A QoS-aware Underwater Optimization Framework …pompili/paper/TW-Jul-13-1203.pdfIEEE TRANSACTIONS...

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, ACCEPTED FOR PUBLICATION 1

A QoS-aware Underwater Optimization Frameworkfor Inter-vehicle Communication using Acoustic

Directional TransducersBaozhi Chen and Dario Pompili,Senior Member, IEEE

Abstract—Underwater acoustic communications consume asignificant amount of energy due to the high transmission power(10−50 W) and long data packet transmission times (0.1−1 s).Mobile Autonomous Underwater Vehicles (AUVs) can conserveenergy by waiting for the ‘best’ network topology configuration,e.g., afavorable alignment, before starting to communicate. Dueto the frequency-selective underwater acoustic ambient noise andhigh medium power absorption – which increases exponentiallywith distance – a shorter distance between AUVs translates intoa lower transmission loss and a higher available bandwidth.Byleveraging the predictability of AUV trajectories, a novel solutionis proposed that optimizes communications by delaying packettransmissions in order to wait for a favorable network topology(thus trading end-to-end delay for energy and/or throughput).In addition, the solution proposed – which is implemented andcompared with geographic routing solutions and delay-tolerantnetworking solutions using an emulator that integrates under-water acoustic WHOI Micro-Modems – exploits the frequency-dependent radiation pattern of underwater acoustic transducersto reduce communication energy consumption by adjusting thetransducer directivity on the fly.

Index Terms—Underwater acoustic sensor networks, au-tonomous underwater vehicles, position uncertainty.

I. I NTRODUCTION

UNDERWATER Acoustic Sensor Networks (UW-ASNs)[2] have been deployed to carry out collaborative moni-

toring tasks including oceanographic data collection, disasterprevention, and navigation. To enable advanced underwa-ter explorations, Autonomous Underwater Vehicles (AUVs),equipped with underwater sensors, are used for informationgathering. Underwatergliders are one type of battery-poweredenergy-efficient AUVs that use hydraulic pumps to vary theirvolume in order to generate the buoyancy changes that powertheir forward gliding. These gliders are designed to rely onlocal intelligence with minimal onshore operator dependence.Acoustic communication technology is employed to trans-fer vital information (data and configuration) among gliders

Manuscript received July 5, 2013; accepted January 5, 2014.The associateeditor coordinating the review of this paper and approving it for publicationwas M.-O. Pun.

A preliminary shorter version of this paper is in Proc. ofIEEE Conferenceon Sensor, Mesh and Ad Hoc Communications and Networks (SECON), SaltLake City, UT, June 2011 [1]. With respect to the conference version, thecurrent manuscript has been substantially extended and revised.

This work was supported by the NSF CAREER Award No. OCI-1054234.The authors are with Department of Electrical and Computer

Engineering, Rutgers University, New Brunswick, NJ (e-mails:baozhi [email protected], [email protected]).

Digital Object Identifier 10.1109/TWC.2013.131203

Glider i’s

posi!on

a"er Δt

Glider i’s current

posi!on

Glider j’s current

posi!on

Des!na!on d’s

posi!on a"er Δt’’

Des!na!on d’s

current

posi!on

Glider j’s posi!on

a"er Δt’

Fig. 1. Glideri delays its transmission by∆t waiting for a better topologyso to improve the end-to-end (e2e) energy and/or throughputto destinationd.Wide arrows represent the packet forwarding routes and dashed/dotted simplearrows represent glider trajectories.

underwater and, ultimately, to a surface station where thisinformation is gathered and analyzed.

Position information is of vital importance in mobile un-derwater sensor networks as the data collected has to beassociated with appropriate location in order to be spatiallyreconstructed onshore. Even though AUVs can surface peri-odically (e.g., every few hours) to locate themselves usingGlobal Positioning System (GPS) – which does not workunderwater – over time inaccuracies in models for derivingposition estimates, self-localization errors, and drifting due toocean currents will significantly increase the uncertaintyinposition of underwater vehicle. Such uncertainty may degradethe quality of collected data and also the efficiency, reliability,and data rates of underwater inter-vehicle communications[3]. Besides the need to associate sensor data with 3D posi-tions, position information can also be helpful for underwatercommunications. For example, underwater geographic routingprotocols (e.g., [4], [5]) assume the positions of the nodesareknown. AUVs involved in exploratory missions usually followpredicable trajectories, e.g., gliders followsawtooth trajecto-ries, which can be used to predict position and, therefore, toimprove communication.

By leveraging the predictability of the AUVs’ trajectory, theenergy consumption for communication can be minimized bydelaying packet transmissions in order to wait for afavorablenetwork topology, thus trading end-to-end (e2e) delay for en-

1536-1276/13$31.00c© 2013 IEEE

2 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, ACCEPTED FOR PUBLICATION

ergy and/or throughput1. For instance, Fig. 1 depicts a scenariowhere glideri waits for a certain time period∆t [s] to savetransmission energy and to achieve higher throughput. Basedon j’s andd’s trajectory, glideri predicts a ‘better’ topologywith shorter links after∆t and postpones transmission in favorof lower transmission energy and higher data rate. This ap-proach differs from that proposed for Delay Tolerant Networks(DTNs), where delaying transmission becomes necessary toovercome the temporary lack of network connectivity [7].

To estimate an AUV’s position, in [8] we proposed a statis-tical approach to estimate a glider’s trajectory. The estimateswere used to minimize e2e energy consumption for networkswhere packets in the queue need to be forwarded right away(delay-sensitive traffic). In this work, we focus on delay-tolerant traffic and propose an optimization framework thatuses acoustic directional transducers to reduce the computationand communication overhead for inter-vehicle data transmis-sion. Moreover, we offer the distinction between two formsof position uncertainty depending on the network point ofview, i.e.,internal andexternal uncertainty, which refer to theposition uncertainty associated with a particular entity/node(such as an AUV) as seenby itself or by others, respectively(see Sect. IV-A for more details).

Based on the estimated external uncertainty, we pro-pose QUO VADIS2, a QoS-awareunderwateroptimizationframework for inter-vehicle communication usingacousticdirectional transducers. QUO VADIS is a cross-layer opti-mization framework for delay-tolerant UW-ASNs that jointlyconsiders the e2e delay requirements and constraints of under-water acoustic communication modems, including transducerdirectivity, power control, packet length, modulation, andcoding schemes. Specifically, the proposed framework usesthe external-uncertainty region estimates of the gliders andforwards delay-tolerant traffic with large maximum e2e delay,which includesClass I (delay-tolerant, loss-tolerant) trafficand Class II (delay-tolerant, loss-sensitive) traffic [5]. More-over, our cross-layer communication framework exploits thefrequency-dependent radiation pattern of underwater acoustictransducers. By decreasing the frequency band, transducerscan change their “directivity” turning from being almostomnidirectional (with a gain of≈ 0 dBi) – which is a desirablefeature to support neighbor discovery and multicasting, geo-casting, anycasting, and broadcasting) – to directional (withgains up to10 dBi) – which is useful for long-haul unicasttransmissions.

The contributions of this work are as follows:

• We offer the distinction between two forms of positionuncertainty (internal and external, depending on the viewof the different nodes). A statistical approach is thenproposed to estimate the position uncertainty and thisestimated uncertainty is then used to improve networkperformance.

• We exploit the frequency-dependent directivity of theacoustic transducer that is originally used as omnidirec-

1Due to the peculiar ‘V’ shape of the underwater acoustic ambient noiseand the high medium power absorption exponentially increasing with distance[6], a shorter distance between AUVs translates into a lowertransmission lossand a higher available bandwidth.

2“Quo vadis?” is a Latin phrase meaning “Where are you going?”.

tional transducer at one frequency to optimize networkperformance.

• We propose a distributed communication framework fordelay-tolerant applications where AUVs can conserveenergy by waiting for a ‘good’ network topology con-figuration, e.g., afavorable alignment, before starting tocommunicate.

The remainder of this article is organized as follows. Wefirst review the related work in Sect. II. Then we present theunderwater communication model in Sect. III and propose oursolution, QUO VADIS, in Sect. IV. In Sect. V, performanceevaluation and analysis are carried out, while conclusionsarediscussed in Sect. VI.

II. RELATED WORK

We review the following areas: geographical routing so-lutions, terrestrial and underwater DTN solutions, solutionsusing directional transducers and underwater cross-layeropti-mization solutions, which are related to our work.

Geographic routing protocols rely on geographic positioninformation for message forwarding, which requires that eachnode can determine its own location and that the source isaware of the location of the destination. Many geographicalrouting schemes, including some well-known ones such asMost Forward within Radius (MFR) scheme [9], Greedy Rout-ing Scheme (GRS) [10] and Compass Routing Method (CRM)[11], have been proposed for terrestrial wireless networks.In MFR, the message is forwarded to the neighbor that isclosest to the destination, while in GRS a node selects theneighbor whose projection on the segment from the source todestination is closest to the destination. In the CRM [11], amessage is forwarded to a neighbor whose direction from thetransmitter is the closest to the direction to the destination. In[12], a scheme called Partial Topology Knowledge Forwarding(PTKF) is introduced, and is shown to outperform otherexisting schemes in typical application scenarios. Based onthe estimate using local neighborhood information, PTKFforwards packet to the neighbor that has the minimal e2erouting energy consumption. These solutions are proposed forterrestrial wireless networks. In UW-ASNs, they may not workwell since propagation of acoustic signals is quite differentfrom that of radio signals. Moreover, localization underwateris generally more difficult than in the terrestrial environment.

