Performance modeling and analysis of void-handling methodologies in underwater wireless sensor networks

Computer Networks 126 (2017) 1–14

Contents lists available at ScienceDirect

Computer Networks journal homepage: www.elsevier.com/locate/comnet

Performance modeling and analysis of void-handling methodologies in underwater wireless sensor networks Rodolfo W.L. Coutinho a,b,∗, Azzedine Boukerche a, Luiz F.M. Vieira b, Antonio A.F. Loureiro b a b

PARADISE Research Lab, School of Engineering and Computer Science (EECS), University of Ottawa, Ottawa, ON, Canada Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG, Brazil

a r t i c l e

i n f o

Article history: Received 7 November 2016 Revised 16 June 2017 Accepted 26 June 2017 Available online 27 June 2017 Keywords: Communication void region problem Underwater sensor networks Greedy upward routing

a b s t r a c t In this paper, we devise an analytical framework for the performance evaluation of communication voidhandling algorithms designed for underwater wireless sensor networks (UWSNs). Geographic and opportunistic routing (GOR) have been shown efficient for multi-hop data delivery in UWSNs. However, georouting suffers from a serious drawback known as communication void region. The communication void region problem occurs when a source node does not have a neighbor node in closer proximity to the destination that can continue forwarding the packet. Whenever a data packet reaches a node in a void region, the data packet should be re-routed from a void-handling procedure or discarded. In this paper, we model the three main methodologies that have been used for the design of void-handling algorithms for geographic and opportunistic routing protocols in UWSNs. Our proposed analytical framework considers the characteristics of underwater sensor networks, network density and traffic load, underwater environment and acoustic channel, as well as the characteristics of the power control, bypassing void region and mobility assisted void-handling paradigms. The devised analytical framework is aimed to fill the gap in the literature of analytical tools that allow the performance evaluation of the trade-offs of each paradigm along different scenarios of UWSNs. The proposed model provides insights for the further design of void-handling algorithms in different underwater application and sensor network configurations. Numerical results show that the widely used bypassing void region approach is not effectively for moderate- and high-density UWSN scenarios. Conversely, topology control-based approaches (power control and mobility-assisted) are preferable as they create additional links. However, the use of voidhandling procedures increased the network energy consumption, which made each paradigm unsuitable for specific scenarios revealed by the proposed modeling. © 2017 Elsevier B.V. All rights reserved.

1. Introduction Nowadays, underwater wireless sensor networks (UWSN) are promising technology for the monitoring of large areas of submarine environments. Traditionally, a UWSN is composed by autonomous sensor nodes deployed underwater and sonobuoys (sinks) deployed at the surface of the ocean. Each underwater sensor nodes is responsible to monitor and track events of interest in its vicinity. Moreover, they act collaboratively to deliver gathered data to sonobuoys. The sonobuoys are responsible for data collection from the sensors. The applicability of UWSNs ranges among scientific, industrial and military applications, such as ocean explo∗

Corresponding author at: Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG, Brazil. E-mail addresses: [email protected] (R.W.L. Coutinho), [email protected] (A. Boukerche), [email protected] (L.F.M. Vieira), [email protected] (A.A.F. Loureiro). http://dx.doi.org/10.1016/j.comnet.2017.06.027 1389-1286/© 2017 Elsevier B.V. All rights reserved.

ration and data collection, disaster prevention, seismic and tsunami monitoring, offshore exploration, pollutant content monitoring, and navigation assistance [1–4]. Despite the potential of UWSNs over traditional wire-line instruments, their wide implementation is still costly, and limited by several networking challenges, mostly arising from the use of underwater acoustic channel. Underwater acoustic communication suffers from large and variable delays, high bit errors, temporary losses of connectivity, limited bandwidth capacity, and high energy cost [5,6]. 1.1. Motivation The aforementioned channel characteristics introduce challenges into the design of routing protocols for UWSNs. Because of them, the use of traditional proactive routing is impractical, as it requires the maintenance of an up-to-date routing table to every node in the network; reactive routing is impractical, as it creates

