SectOR: Sector-Based Opportunistic RoutingProtocol for Underwater Optical Wireless Networks

Abdulkadir Celik, Nasir Saeed, Basem Shihada, Tareq Y. Al-Naffouri, and Mohamed-Slim Alouini

Abstract—Underwater optical wireless communications(UOWC) is an emerging technology to provide underwaterapplications with high speed and low latency connections.However, it suffers from limited range and requires effectivemulti-hop routing solutions for the proper operation ofunderwater optical wireless networks (UOWNs). In thisregard, this paper proposes a distributed Sector-basedOpportunistic Routing (SectOR) protocol. Unlike the traditionalrouting techniques which unicast packets to a unique relay,opportunistic routing (OR) targets a set of candidate relays byleveraging the broadcast nature of the UOWC channel. OR isespecially suitable for UOWNs as the link connectivity can bedisrupted easily due to the underwater channel impairments(e.g., pointing errors, misalignment, turbulence, etc.) and seacreatures passing through the transceivers’ line-of-sight. In suchcases, OR improves the packet delivery ratio as the likelihood ofhaving at least one successful packet reception is much higherthan that in conventional unicast routing. Contingent upon theperformance characterization of a single-hop link, we obtaindistance progress (DP) and expected (DP) metrics to evaluatethe fitness of a candidate set (CS) and prioritize the membersof a CS. Since rateØerror and rangeØbeamwidth tradeoffsyield different candidate set diversities, we develop a candidateselection and prioritization (CSPA) algorithm to find theoptimal sector shaped coverage region by scanning the feasiblesearch space. Moreover, a hybrid acoustic/optic coordinationmechanism is considered to avoid duplicate transmission ofthe relays. Numerical results show that SectOR protocol canperform even better than an optimal unicast routing protocol inwell-connected UOWNs.

I. INTRODUCTION

Emerging underwater applications with ambitious qualityof service demands require high speed, long range, and lowlatency underwater networking solutions [1]. However, suchgoals pose a daunting challenge for most electromagneticfrequencies because of the hostile channel impediments of theaquatic medium. Underwater acoustic communication (UAC)is a proven and widely accepted technology adopted by manycommercial, scientific and military applications. Albeit its om-nidirectional and long communication range, acoustic systemsare not suffucient for many underwater applications due to itslow achievable rates [2]. Due to the low propagation speed ofacoustic waves (1500 m/s), experienced high latency disruptsthe proper functioning of long-range applications especiallyfor real-time operation and synchronization tasks [3].

Recently, underwater optical communication (UOWC) hasgained attention by the advantages of higher bandwidth, lowerlatency, and enhanced security [4]. Nonetheless, the mainrestrictions of UOWC systems are directivity and limited

Authors are with Computer, Electrical, and Mathematical Sciences andEngineering Division (CEMSE) at King Abdullah University of Science andTechnology (KAUST), Thuwal, 23955-6900, KSA.

communication range which is mainly driven by the absorp-tion and scattering effects of the underwater environment.Therefore, provision of multihop underwater optical wirelessnetworks (UOWNs) is of the utmost importance to reapthe full benefits of the broadband UOWC at far distances.Designing advanced routing protocols tops the list of opennetworking problems as it couples medium access controlissues with unique physical layer characteristics of UOWCs.First and foremost, existing routing protocols developed foromnidirectional terrestrial wireless sensor networks (WSNs)and underwater acoustic networks (UASNs) cannot be usedfor UOWNs in a plug-and-play fashion. Due to the directednature of the light propagation, coverage region of a lightsource is in a sector shape whose central angle (i.e., thedivergence angle of the light beam) and radius (i.e., com-munication range) are inversely proportional. Hence, settinga wide divergence angle (e.g., light emitting diodes) allowsreaching many nearby neighbors whereas employing a narrowdivergence angle renders communicating with a distant node.While the latter requires less number of hops to reach thedestination at the cost of equipping the transceivers withaccurate pointing-acquisitioning-tracking (PAT) mechanisms,the former may operate without PAT at the expense of a highernumber of hops and power consumption.