Solutions for DTNs have been proposed for communi-cations within extreme and performance-challenged environ-ments where continuous e2e connectivity does not hold mostof the time [7], [13]. Many approaches such as ResourceAllocation Protocol for Intentional DTN (RAPID) routing[14], Spray and Wait [15], and MaxProp [16], are solu-tions mainly for intermittently connected terrestrial networks.RAPID [14] translates the e2e routing metric requirement suchas minimizing average delay, minimizing worst-case delay,and maximizing the number of packets delivered before adeadline into per-packet utilities. At a transfer opportunity, itreplicates a packet that locally results in the highest increasein utility. Spray and Wait [15] “sprays” a number of copies perpacket into the network, and then “waits” until one of thesenodes meets the destination. In this way it balances the tradeoff

CHEN AND POMPILI: A QOS-AWARE UNDERWATER OPTIMIZATION FRAMEWORK FOR INTER-VEHICLE COMMUNICATION. . . 3

between the energy consumption incurred by flooding-basedrouting schemes and the delay incurred by spraying only onecopy per packet in one transmission. MaxProp [16] prioritizesboth the schedule of packets transmissions and the scheduleof packets to be dropped, based on the path likelihoodsto peers estimated from historical data and complementarymechanisms including acknowledgments, a head-start for newpackets, and lists of previous intermediaries. It is shown thatMaxProp performs better than protocols that know the meetingschedule between peers. These terrestrial DTN solutions maynot achieve the optimal performance underwater as the char-acteristics of underwater communications are not considered.Hence, in the rest of this section, we focus on related solutionsfor UW-ASNs.

Several DTN solutions for UW-ASNs have been proposedin [17]–[20]. In [17], an energy-efficient protocol is proposedfor delay-tolerant data-retrieval applications. Efficient erasurecodes and Low Density Parity Check (LDPC) codes are alsoused to reduce Packet Error Rate (PER) in the underwaterenvironment. In [18], an adaptive routing algorithm exploitingmessage redundancy and resource reallocation is proposed sothat ‘more important’ packets can obtain more resources thanother packets. Simulation results showed that this approachcan provide differentiated packet delivery according to appli-cation requirements and can achieve a good e2e performancetrade-off among delivery ratio, average e2e delay, and en-ergy consumption. A Prediction Assisted Single-copy Routing(PASR) scheme that can be instantiated for different mobilitymodels is proposed in [19]. An effective greedy algorithm isadopted to capture the features of network mobility patternsand to provide guidance on how to use historical information.It is shown that the proposed scheme is energy efficient andcognizant of the underlying mobility patterns.

In [20], an approach called Delay-tolerant Data Dolphin(DDD) is proposed to exploit the mobility of a small numberof capable collector nodes (namely dolphins) to harvest infor-mation sensed by low power sensor devices while saving sen-sor battery power. DDD performs only one-hop transmissionsto avoid energy-costly multi-hop relaying. Simulation resultsshowed that limited numbers of dolphins can achieve gooddata-collection requirements in most application scenarios.However, data collection may take a long time as the nodesneed to wait until a dolphin moves into the communicationranges of these nodes.

Compared to the number of approaches using directionalantennae for terrestrial wireless sensor networks, solutionsusing directional transducers for UW-ASNs are very limiteddue to the complexity of estimating position and directionof vehicles underwater. Moreover, these solutions generallyassume the transducers are ideally directional, i.e., theyas-sume the radiation energy of the transducer is focused on someangle range with no leaking of radiation energy outside thisrange. For example, such transducers are used for localizationusing directional beacons in [21] and for directional packetforwarding in [22]. These solutions also use only one fre-quency. In our work, rather than using the ideal transducermodel, we consider the radiation patterns of existing real-world transducers at different frequencies in order to minimizeenergy consumption for communications.

A cross-layer optimization solution for UW-ASNs has beenproposed in [5], where the interaction between routing func-tions and underwater characteristics is exploited, resulting inimprovement in e2e network performance in terms of energyand throughput. Another cross-layer approach that improvesenergy consumption performance by jointly considering rout-ing, MAC, and physical layer functionalities is proposed in[4]. These solutions, however, do not consider uncertaintyinthe AUV positions and are implemented and tested only bysoftware simulation platforms and are not designed for delay-tolerant applications. On the contrary, we propose a practicaluncertainty-aware cross-layer solution that incorporates thefunctionalities of the WHOI Micro-Modem [23] to minimizeenergy consumption. Moreover, our solution is implementedon real hardware and tested in our emulator integrating WHOIunderwater acoustic modems.

III. N ETWORK MODEL

In this section we introduce the network model that oursolution is based on and state the related assumptions. Supposethe network is composed of a number of gliders, which aredeployed in the ocean for long periods of time (weeks ormonths) to collect oceanographic data. For propulsion, theychange their buoyancy using a pump and use lift on wingsto convert vertical velocity into forward motion as they riseand fall through the ocean. They travel at a fairly constanthorizontal speed, typically0.25 m/s [2]. Gliders controltheir heading toward predefined waypoints using a magneticcompass.

Assume the gliders need to forward the data they sensed toa collecting glider. The slow-varying and mission-dependent(and, for such reasons, ‘predictable’) trajectory of a glideris used in our solution to estimate another glider’s positionusing the position and velocity estimate from some timeearlier. A glider estimates its own trajectory and positionuncertainty using its own position estimates; the parametersof the estimated trajectory and internal-uncertainty region aresent to neighboring gliders. Using these parameters, thesegliders can extrapolate the glider’s current position and aconfidence region accounting for possible deviation from theextrapolated course.

The Urick model is used to estimate the transmission lossTL(l, f) [dB] as,

TL(l, f) = κ · 10 log10(l) + α(f) · l, (1)

where l [m] is the distance between the transmitter andreceiver, andf [Hz] is the carrier frequency. Spreading factorκ is taken to be1.5 for practical spreading, andα(f) [dB/m]represents an absorption coefficient that increases withf [6].

The Urick model is a coarse approximation for underwateracoustic wave transmission loss. In reality, sound propagationspeed varies with water temperature, salinity, and pressure,which causes wave paths to bend. Acoustic waves are alsoreflected from the surface and bottom. Such uneven propaga-tion of waves results inconvergence (or shadow) zones, whichare characterized by lower (or higher) transmission loss thanthat predicted by the Urick model due to the uneven energydispersion.


1

5

Horizontal Distance (x104 m)

23

4

Shadow zones

Fig. 2. Shadow zone scenario: the left subfigure represents the transmissionloss of node 1 located at the origin, while the right subfiguredepicts thesound speed profile used to derive the transmission loss (they-axis is thedepth, which has the same range used in the left; the blue, yellow and redareas denote large, medium and small path losses, respectively)

Due to these phenomena, the Urick model is not sufficientto describe the underwater channel for simulation purposes.The Bellhop model is based on ray/beam tracing, which canmodel these phenomena more accurately. This model canestimate the transmission loss by two-dimensional acousticray tracing for a given sound-speed depth profile or field, inocean waveguides with flat or variable absorbing boundaries.Transmission loss is calculated by solving differential rayequations, and a numerical solution is provided by HLSResearch [24]. Bellhop performs two-dimensional acousticray tracing for a given sound speed profile (or sound speedfield), in ocean waveguides with flat or variable absorbingboundaries, and generates output such as transmission lossand amplitude based on the theory of Gaussian beams [25].Due to space limitation, we cannot give a detailed description,but more details can be found in [26].

An example plotted using the Bellhop model is shown inFig. 2. Interesting enough, if node 1 sends a packet, node4 has higher probability of receiving the packet than node 3even though this node is closer. Because the Bellhop modelrequires more information about the environment than a gliderwill have, such as sound speed profile of the whole 3D regionand depths of receivers and ocean boundary, it is only used tosimulate the acoustic environment for testing (relying on tracefiles with historic data). Hence, the proposed solution usesthe Urick model in the cross-layer optimization (Sect. IV-B),which can be computed online on the glider.

We adopt the empirical ambient noise model presented in[6], where a ‘V’ structure of the power spectrum density (psd)is shown. The ambient noise power is obtained by integratingthe empirical psd over the frequency band in use3.

IV. PROPOSEDAPPROACH

Our proposed optimization is based on the estimation ofthe gliders’ trajectories and their external-uncertaintyregions.Therefore, in this section, we introduce the estimation ofexternal-uncertainty regions for gliders first. We then presentthe cross-layer design of our proposed framework.

3Note that in underwater acoustics, power (or source level) is usuallyexpressed using decibel (dB) scale, relative to the reference pressure levelin underwater acoustics1µPa, i.e., the power induced by 1µPa pressure.The conversion expression for the source levelSL re µPa at the distance of1 m of a compact source ofP watts isSL = 170.77 + 10 log10 P .

