Lecture #5

Wi l Wireless Sensor S N Networks t k Dr. Kun Yang University of Essex, Colchester, UK

Monday 23rd March 2009
1

Agenda

Basics of Wireless Sensor Networks A TMDA-based Control Plane Algorithm Deployment and Power Assignment Problem in WSNs Summary, Q&A
Some materials courtesy to I.F. Akyildiz, etcs paper and slides A Survey on Sensor Networks, IEEE Communications Magazine, August 2002.

Faster, Smaller, Numerous

Moores Law
Bells Law: New computer class every 10 years mainframes (1960s) minicomputers (1970s) PC enabled by LAN (1980s) Internet & Web browser (1990s) Web services, e.g., MS .NET or the Grid (2000s), (c 2000) Smartphones (c. Wireless Sensor Networks, (c. >2005): home and body area networks will form by 2010
3

Wireless Sensor Networks: Features

Recent advances in wireless communications and electronics have enabled the development of low-cost, low-power, multifunctional sensor nodes that are small in size and communicate in short distances. These tiny sensor nodes, which consist of sensing, data processing, and communicating components, leverage the idea of sensor networks. Sensor networks represent a significant improvement over traditional sensors. Sensor nodes are densely deployed either inside or close to the phenomenon. R d Random deployment d l t in i inaccessible i ibl terrains t i or disaster di t relief li f operations.
This also means that sensor network protocols and algorithms must

possess self-organizing capabilities.

Sensor nodes are fitted with an onboard processor.

Therefore certain data processing over raw data is carried out before

transmission.

WSN Applications
A wide range of applications In military, for example, the rapid deployment, self-organization, and fault tolerance characteristics of sensor networks make them a very promising sensing technique for military command, control, communications, computing, intelligence, surveillance and targeting systems. In health, sensor nodes can also be deployed to monitor patients and assist disabled patients. Some other commercial applications include managing inventory, monitoring product quality, and monitoring disaster areas.

WSN Architecture
The sensor nodes are usually scattered in a sensor field. Each of these scattered sensor nodes has the capabilities to collect multihop data and route data back to the sink via multihop. The sink may communicate with the task manager node via Internet or satellite.

Architecture of a Sensor Node -1

Architecture of a Sensor Node - 2

A sensor node is made up of four basic components:
a sensing unit, usually composed of two subunits: sensors and analog-to-

digital converters (ADCs).

a processing unit, which is generally associated with a small storage unit,

manages the procedures that make the sensor node collaborate with the other nodes to carry out the assigned sensing tasks. a transceiver unit: it connects the node to the network. a power unit: might use solar cells.

Or other additional application-dependent components such as:

A location finding system: for sensing or routing purpose Mobilizer: M bili to t move sensor nodes d Power generator

Factors Influencing the WSN Design

The design of a WSN is influenced by many factors.
fault tolerance Scalability production costs operating environment sensor network topology hardware constraints transmission media power consumption

These factors can serve as a guideline to design a protocol or an algorithm for WSNs. In addition, these influencing factors can be used to compare different schemes.

Fault Tolerance
Why fails?
Lack of power, physical h i l damage, d or environmental interference

The reliability Rk(t) of a sensor node is modeled using the Poisson distribution to capture the probability of not having a failure within the time interval (0, t):

where k and t are the failure rate of sensor node k and the time period, respectively.
10

Scalability
The number of sensor nodes deployed may be on the order of hundreds , thousands or even millions. The density can be calculated as

N is the number of scattered sensor nodes in region A; R is the radio transmission range.

Th number b of f nodes d in i a region i can be b used dt i di t The to indicate the node density.

11

Production costs
Since sensor networks consist of a large number of sensor nodes, the cost of a single node is very important to justify the overall cost of the networks. The cost of a sensor node should be much less than 1$ in order for the sensor network to be feasible. Comparison: The state-of-the-art technology allows a Bluetooth radio system to be less than US$10 [4]. Also, the price of a piconode is targeted to be less than US$1. The cost of a sensor node should be much less than US$1 in order for the sensor network to be feasible. The cost of a Bluetooth radio, which is known to be a low-cost device, is even 10 times more expensive than the targeted price for a sensor node.

