Sensor Network II

Sensor Network II

2002. 3. 26Lee, chunseung

Codesign And Parallel processing laboratory, School of Computer Science and Engineering, SNU

[email protected]://peace.snu.ac.kr

Tel. 880-7292

Outline• What is the “sensor networks” ?• Application of sensor network• Characteristic of sensor network• Issues for sensor network

– Transmission Control Scheme• Listening mechanism• Backoff mechanism• Contention Based mechanism• Rate Control mechanism• CSMA schemes• Simulation and Empirical results

Outline• Biomedical Sensor

– Introduction– Biomedical application– Preliminary results– Two dimensional analysis– Three dimensional analysis– Wireless Communication Protocol

• Cluster based approach• Tree based approach

– Performance Comparison• Conclusion• Side dishes

– Low power technology in system level• Referances

What is the “Sensor Networks”

• Ad hoc network of sensors– data event traffic

• Emerging area of mobile computing

• Primary Function– Sensory information

• temperature, humidity

– Propagate this data block into the infrastructure

Application of SN

• Internet Security Systems’s– RealSecure, Solaris, WindowsNT– 보안 침입 탐지

• EUCOM( 미유럽사령부 , European Command)– 합동 원격 생물학 조기 경보 체계– JBREWS(Joint Biological Remote Early Warning System)

• JBPDS(Joint Biological Point Detection System)– 공군기지 , 항구 , 배 , 해외 주둔군에 배치– 개인사병 , 헬기 , 소형 무인 정찰 비행기– 부대에서 개인의 보호 수준을 결정하기 위한 첩보 및 기상자료를

송출– 3kg, 100km

Application of SN

• Virtual Keyboard– Each fingernail has a sensor for finger-moving– Or using Sensing board

Application of SN

• Monitor pollutions

Application of SN

Application of SN

Odyssey is a low-cost AUV specifically developed by the MIT

AUV – Autonomous Underwater Vehicles

Application of SN

A Transmission Control Scheme for Media Access in

Sensor Networks

Characteristics of SN• A. Woo, D. E. Culler, “A Transmission Control Scheme for Media Access in Senso

r Networks”, In MOBICOM, July, 2001

• Unusual application requirements– highly constrained resources

• Small packet size– typically 10 bytes or below

• Deep ad hoc multihop dynamic topology• A correlated operating

– Short periods the traffic may be very intense

• Periodic rendezvous

Issues for SN in this paper

• Motivation– The design space is different from traditional mobile computer

networks– SN needs tight constraints

• computational power• Storage• Energy resource• Radio technology

• Targets– High channel utilization– Communication efficiency on energy– Fair bandwidth allocation

• Propose an adaptive rate control(ARC) mechanism

Listening mechanism

• Ethernet environment– CSMA/CD– Very effective and simple– Collision detection is not possible in wireless network without

additional circuitry– The radio must be on to listen during backoff

• More energy consuming

• To conserve energy, it have to shorten the length of carrier sensing– IEEE 802.11 radio off

Backoff Mechanism

• A widely used in media access control to reduce contention– Binary backoff algorithm

• To restrain a node from accessing the channel

Contention Based Mechanism

• Widely used in many MAC protocols– RTS-CTS-ACKs

• RTS (Request To Send)• CTS (Clear To Send)

• For sensor networks where packet size is small, they can constitute a large overhead– RTS-CTS-ACKs handshaking

• Up to 40%

• A contention control scheme for sensor networks should use a minimum number of control packets

Contention Based Mechanism

• RTS – CTS handshaking– First send a RTS packet to it parent and waits for a CTS reply– If no CTS is received for a timeout period (2 CTS packet tim

e), the node will enter backoff with a binary exponential increasing backoff window

– If no CTS has been received after five retries, the transmission will be dropped

– If a node hears a CTS before any of its own transmission, it will defer transmission for one packet time to avoid corrupting the traffic.

Rate Control Mechanism

• The tension between originating traffic and route-thru traffic has a direct impact in achieving fairness goal.