2

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

routing paths on-demand [7]. In this context, geographic and opportunistic (GOR) routing paradigm has been shown promising for UWSNs [8–15]. In the GOR paradigm, there is no need for either the establishment or maintenance of complete routes to the destination, nor the transmission of routing messages to update states. Conversely, a sensor node uses local information about the position or depth of itself and its neighboring nodes to select a next-hop (or a set of them) node, which can advance the data packet towards the destination. However, geographic-based routing protocols suffer from a serious drawback known as communication void region. Communication void region occurs whenever the current forwarder node (void node) does not have a neighboring node advancing the data packet towards the destination. Whenever a data packet gets stuck at a void node, an alternative path should be determined, or it will be discarded. The communication void region problem degrades the performance of UWSNs. Data packets discarded at void nodes can compromise the application, which is already impaired by packet loss relative to the noisy and harsh nature of underwater wireless communication. Moreover, the overhead required for determining and maintaining alternative routing paths from void nodes to the destination increases data packet collision, retransmission, and energy consumption. This serious shortcoming has been widely investigated in radio frequency-based wireless ad hoc networks, as discussed in [16]. 1.2. Our contributions Several void-handling procedures were proposed in geographic routing protocols designed for UWSNs [8,11,17–19]. However, to the best of our knowledge, there is no analytical model for the performance evaluation of void-handling procedures, allowing scientists and practitioners to select the best paradigm when designing a void-handling procedure for the considered scenario of UWSN. This fact is worrying since analytical- and simulation-based extensive and deep studies are needed before a on-the-field UWSN routing protocol test. This is due to the high cost of ship missions for the deployment and maintenance of UWSNs in the ocean. In this paper, we proposed an analytical framework for the evaluation of the trade-offs of the three different strategies used to design void-handling algorithms for UWSNs: power control-, bypassing void region- and mobility assisted-based strategies. Our devised analytical framework take into consideration the characteristics of underwater sensor networks, such as network density and traffic load, underwater environment and acoustic channel, such as the noise and signal absorption, as well as the characteristics of the power control, bypassing void region and mobility assisted voidhandling methodologies. From our devised analytical framework, one might deeply study the trade-offs of each approach in the desired network scenario and configuration. It can be used to obtain insights for the further design of void-handling algorithms in particular deployments of UWSNs. In our proposed model, we devise the following performance evaluation metrics: percentage of nodes in the void region, which measures the efficiency of geographic data routing in terms of packet delivery, coverage rate, which measures the portion of the environment where sensor nodes can sense and deliver data packets using a void-handling procedure when necessary, network lifetime, which measures the efforts, in terms of energy, of void-handling techniques to re-route data packets from void regions through void-handling techniques. In addition, the performance of the void-handling techniques will be strictly related to their characteristics when re-routing data packets from void regions. Thus, we analyze the percentage of nodes with long-range communication, moved nodes and number of hops when power