Apart from the traditional routing protocols that unicastpackets to a unique next-hop forwarder, opportunistic routing(OR) broadcasts packets to a set of candidate nodes. Tradi-tional routing techniques retransmit lost packets which areeventually discarded after a certain number of retransmission.In such a case, by leveraging the broadcast nature of thecommunication, OR involves other candidate nodes in for-warding the packets toward the destination. For instance, Fig.1 demonstrates two different routes: The former is the routewhen the highest priority node (green) successfully receivesthe packet while the latter is over the second highest prioritynode (red) when the highest priority fails to receive packetcorrectly. Hence, OR improves the packet delivery ratio as thelikelihood of having at least one successful packet reception ismuch higher than that in conventional unicast routing. In thisrespect, OR is especially suitable to UOWNs because of theconnection interruptions caused either by underwater channelimpediments (e.g., pointing errors, misalignment, turbulence,etc.) or sea creatures passing through the transceivers’ line-of-sight. Nonetheless, OR requires cooperation and coordinationamong the candidate nodes in order to avoid duplicate trans-missions and collisions.

Recent efforts on UOWC can be exemplified as follows:In [5], three types of UOWC links are modeled: line-of-sight, modulating retroreflector, and reflective links. Assuminga Poisson point process based spatial distribution, Saeed et.

al. analyzed the k-connectivity of UOWNs [6]. Akhoundiet. al. proposed and investigated an interesting adaptation ofcellular code division multiple access (CDMA) to UOWNs [7],[8]. In [9], authors characterized the performance of relay-assisted underwater optical CDMA system where multihopcommunication is realized by chip detect-and-forward method.Similarly, Jamali et. al. consider the performance anaylsisof multihop UOWC using decode-and-forward (DF) relay-ing [10]. Modeling and end-to-end performance analysis ofmultihop UOWNs is addressed under both DF and amplify-and-forward methods [11]. Albeit their valuable contributions,these works do not deal with the effective UOWN routingprotocols which is of utmost importance to extend the limitedcommunication range of UOWCs.

In this paper, we propose a distributed Sector-basedOpportunistic Routing (SectOR) protocol. Being inspired bythe sector-shaped coverage region of the light sources, theSectOR finds the path by only exploiting the local topologyinformation in a distributed and greedy manner. Contingentupon the performance characterization of a single-hop linkquality (i.e., data rate, packet delivery ratio, maximum range,etc), we obtain DP and EDP metrics to evaluate the fitness ofa candidate set (CS) and prioritize the members of a CS. SincerateØerror and rangeØbeamwidth tradeoffs yield differentcandidate set diversities, we develop a candidate selection andprioritization (CSPA) algorithm. By leveraging an adaptivebeam divergence, CSPA obtains the best CS by scanningthe sector-shaped coverage area. In order to overcome nodesparsity, a route alternation method is also considered. Tothe best of authors’ knowledge, this work is first to considerOR to alleviate hostile and unreliable underwater channelenvironment.

The remainder of the paper is organized as follows: SectionII introduces the system model and performance characteriza-tion of a single hop link. Section III provide the details of theproposed SectOR protocol. Section IV presents the numericalresults and Section V concludes the paper with a few remarks.

II. SYSTEM MODEL

A. Network Model

We consider a two-dimensional underwater optical wire-less network (UOWN) which consists of a single surfacestation/sink and M nodes/sensors as demonstrated in Fig. 1.Nodes are equipped with optical transceiver to enable theunderwater optical wireless communication (UOWC) in bothdirections. Light sources are assumed to be capable of adapt-ing their beamwidth and communication range by adjustingthe divergence angle [12]. Although optical transceivers areprimarily employed to deliver large volume of sensing datavia high speed UOWC links, the limited range and directivityof UOWC hinders its ability to serve as a reliable controlmedium for network management tasks. Thanks to its omnidirectional propagation characteristics, each node also has asingle acoustic transceiver to provide the network with a highlyconnected control links. The surface sink is responsible toaggregate data from sensors and disseminating this informationto mobile or onshore sinks. Since a sensor observation is

