BiSNET: A biologically-inspired middleware architecture for self-managing wireless sensor networks

Computer Networks 51 (2007) 4599–4616

www.elsevier.com/locate/comnet

BiSNET: A biologically-inspired middleware architecturefor self-managing wireless sensor networks

Pruet Boonma *, Junichi Suzuki

Department of Computer Science, University of Massachusetts, Boston, MA 02125, United States

Available online 23 June 2007

Abstract

This paper describes BiSNET (Biologically-inspired architecture for Sensor NETworks), a middleware architecture thataddresses several key issues in multi-modal wireless sensor networks (MWSNs) such as autonomy, scalability, adaptability,self-healing and simplicity. Based on the observation that various biological systems have developed mechanisms to over-come these issues, BiSNET follows certain biological principles such as decentralization, food gathering/storage and nat-ural selection to design MWSN applications. In BiSNET, each application consists of multiple software agents, whichoperate on the BiSNET middleware platform in individual sensor nodes, and each agent exploits certain biologically-inspired mechanisms such as energy exchange, pheromone emission, replication, migration and death. This is analogousto a bee colony (application) consisting of multiple bees (agents). This paper describes the biologically-inspired mecha-nisms in BiSNET, and evaluates their impacts on the autonomy, scalability, adaptability, self-healing and simplicity ofMWSNs. Simulation results show that BiSNET allows sensor nodes (agents and platforms) to be scalable with respectto network size, autonomously adapt their sleep periods for power efficiency and responsiveness of data collection, adap-tively aggregate data from different types of sensor nodes, and collectively self-heal (i.e., detect and eliminate) false positivesensor data. The BiSNET platform is implemented simple in its design and lightweight in its memory footprint.� 2007 Elsevier B.V. All rights reserved.

Keywords: Self-managing wireless sensor networks; Middleware support for Self-managing sensor network applications

1. Introduction

This paper describes a middleware architecture formulti-modal wireless sensor networks (MWSNs),1

1389-1286/$ - see front matter � 2007 Elsevier B.V. All rights reserved

doi:10.1016/j.comnet.2007.06.006

* Corresponding author.E-mail addresses: [email protected] (P. Boonma), jxs@

cs.umb.edu (J. Suzuki).1 An MWSN deploys multiple types of sensor nodes in an

observation area (e.g., temperature, humidity and carbon mon-oxide (CO) sensors). Data from different types of sensor nodesare aggregated, through in-network processing, to provide amulti-dimensional view of collected sensor data.

called BiSNET (Biologically-inspired architecturefor Sensor NETworks), which inherently addressesfive challenges in MWSNs. The first challenge isautonomy. Since sensor nodes can be deployed inan unattended area (e.g., forest and ocean) orphysically unreachable area (e.g., inside a buildingwall), they are required to operate with the mini-mum aid from base stations or human administrators[1,2].

The second challenge is scalability. In order tocover large spatial extents or monitor the extentsat a high-resolution, sensor networks are required

.

mailto:[email protected]

mailto:jxs@ cs.umb.edu

mailto:jxs@ cs.umb.edu

Fig. 1. BiSNET platform architecture.

4600 P. Boonma, J. Suzuki / Computer Networks 51 (2007) 4599–4616

to scale to a large number of sensor nodes2 and alarge amount of data generated by sensor nodes[1,3].

The third challenge is adaptability. Sensor nodesare required to adapt their operations to the envi-ronmental conditions that they monitor (e.g., tem-perature and carbon monoxide (CO)) [3–6]. Forexample, sensor nodes may increase their duty cycleintervals (sleep periods) when there is no significantchange in their sensor readings. This results in lesspower consumption in the nodes. Also, when neigh-boring nodes report environmental changes (e.g.,changes in temperature or CO level), a sensor nodemay draw inference from the reports and decreaseits sleep period to be more watchful for a potentiallocal environmental change in the future. This canincrease responsiveness of the node to transmit itssensor data to a base station. In addition, a sensornode may aggregate data from different types ofsensor nodes (e.g., temperature and CO data) andtransmit the aggregated data to a base station. Thiscan reduce power consumption in the nodes on apath toward the base station.

The fourth challenge is self-healing. Sensor read-ing usually contains some noise; it may be a falsepositive due to, for example, malfunction of sensors.Sensor nodes are required to self-heal (i.e., detectand eliminate) false positives in their sensor read-ings instead of transmitting them to base stations[5,7]. This can reduce power consumption of sensornodes because in-node data processing consumesmuch less power than data transmission does [8].

The fifth challenge is simplicity. Sensor controlsoftware (e.g., applications and middleware) needsto be simple in its design and small in its footprintbecause of limited availability of CPU power, mem-ory and battery.

In order to address the above five issues, BiSNETprovides a middleware platform, called the BiSNETplatform. The BiSNET platform hides low-leveloperating and networking details (e.g., network I/O and state control of sensor nodes) from applica-tions, and implements a series of mechanisms tosupport autonomous, scalable, adaptive and self-healing applications. BiSNET also provides ahigh-level programming abstraction to aid the sim-ple and rapid development of applications. Thedesign of BiSNET is motivated by the observation

2 For example, the DARPA Networked Embedded SystemsTechnology program envisions networks consisting of 100 to100000 simple computing nodes.

that various biological systems have already devel-oped mechanisms necessary to overcome those chal-lenges [9,10]. For example, bees act autonomously,influenced by local conditions and local interactionswith other bees. A bee colony can scale to a massivenumber of bees because all activities of the colonyare carried out without centralized control. A beecolony adapts to dynamic environmental condi-tions. When the amount of honey in a hive is low,many bees leave the hive to gather nectar from flow-ers. When the hive is full of honey, bees expand thehive. Also, bees recover (or self-heal) their phero-mone traces to flowers when a part of them is lost.The structure and behavior of each bee is verysimple; however, a group of bees autonomouslyexhibits desirable system characteristics such asadaptability and self-healing through collectivebehaviors and interactions among bees. Based onthis observation, the authors of the paper believethat, if MWSN applications are designed after cer-tain biological principles and mechanisms, theymay be able to meet the requirements in MWSNs(i.e., autonomy, scalability, adaptability, self-heal-ing and simplicity).

The BiSNET platform operates atop TinyOS ineach sensor node to host applications (Fig. 1). InBiSNET, each application consists of multipleagents,3 which follow several biological principlessuch as decentralization, autonomy, food gather-ing/storage and natural selection. This is analo-gous to a bee colony (application) consisting ofmultiple bees (agents) running on multiple plat-forms (hives). Each agent contains a set of data

3 Agents are software entities (software objects, components ormodules); they do not represent any physical entity.

4 The concept of energy in BiSNET does not represent theamount of physical battery in a sensor node. It is a logicalconcept that affects agent behavior.

P. Boonma, J. Suzuki / Computer Networks 51 (2007) 4599–4616 4601

and program code, which is interpreted by theBiSNET platform at runtime. Each agent readssensor data with the underlying sensor device,and discards or reports it to a base station usingbiological behaviors such as pheromone emission,replication and migration. The BiSNET platformconsists of a container and platform services(Fig. 1). A container provides an execution envi-ronment for agents, and controls the state of thelocal sensor node (e.g., sleep, listen and broadcast).Platform services are used by agents to read sensordata and perform their behaviors.