12

Hardware constraints
Size
matchbox-sized module

consume extremely low power, operate in high volumetric densities, have low production cost and be dispensable, be autonomous and operate unattended, be adaptive to the environment.

13

Sensor network topology

Deploying a high number of nodes densely requires careful handling of topology maintenance. Pre-deployment and deployment phase
Sensor nodes can be either thrown in mass or placed one by one in

the sensor field.

Post-deployment phase
Sensor network topologies are prone to frequent changes after

deployment.

Re-deployment p y of additional nodes p phase

Addition of new nodes poses a need to re-organize the network.

14

Operating Environment
Sensor nodes may be working
y intersections, in busy
in the interior of a large machinery, at the bottom of an ocean, inside a twister, in a battlefield beyond the enemy lines, in a home or a large building.

15

Transmission media
Sensor nodes are linked by a wireless medium. These links can be formed by radio, infrared, or optical media. Much of the current hardware for sensor nodes is based on RF circuit design. design Industrial, scientific and medical (ISM) bands
offer license-free communication in most countries.

Infrared
license-free and robust to interference requirement of a line of sight between sender and receiver. Infrared-based transceivers are cheaper and easier to build.

Optical medium: in Smart Dust mote, which is an autonomous sensing, computing, and communication system. Both infrared and optical require a line of sight (LOS) between the sender and receiver.

16

Power consumption
Sensor nodes are only equipped with limited power source(<0.5Ah, 1.2V). Node lifetime strongly dependent on battery lifetime.
In some application application, replenishment of power resources might be

impossible.

In a multihop ad hoc sensor network, each node plays the dual role of data originator and data router. The malfunctioning of a few nodes can cause significant topological changes and might require rerouting of packets and reorganization of the network. Hence, power conservation and power management take on additional dditi li importance. t The main task of a sensor node in a sensor field is to detect events, perform quick local data processing, and then transmit the data.
Power consumption can hence be divided into three domains: sensing/acurating, communication, and data processing.
17

Communications via Radio

Radio can mean differently:
Motes radio is only a transceiver, and a lot of low-level processing takes

place in the main CPU While, Whil typical t i l 802 802.11b 11b radios di d do everything thi up t to MAC and d li link k level l l encryption in the radio

Energy per bit in radios is a strong function of desired communication performance and choice of modulation
Range and BER for given channel condition (noise, multipath and

Doppler fading)

Transmit, receive, idle, and sleep modes

E.g. E g WINS consumes only 1/6-th 1/6 th the power when MCU is asleep as

opposed to active

Idle radio consumes almost as much power as radio in Rx mode

Radio needs to be completely shut off to save power as in sensor networks idle time dominates MAC protocols that do not listen a lot
18

Energy Consumption: Computation vs. Communication

Energy breakdown for voice
Encode Decode

Energy breakdown for MPEG

E Encode d Decode Transmit

Receive

Transmit

Receive Radio: Lucent WaveLAN at 2 Mbps P Processor: St StrongARM ARM SA SA-1100 1100 at t 150 MIPS

The relative impact of the communication subsystem on the system energy consumption will grow
19

Energy in Radio
Tx: Sender I Incoming i information
Tx Eelec
Transmit electronics

Rx: Receiver Channel Outgoing information

Rx Eelec
Receive electronics

ERF
Power amplifier

Wireless communication subsystem consists of three components with substantially different characteristics Their relative importance depends on the transmission range of the radio
Courtesy to M. Srivastavas slides for Mobicom02 tutorial
20

10

Energy Consumption Examples

GSM
nJ/bit
8000 6000 4000 2000 0 100 200

Nokia C021 Wireless LAN

nJ/bit
300

Medusa Sensor Node (UCLA)

nJ/bit
600

200

400

E RF E

Tx elec

Rx elec

~ 1 km

ERF E E ~ 50 m

Tx elec

Rx elec

Tx Rx E RF Eelec Eelec 0m ~ 10

The RF energy increases with transmission range The electronics energy for transmit and receive are typically comparable
Courtesy to M. Srivastavas slides for Mobicom02 tutorial
21

