Ad Hoc Networks 7 (2009) 837–848

Contents lists available at ScienceDirect

Ad Hoc Networks

journal homepage: www.elsevier .com/locate /adhoc

Software-driven sensor networks for short-range shallowwater applications

Raja Jurdak a,*, Pierre Baldi b, Cristina Videira Lopes b

a CSIRO ICT Centre, Pullenvale QLD 4069, Australiab School of Information and Computer Sciences, University of California, Irvine, United States

a r t i c l e i n f o

Article history:Received 7 January 2008Received in revised form 13 May 2008Accepted 20 July 2008Available online 15 August 2008

Keywords:UnderwaterSensorNetworkSoftwareModemShort-rangeShallowWaterFrequency shift keying

1570-8705/$ - see front matter � 2008 Elsevier B.Vdoi:10.1016/j.adhoc.2008.07.012

* Corresponding author. Tel.: +61 7 3327 4581.E-mail addresses: rjurdak@ieee.org (R. Jurdak), p

Baldi), lopes@ics.uci.edu (C.V. Lopes).

a b s t r a c t

Most existing underwater networks target deep and long range oceanic environments,which has led to the design of power hungry and expensive underwater communicationhardware. Because of prohibitive monetary and energy cost of currently over-engineeredcommunication hardware, dense deployments of shallow water sensor networks remainan elusive goal. To enable dense shallow water networks, we propose a network architec-ture that builds on the success of terrestrial sensor motes and that relies on the coupling ofsoftware modems and widely available speakers and microphones in sensor motes toestablish acoustic communication links. In this paper, we analytically and empiricallyexplore the potential of this acoustic communication system for the underwater environ-ment. Our experimental approach first profiles the hardware in water after waterproofingthe components with elastic membranes. The medium profiling results expose the favor-able frequencies of operation for the hardware, enabling us to design a software FSKmodem. Subsequently, our experiments evaluate the data transfer capability of the under-water channel with 8-frequency FSK software modems. The experiments within a 17 � 8 mcontrolled underwater environment yield an error-free channel capacity of 24 bps, andthey also demonstrate that the system supports date rates between 6 and 48 bps withadaptive fidelity.

� 2008 Elsevier B.V. All rights reserved.

1. Introduction

Water is a crucial resource for most life on earth, cover-ing more than 70% of our planet. Sustainable use andexploitation of our water resources requires a deep under-standing of both oceanic and inland aquatic environmentsthrough long-term monitoring of these environments.Existing aquatic monitoring platforms have attempted tocapture information at high temporal scale (e.g. surfacebuoys with suspended probes in the water, satellites thatobserve large geographic regions) or high spatial scale(e.g. research vessels that survey sea floors) from the oce-anic environment. Monitoring aquatic environments at

. All rights reserved.

fbaldi@ics.uci.edu (P.

both high temporal and spatial scales remains an elusivegoal despite its strategic significance to the social and eco-nomic development of the global population.

An enabling technology for granular monitoring ofaquatic environments is wireless sensor networks, whichhas been successful in many terrestrial applications. Adense deployment of in situ sensor nodes that communi-cate wirelessly in the water can meet the scale require-ments of aquatic monitoring, as the sensor nodes arerelatively cheap (ranging from 10 to 250 US dollars in pricedepending on purchase volume) compared to conventionalplatforms, making it cost-feasible to deploy a large num-ber of nodes in a physical area. Because the nodes residein the aquatic environment, they can supply data at anytemporal scale required by the network user. Adaptingsensor nodes for wireless underwater communicationrequires the addition of hardware for acoustic modulation

838 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

and communication. While the cost of sensor modulesthemselves is low, the cost of add-on hardware foracoustic communication, for modulation and demodula-tion, and for protecting and waterproofing the componentsis prohibitive, typically running into the thousands ofdollars.

Most of the existing work on wireless underwater net-working targets oceanic applications, where the communi-cation range and depth are typically in the order ofhundreds to thousands of meters, and the data transferrates go up to few Kbits per second. Design strategies forwireless underwater communication in the oceanic envi-ronment must also consider harsh conditions and peculiarcharacteristics of the underwater channel that distinguishit from the aerial channel [1]. First, the underwater channelhas severely limited bandwidth that is highly dependenton communication range, which requires bandwidth-effi-cient modulation and data compression techniques. An-other unique characteristic of the underwater channel isthe time-varying multi-path effects, where inter-symbolinterference can span several data symbols. The fadingand outage behavior of the underwater channel is alsonot yet understood. This requires dynamic protocols thatrely on cross-layer optimization to adapt to unpredictablechannel variations. Finally, the speed of sound in water,which is 1500 m/s, is significantly lower than the speedof electro-magnetic signals in air. The relatively slow signalspeed in water causes severe doppler distortion and verylong propagation delays. This emphasizes the importanceof synchronization mechanisms and throughput efficiencyof protocols.

The numerous constraints for wireless underwatercommunications in medium to long range oceanic applica-tions has led companies and researchers to highly engineerspecialized hardware for modulating, transmitting, receiv-ing and demodulating acoustic signals. The specializedmodulation hardware ranges from expensive commerciallyavailable acoustic modems [9,10] to more affordable dedi-cated integrated circuits [6] and dedicated DSP boards[12,11,13]. The communication hardware ranges from spe-cialized underwater acoustic transducers and hydrophones[7] to generic speakers and microphones [6]. The use ofspecialized hardware for establishing acoustic communi-cations underwater typically increases the network cost,the design time spent in interfacing node hardware com-ponents, and the size and weight of individual networknodes.

While using specialized hardware is a necessity foroceanic applications to establish communication links, itrepresents an overkill for short-range shallow water applica-tions, such as the water quality monitoring of lakes, bays,rivers, estuaries, and reservoirs. Water quality monitoringin a river [2] exemplifies an application that requiresmonitoring at a high spatial granularity, which is synony-mous with short inter-node distances in sensor networks,in order to capture the small scale variations in contamina-tion and to identify pollution sources and causes. Nodeslocated at close proximity to each other only requireshort-range wireless communication in the water. Despitethe availability of underwater communication hardware,short-range shallow water monitoring applications have

not adopted existing technologies specifically because oftheir prohibitive cost and over-engineering for this classof applications. Shallow water networks refer to depthsranging from 0 to 50 m, where sound propagation ismostly horizontal except for surface and bottom reflec-tions [3]. Section 5 considers the impact of applying ourarchitecture to deep water networks.

In the past, most underwater deployment efforts havefocused on hardware acoustic modulation because lowprocessing speeds did not allow the modulation of acousticsignals in software. Software modulation and demodula-tion [14] is an alternative approach which overcomes mostof the drawbacks of hardware modems. Recent advances inminiaturization and circuit integration have yielded smal-ler and more powerful processors that are capable of effi-ciently running acoustic modulation and demodulationsoftware. Software modulation also provides a higher levelof flexibility for on-the-fly tuning of modulation parame-ters to suit different environments. The transmission andreception of the software modulated acoustic signal canalso avoid using specialized hardware through genericspeakers and microphones.