This paper describes the biologically-inspiredmechanisms in BiSNET and evaluates their impactson the autonomy, scalability, adaptability, self-heal-ing and simplicity of MWSN applications. Simula-tion results show that BiSNET allows sensornodes (agents and platforms) to autonomouslyadapt their sleep periods for power efficiency, drawinference on potential environmental changes fromsensing activities of neighboring nodes, adaptivelyaggregate data from different types of nodes, andcollectively self-heal (i.e., detect and eliminate) falsepositive sensor data. The BiSNET platform is light-weight thanks to a set of simple biologically-inspired mechanisms.

2. Contributions

This section summarizes the contributions of thiswork.

• Adaptive and decentralized duty cycle manage-

ment: BiSNET is the first attempt to investigatedynamic duty cycle management that adaptivelybalances the tradeoff between power efficiencyand sensing responsiveness for environmentalchanges (i.e., the risk to miss significant environ-mental changes during sleep period). The BiS-NET platform allows each sensor node toautonomously adjusts its sleep period in a decen-tralized manner.

• A simple and generic architectural design: BiS-NET applies a small number of simple biologicalconcepts to design the mechanisms that addresskey challenges in MWSNs (e.g., the mechanismsfor adaptive data transmission, data aggregation,self-healing, power efficiency and inference).Rather than implementing those mechanismsseparately, BiSNET provides a simple and gen-eric solution to implement the mechanismssimultaneously. The simplicity of the biologi-

cally-inspired mechanisms in BiSNET contrib-utes to the simplicity and lightweightness of theBiSNET platform.

3. Design principles for BiSNET agents

In BiSNET, agents are designed after the follow-ing biological principles.

1. Decentralization: Similar to biological systems(e.g., bee colonies), there are no centralized enti-ties in BiSNET to control and coordinate agents.Decentralization allows agents to be scalable andsimple by avoiding a single point of performancebottlenecks and failures [11,12] and by avoidingany central coordination in deploying agents [13].

2. Autonomy: Similar to biological entities (e.g.,bees), agents sense their local environments,and based on the sensed conditions, they auton-omously behave without any intervention from/to other agents, platforms, base stations andhuman administrators.

3. Food gathering and storage: Biological entitiesstrive to seek and consume food for living. Forexample, bees gather nectar from flowers anddigest it to produce honey. In BiSNET, agents(bees) read sensor data (nectar) in each dutycycle, and digest it to energy (honey).4 (Energygain is proportional to an absolute changebetween the current and previous sensor data.)They keep some of the energy and deposit therest in the local platform (hive).

4. Natural selection: The abundance or scarcity ofstored energy in agents affects their behaviorsand triggers natural selection. For example, anenergy abundance indicates a significant changein sensor reading; thus, an agent emits a phero-

mone to stimulate replicating itself and its neigh-boring agents. A replicated agent migratestoward a base station on a hop-by-hop basis toreport sensor data. An energy scarcity (an indica-tion of few changes in sensor reading) eventuallycauses the death of agents. As in biological natu-ral selection where more favorable species in anenvironment become more abundant, the popu-lation of agents dynamically changes based ontheir energy levels (i.e., changes in their sensorreadings).


4. BiSNET

This section describes a programming abstrac-tion for agents, the default agent the BiSNET plat-form provides, and the functions of the BiSNETplatform.

4.1. BiSNET agent

The BiSNET platform provides a high-level pro-gramming abstraction for application developers toimplement agents (i.e., MWSN applications) in aneasy-to-understand manner. In BiSNET, each agentconsists of attributes, body and behaviors. Attributes

carry descriptive information on an agent. Theyinclude agent type (e.g., temperature sensing orCO sensing agent), energy level, sensor data to bereported to a base station, time stamp of the sensordata, and ID/location of a sensor node where thesensor data is collected. Application developerscan define arbitrary attributes for their agents.

Body implements the functionalities of an agent:collecting and processing sensor data. In each dutycycle, each agent gathers sensor data (as food) fromthe underlying sensor device, converts it to energyand processes it (e.g., discards it or reports it to abase station). Depending on their agent types, differ-ent agents collect different types of sensor data.

Behaviors implement actions inherent to all agents.This paper focuses on the following five behaviors.

• Pheromone emission: Agents may emit phero-mones in response to the abundance of storedenergy (i.e., significant changes in their sensorreadings). Different types of agents emit differenttypes of pheromones, each of which carries sen-sor data. For example, temperature sensingagents emit temperature pheromones, whichcarry temperature data. CO sensing agents emitCO pheromones, which carry CO data. Phero-mones stimulate the agents on the local andneighboring nodes to replicate themselves.

• Replication: Agents may make a copy of them-selves in response to the abundance of energyand pheromones. When an agent performs repli-cation, it creates a new set of data and code (anew software agent). Each agent replicates itselfonly when enough types and concentration ofpheromones become available on the local node.Individual pheromones are not independentlytransmitted to base stations. Certain types ofhigh-concentration pheromones are grouped to

stimulate agent replication. For example, anagent may replicate itself only when sufficientconcentration of both temperature pheromonesand CO pheromones are available. A replicated(child) agent retains the same agent type as itsparent’s type, and aggregates multiple sensordata stored in multiple types of available phero-mones. A child agent is placed on the platformthat its parent agent resides on, and it receivesthe half amount of the parent’s energy level. Eachchild agent is intended to move toward a basestation to report (aggregated) sensor data.

• Migration: Agents may move from one sensornode to another in response to energy abundance(i.e., significant changes in their sensor readings).Migration is used to transmit agents (sensordata) to base stations on a multi-hop and short-est-path basis. When an agent perform migrationfrom one sensor node to another, the BiSNETplatform at a source node serializes the agent’sdata and code and transmits them to a destina-tion node. The BiSNET platform on a destina-tion node deserializes the transmitted data andcode to run the agent.

• Energy exchange: Agents on each platformalways share their energy units (honey) with eachother so that their energy levels become equal. Amigrating agent shares its energy units with otheragents on a destination platform. Also, agentsperiodically deposit some of their energy units(honey) to their local platforms (hives).

• Death: Agents die due to lack of energy when theycannot balance energy gain and expenditure. Thedeath behavior is intended to eliminate agents thatcarry false positive sensor data. When an agentdies, the underlying platform removes the agentand releases all resources allocated to the agent.

Every agent expends certain amount of energy toperform pheromone emission, replication and migra-tion behaviors. The energy costs to invoke the behav-iors are constant for all agents.

4.2. Default agent implementation

The BiSNET platform provides the default (ortemplate) agent so that application developers cancustomize it to rapidly develop their own agents.Fig. 2 shows the body (a sequence of actions) thatthe default agent performs in each duty cycle. In thissection, agent behaviors are visualized with the UML(Unified Modeling Language) sequence diagram.

Fig. 3. Agent death behavior.

Fig. 2. Body of the default agent.


First, an agent reads sensor data (nectar) with theunderlying sensor device, and converts it to energy(honey). The energy intake (EF) is calculated withEq. (1). S represents the absolute difference betweensensor data in the current and previous duty cycles.M is the metabolic rate, which is a constant between0 and 1.

EF ¼ S �M : ð1ÞDifferent platforms may have different M values

to prioritize particular types of sensor nodes. Allagents on a platform follow the same M value thatthe platform has. The higher M value a platformhas, the more often agents replicate and migrateon the platform because of higher energy intake.For example, if a MWSN is configured to be moresensitive to CO data than temperature data, themetabolic rate of CO sensor nodes should be greaterthan that of temperature sensor nodes.