A. Internal and External Uncertainty

When an AUV surfaces to synch with the GPS satellitesand obtain its updated position, energy is spent and time iswasted (not to mention the risk that – as it has happened– the vehicle is stolen by pirates or damaged by vandals).In some applications such as coastal tactical surveillance, itis necessary not to surface or rely on surface vehicle. Forthese reasons, in order to maximize the success probabilityof a collaborative mission (and/or to minimize its duration),AUVs need to surface only when strictly needed or requiredby the mission itself. Another way to estimate an AUV’slocation is to rely on nodes or vehicles (autonomous or not)with accurate position and use them as reference nodes forlocalization. Based on these reference locations, the AUVapplys localization algorithms such as range-based ones (e.g.,[27]) to estimate its own location. Some solutions such as[28], [29] are proposed to use a surface vehicle with accurateGPS information to localize a vehicle underwater, which stillrequires a vehicle to stay on the surface. In this work we aimat keeping the surfacing of mobile AUVs minimal withoutusing surface vehicles. Under such constraints, we proposealgorithms to estimate the AUV’s position and associateduncertainty, and we further use the estimate of position anduncertainty to optimize inter-vehicle communications.

In this subsection, we first offer the distinction between twotypes of position uncertainty, followed by the discussion onthe relationship between these two types of uncertainty. Thenwe present the statistical approach for external-uncertaintyestimation when gliders are used as AUVs and ocean currentsare unknown. Since the details have been presented in [1], wejust summarize them here.

Internal uncertainty refers to the position uncertaintyassociated with a particular entity/node (such as an AUV)as seen by itself. Existing approaches such as those usingKalman Filter (KF) [30] may not guarantee the optimalitywhen the linearity assumption between variables does nothold. On the other hand, approaches using non-linear filterssuch as the extended or unscented KF attempt to minimizethe mean squared errors in estimates by jointly consideringthe navigation location and the sensed states/features such asunderwater terrain features, which are non-trivial, especiallyin an unstructured underwater environment.

External uncertainty, as introduced in this work, refersto the position uncertainty associated with a particular en-tity/nodeas seen by others. Let us denote the internal uncer-tainty, a 3D region associated with any nodej ∈ N (N is theset of network nodes), asUjj , and the external uncertainties,3D regions associated withj as seen byi, k ∈ N , asUij andUkj , respectively (i 6= j 6= k). In general,Ujj , Uij , andUkj

are different from each other; also, due to asymmetry,Uij

is in general different fromUji. External uncertainties maybe derived from the broadcast/propagated internal-uncertaintyestimates (e.g., usingone-hop or multi-hop neighbor discoverymechanisms) and, hence, will be affected by e2enetworklatency and information loss.


Pjj(τ)

HL

(- sign)

HU

(+ sign)

R

Des!na!on d

Glider i

j

P1

PN PN

P1

^

^

Uij

Ujj

Uid

(a) Estimated internal-uncertainty region byj: a cylinder with circular bottomradiusR and heightHU −HL

Ujj

Glider i

Time T0

Time T1

Time T2 Time T0

Time T3

Des!na!on

(b) Change of internal-uncertainty region over time.

Fig. 3. External- and internal-uncertainty regions for gliders under the effect of unknown ocean currents.

The estimation of the external-uncertainty region4 Uij of ageneric nodej at nodei (with i 6= j) involves the participationof both i and j. Here we use the receivedUjj as Uij (adelayed version due to propagation delay, transmission delay,and packet loss). Better estimation ofUij involves estimationof the change ofUjj with time and is left as future work.We provide a solution for internal- and external-uncertaintyestimation when1) gliders are used (following a ‘sawtooth’trajectory) and2) ocean currents are unknown.

Internal-uncertainty estimation at j: Assume glidersestimate their own locations over time usingdead reckoning.Given glider j’s estimated coordinates,Pn = (xn, yn, zn)at sampling timestn (n = 1 . . .N ), as shown in [1], itstrajectory segment can be described asP (t) = P +−→v (t− t),where P = (x, y, z) = 1

N

∑Nn=1(xn, yn, zn) and −→v =

‖−−−→P1PN‖

‖(a∗,b∗,c∗)‖·(tN−t1)· (a∗, b∗, c∗). Here,[a∗, b∗, c∗]T is the sin-

gular vector ofN×3 matrix A = [[x1−x, . . . , xN−x]T , [y1−y, . . . , yN − y]T , [z1 − z, . . . , zN − z]T ] corresponding to itslargest absolute singular value,t = 1

N

∑Nn=1 tn is the average

of the sampling times, andPi is the projection of pointPi onthe line segment (Fig. 3(a)).

The internal-uncertainty region ofj is estimated as acylindrical region [1] U described by its radiusR and itsheightHU − HL, whereHU andHL – in general different– are thesigned distances of the cylinder’s top and bottomsurface (i.e., the surface ahead and behind in the trajectorydirection, respectively) to gliderj’s expected location on thetrajectory. In [8] we demonstrate that:

1) HL andHU can be estimated as

HL = H − tα,N−1S(H)

√

1 + 1/N

HU = H + tα,N−1S(H)

√

1 + 1/N, (2)

whereH =∑N

n=1 Hn/N is the mean of theseN samples,S(H) = [ 1

N−1

∑Nn=1(Hn −H)2]1/2 is the unbiased standard

deviation,1 − α is the confidence level, andtα,N−1 is the100(1 − α/2)% of Student’s t-distribution [31] with N − 1

4Note that, “internal uncertainty” is essentially the position probabilitydistribution (with corresponding distribution region) sensed by the vehicleitself, and “external uncertainty” is essentially the position probability dis-tribution (with corresponding distribution region) sensed by other vehicles.For simplicity, we also use “uncertainty region” to represent the probabilitydistribution and the corresponding region where the AUV is distributed for agiven confidence level.

degrees of freedom (hereHn is the n-th sample calculatedfrom Pn’s [8]); and

2) R is estimated by

R =

√N − 1S(R)

√

χα,2(N−1)

, (3)

whereS(R) = [ 1N−1

∑Nn=1(Rn −R)2]1/2, R = 1

N

∑Nn=1 Rn,

and χα,2(N−1) is the 100(1 − α)% of χ-distribution with2(N − 1) degrees of freedom (hereRn is the n-th samplecalculated fromPn’s [8]). As shown in Fig. 3(b),j’s internal-uncertainty region becomes smaller over time (fromT0 to T2),i.e., as more position estimates are acquired. Note that param-eterα in the above expressions gives the error probability ofthe uncertainty estimate and the impact of estimate error willbe evaluated in Sect. V.

External-uncertainty estimation at i: After receivingj’s trajectory and internal-uncertainty region parameters(P , t,−→v , HU , HL, R), glider i can update the estimate ofj’s external-uncertainty region. Because AUVs involved inmissions show predictable trajectories, information about thesawtooth segment can be used to derive the entire glidertrajectory through extrapolation assuming symmetry betweenglider ascent and descent. Due to packet delays and lossesin the network,j’s external-uncertainty regions as seen bysingle- and multi-hop neighbors aredelayed versions of j’sown internal uncertainty (Fig. 3(b)). Hence, when usingmulti-hop neighbor discovery schemes, the internal uncertainty ofa generic nodej, Ujj , provides alower bound for all theexternal uncertainties associated with that node,Uij , ∀i ∈ N .Hence we use the receivedUjj asUij (a delayed version dueto propagation delay, transmission delay, and packet loss).

B. Cross-layer Optimization for Delay-tolerant Applications

With the external-uncertainty regions, a glider needs toselect an appropriate neighbor to forward each packet to itsfinal destination. Because the major part of available energyin battery-powered gliders is generally devoted to propulsion,acoustic communications should not take a large portion ofthe available energy. Our proposed protocol minimizes theenergy spent to send a message to its destination and considersthe functionalities of a real acoustic modem for a practicalsolution. Specifically, we provide support and differentiatedservice to delay-tolerant applications with different Quality


of Service (QoS) requirements, from loss sensitive to losstolerant. Hence, we consider the following two classes oftraffic:

Class I (delay-tolerant, loss-tolerant).It may include mul-timedia streams that, being intended for storage or subsequentoffline processing, do not need to be delivered within strictdelay bounds. This class may also include scalar data or nontime-critical multimedia content such as snapshots. In thiscase, the loss of a packet is tolerable at the current hop, butits e2e PER should still be below a specified threshold.

Class II (delay-tolerant, loss-sensitive).It may includedata from critical monitoring processes that require some formof offline post processing. In this case, a packet must be re-transmitted if it is not received correctly.

Our protocol employs only local information to make rout-ing decisions, resulting in a scalable distributed solution (eventhough the destination information is required for routing,we can use the destination information learned from localneighbors to predict the position of the destination). It isasuboptimal solution instead of a global one since it relies onlocal information. The external-uncertainty regions obtainedas described in Sect. IV-A are used to select the neighborwith minimum packet routing energy consumption. Here, aframework using the WHOI Micro-Modem [23] is presented.This framework can be extended and generalized in such away as to incorporate the constraints of other underwatercommunication modems.