Simulation Tools
Sensor Network-level Simulation Tools
Ns-2 enhancements by ISI Ns Ns-2 2 based SensorSim/SensorViz by UCLA C++-based LECSim by UCLA PARSEC-based NESLsim by UCLA

Node-level Simulation Tools

MILAN by USC for WINS and AMPS ToS-Sim for Motes

Processor-level Simulation Tools

JoulesTrack by MIT

22

11

Protocol Stack

Used by the

sink and sensor nodes

23

Protocol Layers
The physical layer addresses the needs of simple but robust modulation, transmission, and receiving techniques. Since the environment is noisy and sensor nodes can be mobile, the h medium di access control l (MAC) protocol l must b be power-aware and able to minimize collision with neighbors broadcasts. The network layer takes care of routing the data supplied by the transport layer. The transport layer helps to maintain the flow of data if the sensor networks application requires it. Depending on the sensing tasks, different types of application software can be built and used on the application layer.

24

12

Management Planes
These management planes make sensor nodes work together in a power efficient way, route data in a mobile sensor network, and share resources between sensor nodes. Power management plane
manages how a sensor node uses its power. For example, the sensor node may turn off its receiver after receiving a

message.

When the power level of the sensor node is low, the sensor node

broadcasts to its neighbors that it is low in power and cannot participate in routing messages.

Mobility management plane

detects and registers the movement of sensor nodes So a route back to the user is always maintained the sensor nodes can keep track of who are their neighbor sensor nodes.

Task management plane

Balances and schedules the sensing tasks given to a specific region. Not all sensor nodes in that region are required to perform the sensing

task at the same time.

25

Physical Layer
Frequency selection, carrier frequency generation, signal detection, modulation, and data encryption. 915 MHz ISM band has been widely suggested for sensor networks. signal propagation effects
the minimum output power required to transmit a signal over a distance

d is proportional to dn, where 2<= n < 4. The exponent n is closer to four for low-lying antennae and near-ground channels, as is typical in sensor network communication. multihop communication in a sensor network can effectively overcome shadowing and path loss effects, if the node density is high enough.

Modulation schemes: Simple and low-power modulation schemes.

The modulation scheme can be either baseband, as in UWB, or

passband.

Strategies to overcome signal propagation effects Hardware design

Open research issues

26

13

Data Link Layer

The data link layer is responsible for the multiplexing of data stream, data frame detection, medium access and error control It ensures reliable point-to-point and point-tomultipoint connections in a communication network. Media Access Control two goals:
Creation of the network infrastructure Fairly y and efficiently y share communication resources between

sensor nodes

27

Why existing MAC protocol cant be used?

The primary goal of the existing MAC protocol is the provision of high QoS and bandwidth efficiency. In a cellular system, the base stations form a wired backbone.
A mobile node is only a single hop away from the nearest base station. infrastructure- based: base station doing network-wide

synchronization

Power-awareness is not a big deal: base stations have unlimited power supply and the mobile user can replenish exhausted batteries in the handset.

Mobile ad hoc network (MANET), a close peer to WSNs

MAC in a MANET: forming the network infrastructure and maintaining it in the face of mobility. Hence, the primary goal is the provision of high QoS under mobile conditions. Battery can be replaced by the user, and hence power consumption is only of secondary importance.

28

14

Why existing MAC protocol cant be used?

In contrast to these two systems, the sensor network may have a much larger number of nodes. The transmission power (~0 dBm) and radio is much less th than th those of di range of f a sensor node d i hl f MANET. MANET Topology changes are more frequent in a sensor network due to both node mobility and failure. The mobility rate can also be expected to be much lower than in the MANET. In essence, the primary importance of power conservation to prolong network lifetime in a sensor network means that none of the existing MANET MAC protocols t l can b be di directly tl used. d

29

MAC for WSNs

MAC protocol for sensor network must have built-in
p power conservation, mobility management failure recovery strategies

How to achieve energy efficiency:

A variant of TDMA, random medium access such as CSMA constant listening times adaptive rate control schemes.