Given EF, each agent updates its energy level asfollows:

EðtÞ ¼PN

i Eðt � 1ÞN

þ EF: ð2Þ

E(t) is the current energy level of the agent, andE(t � 1) is the agent’s energy level in the previousduty cycle. t is incremented by one at each duty cy-cle. Note that agents always exchange and sharetheir energy units equally with other agents in thesame platform.

If an agent’s energy level (E(t)) becomes very low(below the death threshold: TD), the agent dies dueto energy starvation (see also Figs. 2 and 3).5

5 If all agents are dying on a platform at the same time, arandomly selected agent will survive. At least one agent runs oneach platform.

Then, an agent emits a pheromone if its energylevel (E(t)) exceeds its pheromone emission thresh-old TP (see Fig. 4). Agents continuously adjust theirpheromone emission thresholds as the EWMA(exponentially weighted moving average) of theirenergy levels:

T PðtÞ ¼ ð1� aÞT Pðt � 1Þ þ aEðtÞ; ð3Þ

TP(t) is the current pheromone emission threshold,and TP(t � 1) is the one in the previous duty cycle.EWMA is used to smooth out short-term minoroscillations in the data series of E (energy level ofan agent). The a value is a constant to control thesensitivity of TP against the changes of E. A highervalue of a changes TP more sensitively against therecent changes in E. BiSNET uses a relatively lowa value (0.25) in order to place more emphasis onthe long-term transition trend of E. (Only significantchanges in E have the effect of changing TP.)

When a pheromone is emitted on a platform, allthe agents on the platform can sense it. It may stim-ulate their replications. Each pheromone has itsown concentration (or strength). It decays by halfat each duty cycle. A pheromone completely evapo-rates (disappears) when its concentration becomeszero.

An agent replicates itself when it meets two con-dition: 1. when the agent’s energy level (E(t))exceeds its replication threshold (TR) and 2. whenthe concentration of each type of available phero-mones (Pi

6) exceeds the pheromone’s stimulationthreshold T Si (see Fig. 5). The agent keeps replicat-ing itself until its energy level becomes less than itsTR. Agents continuously adjust their replicationthresholds as the EWMA of their energy levels(Eq. (4)). The stimulation threshold of a pheromonechanges as the EWMA of the pheromone’s concen-tration (Eq. (5)).

6 Pi denotes the total concentration of pheromone type i. i isused to indicate different types of pheromones available on thelocal platform (e.g., temperature and CO pheromones).

Fig. 6. Migration behavior sequence diagram.

Fig. 5. Agent replication behavior.

Fig. 4. Pheromone emission behavior.

7 All agents migrate from a platform whose energy gain isgreater than TB, except a randomly selected agent. If there is onlyone agent in a platform, the agent cannot migrate. At least oneagent remains on a platform.


T RðtÞ ¼ ð1� bÞT Rðt � 1Þ þ bEðtÞ; ð4ÞT SiðtÞ ¼ ð1� cÞT Siðt � 1Þ þ cP iðtÞ: ð5Þ

TR(t) is the current replication threshold, andTR(t � 1) is the one in the previous duty cycle. T Si

is the current pheromone stimulation threshold forthe pheromone type i, and T Siðt � 1Þ is the one inthe previous duty cycle. The b and c values are theconstants to control the sensitivity of TR and T Si

against the changes of E and Pi, respectively. Simi-lar to a, BiSNET uses a relatively low b and c values(0.25) in order to place more emphasis on the long-term transition trend of E and Pi. (Only significantchanges in E and Pi have the effects to change TR

and T Si , respectively.)A replicating (parent) agent splits its energy units

in to halves EðtÞ�ER

2

� �, gives a half to its child agent,

and keeps the other half. ER is the cost (energyunits) for an agent to invoke the replication behav-ior. A replicated (child) agent aggregates the sensordata in the pheromones that stimulated its parentagent to perform the replication behavior.

Each agent deposits a certain amount of energy(EP) to a platform that it resides on. Each agentstrives to keep its energy level (E(t)) close to theone in the previous duty cycle (E(t � 1)).

EP ¼EðtÞ � Eðt � 1Þ if EðtÞP Eðt � 1Þ0 if EðtÞ < Eðt � 1Þ

�ð6Þ

When a platform’s total energy gain ðP

EPÞ isgreater than a threshold (TB), the platform changesits state to the broadcast state. This allows agentsand pheromones to move to neighboring platforms(see Fig. 6).7 In order for agents to determine whichneighboring platforms they move to, each base sta-tion periodically propagates base station phero-mones. (This is a different type of pheromonesfrom those emitted by agents.) The concentrationof base station pheromones decays on a hop-by-hop basis. Using base station pheromones, agentscan sense where base stations exist approximately,and move toward the base stations by climbingpheromone gradients.

As described above, agents replicate themselvesonly when they gain a large amount of energy onthe local node and receive enough types of high-con-centration pheromones from neighboring nodes.This means that sensor data are aggregated andtransmitted to base stations only when significantchanges in sensor data are detected on the localand neighboring nodes. Agents do not respond togradual changes in sensor readings (e.g., tempera-ture change during a day or between different sea-sons). This reduces power consumption in sensornodes and extends the life of a sensor network.

This adaptive data aggregation and transmissionmechanism is designed with a self-healing capabilityin mind. When a sensor node does not work prop-erly due to, for example, malfunctions, each agenton the node may emit the pheromones that contain


false positive sensor data. A large number of falsepositive pheromones may be transmitted to a neigh-boring node. However, they are discarded at theneighboring node because they are not aggregatedwith other types of pheromones. This means thatfalse positive pheromones are not propagated morethan two hops from a malfunctioning node. Also,agents stop emitting false positive pheromones onthe malfunctioning node because their pheromoneemission thresholds increase (Eq. (3)).

4.3. BiSNET platform

Each platform consists of platform services and acontainer (Fig. 1). Platform services hide lower-levelcomputing and networking details, and providehigh-level runtime services for agents to read sensordata and perform their behaviors. Example plat-form services are described below.

• Agent behavior services implement agent behav-iors. Each of the agent behaviors described inSection 4.1 is implemented as a platform service.When agents perform their behaviors, theyinvoke corresponding platform services (Figs.3–6). For example, when an agent replicatesitself, it invokes the replication service that thelocal platform provides (Fig. 5).

• Network I/O service performs low-level wirelessnetwork communication such as data transmis-sion and medium access control.

• Localization service implements a range-freearea-based localization mechanism. This mecha-nism approximates the physical location of thelocal node by using base stations as referencepoints.8 Upon receiving a base station phero-mone from a base station, the localization servicemeasures and records the topological distance(hop count) to the base station. Then, the serviceperiodically transmits the distance information tothe nearest base station. Base stations calculatethe location of each node by performing multi-lateration with their locations.

• Neighboring node maintenance service keeps trackof neighboring nodes. The service maintains atable on neighboring nodes, and updates thetable when it receives agents, pheromones andbase station pheromones from neighboring

8 This paper assumes that each base station keeps track of itslocation with a GPS device and other nodes do not know theirlocations.

nodes. Agents use this service to emit phero-mones to neighboring nodes and migrate towardbase stations. It is also used by the localizationservice.

• Pheromone management unit maintains informa-tion about pheromone concentration at the localnode and also in neighboring platforms.