Cand

idat

e Se

t

Sink

Source

Route 1

Route 1

Route 2

Route 2

Fig. 1: Illustration of UOWNs and SectOR routing protocol.

generally useful only if it is geographically tagged to anaccurate sensing location, we assume that each node is awareof its own location (`i, i P r1,M s) along with some otherneighbors. Underwater location information can be obtainedeither by a fully optic [13], [14] or hybrid acoustic/optic [15]network localization methods. By default, all sensors keep theorientation of their body frames and pointing vector of theoptical transmitter to be directed toward the sink node.

B. Channel Model

According to the Beer’s law, absorption and scatteringeffects of the aquatic medium can be characterized by anextinction coefficient cp�q “ ap�q ` bp�q where �, ap�q,and bp�q denote the carrier wavelength, absorption coefficient,and scattering coefficient, respectively. Based on Beer-Lambertformula, the propagation loss factor between ni and nj isdefined as follows

BLji “ exp

#´cp�q dij

cosp'ji q

+, (1)

where dij is the Euclidean distance between the nodes and�ji is the angle between the receiver plane and the transmitter-

receiver trajectory, as depicted in Fig. 2 where nj is located atpoint A. In case of a perfect alignment, (3) reduces to BLj

i “exp t´cp�qdiju if nj is located at point B. Geometrical lossis a result of spreading the light beam to an area larger thanthe receiver aperture size Aj and can be given for a semi-collimated transmitter emitting a Gaussian beam by

GLji “

˜Aj cosp'j

i q✓i1{edij

¸2

(2)

where ✓i1{e is the is full width beam divergence angle of ni

that is measured at the point where the light intensity drops to

Long RangeNarrow Beam

Divergence Angle

n"T2

Short RangeWide Beam

Divergence AngleswT2

A

AM

A\

Pointing VectorB

dsw

dln

Fig. 2: Illustration of a single-hop link and the tradeoff betweendivergence angle and communication range.

1{e of its peak. In the case of a perfect alignment, (2) reducesto the approximation given in [16]. Accordingly, the path lossbetween ni and nj is given by the product of (1) and (2) as

PLji “

˜Aj cosp'j

i q✓i1{edij

¸2

exp

#´cp�q dij

cosp'ji q

+. (3)

which is merely based on the received ballistic photons whichpropagates without being disturbed by the scattering effects.That is, (3) neglects all of the scattered photons received bynj by assuming their total loss. By modifying [17, Eq. (4)],scattered rays can be taken into account as follows

PLji “

˜Aj cosp'j

i q✓i1{edij

¸2

exp

#´ cp�qdijcosp'j

i q

˜Aj cosp'j

i q✓i1{edij

¸↵+

(4)where ↵ is a correction coefficient which can be determinedbased on parameters such as cp�q, Aj , ✓i1{e, field-of-view(FoV) angle of the receiver, etc. By analyzing (4), one cangain a valuable insight into the tradeoff between divergenceangle and communication range. As illustrated in Fig. 2, awide divergence angle results in a short communication rangeso that the source can reach many neighbor nodes withinits proximity. On the other hand, a narrow divergence anglehelps to reach a distant receiver, which naturally requiresan agile and accurate PAT mechanism to sustain a reliablecommunication link.

C. Performance Characterization of a Single-hop Link

In this section, we characterize the performance of a singlehop link in terms of distance, reliability, and achievable rates.Assuming that photon arrivals follows a Poisson Process,photon arrival rate from ni to nj is given as [18]

fij “ P jrx⌘

jc�

RjiT}c

, (5)

where P jrx “ P i

tx⌘itx⌘

jrxPL

ji is the received power by nj ,

P itx is the transmission power of ni, ⌘itx (⌘jrx) is the trans-

mitter (receiver) efficiency of ni (nj), ⌘jc is the detectorcounting efficiency of nj , Rj

i is the data rate, T is pulseduration, } is Planck’s constant, and c is the underwaterspeed of light. As per the central limit theorem, the Pois-son distribution can be approximated by a Gaussian distri-bution if the number of received photons is large enough.