To be more specific, given the current timetnow [s] and amessagem generated at timet0 [s], glider i jointly optimizesthe time∆t [s] to wait for the best topology configuration,a neighborj∗, a frequency bandfij , transmission powerP

(i,j)TX (t) [W], packet typeξ, and number of frames5 NF (ξ),

so that the estimated energyEid(t) [J] to routem to destinedglider d’s regionUid is minimized and messagem reaches itwithin Bmax [s], the maximum e2e delay from the source tothe destination. We assume power control is possible in therange[Pmin, Pmax] although transmission power is currentlyfixed for the WHOI Micro-Modem. We anticipate more ad-vanced amplifier hardware will make this power optimizationpossible.

Here, Eid(t) is estimated by the energy to transmit thepacket to neighborj in one transmission, the average numberof transmissionsN (i,j)

TX (t) to sendm to j, and the estimatednumber of hopsN (j,d)

hop (t) to reach regionUid via j. We needto estimate the transmission power and the number of hops todestination. The external-uncertainty region is used to estimatethe number of hopsN (j,d)

hop (t) to d via neighborj and thelower bound of the transmission power as follows (Fig. 4). Letli,p1,p2(t) [m] be the projected distance of line segment fromito positionp1 on the line fromi to positionp2, andli,p(t) bethe distance fromi to positionp. N (j,d)

hop (t) is estimated by the

worst case ofli,p(t)/li,p1,p2(t), i.e., (8). The lower bound fortransmission power is estimated by the average transmissionpower so that the received power at every point inUij is abovethe specified thresholdPTH . The transmission power lowerbound is the integral of the product of the transmission power

5Each packet sent by WHOI Micro-Modem consists of a number of frameswhere the maximum number depends onξ.

Uij

Glider i

Des!na!on d

Uid

li,p2

p2

p1

li,p1,p

2

^

PRX (i,j,x,y,z)

Gij at distance

to (x,y,z)

PTX(i,j)

Glider j

pli,p

Fig. 4. Use of external-uncertainty region in the optimization framework.

-3 dB gain

z

horizontal

planez

Fig. 5. Picture of our underwater glider and radiation pattern of the BT-25UFtransducer.

to obtainPTH at a point inUij and the probability densityfunction (pdf) ofj to be at this point.

To estimate the received power, it is necessary to estimatethe transducer gains at the transmitter and receiver. To estimatethe transmitter’s gainGTX(θij , φij , fij), i needs to computethe radiation angles – the horizontal angleθij ∈ [−180, 180]and the vertical angleφij ∈ [−90, 90] with respect toj. Since the transducer is located on top of the underwaterglider (Fig. 5), the relative angles of two transducers can beestimated if the pitch, yaw, and roll angles of the glidersare known. Assume the initial position of the transduceris as shown in the top left corner of Fig. 6 (i.e., uprightposition), theni’s normalized transducer direction vector is−→ni = (0, 0,−1) with the horizontal plane z = z

(i)0 (de-

fined as the plane perpendicular to−→ni). While the glideris moving, its pitch, yaw, and roll angles are denoted byεi, ζi, and ηi, respectively. From geometry, the directionvector after rotation is

−→n′i = Qx(ηi)Qy(εi)Qz(ζi)

−→niT , while

the transducer’s horizontal plane is expressed as[0, 0, 1] ·

Qz(−ζi)Qy(−εi)Qx(−ηi)[x′, y′, z′]T = z

(i)0 , where z

(i)0 is

a constant, andQx(ηi), Qy(εi) andQz(ζi) are

1 0 00 cos ηi − sin ηi0 sin ηi cos ηi

,

cos εi 0 − sin εi0 1 0

sin εi 0 cos εi

,

cos ζi − sin ζi 0sin ζi cos ζi 00 0 1

,

respectively.With the position vector

−−→PiPj from i to j, we can de-

rive cosφij =−−−→PiPj

−−−→PiPj

‖−−−→PiPj‖·‖

−−−→PiPj‖

and cos θij =−−−→PiPj

−→v i

‖−−−→PiPj‖·‖

−→v i‖

,

where−−→PiPj is the projection of

−−→PiPj on the transducer’s

horizontal plane, is the inner product, and−→vi = ‖−→vi‖ ·[cos εi cos ζi, cos εi sin ζi, sin εi] = (a∗i , b

∗i , c

∗i ) is the velocity

vector of glideri as estimated in Sect. IV-A. As−→n′i is perpen-

dicular to the transducer’s horizontal plane, we havesinφij =

cos(90−φij) =−→n

′

i−−−→PiPj

‖−−−→PiPj‖

and−−→PiPj =

−−→PiPj − (

−−→PiPj

−→n′i) ·

−→n′i.

The transducer’s gain at receiverj, GRX(θji, φji, fij), can beestimated in a similar way.

Let Lm(ξ) bem’s length in bits depending on packet typeξ


x

yz

i

jθij

ϕij

εi

ζi

ηi

ηi

i

i’ s axis

Transducer

Glider hull

View from

glider’s front

90º-ϕij

ni→

PiPj

→

PiPj

→

Initial Transducer

Position

centroid

Plane perpendicular

to transducer

Fig. 6. Derivation of transducer angles from glideri to j.

andB(ξ) be the corresponding bit rate. The energy to transmitthe packet to neighborj in one transmission can therefore beapproximated byP (i,j)

TX (t) · Lm(ξ)B(ξ) . Overall, the optimization

problem can be formulated asP(i,d, tnow,∆tp): Cross-layer Optimization Problem

Given: Pmin, Pmax,Ξ,Ωξ, GTX(), GRX(), η, Bmax, PERe2emax

Computed: εi, ζi, εj , ζj ,Uij ,∀j ∈ Ni ∪ d (i.e., R(i)j ,H

(i,j)L ,H

(i,j)H )

Find: j∗ ∈ Ni, P(i,j)∗TX (t) ∈ [Pmin, Pmax],

ξ∗ ∈ Ξ, N∗F (ξ) ∈ Ωξ,∆t∗, f∗

ij ∈ [fL, fU ]

Minimize: Eid(t) = P(i,j)TX (t) ·

Lm(ξ)

B(ξ)· N

(i,j)TX (t) · N

(j,d)hop (t). (4)

In P(i,d, tnow,∆tp), Ni, Ξ, andΩξ denote the set ofi’sneighbors, the set of packet types, and the set of number oftypeξ frames respectively. The objective function (4) estimatesthe energy required to send messagem to the destinationregion Uid. To solve this problem, we need to derive therelationship between these variables. LetLF (ξ) [bit] be thelength of a frame of typeξ, LH [bit] be the length ofmessagem’s header,PER(SINRij(t), ξ) be the PER of typeξ at the Signal to Interference-plus-Noise RatioSINRij(t),TL(lij(t), fij) be the transmission loss for distancelij(t) andcarrier frequencyfij [kHz] – which is calculated using (1)– A\i be the set of active transmitters excludingi, andP

(i,j)TX (t) be the transmission power used byi to reachj, we

have the following formulas,

(class-independent relationships)

t = tnow +∆t; (5)

tTTL = Bmax − (tnow − t0); (6)

Lm(ξ) = LF (ξ) ·NF (ξ) + LH ; (7)

N(j,d)hop (t) =

maxp∈Uidli,p(t)

minp1∈Uij,p2∈Uidli,p1,p2(t)

; (8)

SINRij(t) =P

(i,j)TX (t) · 10Gij (lij (t),fij)/10

∑

k∈A\i P(k,j)TX (t) · 10Gij (lkj(t),fij)/10 +N0

;

(9)

Gij(lij , fij) = GTX(θij , φij , fij) +GRX(θji, φji, fij)

−LAMP (fij)− TL(lij , fij); (10)

θij = arcsin

−→n′i

−−→PiPj

‖−−→PiPj‖; (11)

φij = arccos

−−→PiPj −→v i

‖−−→PiPj‖ · ‖−→v i‖. (12)

Note thatN0 =∫ fUfL

psdN0(f, w)df is the ambient noise,where psdN0(f, w) is the empirical noise power spectraldensity (psd) for frequency band[fL, fU ] andw [m/s] is thesurface wind speed as in [6].tTTL is the remaining Time-To-Live (TTL) for the packet,LAMP (fij) [dB] is the power lossof the power amplifier atfij andPERe2e

max is the maximume2e error rate for packetm. In these relationships, (5) isthe time after waiting∆t; (6) calculates the remaining TTLfor messagem; (7) calculates the total message’s length; (8)estimates the number of hopsN (i,j)

hop (t) to reach destinationd; (9) estimates the SINR atj while (10) estimates the totaltransmission gain indB from i to j, including the transducergain at the transmitter and receiver, loss at the power amplifier,and transmission loss; (11) and (12) estimate the transducer’sradiation angles ofj with respect toi. The constraints forP(i,d, tnow,∆tp) are,

(class-independent constraints)

P(i,j)TX (t) ≥

∫

(x,y,z)∈Uij

PRX(i, j, x, y, z) · 10−Gij (lij (t),fij)/10

(13)

·gR(x, y) · gH(z)dxdydz;