A container maintains a reference table to theagents running on the local platform. Agents usethe table to inspect which agents and how manyagents are running on the local platform. Anotherresponsibility of each container is to dynamicallychange the state of the underlying sensor node forcontrolling its duty cycle. Each sensor node can bein the listen, broadcast or sleep state (Fig. 7). A plat-form and agents can work on a sensor node when itsstate is in the listen or broadcast state. In eitherstate, each agent performs the series of actionsdescribed in Fig. 2.

In the listen state, a platform turns on a radioreceiver to receive data (agents and pheromones)from neighboring sensor nodes. The listen statechanges to the broadcast state if a platform gainsenergy more than the broadcast threshold (

PEP >

T B; see also Figs. 6 and 7). In the broadcast state,a platform turns on a radio transmitter to allowagents and pheromones to move to neighboringnodes.

When a platform gains no energy from agentsðP

EP ¼ 0Þ, the platform goes into the sleep state(Fig. 7). The sleep period is determined as follows.Psleep is a constant, and Pi is the concentration ofeach type of pheromones (the pheromone type i)available on the platform.

sleep period ¼P sleepP

P iifP

P i > 0;

P sleep ifP

P i ¼ 0:

(ð7Þ

The sleep period is inversely proportional to thetotal concentration of pheromones available on aplatform

PP ið Þ. This means that a platform

Fig. 7. Platform state transition.


increases its sleep period to reduce power consump-tion when agents find no significant changes in theirsensor readings on the platform and its neighboringplatforms.

This adaptive duty cycle management mechanismis designed with an inference capability in mind.When a platform receives pheromones from aneighboring node(s) via the Pheromone Manage-ment Unit, it decreases its sleep period even if thereis no change in the sensor reading on the local node(see Eq. (7)). This way, agents can be more watchfulon the node for a future potential change in theirsensor readings so that they do not miss it duringa sleep period.

5. Simulation results

This section shows a series of simulation resultsto evaluate BiSNET in terms of adaptability, scala-bility, self-healing, inference, power efficiency9 andsimplicity. BiSNET is implemented on TinyOSand evaluated in the PowerTOSSIM simulator [14].

5.1. Simulation configurations for a wildfire detection

application

This simulation study emulates a sensor networkdeployed in a forest to detect wildfires. The simu-lated network consists of temperature sensors andCO sensors randomly deployed in a N · N gridtopology (see Fig. 8). Two different size of networksare examined, 7 · 7 (49 nodes) and 25 · 25 (625nodes). Half the sensor nodes equip temperaturesensors, and the other half equip CO sensors. Awildfire moves from southeast to northwest, towardthe base station located at the northwest corner ofthe sensor network. Simulations run with a firespread model that describes wildfire spreading innature [15].

This simulation study evaluates two different setof simulation results: micro evaluation and macro

evaluation. The micro evaluation focuses on twosensor nodes in the network (nodes 21 and 6; seeFig. 8), and evaluates how BiSNET works acrossthe two nodes. Node 21 detects a temperaturechange first, and then node 6 detects a CO levelchange next (Fig. 8). At node 21, temperaturechanges from 80�F to 240�F, and goes back to 80�.

9 Voltage supply in sensor nodes is assumed to be constant;therefore, current in milliamperes is used as an indication ofpower.

At node 6, CO level changes from 80 to 240 ppm(parts per million),10 and goes back to 80 ppm.

The macro evaluation evaluates how BiSNETimpacts the performance of the whole sensor net-work, such as the scalability against network size,success rate of sensor data transmission and net-work life.

5.2. Micro evaluation

This section presents micro evaluation results.Every micro evaluation is conducted with a 7 · 7sensor network.

5.2.1. Adaptive data transmission

Fig. 9 shows the concentration of pheromonesemitted by agents on node 21 as well as the phero-mone emission threshold on node 21. The phero-mone concentration increases when temperaturespikes and drops, because agents emit more phero-mones in response to higher energy intake. As theenergy levels of agents grow, their pheromone emis-sion thresholds increase as well (see Eq. (3)). Thisprevents agents from emitting pheromones. AsFig. 9 shows, agents stop emitting pheromoneswhen their pheromone emission threshold spikes(around 100th, and 160th min), and pheromoneconcentration drops. Note that agents do not emitpheromones at all when there is no temperaturechange. Agents adapt their pheromone emissions(i.e., sensor data transmission) to dynamic changesin their sensor readings.

5.2.2. Inference

Fig. 10 shows the pheromone concentrations onnode 6 and the number of replicating and migratingagents on node 6. Temperature pheromones arrivenode 6 from node 21 before CO level increases atnode 6. This allows the platform on node 6 to drawinference from the temperature pheromones andreduce the node’s sleep period (see Eq. (7)); there-fore, the agents on node 6 can start collecting moreCO data before CO level increases, and the agentscan start emitting CO pheromones immediatelyonce the CO level increases. This inference mecha-nism allows agents to be more responsive to envi-ronmental changes so that they can quicklyreplicate themselves and replicated agents can reach

10 240 ppm is in the range of the FAA CO minimum perfor-mance standard for smoke detectors (200 ± 50 ppm).

Fig. 8. Simulation configurations.

Fig. 9. Pheromone concentration and pheromone emissionthreshold at node 21.


base stations with a shorter delay. In Fig. 10, theresponsiveness (the time lag between an environ-mental change and pheromone emission) on node6 is two times shorter than that on node 21.

Fig. 10. Pheromone concentration and the number of replicat-ing/migrating agents.

Fig. 10 also shows that the agents on node 6 per-form replications only if the concentrations of bothtemperature and CO pheromones are high enough.As described in Section 4.2, agent replication (sen-sor data aggregation) is performed only whenenough types of pheromones exhibit high-concentrations.

Fig. 11 depicts how sleep periods dynamicallychange on nodes 21 and 6. On both nodes, plat-forms decreases the node’s sleep periods whenagents detect environmental changes, and increasesit when agents detect no environmental changes.Platforms adapt their underlying nodes’ sleep peri-ods to environmental changes. Another findingfrom Fig. 11 is that the sleep period of node 6decreases before CO level increases. This is becausethe platform on node 6 receives temperature phero-mones from node 21 and the total pheromone con-centration increases on node 6, which in turndecreases the sleep period of node 6. As describedabove, this inference mechanism increases theresponsiveness of agents to environmental changes.

Fig. 11. Sleep period.

Fig. 12. Power consumption.


Although the inference mechanism allows agentsto collect more sensor data to be watchful for poten-tial environmental changes, the they consume morepower on sensor nodes. Table 2 summarizes thistradeoff. It shows the number of collected sensordata and power consumption at the node that per-forms inference (i.e., node 6) and the node that doesnot perform it (i.e., node 21). The data collectionand power consumption are measured betweenwhen temperature/CO level spikes from 80� to240�/ppm and when it drops back to 80�/ppm (for250 min approximately). As shown in Table 2, bydrawing inference from the pheromones emittedfrom node 21, node 6 collects 16.52% more datawith only 3.55% more power consumption. Theauthors of the paper believe that extra power con-sumption is small enough to perform inferenceand BiSNET balances the tradeoff between sensingresponsiveness and power consumption.

