uSense: A Unified Asymmetric Sensing Coverage Architecture for Wireless Sensor NetworksYu Gu, Joengmin Hwang,Tian He,David Hung-Chang Du

Presented ByJesmin Jahan TithiStd No: 0409052065

Outline•Coverage•Related Works•Motivation•Key Contributions of The Paper•The Architecture•Switching Algorithm•Scheduling Algorithm•Results•Advantages•Limitations•Future Direction

Coverage•To provide Coverage in wireless sensor

network means to provide some functionalities or services to a specific area using sensors. ▫For example to detect some target events

in a specific area using sensors

•Coverage algorithms aim to support flexible and efficient coverage in sensor networks

Related Works•Physical sensing coverage

▫Full coverage Full surveillance coverage Differentiated surveillance k-coverage by approximations and huristics k-barrier coverage

▫Partial coverage partial coverage within a fixed time delay partial coverage with guaranteed worst-case

detection delay

Motivation• To support multiple operating scenarios

Download separate program images and switch between them

• To incorporate flexibility and efficiency in sensing coverage

Problem• Excessive overhead in terms of communication

bandwidth, energy and storage

Solution• Conceptual separation of switching from

scheduling

Key Contributions of The Paper• uSense: Unified Sensing Coverage

Architecture▫Key Features

Asymmetric Architecture Decoupling of sensing coverage into scheduling

and switching▫Global Scheduling

Implemented in a separated powerful computational entity

Support multiple scheduling algorithms Calculates the parameters of a working schedule

for individual nodes▫Generic Switching

Implemented in lightweight sensor nodes Turns on/off the sensors according to the

scheduling parameters

uSense: Asymmetric Sensing Architecture

•Static Network•Scheduling algorithm generates scheduling parameters to be used by the switching algorithm•Bi-directional communication

Generic Switching Algorithm•Generic algorithm to accommodate

various types of schedules•Parameters used

▫Schedule bits S Infinite binary string 1=active state, 0=inactive state

▫Switching rate R R=the rate of toggling between states When R―>∞, infinite string of schedule bits

shows on/off behavior generated by any coverage algorithm

• Takes (S, R) as inputs

• As S is usually periodic and follows certain pattern, S can be expressed with regular expression

• (0010)∗ denotes a repeated off-off-active-off schedule

• Timed Finite Automata is used for state transition (01 or 10)

• Delays of transitions are the gaps between 01 or 10 segments

Generic Switching Algorithm

Scheduling Algorithm•Implemented separately (e.g., at the

second tier)-can support a large number of coverage algorithms

•Need to convert the output of a coverage algorithm into two parameters understandable by the generic switching algorithm

uScan: Global Scheduling Algorithms• Two-level scheduling• uScan divides the area into small regions, and

decides the working schedules for these regions

• Nodes are assign to cover the active regions at different time intervals, using a set-cover technique

• The schedule bits S for individual node is decided by combining first-level schedule and the set-cover assignment

• Outputs of uScan are (S, R)

Fig.2.The Design of uScan

Assumptions• Nodes are time-synchronized and their locations

are precise

• Sensing area of a node is a circle with a nominal radius r centered at the node’s location

Detail Algorithm

Level I: Tessellation•The area under surveillance is partitioned into small rectangle tiles with size smaller than the minimum target size

•Nodes do not have the concept of tiles and partition

Tile Scheduling: Line Scan•Only a column/row of tiles is covered in a certain interval of time during one round of scan

• Covered columns/rows increase or decrease consecutively

•Only a small percentage of tiles are sensed at a specific point of time less energy consumption

Fig.3.Regular Tessellations

Line Scan (Continued)

• Speed of scan =v

• Tile length=Ll, Tile width=Lw

• For horizontal scan, switch rate R=v/Ll

• • For vertical scan, switch rate

R=v/Lw

• A tile with coordinates (row, col) has index of row ∗ colmax + col

• To cover a tile t(i) with a coordinates (row, col) in a scanning round, schedule bits S