• This paper propose an mechanism which adapts the rate of transmission without the use of any MAC control packets

• Fairness channel allocation– Channel Capacity / N , where N is total number of node in

the entire network– The spontaneous ad hoc of sensor networks make

impractical• Proposed transmission rate control mechanism

– Linear increase– Multiplicative decrease

Rate Control Mechanism

• S : application transmission rate• S*p : the actual rate of originating data. p [0, 1]• p : probability of transmission• : a constant• : multiplicative decrease a factor where 0 < < 1

• route = 1.5 * originate

• originate = route / (n + 1)

controls the penalty given a failure of transmission

S : current rate

if(S is acceptance) {p = p + ;

S = S * p; } else {

p = p * ;S = S * p;

}

CSMA schemes

• Packet size – 30 byte– Manchester encoding

• Channel capacity– 10 kbps, 20.8 packet/sec

• 16bit CRC error detection for corrupted packet• 각 노드는 근소한 시작시간을 가지고 주기적으로 초당 5 packet 을 보냄• 채널용량이 20.8packet/s 이므로 traffic load 는 4 번째 노드이상에서

채널용량을 초과할 것이다 .(saturation)

Simulation Setting

Utilization and Bandwidth of Channel

Constraint

- high channel Utilization(bandwidth)

- energy efficiency

- fairness

실험 결과

명시적인 ACK 을 사용하는 802.11 보다 더 좋은 성능을 나타냈다 .

특히 , constant listen period 와 no random delay 를 갖는 3 개의 스킴은 가장 높은 bandwidth 를 나타냄 .

하지만 그들은 모두 robust 하지 않다 .

( 원인 : 2 개의 dip repeated collision)

또 결과들을 살펴보면 , 802.11 을 제외한 모든 스킴들은 채널용량의 75% 정도 (15.6 packet/s 의 bandwidth) 의 이용률을 보였다 .

20.8*0.75 = 15.6 packet/s

결론적으로 , 성능면에서는 no random delay 와 constant listen 이 좋긴 하지만 robust 하지 않기 때문에 매력이 없다 .

Randomness 를 사용하는 backoff 메커니즘은 repeated collision 을 방지하는데 효과가 좋다 .

실험 결과- 모든 노드가 동시 전송 ( 최악의 경우 )

no random delay 와 constant listening window 를 가지는 3 개의 스킴은 zero bandwidth 가 나왔다 . (repeated collision)

802.11 은 초기에 5 packet/s 로 시작하였고 , 그림 4 보다 낮은 throughput을 보였다 .

802.11 이 no random delay 와 constant listening 주기를 가지고 있을지라도 ACK 가 collision 감지와 backoff 메커니즘을 트리거하기 때문이다 .

나머지 커브들은 random delay 를 가지거나 random listening 주기를 가지는 스킴들이다 .결론 ,

1 . Randomness 를 갖는 알고리즘들은 좋은 채널이용률과 high load 를 잘 견뎌낸다 .

2. Performance 는 backoff 메커니즘에 영향을 받지 않는다 .

3. Backoff 메커니즘을 가지지 않는 ND_RAND 조차도 다른 6 개의 스킴들 만큼이나 잘 동작한다 .

4. Pre-collision 단계안에서의 randomness 는 robustness 를 위해 필수적이다 .

Energy Efficiency

Constraint

- high channel Utilization(bandwidth)

- energy efficiency

- fairness

Simulation Setting

• In examining the energy consumed in communication– Transmitting and receiving packet– During listening period

• The former is determined primarily by the traffic load and the latter is primarily determined by the CSMA protocol

대부분 에너지 효율적인 스킴들은 constant listening 주기와 random delay 이고 , robustness 를 제공한다 .

802.11 이 가장 안 좋은 효율을 보임

이유는 , constant listening 주기를 가지고 있을지라도 , backoff 동안 채널을 항상 listening 하고 있어야 되기 때문이다 .

constant listening 주기를 가지는 scheme들이 좋고 , 10 uJ/packet 으로 네트워크의 크기에 관계없다 .

random listening 주기를 가지는 스킴들은 40 uJ/packet 으로 다소 비싸다 .

특이한 점은 ND_RAND 는 네트워크의 크기가 증가할 수록 에너지가 증가한 이유는 , backoff 가 없기 때문이다 .

실험 결과

Fairness

Constraint

- high channel Utilization(bandwidth)

- energy efficiency

- fairness

앞 bandwidth 와 energy 효율성 측면에서 모두 좋은 random delay, constant listening 주기를 갖는 3개의 스킴을 비교해 보자 .

그림 7

위의 3 스킴의 표준편차 : 0.25 packet/s 이고 , 트래픽이 증가 할수록 감소하는 경향을 보인다 .

결국 , backoff 메커니즘의 차이는 uniform 한 부하면에서 보면 , fairness 가 별 중요치 않다 .

그림 8

802.11 은 unfair 하다 . 표준편차가 1 packet/s 이상이다 .

이유는 , 다른 것보다 이른 전송시간을 갖는 노드가 채널을 capturing 하기 때문이다 .(capturing effect)

실험 결과

표 3.

그림 3 에서의 node 0 을 multihop으로 구성했을 때 node 10 에 대한 bandwidth 를 기준으로 표준화시킨 값이다 .