5.2.3. Power efficiency through adaptive duty cycle

management

Fig. 12 shows the power consumption of nodes21 and 6. In the beginning of a simulation, the sen-sor nodes consume power to discover neighboringsensor nodes and set up network topology. Afterthat, they minimize power consumption by increas-ing their sleep periods because there is no significant

Table 1Simulation parameters

Parameters Values

The number of sensor nodes 49 or 626Metabolic rate (M) 1EWMA coefficients (a, b and c) 0.25Agent death threshold (TD) 1Platform broadcast threshold (TB) 25Power consumption in the listen state 10 mAPower consumption in the broadcast state 25 mAPower consumption in the sleep state 5 mAMaximum sleep period 5 minMinimum sleep period 1 min

Table 2Data collection and power consumption with and withoutinference

Number ofcollected data

Power consumption

Without inference(node 21)

230 3795 mA

With inference (node 6) 268 3930 mARate of increase (%) 16.52 3.55

change in their sensor readings. When the tempera-ture/CO level spikes, the power consumption of thesensor nodes spikes too because they immediatelydecrease their sleep periods (see also Fig. 11). Asshown in Fig. 12, the adaptive duty cycle mecha-nism in BiSNET allows sensor nodes (agents/platforms) to effectively reduce their power con-sumption when there is no significant environmentalchange.

Currently, platforms dynamically adjust theirsleep periods between 1 min and 5 min (seeFig. 11). Table 3 compares the number of collecteddata and the power consumption of sensor nodes(agents/platforms) with that of the configurationsin which sleep period is fixed at 1 min or 5 min.The data collection and power consumption aremeasured between when temperature/CO levelspikes from 80 to 240�/ppm and when it drops backto 80�/ppm (for 250 min approximately). Comparedwith the 5 min (fixed) duty cycle management, BiS-NET consumes only 5% more power while collect-ing 16.52% more data. BiSNET sacrifices the 5%power consumption to improve the sensing respon-siveness against environmental changes. A fixedduty cycle management scheme cannot responsivelysense and report environmental changes as BiSNETdoes. Compared with the 1 min (fixed) duty cyclemanagement, BiSNET consumes only 62% of the

Table 3Data collection and power consumption in different configura-tions of duty cycle management

Sleep period(in min)


Power consumption(mA)

1–5 (variable;BiSNET)

268 3930

5 (fixed) 230 37401 (fixed) 268 6340

Fig. 13. Intra-node self-healing of false positive data.


power used by the fixed scheme while collecting thesame amount of data. In particular, the power con-sumption in listen state is much lower than inbroadcast state (see Table 1), so even if the sleepperiod of a sensor node is shorten (i.e., sensor nodespends more time in listen state), it does not affectthe overall power consumption of a sensor nodemuch. Moreover, from Fig. 11, the interval whenthe sleep period of node 6 (with inference, dynami-cally adjusted sleep period between 1 min and5 min) does not overlap with that of node 21 (with-out inference, the sleep period is fixed at 5 min) isrelatively small compared with with the whole sim-ulation time; consequently, this contributes to asmall increase in power consumption. A sensor nodein listen state consumes much less batter power thanin broadcast state; however, it is two times largerthan that of sleep state. Therefore, a sensor nodewith 1 min fixed duty cycle has to waste batterypower when there is no event (e.g., from 0 min to55 min). BiSNET effectively reduces power con-sumption by decreasing duty cycle only whennecessary.

Fig. 14. Inter-node self-healing of false positive data.

5.2.4. Power efficiency through data aggregation

In addition to adaptive duty cycle management,pheromone-based data aggregation contributes toreduce power consumption of sensor nodes. Table4 compares the power consumption of node 6 inthe two configurations that agents perform dataaggregation and do not. When agents do not per-form data aggregation, agents use only energy levelto decide whether they replicate themselves. (Theydo not use pheromones.) Table 4 shows that node6 consumes 4.9% less power when agents performdata aggregation.

5.2.5. Self-healing of false positive data

Figs. 13 and 14 demonstrate how each sensornode self-heals (i.e., detects and eliminates) falsepositive data when it or a neighboring node mal-functions. BiSNET provides two self-healing capa-bilities: intra-node and inter-node self-healing.Fig. 13 shows a result of intra-node self-healing.

Table 4Power consumption with and without data aggregation

Power consumption

Without data aggregation 4123 mAWith data aggregation 3930 mAReduction rate 4.9%

In this case, node 21 is configured to malfunctionand generate temperature data of 0 and 200� repeat-edly. When node 21 starts malfunctioning, agentsemit a large number of temperature pheromonesvery often because sensor data widely swingsbetween 0� and 200�. (The energy intake of agentsis very high on node 21.) However, the pheromoneemission thresholds of agents rapidly grow as theagents’ energy levels increase (see also Eq. (3));within 2 min, agents start suppressing their phero-mone emission. In 5 min, the concentration of pher-omones dramatically drops, and no pheromones areemitted after 5 min. Accordingly, the pheromonesare not transmitted to neighboring nodes in 6 mineven if node 21 keeps malfunctioning.

Fig. 14 shows a result of inter-node self-healing.In this case, node 21 works properly; however, node6 malfunctions. Node 6 periodically propagates alarge number of CO pheromones and transmitsthem to node 21. (Node 6 does not perform intra-node self-healing in this simulation scenario.) Sincenode 21 does not detect temperature changes, theagents on node 21 do not emit any temperature

Table 6Power consumption and the number of reported data withdifferent a values

a Power consumption (mA) Number of reported data

0.1 4105 5480.25 3614.08 5480.5 3213 5320.75 2880 518

Table 7Power consumption and the number of reported data withdifferent b values

b Power consumption (mA) Number of reported data

0.1 3912.3 5480.25 3614.08 5480.5 3113.5 4880.75 2530.3 442


pheromones. As a result, the agents do not replicatethemselves at all. Thus, even if node 21 keepsaccepting CO pheromones from node 6, the agentson node 21 totally ignore those pheromones. Usingintra-node and inter-node self-healing, BiSNETallows sensor nodes (agents) to autonomously self-heal, i.e., detect and eliminate, false positive data(pheromones) and avoid wasting power.

5.2.6. Simplicity: memory footprint

In order to evaluate the simplicity of BiSNET,Table 5 shows the memory footprint of the BiSNETplatform, a MICA2 mote. It also shows the foot-print of Blink (an example program in TinyOS),which periodically turns on and off an LED, and amobile agent platform for sensor networks calledAgilla [16]. As shown in Table 5, the BiSNET plat-form is fairly lightweight in its footprint, and it canbe deployed on sensor devices whose resource avail-ability is severely limited.

5.3. Macro-level evaluation

This section presents macro evaluation results.Macro evaluation is conducted with a 7 · 7 and25 · 25 sensor networks.

5.3.1. Impacts of simulation parameters on BiSNETperformance

As Table 1 shows, this simulation study uses 0.25for the value of a, b and c in Eqs. (3)–(5). Theseparameters control the sensitivity for agents to emitpheromones, replicate themselves and migrate. Thissection discusses how different a, b and c valuesimpact the total power consumption of a sensor net-work (i.e., power consumption by all the sensornodes in a network) and the number of datareported to the bases station throughout a simula-tion. This evaluation is conducted with a 7 · 7 sen-sor network.