Some MAC protocols for WSNs to follow

30

15

Network layer
The networking layer of sensor networks is usually g g to the following gp p designed according principles:
Power efficiency is always an important consideration. Sensor networks are mostly data centric. Data aggregation is useful only when it does not hinder the

collaborative effort of the sensor nodes. location awareness.

An ideal sensor network has attribute-based addressing and

31

Power efficiency
Node T is the source node that senses the phenomena phenomena. PA is the available power is the energy required to transmit a data packet through the related link.
Route 2: Sink-A-B-C-T, total PA=6, total =6, Route 1: Sink-A-B-T, total PA=4, total =3,

Route 3: Sink-D-T, total PA=3, total =4,

Route 4: Sink-E-F-T, total PA=5, total =6 32

16

Power efficiency
Maximum available power (PA) route gy ( ) route Minimum energy (ME) Minimum hop (MH) route
Select Route 2 Select Route 1 (if the same then ME=MH) Select Route 3 (if the same then MH=ME) Select Route 3 Preclude the risk of using up a sensor node with low PA.

Maximum minimum PA node route

33

Data-centric Routing
Interest dissemination is performed to assign the sensing tasks to the sensor nodes. Two approaches used for interest dissemination:
Sinks broadcast the interest Sensor nodes broadcast an advertisement for the available data

and wait for a request from the interested sinks.

Requires attribute-based naming

Querying an attribute of the phenomenon, rather than

querying an individual node.

Eg: the areas where the temperature is over 70F is a more

common query than the temperature read by a certain node

34

17

Data aggregation
A technique used to solve the implosion and overlap problems in datacentric routing. Data coming from multiple sensor nodes with the same attribute of phenomenon are aggregated. aggregated Data aggregation can be perceived as a set of automated methods of combining the data the comes from many sensor nodes into a set of meaningful information.
With this respect, data aggregation is

Known as data fusion.

Sensor network is usually perceived as a reverse multicast tree.

35

Transport layer
The transport layer is needed when the system is planned to be accessed through Internet or other external networks. Few scheme related to the transport layer of a sensor network has been proposed in literature.
Hardware constraints such as limited power and memory. Acknowledgments are too costly.

TCP/UDP

36

18

Application layer
Potential application layer protocols for sensor networks remains a largely unexplored region.

Task assignment and data advertisement protocol (TADAP)

Users send their interest to a sensor node, a subset of the nodes or whole

network. This interest may be about a certain attribute of the phenomenon or a triggering event. Another approach is the advertisement of available data in which the sensor nodes advertise the available data to the users

Sensor q query y and data dissemination p protocol ( (SQDDP) Q )

SQDDP provides user applications with interfaces to issue queries, respond to queries and collect incoming replies. attribute-based or location-based naming the locations of the nodes that sense temperature higher than 70 0C Temperatures read by the nodes in region A Sensor query and tasking language (SQTL) is proposed.
37

Agenda

Overview of Wireless Sensor Networks A Nimble and Adaptive TDMA Control Phase Algorithm for Cluster-based WSNs Deployment and Power Assignment Problem in WSNs Summary, Q&A

38

19

CSMA vs TDMA
Medium Access Control (MAC) protocol plays a critical role in the performance of wireless sensor networks. While led by IEEE 802 802.11 the Whil CSMA/CA as l db 11 being b i th mainstream technique for wireless network MAC, it is not inherently immune to the retransmission caused by collision, and overhearing and this retransmission consumes much sensor energy. In contrast to the random access protocols, the scheduling based MAC protocols such as TMDA bear inherent immunity y to these factors. This paper considers cluster-based sensor networks where the cluster head in a cluster serves naturally as a base station to carry out cluster-level time slot scheduling.
39

TDMA-based MAC
TDMA-based MAC in wireless networks typically involves the following four phases:
f h i i phase, h frame synchronization
control phase, scheduling phase and data transmission phase.

