A Transmission Control Scheme for Media Access in Sensor

Networks

Alec Woo, David Culler

(University of California, Berkeley)

Special thanks to Wei Ye @ USC for some of the ppt slides

Sensor Net Traffic Characteristics

• Sampling of environment for sensory information

• Propagation of time series information to infrastructure

• Duty cycle low until abnormal event sensed

• Periodic sampling creates a high amount of highly correlated traffic

Sensor Networks

• Sensing and Radio capabilities• Limited storage, processing power• Limited ENERGY• Small packet size due to low bandwidth• Multi-hop topologies• Route-thru traffic at sensor nodes• Bi-directional connectivity (in the face of

multipath interference and short irregular transmission ranges)

Related Work

• IEEE 802.11 & MACAW– Last-hop wireless, single cell scenario– Peer to peer communications – Single-hop base station interaction within a cell

• Bluetooth– Centralized TDMA within piconet– Time synchronization requirement

• Home RF– Combo of contention control protocol for synchronization

data traffic and centralized TDMA for synchronous voice– No multihop support

Evaluation Metrics

• Fairness in sensor data coverage– Original vs. route-thru traffic– But, route-thru traffic already has network

resources invested in it

• Energy Efficiency– Aggregate bandwidth per unit of energy– Fairness and high channel utilization are at

odds

Fairness

• Fairness in bandwidth allocation– Each node generates roughly the same amount

of data– Example

• N nodes in a multihop network

• All data are sent to 1 base station

• Ideally, the base station should equally receive 1/N portion of data from each node

Fairness

• Fairness in bandwidth allocation

– If nodes 4, 5, 6 generate too many packets, congestion may happen at node 1.

– Node 1 has no more bandwidth for its own data.

Base station

12

3

4

5

6

Sensor Network Platform

• ATMEL 4MHz 8 bit microprocessor– 8K program memory– 512 bytes data memory

• Single channel RF transceiver– Operating at 916MHz– 10kbps using on-off-keying encoding

• Variety of Sensors – Temperature, photo, etc.

• TinyOS – event-based operating system– 30 byte messages

Design

• Listening Mechanism– CSMA effective when all nodes can hear each other– Periodic listening to conserve energy– CD requires additional circuitry– Problematic for periodic and synchronized traffic– Solution: introduce random delay

• Backoff mechanism– Also can help to break the synchronization by introducing a

phase shift

• Contention based mechanism– Control packets, like RTS/CTS/ACK, are large overhead if

packets are short (up to 40% overhead)

Design

• Rate control mechanism– MAC controls the rate of originating data of a node

S * p, where p = [0, 1], S is application transmission rate

– Without any MAC control packets

– A node periodically tries to inject a packet• If successful, linearly increase transmission rate

p = p + • If unsuccessful, multiplicatively decrease rate

p = p *

Design

• Rate control mechanism without control packets1. How does a node know whether its transmission is

successful or not?• If my parent routes the same packet to my grandparent, I

know my transmission is successful

• Success relies on symmetric links and implicit ACKs

2. How do we deal with collision without RTS/CTS?• Constantly tune transmission rate.

• If I just transmitted a packet, restrain further transmission for duration x, which is the processing time at my parent.

Simulation Environment

• Simple home-grown simulator– Simple radio propagation model (not

specified)– Zero bit error rate

Simulation

• Analysis of CSMA schemes

Simulation

• Single-hop topology

Simulation

• Settings– Channel capacity 10 kbps– Packet length 30 bytes– Node transmission rate 5 packet/s– 16 bit CRC for error detection– Highly synchronized traffic – all nodes start at

the same time

Simulation Results – Single Hop

• Schemes with randomness built into delay or listening achieve both fairness and stable channel utilization.

• Overall, the best CSMA schemes are those that have constant listen period with random delay of transmission

Simulation – Multihop Scenario

• Topology and fairness results

Simulation – Multihop Scenario

• Energy efficiency

Simulation – Multihop Scenario

• Aggregate bandwidth

Conclusions

• Adaptive rate control can effectively achieve fairness of bandwidth allocation

• Energy efficient – no control packets

• More collisions than RTS/CTS schemes

• Lower aggregate bandwidth and yield than RTS/CTS schemes

Critique

• Link symmetry assumption

• Many to one routing assumption

• No aggregation

• Control overhead not so large if per ADU

• Weak multihop results

• Can we trust their simulator?

• Extraordinarily sparse references