Table 6 shows the power consumption of a sen-sor network and the number of reported data whena varies from 0.1 to 0.75. (b and c are 0.25). When aincreases, the number of reported data decreases

Table 5Memory footprint in a MICA2 mote

ROM (KB) RAM (KB)

BiSNET 0.7 18Blink 0.04 1.6Agilla 3.59 41.6

because the pheromone emission threshold (TP)becomes more sensitive to the changes in energylevel; it is less likely that agents produce phero-mones (see Eq. (3) and Fig. 4). As a result, agentsreplicate themselves less often, and less data isreported to the base station. On the other hand,when a is too small, agents produce pheromonestoo often and consume more power.

Table 7 shows the power consumption of a sen-sor network and the number of reported data whenb varies from 0.1 to 0.75. (a and c are 0.25). When bincreases, the number of reported data decreasesbecause the replication threshold (TR) becomesmore sensitive to the changes in energy level; it isless likely that agents replicate themselves (see Eq.(4) and Fig. 5). As a result, agents report less datato the base station. On the other hand, when thevalue of b is too small, agents replicate themselvestoo often, transmit more data and consume morepower.

Table 8 shows the power consumption of a sen-sor network and the number of reported data whenc varies from 0.1 to 0.75. (a and b are 0.25). When cincreases, the number of reported data decreases

Table 8Power consumption and the number of reported data withdifferent c values

c Power consumption (mA) Number of collected data

0.1 3892 5480.25 3614.08 5480.5 3105.3 5050.75 2588.5 482

Table 10Average power consumption of each sensor node

Average powerconsumption (mA)

Standarddeviation

Without pheromones 5371.84 4318.62With pheromones 3614.08 1995.354Reduction rate 32.72%


because the pheromone stimulation threshold ðT SiÞbecomes more sensitive to the changes in phero-mone concentration; it is less likely that agents rep-licate themselves (see Eq. (5) and Fig. 5). As a result,agents report less data to the base station. On theother hand, when c is too small, agents replicatethemselves too often, transmit more data and con-sume more power.

Currently, BiSNET uses a relatively low-value(0.25) for a, b and c in order for agents to placemore emphasis on the long-term transition ofenergy level and pheromone concentration whilemaintaining the sensitivity to them.

5.3.2. Data aggregation

Table 9 shows the total number of sensor datathat all sensor nodes collect and transmit to the basestation throughout a simulation in two differentconfigurations with pheromones enabled and dis-abled. With pheromones enabled, sensor nodes col-lect and transmit more sensor data to the basestation by taking advantage of the inference mecha-nism in BiSNET. The success rate of data transmis-sion from sensor nodes to the base station decreases1.33% with pheromones enabled. When a phero-mone is emitted, it is propagated to neighboringnodes. This propagation can block agents frommigrating toward the base station. (There is no datapropagation with pheromones disabled.) However,the authors of the paper believe this reduction rateis acceptable enough.

Table 10 shows the average power consumptionof each sensor node in two different configurationswith pheromones enabled and disabled. With pher-omones enabled, agents reduce the number of sen-sor data transmitted to the base station byaggregating sensor data. This contributes to37.72% reduction of power consumption on eachnode, compared with the configuration with phero-mones disabled. Moreover, the standard deviationof power consumption among nodes is smaller whenusing pheromones.

Table 9The number of collected and reported sensor data with andwithout pheromones


Number ofreported data

Successrate

Without pheromones 494 490 99.19%With pheromones 560 548 97.86%Change rate 13.36% 11.84% 1.33%

Figs. 15 and 16 show how much power individualsensor nodes consume. (Each intersection of lines ina figure represents a sensor node.) As shown inFig. 16, with pheromones disabled, some nodes(particularly, the nodes close from the base station)consumes much more power than others. Thisincreases a risk that a network separates into islandsand some nodes cannot communicate with the basestation. In contrast, with pheromones enabled,every node consumes a lower and similar amountof power.

Fig. 17 shows the distribution of the number ofsensor nodes against power consumption through-out a simulation. This figure suggests that thenumber sensor nodes is normally distributedagainst power consumption. As Table 10 showsas well, power consumption is better distributedover sensor nodes when using pheromones. Thisavoids network separations and contributes toextend network life.

5.3.3. Network lifetime

Fig. 18 shows how soon sensor nodes go downdue to lack of power with pheromones enabledand disabled. The same battery capacity is assignedto both cases (750 mA h). With pheromones dis-abled, 15 of 49 nodes go down in the first

Fig. 15. Power consumption of sensor nodes with pheromonesenabled.

Fig. 16. Power consumption of sensor nodes with pheromones disabled.

Fig. 17. Distribution of the number of nodes against powerconsumption.

Table 11The total number of collected and reported sensor data in a25 · 25 network


Number ofreported data

Successrate

Without pheromones 670 665 99.25%With pheromones 774 762 98.45%Change rate 15.52% 14.59% 0.81%

Table 12Average power consumption of each sensor node in a 25 · 25network

Average powerconsumption (mA)

Standarddeviation

Without pheromones 5542.4 4148.0With pheromones 4024.104 1674.04Reduction rate 27.39%


150 min, and all nodes go down in 238 min. Withpheromones enabled, 26 nodes go down in250 min, and all nodes go down in 379 min. Usingpheromones contributes to extend the networklifetime.

Fig. 18. Histogram of sensor node lifetime.

5.3.4. Scalability

In order to evaluate the scalability of BiSNETagainst the number of nodes, a set of simulationswas carried out with a larger sensor network con-sisting of 600 nodes. Table 11 shows the total num-ber of sensor data that all sensor nodes collect andtransmit to the base station throughout a simula-tion. Table 12 shows the average power consump-tion of sensor nodes throughout a simulation.These results are very similar to the results obtainedfrom smaller-scale simulations with 49 nodes (seeTables 9 and 10). Simulation results show that BiS-NET is scalable against the increase of network sizeand data volume generated by nodes.

6. Related work

This paper describes the research findings extend-ing several previous works [17,18]. In [17], BiSNET


did not support MWSNs; it supported only one typeof sensor nodes throughout a network. In thispaper, BiSNET is extended to support MWSNs byintroducing the concept of pheromones. Phero-mones are used, in the agent replication behavior,for each agent to aggregate different types of sensordata and self-heal false positive data. Pheromonesare also used for each platform to perform adaptiveduty cycle management for power efficiency andinference on potential environmental changes forsensing responsiveness. In [18], BiSNET was exam-ined through micro evaluation in a smaller networkof 30 nodes. This paper includes additional microevaluation results as well as macro evaluationresults in larger networks of 49 and 625 nodes.

There are several research efforts applying bio-logical mechanisms to sensor networks. For exam-ple, in order to synchronize clocks of sensor nodesin a decentralized manner, [19] applies firefly’s phasesynchronization mechanism in which fireflies syn-chronize their light on/off periods with each other.BiSNET focuses on different issues; it applies bio-logical mechanisms to adaptive duty cycle manage-ment, inference on potential environmental changes,data aggregation and self-healing of false positivedata.

Britton et al. [20] proposes to apply biologicalmechanisms to an operating system for sensor net-works, called kOS, in order to make them robustto topological changes, scalable and self-organizing.However, kOS has not implemented any specificbiological mechanisms yet. In contrast, BiSNETspecifically implements biological mechanisms suchas energy exchange, pheromone emission, replica-tion, migration and death to improve the ability ofsensor nodes for power efficiency, inference andself-healing.