Eliminating the need for specialized hardware foracoustic communication greatly reduces the cost of net-work nodes, which facilitates the dense deployment ofmote-class nodes to form underwater sensor networks.Within this context, this article proposes software-drivenunderwater acoustic sensor networks for dense shallowwater quality monitoring in rivers, bays, estuaries, andlakes. The network consists of affordable off-the-shelf sen-sor modules (motes) that use software modems and gener-ic hardware to communicate acoustically and send thedata towards the base station through multi-hop commu-nication. The motes are placed into elastic latex mem-branes that waterproof the hardware while maintainingacoustic coupling with the water channel. In addition tobeing cost-feasible and satisfying the temporal and spatialscale requirements, sensor motes, which are originally de-signed for terrestrial applications, combine processing andstorage capabilities that provide an intelligent platform fornetwork self-configuration and self-management. Finally,the speakers on board the mote platform have low outputpower, which is favorable for both network longevity andfor minimizing interference with aquatic ecosystems.

Software-driven underwater sensor networks involvethe design and development of the acoustic communica-tion links, communication protocols, and applicationbehavior. This paper investigates the design and develop-ment of reliable acoustic communication links for realiz-ing software-driven underwater sensor networks. Inparticular, the paper first profiles through empiricalexperiments the unique channel of the proposed soft-ware-driven underwater sensor networks that includesthe speaker, the microphone, the latex membranes, andthe water. Based on the underwater profiling results, wederive the theoretical error-free transfer rate of each hard-ware set. To validate the transfer rate projections, our sec-ond set of experiments evaluates the data transfercapability of the underwater channel with 8-frequencyFSK software modems coupled with the mote’s acoustichardware.

R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848 839

The novel contributions of this paper are:

(1) Proposal of a cost-effective low-power sensor net-work architecture for short-range shallow water appli-cations, based on the coupling of software modemsand off-the-shelf motes equipped with generic acoustichardware.(2) Validation of the feasibility of the wireless commu-nication links driving this architecture by empiricallyprofiling the underwater channel to determine the fre-quencies with the highest signal-to-noise (SNR) for thischannel.(3) Proposal and empirical evaluation of FSK softwaremodems for acoustic data modulation, based on thebest frequencies identified in (2), to determine achiev-able bit rates and symbol error rates as a function ofcommunication range.

The remainder of this paper is organized as follows. Sec-tion 2 discusses the related work on underwater networkarchitecture and on both hardware and software acousticmodems. Section 3 presents the network architecture andits main components, with a focus on the physical layerfunctionality, underwater acoustic concepts, and imple-mentation issues. Section 4 demonstrates the acousticcommunication capability of our acoustic communicationsystem through profiling experiments for Tmote hardwarein the underwater medium. The resulting medium profileresults enables the design of a simple FSK modem. Weuse the modem to evaluate the data transfer characteristicsof our acoustic underwater communication system. Sec-tion 5 discusses the wider applicability of the communica-tion system for underwater sensor networks and concludesthe paper.

2. Related work

This section first reviews existing architecture for under-water networks. The latter part of the section discussesacoustic modulation, both in hardware and in software, asan enabler of wireless underwater communication.

2.1. Underwater network architectures

Most underwater network proposals rely on wirelessacoustic communication between a set of underwaternodes, which eventually relay information to a node atthe surface. Akyildiz et al. [5] define both 2-dimensionaland 3-dimensional architectures for underwater networks.The 2-dimensional architecture follows a clustered topol-ogy in which each group of underwater nodes that arefixed to the sea floor communicates with more powerfulnodes, or clusterheads, in their vicinity. The clusterheadsthen relay the data to a surface node, which in turns for-wards the data through satellite or long-range radio linksto a central repository. The 3-dimensional architectureconsiders underwater sensor nodes that are anchored tothe sea-floor, thus allowing them to float at differentdepths depending on currents and tides. Since our paperconsiders shallow water applications with communication

at very short ranges, differences in depth are relativelysmall so the network adheres to the 2-dimensionalarchitecture.

Another recent survey by Cui et al. [4] classifies under-water network architectures according to their deploy-ment duration and the criticality of data they sense,differentiating between two architecture classes: long-term non-critical deployments; and short-term criticaldeployments. Their article also identifies very short-rangeunderwater acoustic modems as a gap in the currentstate-of the art. Our work coincides with the long-termnon-critical monitoring architecture of Cui et al., and inthis paper we specifically target the development andinvestigation of low power acoustic modems for veryshort-range underwater networks.

Vasilescu et al. [6] propose another network architec-ture where autonomous underwater vehicles (AUV) peri-odically visit the network area to collect data from thein situ sensors through ultra short wireless acoustic links.Although our work focuses on the adaptation of stationarymote platforms for the underwater environment, thedevelopment of acoustic communication links in that oper-ate with a range of about 20 m on the motes would en-hance Vasilescu’s architecture, reducing the localizationand navigation requirements, as well as the duration ofthe data collection process by the AUV.

2.2. Hardware modems

Earlier efforts in acoustic communication have focusedon using specialized and dedicated hardware for underwa-ter acoustic modulation and demodulation. Acousticunderwater communication is a mature field and thereare several commercially available underwater acousticmodems [9,10]. The commercially available acoustic mod-ems provide data rates ranging from 100 bps to about40 Kbps, and they have an operating range of up to a fewkm and an operating depth in the order of thousands ofmeters. The cost of a single commercial underwater acous-tic modem is at least a few 1000 US dollars. The prohibitivecost of commercial underwater modems has been anobstacle to the wide deployment of dense underwater net-works, until the recent development of research versions ofhardware acoustic modems.

Researchers at the Woods Hole Oceanographic Institu-tion are developing a Utility Acoustic Modem (UAM) as acompletely self-contained, autonomous acoustic modemcapable of moderate communication rates with low-powerconsumption [12]. This modem uses a single specializedDSP board with on board memory and batteries. The pur-pose of developing the UAM is to make a more affordableacoustic modem available for the research community.Researchers at UC, Santa Barbara are also developing ahardware acoustic underwater telemetry modem [11] forecological research applications, using a DSP board withcustom amplifiers, matching networks, and transducers.Their modem is intended for interfacing to nodes in anunderwater ad hoc network, and it achieves a 133 bps datarate. Whereas both of the efforts reported in [12,11] aim atmaking underwater acoustic modems cheaper and moreaccessible by developing specialized affordable hardware,

840 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

our work aims at driving the cost even lower and at mak-ing acoustic underwater communications even moreaccessible through the development of software acousticmodems that can operate on generic hardware platforms.