For intensity-modulation/direct-detection (IM/DD) with on-offkeying (OOK) modulation, BER of the link between ni andnj is given by [5]

BERji “ 1

2erfc

˜cT

2

´bf1ij ´

bf0ij

¯¸(6)

where erfcp¨q is the complementary error function, f1ij “

fij ` fdc ` fbg and f0ij “ fdc ` fbg are the numbers of

photon arrivals when binary 1 and binary 0 are transmittedrespectively, fdc is the additive noise due to dark counts,and fbg is the background illumination noise. Denoting �jias the underwater environment disruption probability (whichcan be characterized by environmental factors such as animalpopulation, air bubbles, oceanic turbulence, pointing errors,etc.), the packet error rate (PER) can be given as

PERji “ 1 ´ p1 ´ �jiq

´1 ´ BERj

i

¯L(7)

where L is the packet length. Accordingly, the packet deliveryratio (PDR) is given by

PDRji “ 1 ´ PERj

i “ p1 ´ �jiq´1 ´ BERj

i

¯L. (8)

Following from the fact that PERji • 0 and PDRj

i § 1, theupper bound on �ji is given by

max´PERj

i, 1 ´ p1 ´ PERjiq1{L

¯“ PERj

i • �ji (9)

where equality follows from L • 1. The upper bound in(9) constitutes PERj

i as a maximum tolerance towards theconnectivity disruption caused by the underwater environment.For a certain PER threshold, PER

ji, data rate between ni and

nj is then derived by using (5)-(7) as

Rji “ Pj

rx⌘cj �

T}c„´

erfc´1 p2aqa2{T `

bf0ij

¯2´ f0ij

⇢ (10)

where a “ 1 ´´

1´PERji

1´�jiq¯ 1

L

and PERji • �ji . At last, for a

given data rate Rji , the communication range between ni and

nj is obtained by using (3)-(6) as

dij “ˆ

2

p↵ ´ 1qcW0

„p↵ ´ 1qc

´2

21´↵ b

¯↵´12

⇢˙ 11´↵

(11)

where W0p¨q is the principal branch of product logarithm,

b “Rj

iT}c„´

erfc´1p2aq?

2{T`?

f0ij

¯2´f0ij

⇢

P itx⌘

itx⌘

jrx⌘c

j�

ˆ✓i1{e

Aj cosp�ji q

˙2

, and

c “ ´ cp�qcosp'j

i q

ˆAj

✓i1{e

˙↵

.

III. SECTOR: SECTOR-BASED OPPORTUNISTIC ROUTING

Unlike the traditional multihop routing protocols whichselect a unique forwarder at each hop, OR choose a set ofcandidates which can overhear the transmitted packets andforward them to the next-hop in a prioritized and coordinatedmanner. Along the path from the source node ns to thedestination node nd, the CS of the forwarder node ni is

denoted by Cisd. Among the members of Ci

sd, the forwardingcandidates is similarly presented by F i

sd Ñ Cisd. In what

follows, we focus on the design and provision of the proposedSectOR protocol.

A. Opportunistic Routing Metrics

OR metrics play a crucial role in the performance of thedeveloped protocol since it has a direct impact on the candidateset (CS) selection and prioritization of the CS members. Themain objective of the OR metric is to reduce the expectednumber of transmissions (ExNT) such that end-to-end delayand energy consumption is decreased. Based on the availablenetwork state information at each node, the OR metrics canbe classified into local and end-to-end metrics which requireinformation from neighboring nodes and the entire networktopology, respectively. In order to avoid the cost of updatingand storing the entire topology state, we consider local ORmetrics by assuming that each node has the location informa-tion of itself, its search space (i.e., one hop neighborhood),and the destination (i.e., the sink). One of the most commonlocal OR metrics is the distance progress (DP) which selectsand orders the candidate according to their closeness towardthe destination [19]. Accordingly, the DP metric is given by

