An Adaptive Energy-Efficient and Low-Latency MAC for Data Gathering in Wireless Sensor Networks

Gang Lu, Bhaskar Krishnamachari, and Cauligi S. RaghavendraDepartment of Electrical Engineering

University of Southern California

18th Intl. Parallel and Distributed Processing Symposium (IPDPS’04)

Outline

Introduction DMAC Protocol Performance Evaluation Conclusions

Main Concern

Primary goal: Energy efficiency Limited battery power

Secondary goal: Latency, throughput, fairness Source of energy consumption

Idle listening Collisions Overhearing Control packet overheads

Issues in S-MAC

Main idea Eliminate idle listening

Major shortcoming Long delay Fixed duty cycle

Unadapted to the traffic variation Improvement

Adaptive listening Challenge

Data Forwarding Interruption (DFI) problem

DFI Problem

Data forwarding process will stop at the node whose next hop is out of the overhearing range because it is in sleep mode

Packets are queued until the next active period, which increases latency

DMAC Protocol

Application Data gathering

Motivation DFI problem

Objective Design of a MAC protocol, where nodes can

dynamically increase duty cycles and their nearby nodes keep asleep

DMAC Overview

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

…… sleep

…… sleep

…… sleep

…… sleep

data ack

data ack

data ack

data ack

data ack

data ack

time

duty cycleadaptation

dataprediction more-to-send

staggered wakeup schedule

Staggered Wakeup Schedule

Nodes wake up sequentially like a chain reaction An interval comprises receiving, sending, and sleep

periods

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

Rx Tx

…… sleep

…… sleep

…… sleep

…… sleep

data ack

data ack

data ack

data ack

data ack

data ack

time

Staggered Wakeup Schedule (cont’d)

Tx period = Rx period = Tx: transmit data packet, receive ack packet Rx: receive data packet, transmit ack packet

Collision avoidance = BP + CW + DATA + SP + ACK

BP: backoff period (like DIFS in IEEE 802.11) CW: contention window SP: short period (like SIFS in IEEE 802.11) BP > SP

Duty Cycle Adaptation

Scenario A node has multiple packets to send at a sending slot

Idea Increase own duty cycle Request other nodes on the multihop path to increase the

ir duty cycles Solution

slot-by-slot renewal mechanism More data flag (piggybacked in both data and ack packets)

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Tx

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Tx

Multiple packets

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Tx

data(more data)

Multiple packets

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rxack(more data)

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rx

Tx

data(more data)

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rx

Tx

Rxack(more data)

time

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rx Rx

time

data

Tx Tx

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rx Rx

time

ack

Tx Tx

Duty Cycle Adaptation (cont’d)

Rx Tx

Rx Rx

Tx

Tx

sleep

time

Rx

Tx

data

Data Prediction

A

B C

S S S Rx Tx ……Rx

S S S S S ……Rx Tx

S S S Tx S ……Rx Tx

A

B

C1 packet 1 packet

additional sending

slot

additional receiving

slot

data

data ack

ack

3

3

A

More-to-Send Packet

MTS packet Explicit control packet

Dest. local ID and a flag Request MTS (flag=1) /clear MTS (flag=0)

The MTS packet incurs small increase in slot length, and energy consumption

B

A

D

C

More-to-Send Packet (cont’d)

A node sends a request MTS when Channel is busy Receives a request MTS from its children

A node sends a clear MTS when Buffer is empty All request MTSs received from children are cleared It sends a request MTS to its parent before and has not se

nt a clear MTS

Performance Evaluation --- Model

Energy cost The total energy cost of

deliver a certain number of packets from sources to the sink

Latency End-to-end delay of a

packet Delivery ratio

The ratio of the successfully delivered packets to the total packets originating from all sources

Radio bandwidth 100 Kbps

Radio transmission range 250 m

Radio interference range 550 m

Data packet length 100 bytes

Transmit power 0.66 W

Receive power 0.395 W

Idle power 0.35 W

MTS packet length 3 bytes

Receiving/sending slot (w/ MTS)

10 ms (11 ms)

Active period in SMAC 20 ms

Duty cycle 10%

Multihop Chain Evaluation

source

• Scenario (11 nodes)

sink

200 m

Increase linearly

Increaselinearly

Non-next hop nodes haveadditional active periods

Traffic Load Evaluation --- Random Data Gathering Tree

Sensor field: 1000m 500m area 50 nodes randomly deployed Sink is at the right bottom corner 5 sources are at the margin

Traffic Load Evaluation --- Latency

• full active CSMA/CA has the smallest delay for all traffic loads• DMAC/MTS handle the highest traffic load with the smallest delay

Traffic Load Evaluation --- Energy

• DMAC and DMAC/MTS are two most energy efficient MAC protocols• DMAC outperforms DMAC/MTS because of less overheads requested by MTS packets

Traffic Load Evaluation --- Delivery Ratio

• DMAC/MTS is better than other protocols• SMAC and DMAC have lower delivery ratio when the traffic load is heavy

Scalability Evaluation

100 nodes randomly deployed in a 100m 500m area

Traffic generation: 1 message per 3 seconds Sink: at the right bottom corner Sources: at the margin

Scalability Evaluation --- Latency and Energy

Energy Latency

SMAC: a general-purpose energy-efficient MAC protocolDMAC: not suitable for applications that require data exchange between arbitrary sensor nodes

Conclusions

Techniques in DMAC protocol Staggered active/sleep schedule Duty cycle adaptation Data prediction More-to-send packet

Advantages in DMAC protocol Tree-based data gathering Energy-efficient Low latency