PRX(i, j, x, y, z) ≥ PTH ; (14)

0 ≤ ∆t ≤ tTTL

N(i,j)TX (t) · N (j,d)

hop (t). (15)

In these constraints,PRX(i, j, x, y, z) is the received signalpower at the generic 3D location (x, y, z) when i transmitsto j. Last, gR(x, y) and gH(z) are the pdfs of the glider’sposition on the horizontal plane (i.e.,χ-distribution withdegree of2N−2) and on the vertical direction (i.e., Student’st-distribution with N − 1 degrees of freedom), respectively[8], PTH is the received power threshold so that the packetcan be received with a certain predefined probability. (IV-B)estimates the lower bound of the transmission power to coverthe external-uncertainty region so that the received powerisabove a pre-specified threshold, as accounted for in (14); (15)estimates the bounds of∆t, which must be less than themaximum tolerable delay at the current hop. To support thetwo classes of delay-tolerant traffic, we have the followingadditional constraints,

(additional class-dependent constraints)

Class I=

N(i,j)TX (t) = 1

1−[

1− PER(SINRij(t), ξ)]N

(j,d)hop

(t) ≤ PERe2emax

;

(16)

Class II =

N(i,j)TX (t) =

[

1− PER(SINRij(t), ξ)]−1 . (17)

The first constraint for Class I traffic forces packetm to betransmitted only once, while the second constraint guaranteesthe e2e PER ofm should be less than a specified thresholdPERe2e

max. The constraint for Class II traffic guarantees mes-sagem will be transmitted for the average number of times forsuccessful reception atj. By solving this local optimizationproblem every time the inputs change significantly (and notevery time a packet needs to be sent),i is able to select theoptimal next hopj∗ so that messagem is routed (using min-imum network energy) to the external-uncertainty regionUid

where destinationd should be. Obviously different objectivefunctions (e2e delay, delivery ratio, throughput) could beuseddepending on the traffic class and mission QoS requirements.Note that in fact our solution can be extended to serve twoother classes of traffic - 1) delay-sensitive, loss-tolerant traffic,


Solve P(i,d,tnow ,Δtp),

calculate Δtp’

time

tnow tnow+Δtp’ tnow+Δtp’+Δtp’’

i

j

k

Solve P(i,d,tnow , Δtp’),

calculate Δtp’’

Solve P(i,d,tnow , Δtp’’),

calculate Δtp’’’

Fig. 7. SolvingP(i,d, tnow,∆tp) every∆tp at i.

and 2) delay-sensitive, loss-sensitive traffic - by setting∆t to0 (the delay discussed here can be hours or days as theseAUVs move slow in the vast ocean).

To reduce the complexity, we can convertP(i,d, tnow,∆tp) into a discrete optimization problemby considering finite sets ofP (i,j)

TX and ∆t, which canbe taken to be a number of equally spaced values withintheir respective ranges. The problem then can be solvedby comparing the e2e energy consumption estimates ofdifferent combination of these discrete values. Assuming thattransmission power and time are discretized intoNP andNtime values, respectively, for the case of WHOI modem (3frequencies and 14 combinations of packet type and numberof frames [8]), the processor in nodei needs to calculate theobjective value42NP ·Ntime · |Ni| times in each round. Theembedded Gumstix motherboard (400 MHz processor and64MB RAM) attached to the Micro-Modem is adequate to solvesuch a problem. To further reduce the computation, insteadof running the solution for every packet, it will be rerunonly at tnow + ∆tp for the same class of traffic flow that issent fromi to the same destinationd. Here,∆tp is taken asthe minimum of the∆t values of the packets belonging tothe same class of traffic and the same destination, estimatedfrom the previous run. Figure 7 depicts an example of howP(i,d, tnow,∆tp) is solved ati. At time tnow, the problemis solved withj found to be the next hop tod. The minimumof the ∆t values of these packets belonging to the sameclass of traffic and the same destination observed beforetnowis ∆t′p. Packets ford will then be forwarded toj with thecalculated transmission power at the selected frequency banduntil tnow + ∆t′p. Then, the problem is solved again andk is found to be the next hop. The minimum∆t observedso far is ∆t′′p and, hence, the problem will be solved attnow +∆t′p +∆t′′p .

Once the optimal frequency band is selected,i needs tonotify j to switch to the selected band. A simple protocol canbe used as follows. All AUVs use the same frequency bandas the Common Control Channel (CCC) to tell the receiverwhich band is selected. A short packet or preamble with theselected band number is first sent by the transmitter using theCCC, followed by the data packet using selected frequencyband after the time for the transmitter and receiver to finishfrequency band switching. The receiver will first listen onthe CCC, switch to the selected band embedded in the shortcontrol packet or preamble, receive the data packet, and thensend back a short ACK packet to acknowledge the reception.Finally, both sides switch back to the CCC if the transmission

succeeds or the transmission times out. The time out period isset long enough to make sure the ACK packet replied withinthe transmission range will be received with specified proba-bility. Retransmission (with limited number) will be triggeredif the transmission times out. More sophisticated frequency-band switching protocols, which are out of the scope of thispaper, can be designed to improve network performance. Werely on the Medium Access Control (MAC) scheme with theWHOI modem to send the data. Since the speed of acousticwave underwater is very slow when compared with radiowaves, the propagation delay has to be considered in order toavoid packet collisions. However, it is difficult to estimate thepropagation delay since the positions are uncertain. It maynotimprove the performance much as the actual propagation delaymay be different from the estimation. Moreover, the inter-vehicle traffic underwater is generally low. So the problem ofpacket collisions is not severe and hence we can just use theonboard MAC scheme.

C. Discussion

Studying the impact of an unreliable wireless channel onnetworked systems (such as underwater acoustic channel) hasbeen a hot research topic over the years. Many theoreticalworks have been proposed to study the performance boundsof networked control systems or wireless sensor networkswhen the wireless channel is unreliable. Some works [32], [33]focused on the analysis or design of source encoding, channelencoding, decoding, and controller for optimality of systemperformance. In [32], a new concept called anytime capacityis defined to study the problem of communicating the delay-sensitive data of an unstable discrete-time Markov randomprocess through a noisy channel. Source coding, channel cod-ing and delay sensitivity are studied and a new source/channelseparation theorem is given for delay-sensitive data, which isshown to be useful in control systems. There are also worksthat focus on the analysis and design of optimal estimationand control in the networked systems. For example, the workin [34] seeks to synthesize the optimal information flow andcontrol under given communication network constraints. Ajoint design of the information flow and the control to achieveoptimal estimation and control is proposed, and it is shown tobenefit system stability and performance.

In this work, we focus on the optimization of inter-vehiclecommunications among networked mobile AUVs instead ofoptimal control performance. Moreover, we consider theconstraints of existing communication modems, which havelimited modulation and channel coding options. Our goal hereis to have a solution that will be used in practice to optimizethe inter-vehicle communications. A more thorough theoreticalanalysis of the proposed optimization framework, such as thestudy of capacity bounds for inter-vehicle communicationsandthe impact on AUV control performance, is left as a futurework.

V. PERFORMANCEEVALUATION

The communication solution is implemented and testedon our underwater communication emulator [8] as shownin Fig. 8. This underwater acoustic network emulator is


M-Audio Delta

1010LT Audio

Interface

PC #2

(Dell Opplex 755)

PC #1

(Dell Opplex 755)

USB Cables

Bo!om Layer:

Micro-Modem

Middle Layer: Modem

DSP Coprocessor

Top Layer:

Gumsx

Front View of

Micro-Modem

Micro-Modem

System:

Gumsx and Micro-

Modem

Fig. 8. Underwater communication emulator using WHOI Micro-Modems.

composed of four WHOI Micro-Modems [23] and a real-time audio processing card to emulate underwater channelpropagation. The multi-input multi-output audio interface canprocess real-time signals to adjust the acoustic signal gains,to introduce propagation delay, to mix the interfering signals,and to add ambient/man-made noise and interference. Due tothe limited number of Micro-Modems and audio processingchannels, we can only mix signals from up to three trans-mitters at the receiver modem (one as the receiver and theother three as the transmitters). Therefore, we calculate,selectfor transmission, and mix with ambient noise, only the threemost powerful signals the receiver will encounter. We leavethe simulation of more than three simultaneously transmittedsignals as a problem for further research.

We are interested in evaluating the performance of theproposed solution in terms of e2e energy consumption, e2ereliability (i.e., e2e delivery ratio), average bit rate ofa link,and overhead, under an environment that is described by theBellhop model (and the Munk acoustic speed profile as input).