40

20

Main objective
Much work on scheduling, but less on the control p phase. The scheduling phase utilizes the result of the control phase, mainly the number of successful nodes requesting data transmission, to perform time slot allocation. The main work for control phase is to decide its own length in terms of time slots. This length, when contention is utilized, is also termed as contention window (CW).
41

Polling
A commonly used mechanism in control phase is polling, a scheduling-based approach. Polling reserves a slot for each sensor regardless the sensor's sensor s data transmission request. Polling guarantees no collision at a cost of longer control phase length, which as a result leads to longer packet delay time and low channel utilization (refer to the later analysis for details). Improvement to polling has been proposed which adopts random access [7], [8], mainly using CSMA since it is the most influential contention protocol in practice. However, many of these literatures on wireless sensor networks utilize CSMA directly, not focusing on how to find a proper backoff window size as the length of TDMA control phase.

42

21

Backoff Window Size Current Literature

Various researches have been carried out on backoff window size [9]~[12]. ] sets the backoff window size to a constant while leaving g Sift [ [9] nodes to select the probability of successful transmission in different slots. If no node transmits data in the first slot, every node increases its successful transmission probability for the next slot. Y. Xiao [11] introduces the priority into smaller-sized backoff windows. These researches have indicated that a mechanism that dynamic detects and estimates the number of active nodes usually performs better than otherwise. Cali [ [10] ] establishes the relation of the number of active stations and the idle period. Bianchi [12][13] proposes the relation of the number of contention nodes and the collision probability using Kalman and ARMA filter. The results of these works are all presented in average and expectation.
43

Shortcomings of the Current Literature

However, although research [13] uses Kalman and ARMA to get much more accurate results, the information needed for this is complex. calculation is massive and the computation p p Furthermore, these researches all assume that the number of nodes making request is fixed, namely, all nodes always have data ready to be transmitted (i.e. saturation conditions). These assumptions are not realistic in the real functioning of wireless sensor networks. An algorithm called NACPA is proposed in this paper, aiming to calculating the control phase length (i.e., the number of contention nodes) in a much simpler manner and without the above assumptions.

44

22

NACPA Features
This is achieved via a combination of experimental method and the re-use of the information obtainable g hardware ( from an existing (AGC: Automatic Gain Control). Moreover, this paper, for the first time to the best of our knowledge, proposes to calculate the number of contention node (denoted as N) from the contention slot's point of view rather than contention node's perspective. p p This design methodology contributes significantly to the simplification of the contention node estimation process and the control phase algorithm itself, as to be demonstrated in the paper.
45

Fundamentals
Sensor nodes contend for contention slots (CS) in the control phase. The length of a CS can typically accommodate a data transmission request (DTR) from a sensor. sensor When a sensor has data to transmit, it randomly selects a CS to send DTR to base station (i.e., the cluster head in cluster-based sensor networks as is the case of this paper), as illustrated in the figure below. These sensors that successfully obtain a CS are able to be scheduled a data transmission slot in data transmission phase. The failure ones retransmit their DTR in the next round.

46

23

Problem Re-investigation
Given:
N: the number of contention nodes, and M: the number of CS (Contention Slot) in a round

There is:
the higher the probability of a CS being a valid slot (i.e., being selected by

only one sensor) is,

the more sensors are that get their DTR received without collision, i.e.,

the higher the total number of valid slots becomes, and consequently the more r (the success rate of nodes getting their DTR received by the cluster head) and u (the control phase utilization) become.