In a more recent article, Wills et al. [13] propose theirdesign for an inexpensive hardware modem for denseshort-range underwater sensor networks. Their work aimsat borrowing communication concepts, such as wake-upradio, from terrestrial sensor networks. Although we sharethe same end goal as Wills et al. (inexpensive acousticmodems for dense short-range wireless networks), our ap-proach differs in its emphasis on modulation through soft-ware rather than through specialized hardware.

One of the few attempts to deal with generic micro-phones and speakers is Vasilescu et al. [6]. These authorspropose a network that combines acoustic and opticalcommunications, stationary nodes and AUV’s for monitor-ing coral reefs and fisheries with ranges in the order ofhundreds of meters. The work in [6] uses generic micro-phones and speakers along with a specialized integratedcircuit that generates ASK or FSK modulated sound signalin order to demonstrate the acoustic communication capa-bility underwater. Vasilescu et al. achieve a bit rate in theorder of tens of bits per second up to about 10–15 m.Although our work resembles their work in the use of gen-eric microphones and speakers for acoustic communica-tions, it differs in its proposal and implementation ofsoftware modems for off-the-shelf mote platforms ratherthan the use of specialized integrated circuits forcommunication.

2.3. Software modems

With the rapid increase in processor speeds, the idea ofimplementing acoustic modems in software becomes fea-sible and even attractive due to the low cost processingpower. Coupling software acoustic modems with the useof microphones and speakers for transmission and recep-tion can eliminate the need for specialized hardware foracoustic communication, trading off cheap computationalpower for expensive communication hardware. The costof software acoustic modems is limited to the developmentcost, after which the per unit cost is zero.

Underwater sensor nodes

Surface Buo

Wireless acoustic links

Fig. 1. Target netwo

Because of these attractive features, Lopes and Aguiar[14] have investigated using software modems for aerialacoustic communications in ubiquitous computing appli-cations. Building on their work, software acoustic modemscan also eliminate the need for specialized hardware inunderwater acoustic communications, thereby encourag-ing wider deployment of underwater sensor networks. Inpreliminary experiments, we started profiling the under-water acoustic spectrum and data communications capa-bilities with software acoustic modems [16]. Our workused waterproofed generic microphones and speakers,connected to laptops on the surface, for sending andreceiving software modulated acoustic signals. Theachieved bit rates were in the order of tens of bits per sec-ond for distances up to 10 m. Our work here extends ourearlier work by coupling software modems with Tmote In-vent module hardware. Since the experiments in [16] usedmicrophones and speakers of comparable specifications tothe on-board speakers and microphones on the Tmote In-vent module, we expect the underwater acoustic commu-nication capability of autonomous Tmote Invents to yieldcomparable results. Our recent study in [17] presentunderwater experiments results that confirm the commu-nication capability of software modems on Tmote Inventhardware. In this paper, we build on those results to pro-pose a cost-effective software-driven sensor networkarchitecture for short-range shallow water applications.

3. Network architecture

This section describes the target network architecture,focusing on the physical layer aspects. Fig. 1 sheds morelight on the target network application.

3.1. Network overview

Software-driven underwater sensor networks will con-sist of tens to hundreds of motes deployed in a shallowwater environment. The motes can communicate acousti-cally through short-range wireless links, thanks to theiron-board speaker and microphone and to software acous-tic modems. The motes are placed into elastic latex

y

Broadband Radio Connection To Internet

data repository

Interactive Maps

Query and Download tools

rk application.

Fig. 3. The Tmote Invent module.

R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848 841

membranes, which maintain the acoustic coupling of theon-board speaker and microphone with the water whilewaterproofing them. The modules periodically sampletheir sensors, collecting physical indicator data from theirbuilt-in sensors, such as water temperature or tidalstrength, or from add-on sensors, such as salinity or phos-phorous levels, which help determine water contaminationlevels. After sampling their sensors, the nodes send thesensor values to neighboring nodes, which in turn relaythe data through multi-hop links to the nearest collectionpoint. Because each node periodically sends only few2-byte sensor values, the data transfer rate requirements ofthe network are low.

The network architecture supports two embodiments ofthe collection point: (1) an underwater instrumentationstation that relays network data to a data repository onshore through a sub-sea fibre optic cable; or (2) a surfacestation that resides on a buoy and relays network data tothe data repository on shore through a long-range RFcommunication link. For both configurations, the logicalnetwork topology is a tree, or more generally, a multiple-tree topology.

3.2. Software-driven acoustic communication

3.2.1. System ComponentsFig. 2 illustrates the main components of the communi-

cation system. At the sender side, digital sensor data fromthe underwater environment first enter the software mod-ulator, resulting in a modulated acoustic signal. The on--board speaker then transmits this signal into theunderwater channel. At the receiver, the on-board micro-phone captures the signal and the resident software per-forms symbol synchronization through the S4 [18] block,filtering through FFT or wavelet decomposition, and finallydemodulation.

3.2.2. Target platformFor our application, we have selected mote-class com-

puters, which are powerful enough to perform sufficientin-network processing and are affordable enough to enablethe deployment of a dense network at reasonable cost. Inparticular, we have selected the Tmote Invent module(shown in Fig. 3) which has an on-board SSM2167 micro-phone from Analog Devices sensitive to frequencies from100 Hz to 20 kHz, and an on-board TPA0233 speaker

Fig. 2. Block diagram for software modem: (a) Modulator/Transmitter;(b) Demodulator/Receiver.

amplifier from TI with an 8 X speaker that has a range of400–20 kHz. The goal of this work is to exploit the on-board microphone and speaker to establish short rangeacoustic links among Invent modules. The next subsectionreviews the basic concepts of underwater acoustics thatdriven the design of our software modem.

3.3. Underwater acoustics

3.3.1. The passive sonar equationThe passive sonar equation [15] characterizes the signal

to noise ratio (SNR) of an emitted underwater signal at thereceiver:

SNR ¼ SL� TL� NLþ DI; ð1Þ

where SL is the source level, TL is the underwater transmis-sion loss, NL is the noise level, and DI is the directivity in-dex. All the quantities in Eq. (1) are typically in dB re lPa,where the reference value of 1 lPa amounts to0.67 � 10�22 W/cm2 [15]. However, we will use the thresh-old of human hearing at 10�12 W/m2 as our reference sig-nal level, to better compare results with our earlierresults in [16]. In the rest of this paper, we use the short-hand notation of dB to signify dB re 10.�12, unless other-wise mentioned.

For the purpose of this analysis, we examined severalstudies of shallow water noise measurements under differ-ent conditions [21,15]. As a result, we consider an averagevalue for the ambient noise level NL to be 70 dB as a repre-sentative shallow water case. We also consider a targetSNR of 15 dB [15] at the receiver.

The directivity index DI for our network is zero becausewe assume omnidirectional hydrophones. Note that thisis another conservative assumption, since using adirective hydrophone as described in [19] reduces powerconsumption.