Szumel and Owens [21] proposes a generic com-munication primitive for sensor networks. Theprimitive hides lower-level implementation, i.e., net-work communication, while maintain the ability ofprogrammer to control over the communicationbehavior of sensor node. The authors use a biolog-ical communication mechanism, called pheromone,as the communication primitive. A set of phero-mone properties, i.e., type, strength, source, andpayload, and instructions, i.e., deposit and smell isprovided to the application developer to be usedas the communication mechanism between sensornode. Different from BiSNET which is designedbased on biological systems from the bottom up,the pheromone concept in this work is not properly

integrated into the other part of sensor softwaredevelopment. Hence, application developers haveto deal with two different levels of concept, a high-level concept of pheromone, and low-level conceptof sensor node programming. In addition, the pher-omone based communication primitive in this workdoes not provide any direct benefit; applicationdevelopers have to carefully design their own appli-cation to gain benefits from the pheromone mecha-nism. In BiSNET, the data aggregation, inference,and adaptive sleep period are direct benefits fromusing pheromone, application developers do notneed to concern themselves with those issues whenusing BiSNET.

Agilla proposes a programming language toimplement mobile agents for sensor networks, andprovides a runtime system (interpreter) to operateagents on TinyOS [16]. BiSNET does not focus oninvestigating a new programming language for sen-sor networks. BiSNET agents and Agilla agentshave a similar set of behaviors such as migrationand replication. Both of them are also intended tobe used for similar applications (e.g., wildfire detec-tion). However, Agilla does not address the researchissues that BiSNET focuses on; power efficiency,data aggregation, inference and self-healing. Inaddition, BiSNET focuses on its design simplicityand runtime lightweightness. As shown in Table 5,BiSNET is much more lightweight than Agilla.

Hsin and Liu [22], Misic and Misic [23], Ye et al.[24] describe dynamic duty cycle management in sen-sor nodes. Their goal is to improve power efficiency,and they do not consider sensing responsiveness topotential environmental changes (i.e., the risk tomiss significant environmental changes during sleepperiod). Unlike them, the duty cycle managementscheme in BiSNET is designed to adaptively balancethe tradeoff between power efficiency and sensingresponsiveness for potential environmental changes.As a result, BiSNET uses sensor data (i.e., phero-mones) to determine the sleep period of each sensornode, while [22–24] randomly change sleep period oruse other metrics such as the average time for a sen-sor node to process packets.

BiSNET is similar to [25] in that it proposes amechanism that allows users to manually specifythe sleep period of sensor nodes in order to balancethe tradeoff between power efficiency and sensitivityof event detection. Unlike this mechanism, BiSNETallows sensor nodes to autonomously adjust theirsleep periods without any aid from human users.Moreover, in BiSNET, different sensor nodes can


have different sleep periods, depending on their sen-sor readings. This balances the tradeoff betweenpower efficiency and sensitivity of event detectionmore effectively.

Quasar proposes a data collection protocol thatbalances the tradeoff between data accuracy andpower efficiency [26]. In Quasar, each sensor nodeswitches its state between active and idle (sleep) tominimize its power consumption. A central servercontrols the periods of active and idle states basedon the changes in sensor readings. Unlike Quasar,BiSNET does not require a central server; individualsensor nodes locally adjust their duty cycle intervals.In addition, BiSNET implements two ways to triggerdynamic duty cycle adjustment: based on changes insensor reading on the local node and via inferencefrom sensing activities of neighboring nodes. Quasardoes not implement the inference function.

SASHA proposes a self-healing mechanism byapplying immunological mechanisms for base sta-tions to identify faulty sensor nodes [27]. A base sta-tion detects fault nodes by comparing data frommultiple sensor nodes. In BiSNET, individual sen-sor nodes self-heal false positive sensor data in adecentralized manner. Since false positive data arenot transmitted to base stations, BiSNET consumesless power for self-healing than SASHA. In fact,BiSNET does not incur any extra computing andcommunication overhead for self-healing. Self-heal-ing is achieved as a result of agent making decisionson whether they replicate themselves based on theconcentration of pheromones.

Huang et al. [28] proposes a middleware platformfor MWSNs. It is implemented with a scripting lan-guage (Python) to improve the ease of developingapplications. In contrast, BiSNET is implementedwith NesC, instead of a scripting language, so asnot to sacrifice its performance and runtime light-weightness. Huang et al. [28] allows each sensor nodeto aggregate different types of sensor data. Applica-tion programmers are required to explicitly specify(or hard-code) the condition to aggregate sensordata at each node (e.g., temperature > 200 and COlevel > 200). In BiSNET, application programmersdo not have to specify data aggregation conditionfor each node. Rather than using hard-coded dataaggregation conditions, each node aggregates sensordata generated by the node and its neighboringnodes when the sensor data change significantly(see Fig. 2 and Eqs. (4) and (5)). In addition, [28]does not consider the issues that BiSNET focuseson, such as adaptive duty cycle management for

power efficiency, inference on potential environmen-tal changes for sensing responsiveness and self-heal-ing of false positive sensor data.

Brooks et al. [29] proposes to divide the function-ality of a wireless sensor network into three parts:communication, collaborative sensing and opera-tional commands. Each part has a Petri Net to con-trol the behavior of each sensor node. Brooks et al.[29] is similar to BiSNET in that it can detect andeliminate false positive data as a function of collab-orative sensing. However, BiSNET is designedmuch simpler; it has only a few states and transi-tions among them, while [29] has 133 states and232 transitions among them. Brooks et al. [29] doesnot consider design simplicity and runtime light-weightness. In addition, BiSNET operates in adecentralized manner, while [29] organizes a sensornetwork in a hierarchical manner.

7. Concluding remarks

This paper describes a biologically-inspired sen-sor networking architecture, called BiSNET, whichaddresses several key issues in MWSNs such asautonomy, scalability, adaptability, self-healingand simplicity. Inspired from biological systems,BiSNET provides a set of simple yet generic solu-tions that address those issues simultaneously ratherthan focusing on them one by one or in an ad-hocmanner. This paper describes the biologically-inspired mechanisms in BiSNET and evaluates theirimpacts on these issues in MWSNs. Simulationresults show that BiSNET allows sensor nodes(agents and platforms) to scale to network size,autonomously adapt their sleep periods for powerefficiency, draw inference on potential environmen-tal changes from sensing activities of neighboringnodes, collectively self-heal (i.e., detect and elimi-nate) false positive sensor data, and aggregate datafrom different types of nodes. The BiSNET platformis implemented so as to be simple in its design andlightweight in its memory footprint.

Several extensions to BiSNET are planned. Cur-rently, BiSNET assumes a traditional decentralizedrouting mechanism to transmit sensor data (i.e.,agents) toward base stations via shortest paths. Abiologically-inspired routing mechanism will beinvestigated to effectively direct agents to base sta-tions. In addition, BiSNET currently detects andeliminates false positive data. It is planned to inves-tigate an additional mechanism to detect false nega-tive data as well.


References

[1] M. Yu, H. Mokhtar, M. Merabti, A survey of networkmanagement architecture in wireless sensor network, in:Proceedings of the Sixth Annual PostGraduate Symposiumon The Convergence of Telecommunications, Networkingand Broadcasting, 2006.

[2] J. Blumenthal, M. Handy, F. Golatowski, M. Haase, D.Timmermann, Wireless sensor networks – new challenges insoftware engineering, in: Proceedings of the EmergingTechnologies and Factory Automation, 2003.

[3] S. Hadim, N. Mohamed, Middleware challenges andapproaches for wireless sensor networks, IEEE DistributedSystems Online 7 (3) (2006) 853–865.