Therefore, the problem of maximizing r and u is converted to a problem of maximizing the probability of a CS being a valid slot by tuning M. Theorem 1 below shows there exist such a maximum of the probability of a CS being a valid slot and the value of this maximum is highly coupled with the values of M and N.
47

Theorem 1
Theorem 1: the probability of a CS being a valid slot is maximized when M=N.
Proof: Given a CS, let p(x) be the probability of this CS being selected by x sensor nodes simultaneously. simultaneously p(x) is calculated by: (1) When x=0, p(x) is the probability of a CS being idle; when x=1, p(x) is the probability of a CS being a valid slot; when x>1 (and with ceiling N), p(x) is the probability of a CS being an invalid one. NACPA is interested in how to tune M so as to obtain a maximum p(1). Since M is a variable in Equation , p(1) is denoted as p(M) and derived from as follow:
p(M ) = N 1 1 1 M M
N 2
N 1

1 1 p( x) = C x N 1 M M

N x

In order to get the maximum of p(M), the derivative of p(M) is calculated:

dp ( M ) 1 = N ( N M ) 1 dM M 1 M3

Set Equation =0, we have M=N. This means, when M=N, the probability of CS being used successfully is the maximal.
48

24

Problem Conversion
The target of NACPA is to, at the end of round i, calculate the contention window size of next round, i.e., Mi+1. Theorem 1 proves that when M=N the probability of a CS being a valid g Mi+1 is converted into slot is maximized. Then the p problem of calculating the calculation of Ni+1. However, it is practically impossible to know Ni+1 because round (i+1)-th has not occurred yet. A commonly used technique is to use Ni to replace Ni+1 [10]. So NACPA utilizes the result of ith round to decide the contention window size of the (i+1)-th round. An improvement is to use the results of all i rounds to calculate the contention window of the (i+1)-th round, such as in the work by Bianchi [13] where Kalman filter is utilized for this purpose. purpose A similar approach can be integrated into our work. However the focus of this paper is how to estimate a more accurate N for one round with a significantly reduced computation complexity. For presentation simplicity and without ambiguity this paper uses N to denote Ni.
49

Analysis on N
N is composed of two parts: the number of success nodes and the number of failure node, i.e.:
N = mS + ni
i =1 mF

where mS is the number of valid CS's (i.e. the number of successful sensor nodes), mF is the number of invalid CS's, and ni is the number of contention nodes on i-th invalid CS. N h this hi calculation l l i is i carried i d out f h slot's l ' Note that from the point of view rather than node's perspective. Now the problem is how to get N - the total number of nodes making DTR in a given round.
50

25

Further Analysis on N
Using the signal from AGC (Automatic Gain Control) circuit in the wireless communication radio interface, a cluster head can detect if a slot is idle, used by only one sensor, or there is a collision. Namely, l
mS and mF can be obtained.

However AGC hardware cannot tell how many nodes leading to the collisions in an invalid CS. This paper proposes to utilize a m probability-based mechanism to estimate i =F1 ni since sensors randomly select contention slots. To find out the most probable number of contention nodes causing lli i on a given i CS d this hi probability-based b bili b d estimation i i is i a collision CS. A And across all CS's in the control phase. We denote this most probable number of contention nodes as Nc. Then, the equation in the previous slide is converted into: N = mS + mF N c
51

Theorem 2
Theorem 2: Given N and M, there exist one and only one x where x>1 such that p(x) is maximized.
Its physical implication is: when collision occurs on one CS, there is

only l one most t probable b bl number b of f nodes d (i.e. (i Nc) ) that th t have h selected l t d this CS simultaneously.

Proof: refer to the published paper.

Fig. 2 Fig. 2 on p(x) illustrates that when x>1, each graph has only one maximum (i.e. when x=Nc), as proved in Theorem 2.

52

26

Calculation of Nc*
Expectation E(x) is used to compute the average most probable number of contention nodes, denoted as Nc*. * Nc = ( x ) = i xi p ( xi | xi > 1) N = mS + mF N c* Now let's consider the behaviour of E(x) under N=M.
Recalculate p(x)

When N=n, we denote p(x) as p1(x) and p2(x) respectively for N=n and N=n+1. there is: when N is much larger than 1, p1(x) is very close to p2(x).
Nc*=E(x)=2, when N=5, and 10; Nc*=E(x)=2.3, when N=11, 30 and 80.
* = ( x) = Nc

2, when M = N 10 2.3, when M = N > 10

53

M<N
When M<N, NACPA sets M to Mmax where Mmax is the maximum size of a content window allowed by the system system.