Through the above assumptions, we can express thesource level SL intensity as a function of TL:

SL ¼ TLþ 85: ð2Þ

3.3.2. Source levelThe transmitter source level (SL) of underwater sound

relates to signal intensity It, which in turn depends onthe transmission power.

842 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

Given the transmission power Pt, the transmitted inten-sity of an underwater signal at 1 m from the source can beobtained through the following expression [15]:

It ¼Pt

2p� 1 m� Hð3Þ

in W/m2, where H is the water depth in m. The followingequation determines the source level SL relative to thethreshold level of human hearing:

SL ¼ 10 logIt

10�12

� �: ð4Þ

3.3.3. Transmission lossThe transmitted signal pattern has been modelled in

various ways, ranging from a cylindrical pattern to a spher-ical one. The following expression governs acoustic signalspropagation in shallow water [15]:

TL ¼ 10� l log dþ ad� 10�3; ð5Þ

where d is the distance between source and receiver in me-ters, a is the frequency dependent medium absorptioncoefficient in dB/km, and TL is in dB. The variable l de-pends on the signal spreading pattern. If the acoustic signalspreads in all directions from the sound source, then l isequal to 2. If the acoustic signal spreads in a cylindricalpattern from the source (as is the case signals propagatingalong the surface or ocean floor), then l equals to 1. Inshallow water cases, the value of l lies somewhere be-tween 1 and 2, depending on the depth.

Eq. (5) indicates that the transmitted acoustic signalloses energy as it travels through the underwater medium,mainly due to distance dependent attenuation and fre-quency dependent medium absorption. Fisher and Simmons[20] conducted measurements of medium absorption inshallow seawater at temperatures of 4 and 20 �C. Wederive the average of the two measurements in Eq. (6),which expresses the average medium absorption at tem-peratures between 4 and 20 �C:

a ¼

0:0601� f 0:8552 1 6 f 6 6;9:7888� f 1:7885 � 10�3 7 6 f 6 20;0:3026� f � 3:7933 20 6 f 6 35;0:504� f � 11:2 35 6 f 6 50;

8>>><>>>:

ð6Þ

where f is in kHz, and a is in dB/km.Through Eq. (6), we can compute medium absorption

for any frequency range of interest. We use this value fordetermining the transmission loss at various inter-nodedistances through Eq. (5) which enables us to computethe source level in Eq. (2) and subsequently to computethe power needed at the transmitter.

3.3.4. Noise levelFactors contributing to the noise level NL in shallow

water networks include waves, shipping traffic, wind level,biological noise, seaquakes, volcanic activity, and rain, andthe impact of each of these factors on NL depends on theparticular setting. For instance, shipping activity may dom-inate noise figures in bays or ports, while water currentsare the primary noise source in rivers. In a swimming pool

environment, where we conducted our experiments, themain sources of underwater noise are swimmers, peoplewalking near the pool, water pumps, and drains.

3.4. Implementation issues

The implementation of software acoustic modems forgeneric PC hardware, as we described in [16], is relativelystraightforward. The modulated data exists as a wav fileat the sender laptop. Data transmission simply involvesthe playback of the wav file at the sender laptop and therecording of the received signal as wave file at the receiverlaptop, where the modulated data is extracted from thesignal.

In contrast, the implementation of the software mod-ems on Tmote Invent modules involves several resourceand configuration challenges. First, the Tmote Invent mod-ule has a small fraction of the RAM and storage capacity ofcurrent personal computers. The RAM size of current com-puters is in the order of hundreds of Megabytes, enablingan entire wav file to be loaded into RAM for immediateplayback and random access. The Tmote Invent RAM mem-ory size is only 10 KB, so modulated acoustic signals canonly be loaded into RAM in small size chunks at a time.For example, consider the playback or recording of a mod-ulated acoustic signal with a sampling frequency of22,050 Hz and 8 bits/sample. The Tmote Invent RAM canstore up to 10,000/22,050 = 0.4545 s of the acoustic signalat a time. Lowering the signal sampling frequency in-creases the size of the acoustic signal portion that can beloaded into the RAM at one time. Thus, we aim to lowerthe sampling frequency as much as possible while still pro-viding an acceptable margin above the Nyquist frequencyto maintain an acceptable signal quality.

Another challenge is the effective streaming of theacoustic signal between the Tmote Invent external flashmemory and the Tmote Invent RAM. The Tmote Inventhas an external flash memory of 1 MB. Each node can storethe modulated wave signal in flash and stream this signalto the RAM when needed. Our implementation originallyused the Blackbook [28] file system that provides easy ac-cess files on the external flash of motes. The streaming ofchunks of the audio file from flash to RAM requires the pro-gram to open the file on flash, read the required chunk, andwrite it to RAM. These functions involve appreciable, yetpredictable delay, that is proportional to the size of thedata chunk. The flash-related delay complicates thedemodulation process since the symbols in the transmittedsignals have irregular delays that must be accounted for inthe demodulation process. To address this issue, we ac-count for this flash read delay in the demodulation algo-rithm at the receiver so that the time to decode a symbolactually includes the time to decode the symbol plus theknown read delay.

Another issue is that the flash read time and the flashwrite time within Blackbook are not the same, which pre-sents a more involved challenge. Both the read and writetime are proportional to the size of data in the read orwrite operation. Therefore, for a transmitted signal of Xbits, the flash read delay is aX s. At the receiver, the writingof this signal to flash requires bX s, where b is larger a. The

R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848 843

Tmote microphone must wait for the flash write operationto complete before capturing the next chunk of the acous-tic signal. By the time the microphone turns on again, thesender and receiver lose their synchronization. This mis-match in flash reading and writing causes variable dura-tion symbols.

Because of these complications in synchronizing thesender and receiver with unpredictable flash read andwrite delays, we have adopted an alternative implementa-tion that generates the modulated acoustic signal on-the-fly every time the sensor acquires data. This method usesthe PlayTone interface in the Moteiv’s [30] version of Tiny-os [31] which generates frequency tones at any given fre-quency and plays them on the speaker. The softwaremodem simply employs this function by breaking up the16-bit sensor value into 6 chunks of 3 bits each, with 2 bitsof the last chunk left unused. The modem then translatesthe chunk into the corresponding frequency tones, andcalls the PlayTone command for each frequency tone untilall 16 bits are sent. This avoids any access to external flashat the sender. The only shift that remains is the command/event delay at each of the sender and receiver, which canbe accounted for through repetitive profiling experimentsof the application code of both sender and receiver. Thisis a calibration step that must be performed only oncefor each platform architecture, after which the delay valuesand corresponding time shifts can be taken into consider-ation during transmission and reception of symbols.

3.5. Computational requirements

A natural consideration for the design of software mod-ems to run on tiny mote devices is the computationalrequirements of the modems and how well they matchthe limited processing power of the motes. Referring backto Fig. 2, the sender only need to split the sensor valuesinto appropriate size chunks and iteratively call the com-mand that plays the tones on the speaker, namely the Play-Tone command.