Assume that a glider’s drifting (i.e., the relative displace-ment from the glider’s trajectory) is a 3D random processX(t), t ≥ 0 as the following [35]: 1) In the beginningof the deployment, the drifting is 0, i.e.,X(0) = (0, 0, 0);2) The drifting has independent increments, in that for all0 ≤ t1 < t2 < · · · < tn, X(tn) − X(tn−1), X(tn−1) −X(tn−1), . . . , X(t2)−X(t1), X(t1) are independent; 3) Thedrifting has stationary increments, in that the distribution ofX(t + s) − X(t) does not depend ont and is normallydistributed with zero mean and covariance matrixsσ2I3,where I3 is the 3 × 3 identity matrix, andσ is a scalingfactor that decides the magnitude of drifting. Note that thisdrifting model is ideal since the drifting in any of thex, y, zdirections is Gaussian. The consideration of realistic driftingpattern is left as future work. Emulation parameters are listedin Table I. The radiation pattern of the BT-25UF transducer(Fig. 5) is used in the emulations. Every 10 seconds, a packetis generated in each node. A glider is randomly selected as thecollector and half of the other gliders are randomly selectedto forward their packets towards it. For statistical relevance,emulations are run for 50 rounds and the average is plottedwith 95% confidence interval. Note that it actually is a scenariofor deep water. We will also evaluate the performance inshallow water, where acoustic waves propagate differently.

We are interested in evaluating the performance of our

TABLE IEMULATION SCENARIO PARAMETERS

Parameter ValueDeployment 3D region 2500(L)×2500(W)×1000(H)m3

Confidence Parameterα 0.05[Pmin, Pmax] [1, 10] WPacket TypesΞ 0, 2, 3, 5

Glider Horizontal Speed 0.3 m/sGliding Depth Range [0, 100] mCarrier Frequencies 10, 15, 25kHz

Bmax 10 hr

solution for the two classes of traffic in Sect. IV-B, usingeither the BT-25UF transducer or an ideal omni-directionaltransducer (with gain equal to0 dBi). We also want tocompare the performance of our solution, which delays thetransmission for optimal topology configuration, with thesolution without delaying the transmission. For convenience,we denote, respectively, QUO VADIS for Class I traffic usingthe BT-25UF transducer by ‘QUO VADIS I’, for Class I trafficusing the ideal omni-directional transducer by ‘QUO VADISI - OMNI’, for Class II traffic using the BT-25UF transducerby ‘QUO VADIS II’, for Class II traffic using the ideal omni-directional transducer by ‘QUO VADIS II - OMNI’, and thesolution with no delaying of the transmission (i.e.,∆t = 0 forP(i,d, tnow,∆tp)) by ‘QUO VADIS - ND’. We will alsocompare the performance of our solution with geographicalrouting solutions – MFR, GRS, CRM, and PTKF – and DTNsolutions – RAPID, Spray and Wait, and MaxProp – as reviewin Sect. II. To make the comparison fair, we use two variantprotocols for each of these solutions by adding the constraintsof the two classes of traffic to these solution. For example, wedenote the MFR solution with Class I constraints in (16) by‘MFR I’, and the solution with Class II constraints in (17) by‘MFR II’.

The following networking metrics are compared:• e2e energy consumption: the average energy consumed

to route one bit of data to the destination;• e2e delivery ratio: the number of data packets received

correctly over the number of data packets sent;• link bit rate : the average bit rate between a transmission

pair;• overhead: the number of bytes used for position and

control to facilitate the transmission of payload data.Emulations are done for different settings and the results

are plotted with 95% confidence interval and discussed in thefollowing subsections.

A. Comparison With Geographic Routing Protocols

We compare the performance of our solution with geo-graphic routing protocols in Figs. 9 and 10. As shown inthese two figures, we can see that QUO VADIS has betterperformance than QUO VADIS - OMNI and QUO VADIS- ND for the same class of traffic in terms of these threemetrics. By delaying packet transmissions to wait for theoptimal network topology, the e2e energy consumption isreduced while the e2e delivery ratio and link bit rate increase(e.g., with 5 gliders, the energy consumption for QUO VADISI is around 30% of that for QUO VADIS-ND). By exploiting


0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IQUO VADIS I − OMNIGRS IMFR ICRM IPTKF I

(a) Delivery ratio comparison

0 5 10 15 20 25 30 35 40 45 500

100

200

300

400

500

600

700

800

900

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)


(b) Energy consumption comparison

0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)


(c) Link bit rate comparison

Fig. 9. Performance comparison for Class I traffic withgeographic routing protocols.

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IIQUO VADIS II − OMNIGRS IIMFR IICRM IIPTKF II


0 5 10 15 20 25 30 35 40 45 500

100

200

300

400

500

600

700

800

900

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 10. Performance comparison for Class II traffic withgeographic routing protocols.

the frequency-dependent radiation pattern of the transducer,received signal power may obtained a gain of up to 20 dB,which we observed in the simulations. Hence QUO VADISusing the BT-25UF transducer has better performance thanthat using the omni-directional transducer. Due to the QoS re-quirements, retransmissions are needed to recover link errors,resulting in higher e2e delivery ration for Class II traffic thanfor Class I traffic. On the other hand, this leads to more energyconsumption.

Different versions of our QUO VADIS solutions also per-form better than geographic routing protocols GRS, MFR,CRM and PKTF. This is because that uncertainty in locationleads to errors in route selection, packet transmissions andtransmission power estimates. Also these geographic routingprotocols do not consider the propagation delay underwater,which results in degraded communication performance. In-teresting enough, we can see that among these geographicrouting protocols, PKTF offers the best performance. Thisis because it jointly considers the transmission power androuting to minimize the e2e energy consumption. Thereforeit performs better than the other geographic routing protocol,which only consider the distance or angle metrics for routing(not closely related to network performance). GRS gives theworst performance since it generally needs to forward thepacket to the node that is far from the transmitter, whichintroduces bad link performance. Similarly, CRM performsbetter than MFR as the CRM has less probability to forward

packets to node that is far away than MFR does.

B. Comparison with DTN Solutions

We further compare QUO VADIS with the DTN solutions– RAPID, MaxProp and Spray and Wait. As shown in Figs.11 and 12, QUO VADIS gives improved performance overRAPID, MaxProp and Spray and Wait. The is mainly due tothat these DTN solutions transfer packets once the neighborsare in the transmission range. Such schemes may be good forscenarios where the connectivity is intermittent. However, theperformance may not be optimal since this may not be thetime to achieve the best link performance. In contrast, QUOVADIS predicts and waits for the best network configuration,where nodes move closer for the best communications. Sothe e2e delivery ratio and link bit rate of QUO VADIS is thehighest while its energy consumption is minimal. Note thatamong these compared DTN solutions, RAPID performs thebest. This is because RAPID prioritizes old packets so theywon’t be dropped. MaxProp gives priority to new packets;older, undelivered packets will be dropped in the middle.Spray and Wait works in a similar way, which does not givepriority to older packets. On the other hand, Spray and Wait isslightly better than MaxProp. This is because in our scenario,the network connectivity is not disrupt. The way MaxProproutes based on the e2e delivery ratio estimation will be verydifferent from that Spray and Wait does, i.e., just transmits thepacket to a neighbor then lets the neighbor continue to forward


0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IQUO VADIS I − OMNIRAPID IMaxProp ISpray & Wait I


0 5 10 15 20 25 30 35 40 45 500

100

200

300

400

500

600

700

800

900

1000

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 11. Performance comparison for Class I traffic withDTN protocols.

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IIQUO VADIS II − OMNIRAPID IIMaxProp IISpray & Wait II


0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 12. Performance comparison for Class II traffic withDTN protocols.

0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

2000

2500

3000

Number of Gliders

Del

ay [s

]


(a) Class I traffic: e2e delay

0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

2000

2500

3000

3500

Number of Gliders

Del

ay [s

]


(b) Class II traffic: e2e delay

0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

2000

2500

3000

3500

4000

4500

Number of Gliders

Ove

rhea

d [b

ytes

per

hou

r]

QUO VADISRAPIDMaxPropSpray & WaitPKTFGRS/MFR/CRM

(c) Overhead per node

Fig. 13. Comparison of e2e delay and overhead.

it. Moreover, MaxProp still needs to pay for the overhead toobtain the global e2e delivery ratio information.

C. End-to-end Delay Comparison

To see QUO VADIS can meet the delay requirement of thedelay-tolerant traffic, we also calculate and plot the e2e delaysof these solutions. As shown in Figs. 13(a) and 13(b), QUOVADIS - ND gives the least e2e delay. Compared to QUOVADIS and QUO VADIS - OMNI, QUO VADIS - ND doesnot wait for the vehicles to move to the optimal configurationyet more retransmissions are necessary. As the vehicle speedis much slower than the acoustic speed, QUO VADIS - ND

still needs much less time than QUO VADIS and QUO VADIS- OMNI even though more retransmissions are needed (thusresulting in more communication delay). Similarly, the hugedifference between vehicle speed and acoustic speed leads tothe result that QUO VADIS and QUO VADIS - OMNI needmore time than the DTN protocols (RAPID, MaxProp, andSpray and Wait), especially when the number of vehicles issmall (where average inter-vehicle distance is large). On theother hand, by taking the position uncertainty into account,communications using QUO VADIS - ND is more reliablethan those using RAPID, MaxProp or Spray and Wait soless delay is incurred. QUO VADIS has less delay than QUO


VADIS - OMNI due to the improvement in communicationsby exploiting the directional transducer gain. Also Class IItraffic generally has more e2e delay than Class I due to theneed for retransmissions. Last, note that as the number ofgliders increases, the delays of QUO VADIS and QUO VADIS- OMNI drop quickly. This is because average inter-vehicledistance becomes smaller and the number of close neighborsincreases, which reduces the need for a glider to wait a longtime until a neighbor moves close.