[4] I. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci,Wireless sensor networks: A survey, Elsevier Journal ofComputer Networks 38 (2002) 393–422.

[5] J. Ledlie, J. Schneidman, M. Welsh, M. Roussopoulos, M.Seltzer, Open problems in data collection networks, in:Proceedings of European SIGOPS, 2004.

[6] K. Romer, O. Kasten, F. Mattern, Middleware challengesfor wireless sensor networks, ACM SIGMOBILE MobileComputing and Communications Review 6 (4) (2002) 59–61.

[7] P. Rentala, R. Musunuri, S. Gandham, U. Sexena, Surveyon sensor networks, in: Proceedings of International Con-ference on Mobile Computing and Networking, 2001.

[8] G. Mathur, P. Desnoyers, D. Genesan, P. Shenoy, Ultra-lowpower data storage for sensor networks, in: Proceedings ofIEEE/ACM Conference on Information Processing in Sen-sor Networks, 2006.

[9] D. Gordon, Ants At Work: How an Insect Society isOrganized, The Free Press, 1999.

[10] T. Seeley, The Wisdom of the Hive, Harvard UniversityPress, 2005.

[11] R. Albert, H. Jeong, A. Barabasi, Error and attack toleranceof complex networks, Nature 406.

[12] N. Minar, K.H. Kramer, P. Maes, Cooperating mobileagents for dynamic network routing, Software Agents forFuture Communications Systems.

[13] G. Cabri, L. Leonardi, F. Zambonelli, Mobile-agent coor-dination models for internet applications, IEEE Computer.

[14] V. Shnayder, M. Hempstead, B. rong Chen, G. Werner-Allen, M. Welsh, Simulating the power consumption oflarge-scale sensor network applications, in: Proceedings ofThe Second ACM Conference on Embedded NetworkedSensor Systems (SenSys’04), 2004.

[15] B. Drossel, F. Schwabl, Self-organized critical forest-firemodel, Physical Review Letters 69 (11) (1992) 1629–1632.

[16] C.-L. Fok, G.-C. Roman, C. Lu, Rapid development andflexible deployment of adaptive wireless sensor networkapplications, in: Proceedings of International Conference onDistributed Computing Systems, 2005.

[17] P. Boonma, P. Champrasert, J. Suzuki, Bisnet: A biologi-cally-inspired architecture for wireless sensor networks, in:Proceedings of The Second IEEE International Conferenceon Autonomic and Autonomous Systems, 2006.

[18] P. Boonma, J. Suzuki, Biologically-inspired data aggregationfor multi-modal wireless sensor networks, in: Proceedings ofThe 31th IEEE Conference on Local Computer Network(LCN), 2006.

[19] G. Werner-Allen, G. Tewari, A. Patel, M. Welsh, R. Nagpal,Firefly-inspired sensor network synchronicity with realistic

radio effects, in: Proceedings of The Third InternationalConference on Embedded Networked Sensor Systems, 2005.

[20] M. Britton, V. Shum, L. Sacks, H. Haddadi, A biologicallyinspired approach to designing wireless sensor networks, in:Proceedings of The Second European Workshop on WirelessSensor Networks, 2005.

[21] L. Szumel, J.D. Owens, The virtual pheromone communi-cation primitive, in: P.B. Gibbons, T. Abdelzaher, J. Aspnes,R. Rao (Eds.), Proceedings of the Second IEEE Interna-tional Conference on Distributed Computing in SensorSystems, Lecture Notes in Computer Science, vol. 4026,Springer, 2006, pp. 135–149.

[22] C.-F. Hsin, M. Liu, Network coverage using low duty-cycledsensors: Random & coordinated sleep algorithms, in: Pro-ceedings of The Third International Symposium on Infor-mation Processing in Sensor Networks, 2004.

[23] J. Misic, V.B. Misic, Duty cycle management in sensornetworks based on 802.15.4 beacon enabled MAC, Ad Hocand Sensor Wireless Networks Journal 1 (3) (2005) 207–233.

[24] W. Ye, J. Heidemann, D. Estrin, Medium access controlwith coordinated adaptive sleeping for wireless sensornetworks, IEEE/ACM Transactioncs on Networking 12 (3)(2004) 493–506.

[25] T. He, S. Krishnamurthy, J.A. Stankovic, T. Abdelzaher, L.Luo, R. Stoleru, T. Yan, L. Gu, J. Hui, B. Krogh, Energy-efficient surveillance system using wireless sensor networks,in: Proceedings of the Second International Conference onMobile System, Applications, and Services, 2004.

[26] Q. Han, S. Mehrotra, N. Venkatasubramanian, Energyefficient data collection in distributed sensor environments,in: Proceedings of The Twenty Fourth IEEE InternationlConference on Distributed Computing Systems, 2004.

[27] T. Bokareva, N. Bulusu, S. Jha, SASHA: Towards a self-healing hybrid sensor network architecture, in: Proceedingsof The Second IEEE International Workshop on EmbeddedNetworked Sensors, 2005.

[28] J. Huang, W.-C. Feng, N. Bulusu, W.-C. Feng, Cascades:Scalable, flexible and composable middleware for multi-modal sensor networking applications, in: Proceedings ofThe ACM/SPIE Multimedia Computing and Networking,2006.

[29] R.R. Brooks, M. Zhu, J. Lamb, S.S. Iyengar, Aspect-oriented design of sensor networks, Journal of Parallel andDistributed Computing 64 (7) (2004) 853–865.

Pruet Boonma is a second year Ph.D.student at the University of Massachu-setts, Boston. His research interestsinclude adaptive distributed systems andbiologically-inspired wireless sensor net-works. He received his B.E. in ComputerEngineering from Chiangmai University,Thailand, and M.IT. in Computer Sci-ence from Monash University, Australia.He worked for the Department ofComputer Engineering, Chaingmai Uni-

versity, as a faculty member since 1997. In 2001, he receivedAustralian Government Scholarship (AusAID) for his master
program in Australia. In 2005, he received Thai GovernmentScholarship for his PhD. work in the United States.

ter Networks 51 (2007) 4599–4616

Junichi Suzuki received the Ph.D. incomputer science from Keio University,
Japan, in 2001. He joined the Depart-ment of Computer Science, University ofMassachusetts, Boston in September2004, where he is currently an assistantprofessor. From 2001 to 2004, he waswith the School of Information andComputer Science, University of Cali-fornia, Irvine (UCI), as a postdoctoralresearch fellow. Before joining UCI, he
was with Object Management Group Japan, Inc., as the Tech-nical Director. His research interests include autonomous adap-

4616 P. Boonma, J. Suzuki / Compu

tive distributed systems, biologically-inspired (e.g., ecological,genetic, immunological and developmental) software designs, andmodel-driven software engineering. In these areas, he hasauthored two books, published over 70 refereed papers includingfive award papers, gave over 100 invited talks, and served ontechnical program committees of over 25 conferences includingIEEE AINA 2008 and IEEE/ACM/Create-Net/ICST BIONET-ICS 2007.

He is an active participant and contributor of the InternationalStandard Organization SC7/WG19 and the Object ManagementGroup, Super Distributed Objects SIG. He is a member of IEEEand ACM.

BiSNET: A biologically-inspired middleware architecture for self-managing wireless sensor networks

Documents

Transcript of BiSNET: A biologically-inspired middleware architecture for self-managing wireless sensor networks