The receiver computational requirements are moreextensive. The receiver has to first perform symbol syn-chronization, followed by filtering and finally demodula-tion. Each of these blocks must minimize thecomputational load on the 16-bit processor of the nodes.For this purpose, we have custom-designed a symbol syn-chronization method [18] called S4 that targets platformswith low computational power. The method uses simplearithmetic operations to establish the sample in the re-ceived audio signal that corresponds to the start of thetransmitted data.

For the filtering block, the design specifies three typesof filters that map to three device classes. The first filtertype is narrowband filters, that have high computationalpower and high accuracy. These filters are suitable foruse at a PC-class device, such as at the base station, whichmust collect and filter many streams in a limited time. Thesecond type of filter is FFT, which is suitable for micro-ser-ver or PDA-class devices. This filter has lower computa-tional requirements than the narrowband filter and lessaccuracy, and it fits well for local cluster heads in an under-water network that are typically more computationally

capable than normal nodes. The final type of filters thatthe system supports is wavelet decomposition, which sep-arates lower frequency components from higher frequencycomponents, enabling easier identification of the transmit-ted symbol. Because these filters have the lowest computa-tional requirements and they still serve the purpose ofsuccessfully identifying the transmitted symbol at the re-ceiver, they are suitable for mote-class devices with lim-ited computational power.

The final block at the receiver is demodulation. Thisblock is relatively simple once the signal is filtered, as it re-duces to identifying the highest component in the filteredsignal that corresponds to a useful data symbol, convertingthis symbol to bits, and concatenating the bits to obtain theoriginal sensor value. These operation can all be achievedthrough simple arithmetic and non-arithmetic functioncalls in Tinyos.

4. Experimental results

This section presents empirical experiments and resultsinvolving the Tmote Invent speakers. We first conductunderwater medium profiling experiments in which a thinelastic latex membrane waterproofs the speakers. Theunderwater medium profiling results enable the projectionof error-free data transfer rates, which we then validatethrough underwater data transfer experiments.

4.1. Medium profiling

Our target application has unique channel characteris-tics that differ from underwater channels in the related lit-erature [15,25,26,8], as it includes several components: theunderwater medium; the generic speakers and micro-phone (whose response and coupling with the underwaterenvironment is unknown); the waterproofing membranes(which may amplify or attenuate certain frequencies).

We have performed experiments to assess the fre-quency profile of the channel. In these experiments, weuse a Tmote Invent module, with a built-in audio speaker,as the transmitter of the acoustic signal. The signal is thencaptured by a submerged microphone attached to a laptopPC, where the signal is recorded into a wav file for analysis.We conduct the experiments for the Tmote Invent speakersfor frequencies between 400 and 3500 Hz, for which thehardware exhibited the best response in our earlier exper-iments [16].

Our previous experiments with generic PC speakersused vinyl membranes for waterproofing. Here, we usethinner and more elastic latex membranes that outperformvinyl membranes in preserving the signal acoustic proper-ties underwater, keeping in mind that the change in water-proofing membrane may contribute to the differencesbetween the results for the current experiments with theTmote Invent module and the previous experiments withthe generic speakers.

To obtain the signal quality of each frequency fi of a sig-nal received from distance dj meters away, we apply a100 Hz Equiripple [24] band pass filter centered at fi tothe received signal. The filtered signal shape includes the

Fig. 5. Noise level of the underwater channel as observed by the PCmicrophone.

844 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

transmitted tone at fi along with all the background noisewithin the frequency range fi � 50 to fi + 50. The back-ground noise is distinguishable in all the temporal compo-nents of the signal during which the tone at fi is nottransmitted. Through this filtering process, we can obtainthe signal-to-noise ratio SNR(fi,dj) of the channel for eachfrequency fi and distance dj.

We performed the underwater experiments for dis-tances dj ranging from 1 to 13 m at 1 m increments. At eachdistance dj, we conducted the measurements three timesand obtained the average SNR(fi,dj) of the three samplesfor fi. The value of l is set to 1.5 as a compromise betweencylindrical and spherical spreading.

Fig. 4 illustrates the Tmote Invent measured and theo-retical SNR(fi,dj) values for each frequency fi at eachdistance dj. The measured SNR in Fig. 4 for certain frequen-cies between 1000 and 2000 Hz, such as 1300 Hz and1500 Hz, is higher than the projected SNR for those fre-quencies. This effect is a reflection of the choice of 1.5 forthe l variable for the computation of the transmission loss.The higher values of the measured SNR indicate that thesignal spreading for these particular frequencies is closerto the cylindrical model than the spherical model, high-lighting favorable multi-path and reflection effects forthese frequencies in the closed testing environment.

Another observation on Fig. 4 is the frequency selectivebehavior of the channel (consisting of the Tmote Inventspeaker, latex membranes, underwater medium, and PCmicrophone).

To further investigate this frequency selectiveness ofthe channel, we compute the noise level of each frequencydistance pair. To compute the noise level, we can use thetransmission power Pt of the speakers to obtain It throughEq. (3). We can then compute the source level SL throughEq. (4). We can also get the transmission loss TL(fi,di)through Eq. (5) for every fi and dj. Using the measured val-ues of SNR(fi,dj), we can compute NL(fi,dj) through Eq. (1).

Fig. 5 illustrates the noise level in the water at each fre-quency fi as observed at a receiver that is dj meters awayfrom the sender. The apparent noise level is high at very

Fig. 4. SNR profile of the underwater channel with the Tmote Inventspeakers.

low frequencies, and it drops steadily up until 1000 Hz. Be-tween 1000 and 2000 Hz, the noise level is generally sim-ilar except for minor fluctuations. The noise level thenstarts to increase again for frequencies above 2000 Hz. Thishighlights the favorability of the channel to frequencies be-tween 1000 and 2000 Hz, indicating that response of thespeaker/microphone pair and the acoustic coupling of themembrane with the water medium is best at these fre-quencies. We also note the slight but steady drop in noiselevel as the receiver moves further away from the sender.This effect is attributed to a form of self-interference, sim-ilar to terrestrial networks, where a sender and receiverthat are too close may exhibit high noise levels due tothe high relative amplitude of the multi-path signal com-ponents at the receiver.

The underwater noise level results reflect not only theinterference in the underwater channel, but also the re-sponse and coupling of the microphone, speaker, and elas-tic membrane with the underwater medium. This includesany distortions related to impedance mismatching causedby placing the components in an underwater medium,whereas they are typically designed for the aerial channel.

To put the underwater noise level results in perspective,we have performed similar medium profiling experimentsin the aerial channel and derived the noise level of themicrophone/speaker pair in air using the same method de-scribed above. The parameters for computing the SNR inair can be found in our earlier work in [29].