• Sh (i) = (000..000 1 000..000 )∗ (Hscan)

col−1 colmax−col• Sv (i) = (000..000 1 000..000)* (V scan) row−1 rowmax−row

Fig.4.Horizontal Scan

Schedule bits of a two-way scanS(i)= Sh(i)|Sv(i)

Tile Scheduling: Systolic Scan• Scanned from inner layer to the outer layer

• For the first time interval, the tiles at the center of the area set their first digit of schedule bits to 1: (1 000..000)*

N/2−1

• For the nth time interval, schedule bits

where n = 0, 1, 2..., N/2 − 1 and i is the index of tiles which satisfies specific conditions

Fig.5.SystolicScan

//colmax=rowmax =N

Level II: Node Scheduling• Translates a known tile schedule TSi into a corresponding

node schedule bits S, interpretable by a generic switching algorithm

Assumptions• A tile is said to be covered as long as a portion of the tile is covered

• Let a column of tiles TS= {T1, T2, T 3, T 4, T 5} is covered by a set of nodes NS={N1, N2, N3, N4, N5}

Fig

.6.

Ph

ysic

alC

ove

rag

e

Node Scheduling•Physical coverage is mapped to a

Coverage Bipartite Graph according to the coverage relationship

Fig

.7.B

ipart

iteG

rap

h

Fig

.6.

Ph

ysic

alC

ove

rag

e

Node Scheduling•All one-cover set with minimal number of

nodes are identified, until the size of one-cover set is above a certain threshold

Node Scheduling

• Three one-cover sets for TS: CS1={N1, N5}, CS2={N2, N3} and CS3={N1, N4}

• CS1, CS2 and CS3 can provide coverage to the tile set TS in a round-robin fashion

Fig.6. PhysicalCoverage

Node Scheduling• For CS1, CS2 and CS3, node schedule will have

three segments, each of which has a length of the tile schedule

• If a node belongs to the CSk set, the kth segment has the same value as the tile schedule

• Otherwise, the kth segment has an all-zero value

• If TS=0010, the final schedules for nodes N1, N2, N3, N4, N5 are

S1=(0010 0000 0010)∗ S2=(0000 0010 0000)∗

S3=(0000 0010 0000)∗ S4=(0000 0000 0010)∗

S5=(0010 0000 0000)∗

Node Scheduling (Why Polynomial?)• Identify a minimum set of nodes to cover an

active tile set

• Generic Minimum Set Cover(MSC) problem is NP-Hard

• Line scan coverage is a special case of the generic set cover problem: a node covers only a continuous segment of tiles

• Coverage Bipartite Graph can be mapped into a DAG in polynomial time

One-to-one mapping rules

Fig.7.BipartiteGraphFig. 8. MSC using DAG

•N tiles in TSi into N vertices V={v1, ...,vN } and add one extra vertex vN+1 •If a node covers a set of tiles {Ti,...,Ti+n}, create n directional edges (vi ,vj) where vj= vi+1,..., vi+n+1

•Each edge has a unit cost•Tile set cover problem ≈ problem of finding out the shortest paths from v1 to vN+1

Illustration

• All the tiles are covered using one of the following node sets: {N1,N3}, {N1,N4}, {N1,N5}, {N2,N3} or {N2,N4}, which are five corresponding shortest paths from v1 to v6

Selecting Cover Sets for Multiple TS• In a 2-D space a node may need to cover multiple tile

sets TSi• To cover the area

▫ Each node maintains a counter SC to record how many times it has been selected into final Cover Sets

▫ For a tile set TSi, calculate the minimum cover set MCSi among the nodes with minimum SC values. If the nodes with minimum SC values can not form a complete cover set, nodes with higher SC values are used

▫ The smallest eligible MCSi (SMCS) is selected and recorded for the purpose of node scheduling, and the SC values of nodes within this SMCS set are incremented

▫ Each TSi has a coverage threshold M× 2π /√27, denoting the maximum number of nodes that can be used in a selected MCSi

▫ The SMCS selection process is repeated until the size of all MCSi are larger than their thresholds

Create Node Schedule Bits• Let, k one-cover sets are selected for a tile set

TSi with a tile schedule STi

• Node schedule Si for node N is created with k segments

• If a node belongs to the kth one-cover set, the value of the kth segments is STi

