Distributed Control over Wireless Networks

Transcription

1 Technical Information Systems Seminar Technische Universität Dresden June 19, 2008

2 Outline 1 Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance 2 Delay and Jitter Packet Loss Throughput Control versus Communication Performance 3 4

3 The Problem Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance Physical System Controller Reference reference input r(τ)

4 The Problem Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance Actuator Physical System Sensor control input u(τ) Controller system output reference input y(τ) r(τ) Reference

5 Example: Lighting Control The Problem Wireless Sensor and Actuator Networks Control Performance Dimmer Gallery Photodiode target voltage u(τ) [V] Controller Reference current light intensity y(τ) [Lux]

6 The Problem Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance Actuator Physical System Sensor spatially distributed Controller Reference

7 The Problem Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance Actuator Physical System Sensor spatially distributed wireless links Controller Reference

8 The Problem Introduction and Motivation The Problem Wireless Sensor and Actuator Networks Control Performance Actuator Physical System Sensor spatially distributed wireless links Controller Reference lossy & time-varying unpredictable

9 The Problem Wireless Sensor and Actuator Networks Control Performance Control versus Communication Theory Control theory: ideal channels No (or at least deterministic) communcation and computation delays Synchronous, periodic sampling and actuation No packet drop-outs Communication theory: imperfect channels Non-deterministic communication delays (time-varying) Asynchronous communication Packet loss unavoidable in wireless communication

10 The Problem Wireless Sensor and Actuator Networks Control Performance Wireless Sensor and Actuator Networks (WSANs)

11 The Problem Wireless Sensor and Actuator Networks Control Performance Wireless Sensor and Actuator Networks (WSANs) MICAz (Crossbow) 16-bit RISC, 8 MHz, 48 kb flash IEEE , 250 kbps, 2.4 GHz $ 99 (USD)

12 WSAN Possibilities The Problem Wireless Sensor and Actuator Networks Control Performance Battery-powered independence Flexible deployments flexible network architectures Application in inaccessable, hazardous areas Different sensors on a single module multi-purpose Local (pre-)processing distributed computation Data exchange using low-power radio collaboration Low price many sensors (>1000) heaps of data

13 The Problem Wireless Sensor and Actuator Networks Control Performance WSAN Problems from Control Perspective Problems: delay, jitter, packet loss Control performance degrades Even system instabilility

14 The Problem Wireless Sensor and Actuator Networks Control Performance WSAN Problems from Control Perspective Problems: delay, jitter, packet loss Control performance degrades Even system instabilility Integral absolut error (IAE): J(t) = t 0 r(τ) y(τ) dτ time

15 The Problem Wireless Sensor and Actuator Networks Control Performance Good versus Bad Control Performance ideal without delay comp. with delay comp.

16 Causes and Types of Delay/Jitter Delay and Jitter Packet Loss Throughput Control versus Communication Performance Causes of communication delay/jitter: Node mobility varying transmission distances Sleep scheduling dynamic network topology Adaptive MAC and routing protocols variation in paths, number of hops,... External disturbances Causes of computation delay/jitter: Varying execution times of different control computations Competing processes Types of delay/jitter: Sampling jitter: sampling interval varies Sampling-actuation jitter: delay between sampling and actuation varies

17 Delays in Control Loop Delay and Jitter Packet Loss Throughput Control versus Communication Performance Actuator Physical System Sensor Controller Reference

18 Delays in Control Loop Delay and Jitter Packet Loss Throughput Control versus Communication Performance Actuator Physical System Sensor communication delay/jitter (1) Controller Reference

19 Delays in Control Loop Delay and Jitter Packet Loss Throughput Control versus Communication Performance Actuator Physical System Sensor communication delay/jitter (1) (2) Controller computation delay/jitter Reference

20 Delays in Control Loop Delay and Jitter Packet Loss Throughput Control versus Communication Performance Actuator Physical System Sensor (3) communication delay/jitter (1) (2) Controller computation delay/jitter Reference

21 Packet Loss Introduction and Motivation Delay and Jitter Packet Loss Throughput Control versus Communication Performance Causes of packet loss: Transmission errors Buffer overflows due to congestion Long delays (outdated packets are discarded) Consequences: Controller is not up-to-date Control inputs must be estimated Unavoidable in presence of radio interference, multi-hop communication, variable transmit power, node mobility Packet loss is highly dependent on transmission distance

22 Delay and Jitter Packet Loss Throughput Control versus Communication Performance Packet Loss versus Transmission Distance

23 Throughput Introduction and Motivation Delay and Jitter Packet Loss Throughput Control versus Communication Performance High data rate fine-grained control How to increase the data rate? Deploy more sensors Increase sampling rate of sensors However, this can cause Network congestion many retransmissions delay/jitter High energy consumption at nodes nodes die quickly reduced visibility or even network partition Trade-off between control and communication performance

24 Delay and Jitter Packet Loss Throughput Control versus Communication Performance Control versus Communication Performance

25 Impact of Network Architecture Delay and Jitter Packet Loss Throughput Control versus Communication Performance Integrated controller Stand-alone controller

26 Example: Networked Control System (NCS) Process time-driven h Sensor Actuator event-driven Wireless Network PD Controller event-driven Disturbance long random delays

27 Simulation and Analysis using TrueTime Based on MATLAB/Simulink Simulation of task execution on nodes Simulation of simple communication models Facilitates to study impact of different Communication models Delay/jitter compensation strategies Task scheduling policies

28 NCS Model in TrueTime

29 NCS Model in TrueTime

30 NCS Model in TrueTime Process h = 10 ms Sensor IDEAL Actuator Wireless Network 0 ms no interf. task PD Controller 0 ms Disturbance no disturbance

31 NCS Model in TrueTime Process h = 10 ms Sensor DELAY Actuator Wireless Network 1.5 ms interfering task PD Controller 0.5 ms Disturbance 50% high priority packets

32 Delay/Jitter Compensation through Gain Scheduling Delay sensor controller: Assumption: packets from sensor are timestamped Controller can determine actual delay Delay controller actuator: Assumption: probability distribution of delay is known Controller can estimate expected delay (Rough) idea: Precalculated controller parameters for different delays Proportional gain: K p Derivative gain: K d Select corresponding parameters based on current delay Compute control signal via linear interpolation using Selected controller parameters Current error (r(τ) y(τ))

33 Control Performance ideal without delay comp. with delay comp.

34 WSANs allow for highly embedded, ubiquitous sensing, actuation, and control Delay, jitter, and packet loss degrade control performance Adaptive control algorithms for compensation Joint design of communication and control infrastructure Early, iterative analysis throughout design process with powerful tools (Jitterbug, TrueTime)

35 Appendix Further Reading Further Reading I Akyildiz and Kasimogul. Wireless Sensor/Actuator Networks: Resear. Challenges. Ad Hoc Networks, 2(4): , October Murray, Åström, Boyd, Brockett, and Stein. Future Directions in Control in an Information-Rich World. IEEE Control Syst. Mag., 23(2):20 33, April Carvin, Henriksson, Lincoln, Eker, and Årzen. How Does Control Timing Affect Performance? IEEE Control Syst. Mag., 23(3):16 30, June Hespanha, Naghshtabrizi, and Xu. A Survey of Recent Results in Networked Control Systems. Proceedings of the IEEE, 95(1): , January 2007.