Communication Support for Location-Centric Collaborative

Signal Processing in Sensor Networks

Parmesh RamanathanUniversity of Wisconsin, Madison

Acknowledgements:K.-C. Wang, K. K. Saluja, T. Clouqueur

What is a sensor network? A large ad hoc network of low-cost,

smart devices Devices communicate over

wireless channels Devices can sense only a small

area around them Need collaboration among devices to

carry out most meaningful tasks

Sensor Network Characteristics Commands/queries are typically issued to

a geographic region and not to specific nodes Compute the average temperature in a region Are there any unidentified objects in a given

region? Track an object within a region

Only devices in the specified geographic region need to participate in executing a command/query

Ad Hoc Wireless LANs Nodes are the addressable entities Commands are issued to specific

nodes Typical challenge is to how to

maintain ongoing interactions between a given set of nodes even as they move

Open Question What is the best programming

abstraction and the underlying communication support suited for sensor networks?

Prior Work Programming abstraction

Subscribe-Publish model [USC/ISI/MIT/LL]

Communication support Directed Diffusion [USC/ISI]

Subscribe-Publish Model Nodes disseminate the attributes of the

information they need (Subscribe) Nodes also disseminate the attributes of

the information they can provide (Publish)

An interaction between nodes is established when there is match between their respective subscription and publication

Directed Diffusion Nodes diffuse interest messages identifying

the attributes of the information they need Nodes with the data respond over one or

more routes identified by the interest messages

Reinforcement messages are used to converge a good route

Intermediate nodes may use filters to aggregate information as it passes through the network

Our Programming AbstractionLocation-centric Computing

All nodes are aware of their current location Addressable entity is a geographic region Regions play the traditional role of a node A region must be created before a command

or query can be issued Each region has a manager region responsible

for coordinating intra-region activities Each node maintains the list of regions to

which it belongs and participates only in the activities of its regions

Location-centric Model

Location-centric Communication Primitives

Data exchange primitives Motivated by the well-known

distributed computing library called MPI 1.1

Send, Receive, Reduce, Barrier, Multicast, Broadcast,…

Administrative primitives Create region and delete region

Location-centric Communication PrimitivesExample: SN_Send Sends a message from a node to all

nodes in the addressed region Used to send commands and dataExample: SN_Reduce Aggregates data within a region at the

manager region Aggregation is in the form of min,

max, average, sum, …

Location-based Routing Each node maintains a routing table

identifying next hop to reach a destination region

Routing entry for a region is created on demand using RouteRequest (RREQ) and RouteReply (RREP)

RREQ and RREP use an approach similar to Location-aided Routing [Vaidya] to limit the scope of flooding

Location-Based Routing: Inter-Region

Send:1. message

sent from a source node to a region.

Location-Based Routing: Inter-Region

Send:1. message

sent from a source node to a region.

2. Message flooded to all nodes in the region.

Target Detection and Trackingt=0

t=20

t=60

t=40

81 sensors evenly spaced over 800mx800m square area

Target emits power sensible within 100m

Location-centric solution

Create region(s) at expected entry area(s)

Send detect & track command to the region(s)

Location-centric solutionManager

Nodes in the region use Reduce to aggregate sensor readings

Location-centric solution

Predict the track and create the next region Initiate detection & tracking in the created region

Subscribe/Publish Solution I

Each node subscribes to track info from neighboring nodes

Subscribe/Publish solution

Duplicate copies of track info from same node are suppressed.

Subscribe/Publish solution

Track info from different nodes are not suppressed.

Regional Subscribe/Publish Solution

subscribe

subscribesubscrib

e

subscrib

e

Each node subscribes to track info from neighboring regions.

Regional Solution with Subscribe/Publish Model

Track info propagates to subscribers using directed diffusion

Data suppression more effective.

Evaluation We implemented the three

approaches in ns-2. Counted the total number of

messages exchanged for the tracking scenario Routing messages Application payload

Initial Results

010002000300040005000600070008000

Message count

Loc-Cen SP-I SP-II

PayloadRouting

Summary Collaboration in sensor networks are quite

different from that in conventional wireless ad hoc networks

Compared performance of two different approaches for collaboration in sensor networks

Initial results show that a location-centric based approach is better in terms of number of messages for target tracking application

Ongoing Work Physical and link layer aware

communication schemes to improve energy and bandwidth usage

Incorporate fault-tolerance in sensor fusion algorithms

Develop collaborative signal processing algorithms for typical sensor network applications