DPijsd “ }`s ´ `d} ´ }`j ´ `d},@j P Ci

sd. (12)

which can be calculated using (11). In terrestrial WSNs andUASNs, the DP metric is limited to the scenario where avery far away candidate is selected merely based on itsproximity without accounting for the link quality. In UOWNs,the negative consequences of this limitation can be mitigatedby restricting the candidate set to the sector-shaped coverageregion. A more advanced OR metric is the expected distanceprogress (EDP) which considers the average DP [20]. Accord-ingly, the EDP metric can be formulated as

EDPisd “

ÿ

jPCisd

DPijsd

«PDRj

i

j´1π

k“1

PERki

�(13)

where the multiplicative term is the probability that nj cor-rectly delivers the packet when candidates with higher priorityfail to do so. Therefore, EDP accounts for both link qualityand distance advancement toward the sink at the same time.

B. Candidate Selection and Prioritization Algorithm (CSPA)

Prior to developing the CSPA, it is important to discussrateØreliability and rangeØbeam-width tradeoffs. The formeris common for any communication systems as data rate andPDR are inversely proportional to each other, which followsfrom (10). The latter is specific to OWC as link distancereduces as the divergence angle increases. Following from(11), these two tradeoffs are also coupled as the range is afunction of rate, PDR, and divergence angle. Therefore, vary-ing these parameters yields different neighborhood diversities,CS combinations, thus end-to-end routing performance.

By treating the rate and PDR as design parameters, SectORmanipulates the rangeØbeamwidth tradeoff to obtain a CSwhich delivers the best OR metric. Indeed, a CS can be

MaximumDivergenceAngle

MinimumDivergenceAngle

dmax

dmin

Fig. 3: Candidate Set Search.

identified as pair of range and divergence angle. For a givendivergence angle range ✓min § ✓1{e § ✓max, the maxi-mum and minimum distance can be obtained by substituting✓min and ✓max into (11), respectively. Denoting ?pd, ✓q as asector shaped communication region, the feasible candidatesearch space is given by S “ ?pdmax, ✓maxq ´ t@pd, ✓q P?pdmax, ✓maxq| ✓ R r✓pdq, ✓pdqsu which excludes the pointsthat does not fall within the feasible distance-angle pair, asshown by grey colored nodes in Fig. 3. Denoting the set offeasible CSs by CSi

sd “ tC1sd, ..., C

Kisd u, the first and the last

CSs can be given by ?pdmax, ✓minq and ?pdmin, ✓maxq whichare shown in black colored solid and dashed sectors in Fig. 3,respectively. The rest of CSs may be obtained by quantizingthe interval ✓1{e P r✓min, ✓maxs which may be computationallycomplex for higher resolution. However, this complexity canbe augmented by selecting only necessary quantization pointsbased on the node locations within the search space. As shownby colored sectors in Fig. 3, we widen ✓1{e starting from ✓min

and create a new CS whenever a new node is covered by thesector shaped coverage region. Accordingly, the best CS of ni

is determined by

Ci‹sd “

#Cksd P CSi

sd

ˇˇˇk “ argmax

1§`§Ki

#maxjPC`

sd

´Xij

sd

¯++(14)

where the OR metric X can be set either to DP or EDP. Incase of DP metric, forwarding strategy of Ci‹

sd is determinedby giving a higher priority to nodes closer to the destination.Thus, prioritization set can be given by

Pisd “ tk|DPik

sd ° DPilsd, k † l,@pk, lq P Ci‹

sdu. (15)

Similar to DP, EDP metric can also prioritize the nodes basedon distance as in (15). Alternatively, we propose an iterativeprioritization method which prefers the node with the highestexpected distance advancement, i.e.,

Pi,lsd “

$’’&

’’%

argmaxjPCi‹

sd

´DPij

sdPDRji

¯, l “ 1

argmaxjPCi‹

sd´îl´11 Pi,l

sd

´DPij

sd

”PDRj

i

±j´1k“1 PER

ki

ı¯, o.w.