Fig. 6 shows the aerial noise level for communicationsystem. As for the underwater channel, the noise level re-flects ambient interference as well as how well the compo-nents couple with the aerial medium. The results in Fig. 6exhibit similar behavior to the underwater channel for fre-quency at or below 2000 Hz, where very low frequencyexperience high noise levels and frequencies between1000 and 2000 Hz experience reduced noise levels. Asimilar correlation between the apparent noise level anddistance is also observed for the aerial channel, whereself-interference decreases with distance. The main differ-ence between the aerial and underwater noise levelsoccurs at short transmission distance and frequencies

Fig. 6. Noise level of the aerial channel as observed by the PCmicrophone.

0

5

10

15

20

25

30

35

40

1 2 3 4 5 6 7 8 9 10 11 12 13Distance (m)

Bitr

ate

(bps

)

Fig. 7. Error-free bitrate of Tmote Invent speakers in the underwaterenvironment.

R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848 845

above 2000 Hz, where the underwater noise level increasesteadily. In contrast, the aerial noise level increases slightlyfor frequencies just above 2000 Hz, and it remains rela-tively constant for higher frequencies. At transmission dis-tances of 3 m or higher, the self-interference exhibits asharp drop for frequencies above 2000 Hz, resulting in sig-nificantly lower noise levels than the underwater channelfor the same frequencies. This indicates that our system’sacoustic coupling is mostly unaffected for frequencies be-tween 1000 and 2000 Hz when the nodes are placed inwater, whereas it degrades for high frequencies.

Based on this frequency selectivity, the software acous-tic modem can use the following 8 frequencies that exhibitthe highest signal quality at all distances: 1000, 1200,1300, 1500, 1600, 1700, 1800, and 2000 Hz. The minimumSNR for all of these frequencies within a range of 13 m is6.95 dB. It is worth noting that only one of the above fre-quencies, 1300 Hz, is common with the top 8 frequenciesfor the PC speakers. We attribute this difference to the dif-ference in waterproofing membranes, as well as the poten-tial hardware differences of the two speakers. An attractivefeature of implementing modulation in software is thecapability for self-calibrating modems that can profile themedium and determine the best set of frequencies andthe suitable modulation rate for the channel.

4.2. Data communication

Prior to conducting data communication experiments,we can compute the achievable error-free bit rate C foreach communication distance using the Shannon–Hartleyexpression [27]:

C ¼ Blog2ð1þ SNRÞ bps: ð7Þ

Based on the measured SNR of the selected frequencies ofthe software modem, we can project the achievable er-ror-free bitrate of the Tmote Invent speakers for differentcommunication ranges, using the lowest value of SNR forall the selected data frequencies. Fig. 7 illustrates the pro-jected error-free bitrate of the Tmote Invent speakers. Forall distances up to 13 m, the measured SNR indicates that

the speakers can support an error-free bitrate of at least24 bps.

In our software modem design and experiments, wechoose to explore bit rates that exceed the expected chan-nel capacity within a system that is tolerant of some com-munication errors. Our software modem is based on astructure of time slots. Each time slot of length T millisec-onds contains one FSK symbol, which has a duration of T/2milliseconds, in addition to a guard time of T/2 millisec-onds. Guard times between adjacent FSK symbols are nec-essary to avoid inter-symbol interference which may ariseas a result of multi-path propagation effects. To evaluatethe impact of the length of the time slot on the data recep-tion capability at different distances, we consider 4 casesfor the time slot lengths: (1) 500 ms; (2) 250 ms; (3)125 ms and (4) 62.5 ms. The above time slot lengths corre-spond to data bit rates of 6, 12, 24, and 48 bits per secondrespectively. Note that we refrain from conducting experi-ments for the 96 bps data rate, because of its poor perfor-mance in PC speaker experiments.

The second set of experiments directly evaluates thedata transmission capability of the Tmote Invent speakers.The encouraging results of the Tmote Invent medium pro-filing experiments (mainly the reduced dependence of sig-nal loss on distance) have motivated the expansion of therange of the data transmission experiments to 17 m.Experiments at each distance between 1 and 17 m at 1 mincrements are conducted 4 times, and the results repre-sent the average of the four experiment instances.

Fig. 8 shows the percentage of symbols correctly re-ceived with the Tmote Invent speakers. Although we con-ducted the experiments at 1 m increments, we displaythe data transfer results for 2 m increments for clarity ofpresentation. For all bit rates in the experiments, the recei-ver could decode at least 79% of the transmitted symbols.Also, the decoding capability of the receiver with theTmote Invent speaker is as good as or better than the PCspeaker experiments, as Table 1 reveals. Note that the sig-nal with a bit rate of 48 bps, where the signal decodingcapability improves from 35% to 79%.

For data transfer rates of 12 and 24 bps, the percentageof correctly received symbols is the same for the Tmote In-vent and our earlier PC speaker experiments. However, wenote that the Tmote Invent speaker with the latex mem-brane provides the same data transmission capability at

Table 1Comparison of the percentage of symbols correctly received for the PCspeakers and Tmote Invent experiments

Transfer rate (bps) 6 12 24 48Tmote + Latex membrane up to 17 m P95% P90% P81% P79%Generic + Vinyl membrane up to 10 m N/A P90% P78% P35%

0102030405060708090

100

1 3 11 13 15 17Distance (m)

Sym

bols

Rec

eive

d (%

)48bps

5 7 9

6bps 12bps 24bps

Fig. 8. Percentage of symbols correctly received for the Tmote Inventspeakers in the underwater environment.

846 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

17 m as the PC speakers with the vinyl membrane at 10 m.Again, this confirms the superior performance of the latexmembrane to the vinyl membrane as a waterproofing solu-tion for our application.

5. Discussion and conclusion

5.1. Communication range

The empirical results from this study serve as a proof-of-concept for software-driven underwater communica-tion through generic acoustic hardware with data rates inthe order of tens of bits per second. It is worth noting thatthe communication range of 17 m in this study is a limita-tion of the available physical space and not a technicalcommunication limitation. An interesting direction for fu-ture work is to push the technology further to evaluatethe maximum achievable communication and detectionranges, and the associated data transfer rates. In fact, anextrapolation of the experimental results reveals that thetheoretical channel capacity at 50 m still remains above16 bps.

5.2. Network topology

The current implementation of our software-drivenacoustic underwater networks communication system,which relies on the Tmote Invent speakers as transmit-ters, and a generic PC microphone as receiver, enablesthe deployment of a single-hop network centered at thebase station with a theoretical circular shape and a radiusof at least 17 m. The Tmote Invent modules send theirdata through direct acoustic links to the base station,forming a cluster of underwater motes that provides envi-ronmental data with high spatial granularity within thecoverage area. The deployment of several 1-hop clusterscan cover wider geographic areas through a hierarchical

topology in which the clusterheads send the collecteddata to a central repository for dissemination and analy-sis. Such a topology is especially useful for pinpointingthe propagation areas of oil patches along a coastline inthe aftermath of oil spills, such as the catastrophic oil spilloff the coast of Lebanon resulting from the recent hostil-ities there.