• Otherwise, the kth segment has an all-zero value

• To cover M different tile sets in a single round, the final node schedule S:

Differentiated/Robust Surveillance

•Instead of turning on one set of nodes to cover a column/row, uScan can turn on multiple disjoint set of nodes to increase the degree of coverage

•Fault tolerance can be achieved by turning multiple sets on

•To fix the failure of nodes the schedule bits S of the nodes in the neighborhood of failed node is modified

Design Analysis

•uScan covers only a part of a network•Increases network lifetime•Introduces a certain delay in target

detection

Analytical Results•Detection Delay for Static Targets

▫The minimal detection delay happens when a target shows up in a tile right before this tile is turned on {1/R}

▫The maximum detection delay happens when a target shows up in a tile right after this tile is turned on▫ (1+N)/R for line scan▫ (1+N/2)/R for systolic scan

▫Detection delay for full coverage algorithms is zero

▫Detection delay can be reduced by dividing a network into sub-networks

Breached Area for Mobile Targets • Assumptions:

▫A target can only enter from outside of the network

▫The maximum speed of any target is r tiles per second

• Worst-Case Breach (WCB)=largest percentage of the area that a target can reach without being detected

For systolic scan WCBs (r, R)=(2R+r)r/(R + r)2

For line scan

• The whole area is breached if r>=(√2-1)R• In a full coverage scenario, the worst-case

breach area is zero

Comparison between Line and Systolic Scan • For a given switching rate R, systolic scan consumes

twice energy than line scan

• WCBl (r, 2R) ≥ WCBs(r, R) at all target speeds

• When the target speed is half of scanning speed (r=50),

systolic scan protects about half of the area, while line scan cannot protect any portion of the network

Experimental Setup

Experimental Results

•Grid placement provide full coverage until all of them run out of energy simultaneously •Random placement still keeps about 40% coverage when coverage reduces to zero in the grid placement

Detection Probability Over Time

Detection Delay for Static TargetsC

um

ula

tive

Den

sity

Fu

nct

ion

A larger target size leads to smaller delays

Comparison of Detection DelayC

um

ula

tive

Den

sity

Fu

nct

ion

Under the same switching rate, the detection delay of systolic scan is about one-half of line scan

Impact of the Network Size and Scan Direction

•As the network size reduces, the detection delay decreases accordingly

•To guarantee a certain detection delay, large area should be partitioned and perform scans within the sub-areas

Impact of the Switching Delay and Scan Direction

•Systolic scan has the smallest detection delay at all switching rates

•Line scan in the opposite direction of a target moving direction provides the second smallest detection delay

•The longest delay happens when we scan at the same direction as the target moving direction

•When the switching delay increases, the detection delay increases linearly

Simulation Results•Performance of uScan is evaluated on the

basis of Network Half-lifeThe time from the beginning of the deployment

until exactly half of the nodes are still alive

• Performance under Full Coverage Mode

The half lives for all cases increase linearly when the node density increases

Performance under Scanning Mode

The system half life of the uSense increases almost linearly as the node density increases

Advantage

• Allows many more nodes to activate in turn rather than the localized ones

• Leads to a significant energy savings

• Separation of sensing pattern from the underlying node scheduling

• Application only needs to specify the desired sensing behavior on the field

• When targets move oppositely to the direction of line scan guarantees 100% detection of mobile targets

• Burden on sensor nodes is comparatively less

Limitations• When a tile set does not form a continuous curve or a

node can cover multiple segments of a tile set simultaneously, the polynomial algorithm can not guarantee the complete coverage of active tiles

• Line scan misses the targets when the scan speed is below twice the target moving speed (when both move in the same direction)

• If the scanning direction is the same as the target moving direction, the detection probability drops to 45% at the long switching delay of 600ms

• Algorithm will not work for obstacles

• Does not support dynamic networks

Future Directions• The algorithm can be modified for non-

continuous coverage areas

• Considerations for obstacles in the coverage area can be incorporated

• Can be modified for dynamic network

Thank you