(16)

Starting from the first node Pi,1sd , (16) selects the next priority

node as the one that delivers the maximum EDP among those

are not prioritized yet. In sparse networks, it is possible notto find any CS after the above CSPA steps. In such a case,SectOR checks the remaining nodes within the S (e.g., greynodes in Fig. 3) and align the transmitter toward the node withthe highest OR metric.

C. Candidate Coordination Method

For the candidate coordination method, we consider acompressed slotted acknowledgment for orchestrating thecandidate priority. During the data transmission, forwardingnodes are enforced to send ACK messages with a period ofdistributed interframe space (DIFS), ⌧DIFS , in order to informthe neighboring nodes regarding its use of the optical channel.The source node first emits a request to send (RTS) messagelight beacon to its candidates and listen to the acoustic controlchannel for a duration of ⌧DIFS to conclude on a clear to send(CTS) message. That is, all the candidate nodes will remainsilent if none of the candidate nodes detect a data transmissionactivity within their proximity, which will be inferred by the

source node as a CTS message. If at least one of the candidatesdetect an acoustic ACK message, it also sends an acousticACK message, which will be concluded by the source as anot CTS message. In case of a CTS situation, the sourcenode emits the data packet through the optical channel. Ifthe first priority candidate receives the packet, it broadcastsan ACK message with a delay of short interframe space(SIFS), ⌧SIFS , to inform others its successful reception andprevent lower-priority candidates from attempting a duplicatetransmission. In case of a failure, the highest priority node juststays idle. The second priority candidate waits for 2⌧SIFS aftersuccessful reception of the data packet. Then, it proceeds withdata transmission if does not detect the ACK message of thehigher priority candidate, otherwise drops the packet.

IV. NUMERICAL RESULTS

Throughout simulations, source and sink nodes are locatedat reference points of p0, 0q and p100, 100q, respectively. Theremaining nodes are uniformly distributed over the 100m ˆ

(a) (b)

(c) (d)

Fig. 4: Illustration of SectOR in comparison with the unicast DSP algorithm.

30 40 50 60 70 80 90 100# Nodes

4.5

4.6

4.7

4.8

4.9

5

5.1

5.2

5.3

5.4

5.5A

vg. T

ota

l ExN

T

DSPDPEDP

(a) Total ExNT

30 40 50 60 70 80 90 100# Nodes

5.5

5.6

5.7

5.8

5.9

6

6.1

6.2

6.3

6.4

6.5

Avg

. # H

ops

DSPDPEDP

(b) Number of Hops

30 40 50 60 70 80 90 100# Nodes

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Avg

. # F

ails

DSPDPEDP

(c) Number of Failures

Fig. 5: Performance of DP and EDP based SectOR in comparison with the unicast DSP.

100m simulation area. Obtained results are averaged over10, 000 random realizations. Unless it is stated explicitlyotherwise, we use the parameters listed in Table I whichis mainly drawn from [5]. Before presenting the numericalresults, let us illustrate the operation of SectOR in comparisonwith a benchmark which is set as the unicast Dijsktra’s shortestpath (DSP). Unlike the SectOR, the DSP calculates the shortestpath based on the available global topology information ofthe entire network. Therefore, the DSP sets the divergenceangle at minimum to reach maximum range with low errorperformance.

In Fig. 4, black solid lines represent the route calculatedby the DSP. Blue/red/green colored sectors correspond to thesearch space, maximum divergence angle with the minimumrange, and minimum divergence angle with the maximumrange, respectively. Yellow colored sectors depict the candidatesets obtained by the proposed CSPA. Fig. 4a shows an instancewhere the DP metric yields the same path with the DSPbenchmark. Fig. 4b is a clear example of how the SectORcan leverage the proposed CSPA to find out potential CSs toreach the destination. Notice that the fourth hop starting fromnode 32 cannot find any feasible CS if the pointing vector isaligned to the sink node. Therefore, as explained in the lastpart of Section III-B, node 32 points the transceiver toward theonly node (node 47) that lies within the SS, which is shownby cyan colored sector. By doing so, the SectOR was able toreach to the sink node via node 47. A better example of thiscase is shown in Fig. 4c where the second and third hop ishandled by nodes 14 and 12 as there are no nodes within thefeasible region of coverage. Finally, Fig. 4d demonstrate thenegative impact of the lack of global topology information onthe routing performance. Even though the path is routed overnode 24 by changing the pointing vector node 22, the SectORwas not able to reach to the destination.

