Wireless Network Design for Distributed Control

Liu and Goldsmith - Appeared at CDC 2004

Presented by Vinod Namboodiri

Networked Control is Universal

Information among distributed sensors controllers actuators

needs to be exchanged to achieve a certain control objective

Automated Highway System

Wireless networks become important

Wireless Network for Distributed Control

Multiple control systems co-exist Performance tradeoffs

Control wants accurate, timely, lossless data Random delay, losses acceptable in network design – esp. wireless

Optimize tradeoffs to achieve best end-to-end control performance

Joint Design is Important Tradeoff exists between communication

and controller performance More the controller knows about system,

better the performance More sensors Transmit sensor measurements more frequently

However, this increases communication burden Congestion can result in longer delays or packet

losses – degrades control performance

Goals of Joint Design

Controller design robust and adaptive to communication faults like random delays and packet losses

Communication design optimized for control performance

Cross-Layer Design

Each layer of network stack optimized relative to end-to-end controller performance

Cross Layer Design Parameters Packet Delay Packet Loss

Data Resolution Affects network traffic, which in turn

affects delay and losses

Network, MAC and Link Layer

Application Layer

Linear Quadratic Cost Function (H2 norm), f(delay, loss, data resolution)

Control Performance Measure

Strategy Difficult to simultaneously optimize all the

layers Use sub-optimal iterative method over

physical, MAC and application layers Fix data resolution and MAC protocol, and

choose best physical/link layer For this link design and data resolution, choose

the best MAC protocol Optimize data resolution for chosen link and MAC

protocol Repeat till convergence

Layer Parameters

Physical Layer QPSK, BPSK Modulation

MAC Layer TDMA Random Access (RA) w/ and w/o ACK CSMA/ CA

Control Layer Sample Period

Physical Layer Design choice of modulation

BPSK QPSK

Data Rate x 2x

Larger probability of

error

y c.y, c>1

Error Coding can help reduce probability of error - (15,7) code means 15 coded bits for 7 bits of information- Code rate = information bits/coded bits

Choice of MAC TDMA

Collision free protocol based on time slots

Time slot can be wasted if pre-assigned transmitter has no packet to send

Random Access (RA) Grab channel for transmission

independently with a probability p Variations include with or without ACK

CSMA/CA Sense channel before transmitting Exponential backoff after a collision

Loss free, can add significant delay

Loss possible, delay depends on contention

Control Parameter – Sample Period

Greater sample period Lower data resolution Lesser traffic, possibly lesser delay and

loss High data rate not required; impacts

choice of physical layer design

Iteration 1, Step 1 – Link Design

RA with ACK, p = 0.166 Sample Period = 5 ms

Select QPSK, (15,11) code

Iteration 1, Step 2 – MAC Design

QPSK, (15,11) code Sample Period = 5 ms

Select p = 0.28

Iteration 1, Step 3 – Sample Period

QPSK, (15,11) code RA with ACK, p = 0.28

Select sample period = 6ms

Iteration 2, Step 1: Link DesignRA with ACK, p = 0.166 Sample Period = 6 ms

Select QPSK, (15,11) code

Next Steps

p = 0.28 is again optimal

Thus, algorithm converges

Apparent Advantages of Cross-Layer Design

We chose QPSK even though it is more unreliable

Without joint design, we could have chosen BPSK and may have lead to system instability

Things to ponder about

Are all network faults considered in control design? Currently only random packet drops,

random delay, data rate limits considered

Bursty packet losses instead of independent packet losses

Time varying channels instead of static channels

Thank You

Link Layer Design with TDMA MAC

Example control system