2, when M = N 10 * N c = ( x ) = 2.3, when M = N > 10 xi p ( xi | xi > 1, p ( xi ) > ), when M N i

54

27

*( i +1) Nc = ( x)

( i +1)

2, when M = N 10 = 2.3, when M = N > 10

*( i +1) Nc = ( x)

( i +1)

= x j p ( x j | x j > 1, p ( x j ) > )
j

55

Analytical Evaluation: Bounds of Average Delay

56

28

Analytical Evaluation: Bounds of Channel Utilization

57

Simulation Results: Average Delay

Q=250 bytes payload in the data transmission phase: Q

58

Q=125 bytes

29

Simulation Result: Channel Utilization

59

Conclusions
A dynamic and adaptive algorithm called NACPA to control the control phase of TDMA-based MAC in sensor networks. NACPA firstly proves that the best performance is achieved when contention window size is set equal to the number of contention nodes in one round. Then, by taking advantage of the AGC hardware feature, NACPA proposes a more accurate way to calculate the number of contention nodes in one round and discards the non-practical assumptions of a fixed number of contention nodes. By y further analyzing y g the features of contention p probability y against g the number of contention nodes, NACPA significantly reduces its computation complexity making it feasible for resourceconstrained sensor networks. The analytical evaluation and simulation results both showed its effectiveness and efficiency in comparison with two typical MAC algorithms: polling and CSMA.
60

30

Main References
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.

I. Akyildiz, W. Su, and Y. Sankarasubramaniam, 'A Survey on Sensor Networks', IEEE Communication Magazine, August 2002, pp. 102-114. F. Liu, K. Xing, and X. Cheng, 'Energy-efficient MAC layer protocols in ad hoc networks', Kluwer Academic Publishers, 2004. IEEE Computer Society. ANSI/IEEE Standard 802.11, 1999. K. Sivalingam, J. Chen, P. Agrawal, and M. Srivastava, 'Design and analysis of low-power access protocols for wireless and mobile ATM networks', Wireless Networks, 6(1):73-87, 2000. K.T. Jin and D.H. Cho, 'A new MAC algorithm based on reservation and scheduling for energy-limited ad hoc networks', IEEE Transactions on Consumer Electronics, Vol.49, pp. 135-141, Feb. 2003. J. Chen, K. Sivalingam, and P. Agrawal, 'Scheduling multimedia services in a low-power MAC for wireless and mobile ATM networks', IEEE Transaction on Multimedia, Vol. 1, pp. 187-201, Jun. 1999. L. Bao and J. Garcia-Luna-Aceves, 'A new approach to channel access scheduling for ad hoc networks', Proc 7th Annul Int'l Conf on Mobile Computing and Networking, Rome, pp. 210-221, July 2001. V. Rajendran and K. Obraczka, 'Energy-efficient, collision-free medium access control for wireless sensor networks', Proc 1st Int'l Conf on Embedded Networked Sensor Systems, Los Angeles, CA. 181~192, November 2003. Y. C. Tay, y K. Jamieson and H. Balakrishnan, 'Collision-Minimizing g CSMA and Its Applications pp to Wireless Sensor Networks k ', IEEE Journal l on Selected l d Areas in Communications, Vol. l 22, No. 6, Aug. 2004. F. Cali, M. Conti and E. Gregori, 'Dynamic tuning of the IEEE 802.11 protocol to achieve a theoretical performance limit', IEEE/ACM Transaction on Networking, Vol. 8, No. 6, December 2000. Y. Xiao and Y. Pan, Differentiation, 'QoS Guarantee, and Optimization for Real-Time Traffic over One-Hop Ad Hoc Networks', IEEE Transaction on Parallel and Distributed Systems, Vol. 16, No. 6, June 2005. G. Bianchi, 'Performance analysis of the IEEE 802.11 distributed coordination function', IEEE Journal on Selected Areas in Communications, 2000, 18(3):535~547. G. Bianchi and I. Tinnirello, 'Kalman filter estimation of the number of competing terminals in an IEEE 802.11 network', Proc. of the INFOCOM 2003. Vol.2, 2003. 844~852. 61