5.3. Low-power communication

The limited communication range of our system stemsfrom the ultra lower output power of the speaker. Our net-work architecture adopts a multi-hop topology of low-power communication links for enabling dense underwa-ter sensor networks. The multi-hop topology of our net-work aims at limiting disruption to marine wildlife.Sending sound waves underwater can cause disorientationof the marine wildlife, such as whales and dolphins. Re-cently, there have been several incidents in which whalesor dolphins were disoriented and stranded because of hu-man noise pollution resulting from sonar, oil exploration,and shipping [32]. Avoiding adverse effects on marine biol-ogy is a major consideration for environmental preserva-tion. Because our network relies on multi-hop shortrange low power links between sensor nodes, it minimizessound interference with the marine wildlife.

5.4. Data transfer rate

The achievable bit rate of our system, in the order oftens of bits per second, is sufficient for long-term monitor-ing sensor networks, such as for environmental or habitatmonitoring. The nodes sample their sensors and send thedata once during each update period, typically in the orderof minutes. Since each node must send only a handful ofsensor values during each update period, a data transferrate in the order of tens of bits per second provides morethan enough throughput to communicate all the sensorvalues during an update period.

5.5. Communication protocols

Deploying our communication system within an unat-tended underwater sensor network involves several openresearch challenges, such as the design of Medium AccessControl and routing mechanisms. For instance, the under-water nodes should ensure that their data packets do notcollide, which requires the use of carrier sensing or hand-shake messaging prior to data transmission. Such collisionavoidance mechanisms include either energy overhead ordelay overhead or both. Energy efficiency is always a majorconcern for in situ sensor networks because recharging orreplacing node batteries is difficult, especially in hard-to-access areas such as the underwater environment. AnyMAC or routing layer mechanisms for ensuring proper datadelivery must also provide energy-efficient behavior. Themain challenge is therefore the design of higher layer com-munication mechanisms that serve the needs of the userapplication while balancing the energy-throughput-delaytradeoff. A prominent trend for managing this tradeoff inad hoc and sensor networks is the design of cross-layer

Fig. 9. Cost comparison for covering a 1-km stretch of water.

R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848 847

mechanisms that cut across traditional layer boundaries toachieve fine-grained optimizations [33].

5.6. Deeper waters

Our proposed solution currently relies on latex mem-branes to waterproof the sensor modules, so the volume,the shape, and the resonance frequency range of theenclosing latex membrane could change in deeper waters,where vertical pressure increases. The increased pressureat higher depths also causes a change in bulk modulusand density of the water, leading to increases in soundspeed of 0.017 m/s for every meter of depth increase[22]. This slight change causes a sound beam to bend up-ward at great depths, whereas our model considersstraightforward propagation [15]. Furthermore, the noiseprofile and channel model for deeper waters is not yetdeterministic, so the exact SNR profile and data communi-cation range and rate remain open issues for furtherinvestigation.

5.7. Cost and scale considerations

A main motivator for using software modems is thereduction in cost and the increase in spatial scale of thenetwork. This section considers a typical scenario that re-quires the coverage of a 1 km stretch of river or coastlinewith sensor nodes. We explore the associated cost and spa-tial scale of covering this stretch with hardware and soft-ware modems.

In the case of hardware acoustic modems that havecommunication ranges in the order of kilometers, a singlenode is sufficient to cover the entire 1 km stretch and com-municate the data to shore. The average cost for a singleunderwater node that combines a hardware modem, trans-ducer, and casing is about US$7500 according to our surveyof multiple vendors [9,10].

In the case of software acoustic modems, the currentprice per Tmote Invent node is US$250, and mass produc-tion price is expected to drop prices to below $10 per node[23]. The transmission coverage range of software modemson Tmote Invent modules is in the order of tens of meters,realistically running between 10 and 100 m at best (seeSection 4). Given their limited range, multiple nodes are re-quired for covering the 1 km stretch. The need for multiplenodes to cover the same physical area covered by a singlenode with a hardware modem actually improves the spatialscale of the network, as it provides more data points fromthe same physical area.

Fig. 9 plots the network cost per km as a function of In-ter-node distance and price per node. The transparent plotin Fig. 9 corresponds to the software modem cost, and thesolid plot corresponds to the hardware modem cost, whichis constant at US $7500. In the worst case, the network costfor the software modems is about $25,000 per km for an in-ter-node distance of 10 m and a unit price of $250. In thiscase, the network cost is about 3.5 times the cost of hard-ware modem networks, but the network provides a 100-fold increase in spatial granularity. When the unit pricedrops to $100, the software modem network cost is only33% higher than the hardware modem network, while still

providing a 100-fold increase in spatial scale. For an inter-node distance of 20 m (which conforms with the commu-nication range results in this paper) and a unit cost of$100, the software modem network cost is 33% less thanhardware modem network cost, and it still provides a 50fold increase in spatial scale. In fact, any unit price below$220 or any inter-node distance above 30 m result in sig-nificant increases in network spatial scale and in reduc-tions in network cost. In the most optimistic case of $10for unit cost, savings range between $6500 and $7400depending on inter-node distance.

In sum, this paper has proposed software-driven acous-tic underwater sensor networks for short-range shallowwater monitoring. The network architecture provides bothhigh spatial scale and high temporal scale at relatively lowcost through the use of generic acoustic hardware on boardoff-the-shelf sensor motes. Because the cost of our systemis limited to the relatively cheap sensor module, we expectthe system to promote wider and denser deployments ofunderwater sensor networks.

Acknowledgements

The work of R.J. has been supported in part by theIRCSET Embark Award at University College Dublin. Thework of C.V.L. has been supported in part by NSF GrantNo. CCF-0347902.

References

[1] M. Stojanovic, Underwater wireless communications: currentachievements and research challenges, IEEE Oceanic EngineeringSociety Newsletter (Spring) (2006).

[2] B. Sauser, Networking the Hudson river, in: Technology Review, MIT,2007. Available: http://www.technologyreview.com/Infotech/19309/.

[3] J.G. Proakis, E.M. Sozer, J.A. Rice, M. Stojanovic, Shallow wateracoustic networks, IEEE Communications Magazine (2001).

[4] J. Cui, J. Kong, M. Gerla, S. Zhou, Challenges: building scalable mobileunderwater wireless sensor networks for aquatic applications, IEEENetwork 20 (3) (2006) 12–18.

[5] I.F. Akyildiz, D. Pompili, T. Melodia, Underwater acoustic sensornetworks: research challenges, Ad Hoc Networks (Elsevier) 3 (3)(2005) 257–279.

