1km/h and 1km/h
Pure Time Diversity (0.1km/h)
Piggybacking (0.1km/h)
Switch if No Acknowledgement (0.1km/h)
Piggybacking (1km/h)
Pure Time Diversity (1km/h)
Switch if No Acknowlodgement (1km/h)
12
Elektrobit’s: Channel Emulator PropSIM-c2
TABLE I
CHANNEL PARAMETERS
Tap CHANNEL 1 CHANNEL 2
Relative
tap delay
[ns]
Relative tap
amplitude
[ns]
Relative tap
delay [ns]
Relative tap
amplitude
[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
Multiple receiving antennas:
26% increase in packet delivery ratio
Royal Institute of Technology KTH
• Antenna switching and receiver selection diversity
WP 1/T1.2: Spatial diversity
1 2 3 4 5 6 7 80
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Links (1-8)
PA
CK
ET
DE
LIV
ER
Y R
AT
IO
Bridge 23.3m, Receiver sensitivity = -94 dBm
Receiver Array
• 10 packets/s
• Dual Antenna System
• Receivers Array (4)
Royal Institute of Technology KTH
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
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
Control over WirelessHART networks
P
C
WirelessHART network
Data stream characteristics:
• Slotted time
• Minimum transmission delay
• Time-varying latency, loss
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.
WP 2/T2.1 communication aware data fusion and control
Royal Institute of Technology KTH
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
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
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
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
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
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
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
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
• Measurement filter design based on the network delays
Filter design
Step experiment Filtering Extended plant response
AMIGO design
on extended plant,
tuning rules
G(s)
1( )
1f n
f
G ssT
max,fT f n max0 ( )t
max
(1 )/ 2
max
1, 1
3
1, 1.
13
nf
n
Tn
nnn
Royal Institute of Technology KTH
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 %
0 200 400 600 800 1000 12000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Time [s]
Qo
S
0 200 400 600 800 1000 12000
20
40
Time [s]
0 200 400 600 800 1000 12000
2
4
6
Time [s]
h [s]
Royal Institute of Technology KTH
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
– Automatic code generation, and
code reusability
• All in one tool
• Released as open-source to
researchers
• wsn.tkk.fi/en/software/piccsim
Control design
GUI
Royal Institute of Technology KTH
WP 3: PiccSIM
Royal Institute of Technology KTH
SINK
NODES
40M
20M
20M
10M
1 2 30
10
20
30
40
50
60
70
80
90
Mobility Models
Pa
cke
t D
elv
ery
Ra
tio
(%
)
FreeSpace
Light Machinery
Medium Machinery
Heavy Machinery
• 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
1 2 3 4 5 6 7 80
10
20
30
40
50
60
Link [#]
Packet
dro
p [
%]
Good Bad0
5000
10000
Good Bad0
5000
10000
Good Bad0
200
400
Good Bad0
100
200
Good Bad0
2000
4000
Good Bad0
100
200
Good Bad0
50
100
Good Bad0
2000
4000
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
1 2 3 4 5 6 7 80
10
20
30
40
50
60
70
80
90
100
Link [#]
Packet
dro
p [
%]
Good Bad0
200
400
Good Bad0
1000
2000
Good Bad0
500
1000
Good Bad0
100
200
Good Bad0
1000
2000
Good Bad0
100
200
Good Bad0
2000
4000
Good Bad0
100
200
0
0.5
1
0
0.5
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
Simulation: Crane Control
WP3: Using Field data in Simulations
Heavy Machinery Light Machinery
Residence time
Packet drop probability
Royal Institute of Technology KTH
WP3: Automatic Code Generation
• Simulation -> Implementation and testing on real
hardware
• Generic node block in PiccSIM library
– Make implementation in block
• Simulink blocks, Matlab code...
• Radio blocks for communication between nodes
• Matlab Real-Time Workshop
– Target Language Compiler (TLC)
– Generate code from Simulink block
• Wrapper main file for Sensinode node hardware
– Other wrappers can easily be implemented
Synchronize with Ns -2
do { ... } while
Stop bit
Process_interface
AD 0
Radio timestamp 1
Radio recv 1
DA 6
DA 7
Radio send 1
Process
Process
Pendulum _controller
Radio timestamp 1
Radio recv 1
Radio trigger 1
Radio send 1
Node _Controller
Send to N 0 T 3
Send enable
Timestamps
Data N 1 T 2
Node
ID = 2
Node _ KF
Send to N 2 T 2
Timestamps
Data N 0 T 1
Node
ID = 1
Kalman _filter
Radio timestamp 1
Radio recv 1
Radio send 1
Interface node
Send to N 1 T 1
Timestamps
Data N 2 T 3
Node
ID = 0u
Royal Institute of Technology KTH
WP3: Automatic Code Generation
Radio send 1
3
DA 7
2
DA 6
1
Tapped Delay
4
Delays
Signal Specification
D : 3 Saturation
Rate Transition
Gain 3
2 . 5
Gain 2
2 . 5 / 0 . 8 Gain 1
90 / 2 . 5
Detect
Change
U ~ = U / z
Data Type Conversion
double
Constant
L
Bias 2
u - 2 . 5 / 2
Bias 1
u + 0 . 4
Add
Radio recv 1
3
Radio timestamp 1
2
AD 0
1
Royal Institute of Technology KTH
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
PROSE communication details
Royal Institute of Technology KTH
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
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
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
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
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: [email protected]
http://wsn.tkk.fi