Agenda

Overview of Wireless Sensor Networks A Nimble and Adaptive TDMA Control Phase Algorithm for Cluster-based WSNs Deployment and Power Assignment Problem in WSNs Summary, Q&A

62

31

Objectives
Deployment: to find the initial location of a sensor Power assignment: to specify the initial transmission power of f a sensor In order to maximize two (conflicting) objectives: coverage & lifetime Challenge: how to achieve these two goals simultaneously (and may subject to some constraints)? Solution: using g modern evolutionary y algorithms g and meta-heuristic. Offline, provide guidance to WSN designers

63

System Model
Consider a 2-D static WSN formed by:
a rectangular sensing area A, a number of homogeneous sensors N and a static sink H with unlimited energy, placed at the center of A. The sensors are responsible to monitor and periodically report an

event of interest to H.

We assume a perfect medium access control and adopt the simple path loss communication mode (square distance-based). Residual Energy: A is composed of rectangular grids of identical dimensions centered at (x, y) and we adopted a binary sensing model. A grid at (x, y) is covered, denoted by g(x, y) = 1, if it falls within a sensors sensing range Rs, otherwise g(x, y) = 0. We consider unit-size grids, which are several times smaller than Rs, for a more accurate placement 64

32

Problem formulation
Given: A: 2-D plane of area size x y N: number of sensors to be deployed in A. A E: initial power supply, the same for all sensors. Rs: sensing range, the same for all sensors. The design variables set (X) is composed by: Lj : the location of sensor j. Pj : the transmission power level of sensor j. Objectives: Maximize coverage Cv(X) and lifetime L(X). The network coverage Cv(X) is defined as the percentage of the covered grids over the total grids of A and is evaluated as follows:

The network lifetime is defined as the time that the first sensor dies.

65

Proposed Algorithm

66

33

Genetic Operators
Encoding

Selection operator

Crossover

67

Mutation

Results
Pareto Front

coverage

Algorithm convergence and other MOP (Multi-objective Problem) factors are also evaluated.

ual energy Residu

lifetime # of cycles
68

34

References for more details

A. Konstantinidis, Q. Zhang, K. Yang. A Subproblem-dependent Heuristic in MOEA based on Decomposition for the Deployment and Power Assignment Problem in Wireless Sensor Networks, 2009 IEEE g y Computation p ( ) Norway. y 2009. Congress on Evolutionary (CEC 2009), A. Konstantinidis, K. Yang, Q. Zhang. Problem-specific Encoding and Genetic Operation for a Multi-Objective Deployment and Power Assignment Problem in Wireless Sensor Networks, IEEE ICC 2009 (Int. Conf. on Communications), Germany, June 2009. A. Konstantinidis, K. Yang, Q. Zhang. An Evolutionary Algorithm to a Multi-Objective Deployment and Power Assignment Problem in Wireless Sensor Networks, Globecom08, Dec. 2008, New Orleans, USA. A. Konstantinidis, K. Yang, H H-H H Chen, QF. Zhang. Energy Energy-aware aware Topology Control for Wireless Sensor Networks Using Memetic Algorithms, Elsevier Journal of Computer Communications, Vol. 30, 2007. Pages: 2753-2764. A. Konstantinidis, Q. Zhang, K. Yang, I. Henning. On Energy-aware Topology Control of Wireless Sensor Networks Using Modern Heuristics, IEEE Globecom 2006, San Francisco, USA, Nov. 2006
69

ContactQ&A
Dr Kun Yang School of Computer Science & Elec. Eng. (CSEE), University of Essex, Wivenhoe Park, Colchester, CO4 3SQ, UK Email: kunyang@essex.ac.uk http://privatewww.essex.ac.uk/~kunyang/

70

35