[6] I. Vasilescu, K. Kotay, D. Rus, M. Dunbabin, P. Corke, Data collection,storage, and retrieval with an underwater sensor network, in: Proc.Sensys’05, San Diego, CA, 2005.

848 R. Jurdak et al. / Ad Hoc Networks 7 (2009) 837–848

[7] J. Heidemann, Y. Li, A. Syed, J. Wills, W. Ye, Underwater sensornetworking: research challenges and potential applications, USC/ISITech. Rep. ISI-TR-2005-603, 2005.

[8] X. Yang et al, Design of a wireless sensor network for longterm, in-situ monitoring of an aqueous environment, Sensors 2 (2002) 455–472.

[9] Linkquest Inc., Available at: www.link-quest.com.[10] DSPComm., Available at: www.dspcomm.com.[11] R.A. Iltis, H. Lee et al., An underwater acoustic telemetry modem for

eco-sensing, in: Proc. Oceans’05, September 2005.[12] Utility Acoustic Modem, Available at: www.auvlab.mit.edu.[13] J. Wills, W. Ye, J. Heidemann, Low power acoustic modem for dense

underwater sensor networks, in: Proc. IEEE (WUWNet06), 2006.[14] C.V. Lopes, P. Aguiar, Acoustic modems for ubiquitous computing,

IEEE Pervasive Computing (2003).[15] R.J. Urick, Principles of Underwater Sound, McGraw-Hill, 1983.[16] R. Jurdak, C.V. Lopes, P. Baldi, Software acoustic modems for short

range Mote-based underwater sensor networks, in: Proc. of IEEEOceans, Singapore, May 2006.

[17] R. Jurdak, P.M.Q. Aguiar, P. Baldi, C.V. Lopes, Software modems forunderwater sensor networks, in: Proc. of Oceans’07, June 2007.

[18] R. Jurdak, A.G. Ruzzelli, G.M.P. O’Hare, C.V. Lopes, Reliable symbolsynchronization in software-driven acoustic networks, in: Proc. ofIEEE GlobeCom, Washington, DC, November 2007.

[19] N. Fruehauf, J.A. Rice, System design aspects of a steerabledirectional acoustic communications transducer for autonomousundersea systems, in: Oceans, vol.1, IEEE, 2000, pp. 565–573.

[20] F.H. Fisher, V.P. Simmons, Sound absorption in sea water, Journal ofAcoustic Society of America 62 (1977) 558.

[21] S.A.L. Glegg, R. Pirie, A. LaVigne, A study of ambient noise in shallowwater, Available: http://www.oe.fau.edu/acoustics/.

[22] F.H. Fisher, Effect of high pressure on sound absorption and chemicalequilibrium, Journal of Acoustic Society of America 30 (1958) 442.

[23] M. Hatler, C. Chi, Wireless sensor networks: growing markets,accelerating demand, On World Report, 2005, Available: http://www.onworld.com/html/wirelesssensorsrprt2.htm.

[24] A.E. Cetin, O.N. Gerek, Y. Yardimci, Equiripple FIR filter design by theFFT algorithm, IEEE Signal Processing Magazine 14 (2) (1997) 60–64.

[25] M. Stojanovic, Recent advances in high speed underwater acousticcommunications, Oceanic Engineering 21 (4) (1996).

[26] P. Chapman, D. Wills, G. Brookes, P. Stevens, Visualizing underwaterenvironments using multi-frequency sonar. IEEE Computer Graphicsand Applications, 1999.

[27] C.E. Shannon, A mathematical theory of communication, Bell SystemTechnical Journal 27 (1948) 379–423. 623–656.

[28] Rincon Research Corporation, BlackBook, Available: http://tinyos.cvs.sourceforge.net/tinyos/tinyo-1.x/contrib/rincon/apps/Blackbook5/.

[29] R. Jurdak, P.M.Q. Aguiar, C.V. Lopes, P. Baldi, A comparative analysisand experimental study on wireless aerial and underwater acousticcommunications, in: Proceedings of the IEEE InternationalConference on Digital Telecommunications (ICDT), Cap Esterel,France, August 2006.

[30] MoteIV Corporation, Available at: www.moteiv.com.[31] Tiny Operating System, UC, Berkeley, Available at: www.tinyos.net.[32] R. Black, Research needed on marine sound, BBC News Article,

Available: news.bbc.co.uk/2/hi/science/nature/4706670.stm.[33] R. Jurdak, Wireless Ad Hoc and Sensor Networks: A Cross-layer

Design Perspective, Springer-Verlag, 2007.

Raja Jurdak is a senior research scientist atCSIRO ICT centre in Brisbane Australia. Hereceived the BE degree in Computer andCommunications Engineering from theAmerican University of Beirut in 2000, the MSdegree in Computer Networks and DistributedComputing from University of California,Irvine in 2001, and the PhD degree in infor-mation and computer sciences at the Univer-sity of California, Irvine in 2005. From 2005 to2006, he was a postdoctoral researcher at theUniversity of California, Irvine. He was an Irish

Research Council on Science Engineering and Technology Embark fellow

from 2006–2008. His research interests sensor network communication

protocol design and application development. His current and pastresearch projects include acoustic positioning systems, underwater sen-sor networks, patient monitoring, pet tracking, and fire prevention ingranaries. Much of his work has incorporated cross-layer protocol design,which led him to author a book that was published by Springer in 2007titled ‘‘Wireless Ad Hoc and Sensor Networks: A Cross-layer Design Per-spective”. He has also authored or co-authored over 35 peer reviewedjournal and conference articles, and a pending patent. He is a member ofIEEE Communication Society.

Pierre Baldi received the PhD degree inmathematics from the California Institute ofTechnology in 1986. From 1986 to 1988, hewas a postdoctoral fellow at the University ofCalifornia, San Diego. From 1988 to 1995, hewas a member of the faculty and technicalstaff at the California Institute of Technologyand at the Jet Propulsion Laboratory (JPL). Hewas CEO of a startup company from 1995 to1999 and joined the University of California,Irvine (UCI) in 1999. He is currently a chan-cellor professor in in the School of Informa-

tion and Computer Sciences, director of the Institute for Genomics and

Bioinformatics, and member of the California Institute for Telecommu-nications and Information Technology (Calit2) at UCI. He is the recipientof a 1993 Lew Allen Award at JPL and a Laurel Wilkening Faculty Inno-vation Award at UCI. He is the author of more than 200 research articlesand four books, including Modeling the Internet and the Web-Probabi-listic Methods and Algorithms (Wiley, 2003). His research focuses onprobabilistic modeling and statistical inference, machine learning, bio-informatics, data mining, and communication networks. He is a seniormember of the IEEE.

Cristina Videira Lopes is an associate pro-fessor at the University of California, Irvine.She conducts research in programming lan-guages, software engineering, and pervasivecomputing. She is the recipient of a USNational Science Foundation CAREER Award.She is a member of the IEEE.