D. Overhead Comparison

We plot and compare the overheads (per node) of theseprotocols in Fig. 13(c). Note that as QUO VADIS, QUOVADIS - ND, and QUO VADIS - OMNI work almost thesame way, i.e., the uncertainty region information is broadcastperiodically (here the period is taken to be 60s), their over-heads are the same and thus we use QUO VADIS in the figureto represent these variant versions. Similarly, nodes runningthe geographic routing protocols GRS, MFR and CRM onlyneed to periodically broadcast the position information sotheiroverhead is basically the same. Hence we use GRS/MFR/CRMto represent them.

Surprisingly, even though QUO VADIS achieves the bestnetwork performance, its overhead is not the biggest. Theprotocols with the larger overhead are RAPID and MaxProp.In order to work, RAPID needs the following control infor-mation: average size of past transfer opportunities, expectedmeeting times with nodes, list of packets delivered sincelast exchange, the updated delivery delay estimate based oncurrent buffer state, and information about other packets ifmodified since last exchange with the peer, which takes alarge number of bytes. MaxProp needs to exchange a listof the probabilities of meeting every other node on eachcontact, which is basically global information. It also hasthe neighbor discovery overhead. Compared to RAPID andMaxProp, QUO VADIS only needs to exchange the externaluncertainty information of itself and the destination node,which is obviously less. On the other hand, PKTF needs aprobe message that has five data fields. Only the nodes in theselected path are required to respond with a probe – whetherit is sent for the forwarding or reverse direction. The Sprayand Wait protocol reduces transmission overhead by spreadingonly a few number of data packets to the neighbors. The sourcenode then stops forwarding and lets each node carrying a copyperform direct transmission. In our emulation, we select thenumber to be one to make the comparison fair and hencethe overhead is small. Lastly, for the other geographic routingprotocols GRS, MFR and CRM, the nodes just need to knowthe geographic locations of the neighbors and the destination.Therefore the overhead required is the least. Note that hereitis not necessary to differentiate the two classes of traffic sincethe overhead difference is small.

E. Performance in Shallow Water

So far the results are obtained using the setting in Table I,which is for the deep water. We change the network scenarioto the shallow water scenario by setting the depth of the 3Dregion to 200m. In this shallow water scenario, the path

loss estimated by the Urick’s model is very different fromthat estimated by the Bellhop model. We had anticipated theperformance will degrade because of this mismatch. Surprisingenough, as shown in Fig. 14 and 15, we find the performance(in terms of e2e delivery ratio, energy consumption, and linkbit rate) in the shallow water is actually better. A more carefulanalysis reveals the reason – the existence of thesurface ductin the shallow water. Surface duct is basically a zone belowthe sea surface where sound rays are refracted toward thesurface and then reflected. The rays alternately are refractedand reflected along the duct out to relatively long distancesfrom the sound source. Hence the acoustic waves are relativelyconcentrated in the surface duct, leading to less path loss.Thisconsequently leads to improved network performance.

F. Performance using Different Uncertainty Update Intervals

Emulations so far have been fixing broadcast interval ofuncertainty region to 60s. Our last interest is to evaluatethe performance of the QUO VADIS variants when differentbroadcast intervals are used. Therefore we re-run the emula-tions for two more cases: i) half of interval (i.e., 30s); andii) double of interval (i.e., 120s). From Fig. 16 and 17, wecan see that the performance of the QUO VADIS variantsbecomes worse when the update interval is doubled. This isbecause when the interval is doubled, the position uncertaintyinformation becomes less accurate. This leads to larger errorin selection of neighbor for packet forwarding and estimationof transmission power. On the other hand, halving the intervalleads to improvement of performance due to the uncertaintyinformation is updated in a more timely manner (so rout-ing error becomes smaller and transmission power is betterestimated). However, this obviously leads to the overheadincrease. Therefore the tradeoff between overhead and metricssuch as delivery ratio, energy consumption and link bit rateshould be carefully considered for different applications. Herewe use “QUO VADIS - Half”, “QUO VADIS”, and “QUOVADIS - Twice” to denote the cases with update interval of30 s, 60 s and 120s, respectively.

To find out the optimal update interval, depending on theneed, we can define an objective function that jointly considersthe tradeoff between performance metrics such as the e2eenergy consumption and overhead. For example, to find theoptimal update interval for e2e energy consumption, we candefine an objective function asfobj(Ee2e, Re2e, O, |N |) =Ee2e · Re2e/(O · |N |), which characterizes the e2e energyconsumption per overhead bit per node. In this objectivefunction, Ee2e [J/bit] is the e2e energy consumption aspreviously defined,Re2e [bit/s] is the e2e bit rate,O [bit/s]is the overhead as previously defined, and|N | is the numberof gliders. Emulations are run for different update intervalsfor class I traffic and the results are plotted in Fig. 18. Fromthis figure, we can see that as update interval increases,fobjdecreases first and then increases. This is because that whenthe update interval is increasing from a small value, the redun-dant overhead generated is decreased, leading to decrease inthe energy spent in overhead and the decrease in e2e bit rate(due to the decrease in estimation accuracy). As the updateinterval increases more, the increase in uncertainty estimation


0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IQUO VADIS I − OMNIQUO VADIS − ND (200 m)QUO VADIS I (200 m)QUO VADIS I − OMNI (200 m)


0 5 10 15 20 25 30 35 40 45 500

100

200

300

400

500

600

700

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

2000

2500

3000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 14. Shallow water: performance comparison for Class I traffic.

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS − NDQUO VADIS IIQUO VADIS II − OMNIQUO VADIS − ND (200 m)QUO VADIS II (200 m)QUO VADIS II − OMNI (200 m)


0 5 10 15 20 25 30 35 40 45 500

100

200

300

400

500

600

700

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

500

1000

1500

2000

2500

3000

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 15. Shallow water: performance comparison for Class II traffic.

leads to more data retransmission (and therefore more energyconsumption). Hence the e2e energy consumption increasesagain. From Fig. 18(a), we can find the optimal update intervalfor different versions of our solution. We can see that QUOVADIS I has the largest optimal update interval and QUOVADIS I - ND has the smallest optimal update interval, sincethe increase of position estimation error (due to increase inupdate interval) can be offset by good communication perfor-mance. The optimal update intervals for different number ofgliders are also shown in Fig. 18(b). We can see that as|N |increase the optimal update interval also increase for QUOVADIS I. This is because the increase in estimation errorcan be offset by the increase of possible neighbors for packetforwarding. Similar results can be observed for class II traffic.Due to space limitation, we skip plotting them here.

VI. CONCLUSION

We proposed QUO VADIS, aQoS-aware underwateroptimization framework for inter-vehicle communication us-ing acousticdirectional transducers. Based on the trajectoryand position uncertainties of the AUVs, an AUV predicts afavorable network topology with relatively short links in thefuture and postpones transmission in favor of a lower trans-mission energy and a higher data rate. Communication energyconsumption is further reduced by exploiting the frequency-dependent radiation pattern of underwater acoustic transduc-ers. The proposed solution is implemented and tested in our

underwater communication emulator, showing improvementover some well-known geographic routing protocols and DTNprotocols in terms of e2e energy consumption, reliability,andlink bit rate.

REFERENCES

[1] B. Chen and D. Pompili, “QUO VADIS: QoS-aware underwaterop-timization framework for inter-vehicle communication using acousticdirectional transducers,” inProc. 2011 IEEE Communications SocietyConference on Sensor, Mesh, and Ad Hoc Communications and Net-works.

[2] I. F. Akyildiz, D. Pompili, and T. Melodia, “Underwater acoustic sensornetworks: research challenges,”Ad Hoc Networks (Elsevier), vol. 3,no. 3, pp. 257–279, May 2005.

[3] X. Xiang, Z. Zhou, and X. Wang, “Self-adaptive on demand geographicrouting protocols for mobile ad hoc networks,” inProc. 2007 IEEEInternational Conference on Computer Communications.

[4] J. M. Montana, M. Stojanovic, and M. Zorzi, “Focused beamroutingprotocol for underwater acoustic networks,” inProc. 2008 ACM Inter-national Workshop on Underwater Networks.

[5] D. Pompili and I. F. Akyildiz, “A cross-layer communication solutionfor multimedia applications in underwater acoustic sensornetworks,”in Proc. 2008 IEEE International Conference on Mobile Ad-hoc andSensor Systems.

[6] M. Stojanovic, “On the relationship between capacity and distance inan underwater acoustic communication channel,” inProc. 2006 ACMInternational Workshop on UnderWater Networks.

[7] K. Fall, “A Delay-Tolerant Network Architecture for Challenged Inter-nets,” inProc. 2003 ACM Special Interest Group on Data Communica-tion.

[8] B. Chen, P. Hickey, and D. Pompili, “Trajectory-aware communicationsolution for underwater gliders using WHOI micro-modems,”in Proc.2010 IEEE Communications Society Conference on Sensor, Mesh, andAd Hoc Communications and Networks.