Send rate 를 보면 노드가 10일때 , 2 packet/s 이고 , 노드수가 1일때 10이므로 산술적으로 노드 1,2,3 은 노드 10 에 비해서 500% 정도가 측정되야 하나 802.11 을 제외한 3 scheme 들은 500% 이상이 측정되었다 .

실험 결과

Backoff 메커니즘은 proportional fairness 에 영향을 준다 .

Phase-shift 를 적용한 802.11 스킴은 bandwidth, fairness 모두 많이 향상되었다 .

Phase-shift

The phase of the sensor sampling interval is shifted by a random amount in response to transmission failure.

실험 결과

Empirical Results on singlehopRandom delay 를 가지는 3 개의 스킴 실측하여 시뮬레이션과 비교

실험한 실측치도 채널 용량의 70% 까지 상승했다 . (performance 측면에서 )

Figure 12. The average energy consumption

- the prediction of around 10uJ/packet ( 10 ~ 40 uJ/packet at simulation)

Figure 13. The fairness comparison of throughput

- the deviation among the three schemes vary from 0.3 ~ 0.5 packet/s

(0.25 packet/s at simulation)

Multihop Scenario

Under simulation Maximum uniform origination rate

20/24 = 0.83 packet/s

In the implementation, limit origination rate 15.7/24 = 0.66 packet/s

• The simulation runs with each node sending packets to the base station at rate of 4 packet/s

• Runs at the same time

Simulation Measurements

그림 15.

ARC 스킴이 가장 좋은 성능을 나타냄 .

802.11 과 RTS/CTS 스킴은 레벨이 2 이하 일 때 전혀 packet 을 발생치 않았다 .(hidden node 문제 )

그림 16.

fairness 에 대한 편차를 , 값을 변경하면서 측정

ARC 가 가장 좋다 .

가 편차에 많은 영향을 준다 .(?)

그림 17.

값이 낮을수록 ( 큰 페널티를 부과 ) 낮은 bandwidth 를 나타낸다 .

802.11 과 D_CONST_FIX 가 더 높은 bandwidth 를 제공 ( 단 , base station 과 근접해 있어야 한다 )

그림 18.

값이 낮을수록 에너지 효율성은 증가

Empirical Result on Multihop

그림 20.

비교적 시뮬레이션 결과와 비슷

ARC 도 D_CONST_FIX보다 더 fair 하다 .

Conclusion

결론적으로 , randomness 를 사용하는 backoff 메커니즘은 repeated collision 을 방지하는데 효과가 좋고 ,

에너지 효율적인 면에서는 constant listening 주기와 random delay 이고 , robustness 를 제공한다 .

Phase-shift 방법을 사용하면 bandwidth 와 fairness 를 보다 향상 시킬 수 있다 .

ARC(Adaptive Rate Control) 방법은 명시적인 control packet 없이 fairness 를 효과적으로 동작시킬 수 있다 .

Research Challenge In Wireless Networks of Biomedical Sensors

Introduction• Constraint

– Limited power, computational capacities• Issues

– Bio-compatibility– Fault tolerant– Energy efficient, – Scalable design

• A smart sensor– Physical, chemical, biological sensors combined with integrated

circuits• The examples of smart sensors

– Combat scenarios to track troop movements– Mine robots, Pollution detection – Biomedical application– ‘A drop in the bucket’

Biomedical applications

• Artificial Retina

Biomedical applications

Biomedical applications

• Other applications– Glucose Level Monitors– Organ Monitors– Cancer Detectors– General Health Monitors

Preliminary results

• Power-efficient Topology– A function of the distance and number of bits

transmitted.– Trade off

• The number of neighbors and the total power dissipation

– Two Dimensional Analysis– Three dimensional Analysis

• Wireless Communication Protocols– Cluster-based Approach– Tree-based Approach– Performance Comparison

Two dimensional analysis

• As the number of neighbors increases the number of alternative paths increases.

• Interior routing– A route across the diameter of the network

• Edge routing– The path travels along the edges of the network

Two dimensional analysis

Edge routing dissipates less power than interior routing in all cases except for 3 neighbors

Two dimensional analysis

The power dissipated between the source and destination for a message spanning the diameter of the network for networks with 3 and 6 neighbors

Increasing the number of neighbors decrease the number of transmissions and the total power dissipated in the system

Three dimensional analysis

1. 3D 네트워크의 경우 edge routing 이 interior routing 보다 power 를 적게 소비한다 .

2. 3D 네트워크에서 interior, edge routing 둘 모두 2D 네트워크보다 전송수 , 수신수 , 총 power 소비량이 더 작다 .

