Lei Tang, Yanjun Sun, Omer Gurewitz, and David B. Johnson - - PowerPoint PPT Presentation

PW-MAC: A Predictive-Wakeup MAC Protocol for Wireless Sensor Networks Lei Tang, Yanjun Sun, Omer Gurewitz, and David B. Johnson

Presentation at IEEE INFOCOM 2011, April 2011

PW-MAC objectives

2

Minimize energy consumption both at senders and at receivers while maintaining:

• High packet delivery ratio
• Low delivery latency

Related work: duty cycling

• Duty-Cycling MAC protocols:

– Synchronous: e.g., S-MAC, DW-MAC. – Asynchronous: e.g., B-MAC, PW-MAC, EM-MAC.

3

Duty cycle: The percent of time a node is active.

Related work: synchronous protocols

• Problems of synchronous protocols:

– Global time synchronization. – Contention is packed to DATA period.

4

SYNC DATA SLEEP RTS RTS DATA Cycle

Related work: asynchronous protocols

• Asynchronous

– Nodes wake up asynchronously. – No global time synchronization.

• How does sender rendezvous with receiver?

– Sender-initiated or receiver-initiated.

5

Picture from http://www.fleetcouriers.com/blog/

Related work: B-MAC (sender-initiated)

6

S

Send Receive Node awake Preamble Data Data

R

Time

Problem:

• 1. Sender has large duty cycle.
• 2. High channel contention.

wakeup interval

Related work: X-MAC (sender-initiated)

• Preamble is replaced by repeating data packets.
• Receiver sends an ACK so the sender can stop.

7

S R

DATA DATA DATA DATA A A wakeup interval

Problem:

• 1. Sender has large duty cycle.
• 2. High channel contention.

Time

Related work: WiseMAC (sender-initiated)

8

S

shortened preamble P . P. Data Data

R

Time

Problem:

• 1. Fixed wakeup interval can cause collisions.
• 2. Use a fixed clock drift ratio.

fixed wakeup interval fixed wakeup interval

Other problem of these protocols

• No efficient retransmission mechanism  large duty

cycle and high wireless contention.

9

S1 R

packet packet packet

S2

packet packet packet packet packet collide collide collide collide Time

S R

Send wake-up beacon DATA Wake up to wait for receiver B B DATA Send DATA Send ACK A A

Problem: sender still has large duty cycle due to idle listening and overhearing.

Related work: RI-MAC (receiver-initiated)

Time

10

Overview of PW-MAC

11

DATA DATA Sender S1 DATA DATA Receiver R1

• High energy efficiency at both senders and receivers.
• High packet delivery performance through reducing

collisions and efficient packet retransmission.

DATA DATA Sender S2 DATA DATA Receiver R2 Time

Predictive Wake-up mechanism of PW-MAC

12

pseudorandom time R1 wake-up B B B pseudorandom time R1 wake-up R1 wake-up time

Next wakeup time = now+ pseudorandom(0.5 interval, 1.5 interval)

pseudorandom time R2 wake-up B B R2 wake-up time

R2: R1:

Predictive Wake-up mechanism of PW-MAC

13

Why pseudorandom wakeup?

• Enable sender to predict receiver reduce energy consumption
• Spread traffic to different times mitigate wireless contention

pseudorandom time R wake-up B B B pseudorandom time R wake-up R wake-up time

R:

Predictive Wake-up mechanism of PW-MAC

14

Prediction state obtained by a node S to predict a node R’s wakeups includes:

• Pseudorandom number generator parameters and

current seed of R.

• The time difference between S and R.

pseudorandom time R wake-up B B B pseudorandom time R wake-up R wake-up time

R:

PW-MAC packet transmission

15

S R

wake up at predicted time pseudorandom time DATA B A B DATA A B Through prediction, sender and receiver wake up at the same time  high energy efficiency. DATA is generated time

The problem of packet retransmission

16

S2 R2

DATA Existing work: stay awake and repeat the packets. DATA DATA DATA

S1 R1

DATA DATA DATA Large duty cycle and high channel contention. time DATA DATA DATA

PW-MAC: prediction-based retransmission

17

go to sleep DATA B A B DATA A

S R

DATA B B DATA retransmit at the next predicted receiver wake-up time failure Retransmit packets with high energy efficiency and low wireless contention. time

Outline

1. PW-MAC 1.1 Predictive Wake-up MAC (PW-MAC) 1.2 Prediction-based retransmission 1.3 On-demand prediction error control

18

Prediction error on a pair of MICAz motes

19

Small clock drift ratio

Prediction error on another pair of MICAz motes

20

Much larger clock drift ratio

PW-MAC: on-demand prediction error control

21

DATA B A B DATA A

S R

Wakeup advance time

Prediction error: the difference between predicted and actual wakeup time.

time

Use a wakeup advance time to compensate prediction error.

PW-MAC: on-demand prediction error control

22

DATA B A B DATA A

S R

Detect prediction error Request current prediction state Update prediction state Send prediction state time

Ensure prediction error to be within sender wakeup window

23

• PW-MAC effectively controls the prediction error

to be ≤ wake-up-advance time

• Low cost: average 1 update per 1400 seconds

Advantages:

• Effective
• Low cost: average 1 update per 1800 seconds

PW-MAC: on-demand prediction error control

24

DATA B A B DATA A

S R

Detect prediction error Request current prediction state Update prediction state Send prediction state time

Hidden terminal experiment

25

T wo MICAz senders are hidden to each other With WiseMAC, two senders’ repeated retransmissions cause persistent collisions.

Experiment of wake-up schedule conflicts

26

T wo receivers have the same first wake-up time.

S1 S2 R1 R2

Pseudorandom wakeup scheduling of PW-MAC avoids persistent wakeup collisions.

Multihop network performance

27

• A testbed of 15 MICAz motes.
• Up to 3 multihop traffic flows.

Sender duty cycle

28

PW-MAC has the smallest duty cycle

Packet delivery latency

29

PW-MAC achieved lowest latency

Packet delivery ratio (PDR)

30

Packet Delivery Ratio (PDR)

PW-MAC achieved 100% PDR

Conclusions

• Predictive-wakeup mechanism:

– High energy efficiency at both senders and receivers. – Low channel contention.

• Prediction-based retransmission mechanism.
• On-demand prediction error control.
• PW-MAC outperformed other tested single-channel

protocols on a testbed of MICAz motes.

31