Fig. 5 compares the performance of DP and EDP basedSectOR in terms of the total expected number of transmissions(ExNT), number of hops, and failures. While ExNT of theDSP from ni to nj is calculated by ExNTj

i “ 1PDRj

i

, the

ExNT of SectOR is given by ExNTji “ 1

1´±jPCi

sdPERj

i

. Fig.

5a shows that EDP based SectOR performs always betterthan DP as it combines distance progress and link reliability.

TABLE I: Table of Parameters

Par. Value Par. Value Par. ValuePtx 0.1 W } 6.62E ´ 34 T 1 ns⌘x 0.9 c 2.55E8 m{s R 1 Gbps⌘r 0.9 � 532E ´ 9 L 124 B⌘c 0.16 ep�q 0.1514 PER 0.1A 5 cm fbg 1E6 fdc 1E6� 0.01 ✓min 0.336 rad ✓max 2{3 rad

Although OR does not deliver a better performance than theDSP in sparse networks, it yields desirable results startingfrom 60 nodes. Fig. 5b shows that DP and EDP reaches tothe destination with more and less number of hops than theDSP before and after 60 nodes, respectively. Although it isnot distinguishable in the figure, one important detail to pointout is that DP requires always less number of hops than theEDP since it only cares for the distance progress. Finally,Fig. 5c clearly shows that DP and EDP suffers from the lackof global topology information in sparse network scenarios.Even though the DSP is also subject to the failures becauseof the limited network connectivity, its impact is much moresignificant for DP and EDP based SectOR as it operates ina greedy and distributed fashion. Indeed, Fig. 5c presentsan implicit but a crucial behavior which can be describedas follows: Number of failures becomes negligible startingfrom the 60 nodes where connectivity of the networks is ina desirable state. 60 nodes are also the point where DP andEDP starts delivering a better performance than the DSP. Thatis, having a global network information is not required forthe SectOR to provide better performance once the networkconnectivity is well established. Overall, Fig. 5 shows themerits and shortcoming of the SectOR protocol which canbe regarded as a potential solution for UOWNs.

V. CONCLUSIONS

In this paper, we developed a sector-based opportunisticrouting protocol for UOWNs. OR is especially suitable forUOWCs as hostile aquatic channel impairments can disruptthe established link connectivity. By leveraging the broadcastnature of the communication, backing up the broken linkconnectivity by engaging other users who also received the lostpacket may improve the system performance in a significant

extend. SectOR exploits the sector-shaped coverage regionof the light source and finds the path only exploiting thelocal topology information in a distributed manner. Numericalresults show that it can reach the performance level of a unicastoptimal routing especially for high node density levels.

REFERENCES

[1] N. Saeed, A. Celik, T. Y. Al-Naffouri, and M.-S. Alouini, “Underwateroptical wireless communications, networking, and localization: A sur-vey,” arXiv preprint arXiv:1803.02442, 2018.

[2] T. Ballal, T. Y. Al-Naffouri, and S. F. Ahmed, “Low-complexity bayesianestimation of cluster-sparse channels,” IEEE Trans. Commun., vol. 63,no. 11, pp. 4159–4173, Nov. 2015.