0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS I − HalfQUO VADIS IQUO VADIS I − Twice


0 5 10 15 20 25 30 35 40 45 500

50

100

150

200

250

300

350

400

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 16. Uncertainty update interval: performance comparison for Class I traffic.

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Gliders

Del

iver

y R

atio

QUO VADIS II − HalfQUO VADIS IIQUO VADIS II − Twice


0 5 10 15 20 25 30 35 40 45 500

50

100

150

200

250

300

350

400

Number of Gliders

Ene

rgy

Con

sum

ptio

n (m

J/bi

t)



0 5 10 15 20 25 30 35 40 45 500

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

Number of Gliders

Link

Bit

Rat

e (b

its/s

)



Fig. 17. Uncertainty update interval: performance comparison for Class II traffic.

0 50 100 150 200 25040

50

60

70

80

90

100

Update Interval [s]

f obj [J

/bit/

node

]

QUO VADIS I − NDQUO VADIS IQUO VADIS I − OMNI

(a) Optimal update interval for different versions (N = 10)

0 50 100 150 200 25010

20

30

40

50

60

70

80

Update Interval [s]

f obj [J

/bit/

node

]

QUO VADIS I (N=10)QUO VADIS I (N=20)QUO VADIS I (N=30)QUO VADIS I (N=40)

(b) Optimal update interval for differentN

Fig. 18. Optimal update interval.

[9] H. Takagi and L. Kleinrock, “Optimal transmission ranges for randomlydistributed packet radio terminals,”IEEE Trans. Commun., vol. COM-32, no. 3, pp. 246–257, Mar. 1984.

[10] G. G. Finn, “Routing and addressing problems in large metropolitanscale Internetworks,” ISI, Storrs, CT, Tech. Rep., 1987.

[11] E. Kranakis, H. Singh, and J. Urrutia, “Compass routingon geometricnetworks,” in Proc. 1999 Canadian Conference on ComputationalGeometry.

[12] T. Melodia, D. Pompili, and I. F. Akyildiz, “On the interdependenceof distributed topology control and geographical routing in ad hoc and

sensor networks,”IEEE J. Sel. Areas Commun., vol. 23, no. 3, pp. 520–532, Mar. 2005.

[13] S. Burleigh, A. Hooke, L. Torgerson, K. Fall, V. Cerf, B.Durst, K. Scott,and H. Weiss, “Delay-tolerant networking: an approach to interplanetaryinternet,” IEEE Commun. Mag., vol. 41, no. 6, pp. 128–136, June 2003.

[14] A. Balasubramanian, B. Levine, and A. Venkataramani, “DTN routingas a resource allocation problem,” inProc. 2007 ACM Conference onApplications, Technologies, Architectures, and Protocols for ComputerCommunications.

[15] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and wait: an


efficient routing scheme for intermittently connected mobile networks,”in Proc. 2005 ACM SIGCOMM Workshop on Delay-Tolerant Network-ing.

[16] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine, “MaxProp:routing for vehicle-based disruption-tolerant networks,” in Proc. 2006Conference on Computer Communications.

[17] C. Y. Chan and M. Motani, “An integrated energy efficientdata retrievalprotocol for underwater delay tolerant networks,” inProc. 2007 IEEEInternational Conference on Engineering in the Ocean Environment(OCEANS).

[18] Z. Guo, G. Colombi, B. Wang, J.-H. Cui, and D. Maggiorini, “Adaptiverouting in underwater delay/disruption tolerant sensor networks,” inProc. 2008 IEEE/IFIP Conference on Wireless on Demand NetworkSystems and Services.

[19] Z. Guo, B. Wang, and J.-H. Cui, “Prediction assisted single-copy routingin underwater delay tolerant networks,” inProc. 2010 IEEE GlobalCommunications Conference.

[20] E. Magistretti, J. Kong, U. Lee, M. Gerla, P. Bellavista, and A. Corradi,“A mobile delay-tolerant approach to long-term energy-efficient under-water sensor networking,” inProc. 2007 IEEE Wireless Communicationsand Networking Conference.

[21] H. Luo, Z. Guo, W. Dong, F. Hong, and Y. Zhao, “LDB: localizationwith directional beacons for sparse 3D underwater acousticsensornetworks,”J. Networks, vol. 5, no. 1, pp. 28–38, Jan. 2010.

[22] P. Xie, J. H. Cui, and L. Lao, “VBF: vector-based forwarding protocolfor underwater sensor networks,” inProc. 2005 IFIP Networking.

[23] L. Freitag, M. Grund, S. Singh, J. Partan, P. Koski, and K. Ball,“The WHOI micro-modem: an acoustic communcations and navigationsystem for multiple platforms,” inProc. 2005 IEEE InternationalConference on Engineering in the Ocean Environment (OCEANS).

[24] M. Porter, “BELLHOP Gaussian Beam/Finite Element BeamCode,”http://oalib.hlsresearch.com/Rays/index.html.

[25] F. B. Jensen, W. A. Kuperman, M. B. Porter, and H. Schmidt, Com-putational Ocean Acoustics, 2nd ed. American Institute of Physics,1997.

[26] W. S. Burdic, “Underwater acoustic system analysis,” in Prentice-HallSignal Processing Series, A. V. Oppenheim, Ed. Prentice-Hall, 1984,ch. 2, p. 49.

[27] E. Olson, J. J. Leonard, and S. Teller, “Robust range-only beaconlocalization,” IEEE J. Ocean. Eng., vol. 31, no. 4, pp. 949–958, Oct.2006.

[28] M. F. Fallon, G. Papadopoulos, J. J. Leonard, and N. M. Patrikalakis,“Cooperative AUV navigation using a single maneuvering surface craft,”International J. Robotics Research, vol. 29, no. 12, pp. 1461–1474, Oct.2010.

[29] A. Bahr, J. J. Leonard, and M. F. Fallon, “Cooperative localization forautonomous underwater vehicles,”International J. Robotics Research,vol. 28, no. 6, pp. 714–728, June 2009.

[30] M. Blain, S. Lemieux, and R. Houde, “Implementation of aROVnavigation system using acoustic/Doppler sensors and Kalman filtering,”in Proc. 2003 IEEE International Conference on Engineering in theOcean Environment (OCEANS).

[31] G. Casella and R. L. Berger,Statistical Inference, 2nd ed. DuxburyPress, 2001.

[32] A. Sahai, “Anytime information theory,” Ph.D. dissertation, Mas-sachusetts Institute of Technology, Cambridge, MA, Feb. 2001.

[33] D. Teneketzis, “On the structure of optimal real-time encoders anddecoders in noisy communication,”IEEE Trans. Inf. Theory, vol. 52,no. 9, pp. 4017–4035, 2006.

[34] V. Gupta, “Distributed estimation and control in networked systems,”Ph.D. dissertation, California Institute of Technology, Pasadena, CA,2006.

[35] S. M. Ross,Introduction to Probability Models, 8th ed. AcademicPress, 2003.

Baozhi Chenreceived his B.S. degree from BeijingUniversity of Posts and Telecommunications, Bei-jing, China, in 2000. He received his M.S. degreefrom Columbia University in New York City in2003. In 2012, he received his Ph.D. degree inElectrical and Computer Engineering at Rutgers, theState University of New Jersey - New Brunswick,under the supervision of Prof. D. Pompili. Hisresearch interests lie in underwater communicationsand networking, communication and coordinationof autonomous underwater vehicles, and wireless

communications and networking, wireless sensor networks,and wirelessbody-area networks. He was a co-chair of work-in-progress poster session inACM WUWNet’12. He also served as technical program committee memberin BodyNets’10, GreenNets’11, and GreenNets’12, and as reviewer for ACMand IEEE journals and conferences.

Dario Pompili is an Assoc. Prof. in the Dept.of Electrical and Computer Engineering (ECE) atRutgers University, where he is the director of theCyber-Physical Systems Laboratory (CPS Lab) andthe site co-director of the NSF Center for Cloud andAutonomic Computing (CAC). He received a Ph.D.in ECE from the Georgia Institute of Technology(GaTech) in 2007. In 2005, he was awarded GaTechBWN-Lab Researcher of the Year. He had previ-ously received his ‘Laurea’ (integrated B.S./M.S.)and Doctorate degrees in Telecommunications and

Systems Engineering from the University of Rome “La Sapienza,” Italy, in2001 and 2004, respectively. In 2011, Dr. Pompili received the NSF CAREERaward to design efficient communication solutions for underwater multimediaapplications and the Rutgers/ECE Outstanding Young Researcher award. In2012, he received the ONR Young Investigator Program (YIP) award todevelop an uncertainty-aware autonomic mobile computing grid framework aswell as the DARPA Young Faculty Award (YFA) to enable complexreal-timeinformation processing based on compute-intensive modelsfor operationalneuroscience. He is a Senior Member of both the IEEE CommunicationsSociety and the ACM.

A QoS-aware Underwater Optimization Framework …pompili/paper/TW-Jul-13-1203.pdfIEEE TRANSACTIONS...

Documents

Transcript of A QoS-aware Underwater Optimization Framework …pompili/paper/TW-Jul-13-1203.pdfIEEE TRANSACTIONS...