1. Neighbor 가 적은 네트워크에서 power 를 적게 소비

2. Neighbor 가 많은 토폴로지라도 적은 홉수가 필요하다 .(같은 neighbor수라도 홉수가 적은 것이 power 소비가 적다 .

Wireless Communication Protocol

• Two communication protocols that were designed to reduce energy consumption

• Using a bi-directional transceiver• The placement of node is predetermined and fixed• All communication is wireless• TDMA is used for media access

– Nodes can sleep when they are not sending/receiving data– Less power usage and extended battery life

• All nodes are perfectly synchronized and are aware of the beginning of each slot

• The communications pattern is deterministic and periodic

• Each node has to transmit its data once in 250ms

Cluster-based approach

• Stipulating• Only a small fraction of the nodes are allowed to

communicate with the base station– These nodes are called leaders – Each leader collects data from nodes in its cluster,

compressing data and transits the data to the base station

• Two major advantages– Only relevant data is transmitted to the external processor– Only a small subset of the nodes make a long distance

transmission and hence energy is conserved

Cluster-based approach

• There are several issues– Which nodes should be leaders?

• Base station– Which cluster to join?

• The best signal-to-noise ratio• Using a TDMA scheme for collision

Tree-based approach

• To make a spanning tree– The base station selects one or more nodes to be its children

based on proximity and node density– These selected nodes then make a low intensity transmission,

each at a different frequency– If nodes that receive this transmission at a predefined minimum

signal-to-noise ratio can then request the transmitting node to be its parents using the subscribing protocols

– This continues until all nodes in the system the covered

• Data transmission– When a node wants to send data, it will send it to its parent node

by a low-energy transmission– The parent will collect data from all its children, compress the data

if required, and it turn transmit to the base station

Performance Comparison

- Two dimensional array of nodes with the base station

- The distance of the base station from the center of the arraywas assumed to be twice tat of the inter-node distance

- The variation of power consumption with respect to distance between nodes as well as the number of nodes in the network

Performance Comparison

Cluster based approach shows better results

Conclusion

• Smart sensor offer the promise of significant advances in medical– Artificial Retina– Glucose Level Monitors, Organ Monitors, Cancer Detectors

• Total power consumption is reduced for topologies with fewer neighbors

• Cluster based approach shows better energy-efficiency on distance between nodes and number of nodes than tree based one’s

Conclusion on SNII

• Existing communication protocols are not necessarily sufficient for the sensor network– Due to many different constraints and requirements– New network topology or application requires a new

regime

• Especially, power management will be more significantly

Side dishes

• Low power technology in system level– DPM(Dynamic Power Management)

• idle 상태에서 device 의 전압을 낮춘다 .• 최근 device 의 사용 history 에 근거하여 다음 idle 주기를 예측하는 것이 중요

• DLPM(Device level power management) : old approach• TBMP(Task Based Power Management) : new approach

– DVS(Dynamic Voltage Scaling)• 전압 스케쥴러 (voltage scheduler) 가 시간 제약 조건을

만족하는 범위에서 프로세서의 동작 전압을 조절하여 최소한의 에너지를 소모하도록 하는 방식 (voltage scalable processor)

• PDA 부류의 디바이스에 장착

• MAC Schemes– A. Woo, D. E. Culler, “A Transmission Control Scheme for Media Access in Sensor Netw

orks”, In MOBICOM, July, 2001

• Biomedical Sensors– L. schwiebert, S. K. S. Gupta, J. Weinmann, “Research Challenges in Wireless Networks

of Biomedical Sensors", In MOBICOM,

• Cooperative Sensing Networks– J. Agre, L. Clare, “An Integrated Architecture for Cooperative Sensing Network”, Rockwell

Science Center, May, 2000, COMPUTER, IEEE/ACM p106~p108

• Distributed Surveillance Sensor Network– http://www.spawar.navy.mil/robots/undersea/dssn/dssn.html

References

References

• Low power– DPM

• M. B. Srivastava, A. P. Chandrakasan, R. W. Brodersen, “Predictive System Shutdown and Other Architectural Techniques for Energy Efficient Programmable Computation”, IEEE Transaction on VLSI System, Vol. 4, No1, Mar. 1996

• Y. H. Lu, E. Y. Chung, T. Simunic, L. Benini, G. De Micheli, “Quantitative Comparison of Power Management Algorithms”, Proc. Of Design Automation and Test in Europe, pp. 20~26, 2000

– DVS• T. Pering, T. Burd, R. Brodersen, “The Simulation and Evaluation of Dynamic Voltag

e Scaling Algorithms”, Proc. Of International Symposium on Low Power Electronics and Design, Aug. 1998

• T. Burd, R. Brodersen, “Design Issues for Dynamic Voltage Scaling”, Proc. Of International Symposium on Low Power Electronics and Design, Jul, 2000