[3] Z. Zeng, S. Fu, H. Zhang, Y. Dong, and J. Cheng, “A survey ofunderwater optical wireless communications,” IEEE Commun. Surveys

Tuts., vol. 19, no. 1, pp. 204–238, Firstquarter 2017.[4] H. Kaushal and G. Kaddoum, “Underwater optical wireless communi-

cation,” IEEE Access, vol. 4, pp. 1518–1547, 2016.[5] S. Arnon, “Underwater optical wireless communication network,” Opti-

cal Engineering, vol. 49, pp. 49 – 49 – 6, 2010.[6] N. Saeed, A. Celik, T. Y. Al-Naffouri, and M. Alouini, “Connectiv-

ity analysis of underwater optical wireless sensor networks: A graphtheoretic approach,” in IEEE Intl. Conf. Commun. Workshops (ICC

Workshops), May 2018, pp. 1–6.[7] F. Akhoundi, J. A. Salehi, and A. Tashakori, “Cellular underwater wire-

less optical cdma network: Performance analysis and implementationconcepts,” IEEE Trans. Commun., vol. 63, no. 3, pp. 882–891, March2015.

[8] F. Akhoundi, M. V. Jamali, N. B. Hassan, H. Beyranvand, A. Minoofar,and J. A. Salehi, “Cellular underwater wireless optical cdma network:Potentials and challenges,” IEEE Access, vol. 4, pp. 4254–4268, 2016.

[9] M. V. Jamali, F. Akhoundi, and J. A. Salehi, “Performance character-ization of relay-assisted wireless optical cdma networks in turbulentunderwater channel,” IEEE Trans. Wireless Commun., vol. 15, no. 6,pp. 4104–4116, June 2016.

[10] M. V. Jamali, A. Chizari, and J. A. Salehi, “Performance analysis ofmulti-hop underwater wireless optical communication systems,” IEEE

Photon. Technol. Lett., vol. 29, no. 5, pp. 462–465, March 2017.[11] A. Celik, N. Saeed, T. Y. Al-Naffouri, and M. Alouini, “Modeling and

performance analysis of multihop underwater optical wireless sensornetworks,” in IEEE Wireless Commun. Netw. Conf. (WCNC), Apr. 2018,pp. 1–6.

[12] P. LoPresti, H. Refai, J. Sluss, and I. Varela-Cuadrado, “Adaptivedivergence and power for improving connectivity in free-space opticalmobile networks,” Appl. Opt., vol. 45, no. 25, pp. 6591–6597, Sep. 2006.

[13] N. Saeed, A. Celik, T. Y. Al-Naffouri, and M. Alouini, “Underwateroptical sensor networks localization with limited connectivity,” in IEEE

Intl. Conf. Acoust, Speech and Signal Process. (ICASSP), Apr. 2018, pp.3804–3808.

[14] ——, “Robust 3d localization of underwater optical wireless sensornetworks via low rank matrix completion,” in IEEE 19th Intl. Workshop

Signal Process. Adv. Wireless Commun. (SPAWC), Jun. 2018, pp. 1–5.[15] ——, “Energy harvesting hybrid acoustic-optical underwater wireless

sensor networks localization,” Sensors, vol. 18, no. 1, 2017.[16] J. Poliak, P. Pezzei, E. Leitgeb, and O. Wilfert, “Link budget for high-

speed short-distance wireless optical link,” in 2012 8th International

Symposium on Communication Systems, Networks Digital Signal Pro-

cessing (CSNDSP), Jul. 2012, pp. 1–6.[17] M. Elamassie, F. Miramirkhani, and M. Uysal, “Performance char-

acterization of underwater visible light communication,” IEEE Trans.

Commun., pp. 1–1, 2018.[18] S. Arnon and D. Kedar, “Non-line-of-sight underwater optical wireless

communication network,” J. Opt. Soc. Am. A, vol. 26, no. 3, pp. 530–539, Mar 2009.

[19] M. Zorzi and R. R. Rao, “Geographic random forwarding (geraf) for adhoc and sensor networks: multihop performance,” IEEE Trans. Mobile

Comput., vol. 2, no. 4, pp. 337–348, Oct. 2003.[20] A. Darehshoorzadeh and L. Cerda-Alabern, “Distance progress based

opportunistic routing for wireless mesh networks,” in 2012 8th Inter-

national Wireless Communications and Mobile Computing Conference

(IWCMC), Aug 2012, pp. 179–184.