control-, mobility assisted- and bypassing-based void-handling technique are used, respectively. This investigation allows us to better understanding in which UWSN configurations each voidhandling technique is more suitable. This work significantly enhances our previous work [20] by making the following contributions: • Based on our literature review, we propose a thorough classification of void handling algorithms for UWSNs in three main categories: power control-, bypassing void region-, and mobility assisted-based strategies. • We propose an analytical framework for evaluating the performance and trade-offs of void-handling algorithms in UWSNs. Our model take into consideration the characteristics of the network architecture, underwater acoustic communication and recovery strategies. • We performed numerical experiments for an extensive performance evaluation of the three commonly used strategies for the design of void-handling algorithms in underwater sensor networks. In our numerical analysis, we considered different network density scenarios, which are a critical factor impacting on the number of void nodes. Surprisingly, numerical results show that the widely used bypassing void region approach is not effectively for moderate- and high-density UWSN scenarios. Conversely, topology control-based approaches (power control and mobility-assisted) are preferable as they create additional links. However, the use of void-handling procedures increased the network energy consumption, which made each paradigm unsuitable for specific scenarios revealed by the proposed modeling. The results obtained provide important guidelines for the design of energy-efficient geographic routing protocols and void-handling algorithms for underwater sensor networks over several network density and load traffic conditions. The remaining of this paper is organized as follows. Section 2 presents our proposed classification of void handling algorithms for UWSNs, and reviews the works encountered in the literature. Section 3 describes some preliminary concepts used by our proposed model. Section 4 presents the analytical framework proposed to study the trade-offs of the void-handling algorithms in UWSNs. Section 5 presents the performance evaluation of the analytical model. Finally, Section 6 presents our conclusion and future work. 2. Literature review and void-handling algorithms classification In this section, we review the void-handling algorithms proposed for geographic and opportunistic routing protocols in underwater wireless sensor networks. Based on the main approach used to overcome void regions, we propose classify void-handling algorithms designed for UWSNs in three major categories: power control-, bypassing void region- and mobility assisted-based approach. However, before surveying these protocols, we properly define the communication void region problem and its related concepts that are recurrently used throughout the text. Definition 1 (Communication void region). This occurs when the current forwarder node cannot deliver data packets directly to the destination, and does not have a neighbor node closer to the destination than itself to continue forwarding. Definition 2 (Void node). These are the nodes located within communication void regions. When a packet becomes stuck in the void nodes, it should be either routed using a void handling algorithm, or discarded. Definition 3 (Void handling algorithm). This is the mechanism employed by the geographic routing protocol to route data pack-

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

3

drawbacks of each these categories, highlighting the main protocols encountered in the literature. These aspects are summarized in Table 1. 2.2. Bypassing void region-based approaches The majority of void handling algorithms for UWSNs are designed from the bypassing void region-based strategy, as shown in the example of Fig. 1b. Void handling algorithms using this strategy should discover and maintain an alternative routing path from the void node to a non-void node that can resume the greedy upward forwarding routing. The main disadvantages of this strategy are twofold. First, the network can experience excessive energy consumption of control packets. Second, data packets are routed along more hops, increasing the delay for delivering the packet, as well as network interference. Algorithm 2 shows a simple heuristic to route data packets along void regions based on the VAPR [10] routing protocol. In this algorithm, packets are routed circumventing the communication void region known by means of periodic beacons. Fig. 1. Void handling strategies. a) Communication void region scenario. b) Power control-based approach. c) Bypassing void region-based approach. d) Mobility assisted-based approach.

Algorithm 2 Bypassing void region–based void node recovery algorithm. 1:

ets to their destinations when they cannot be delivered using the greedy forwarding strategy. In the following, we present the traditional pressure-based routing without any void-handling procedure. Moreover, we provide a detailed discussion about geo-based routing protocols, designed for UWSNs, that implement a power control-, bypassing void region- and mobility assisted-based void-handling procedure, respectively.

2: 3: 4: 5: 6: 7: 8: 9: 10:

Algorithm 1 Greedy upward routing algorithm. 1: 2: 3: 4: 5: 6: 7: 8:

next_hopv ← undefined depthnh ← ∞ for u ∈ Ne (v ) do if depth(u ) < depthnh then next_hopv ← u depthnh ← depth(u) end if end for

11: 13: 14: 15: 16: 17: 18: 19: 21:

In traditional geographic routing using the greedy upward (pressure) forwarding strategy, the next-hop forwarder selection can be made according to Algorithm 1. Accordingly, each node selects its neighboring node closer to the surface of the sea (Lines 3–8). When a node has a data packet to send, it inserts the unique address of its next-hop forwarder into the packet header and broadcasts it. Nodes that have correctly received the packet compare whether they are the next-hop forwarder of the sender. The forwarder node will relay the packet towards the surface, to be received by a sonobuoy. The packet will be discarded by the other nodes. When a node is in a communication void region (Fig. 1a), it should employ a void handling algorithm or discard the packet. When a node is in a communication void region and cannot determine a next-hop forwarder according to Algorithm 1, a void handling algorithm should be used. In the following, we classify the proposed recovery algorithms in three categories as: bypassing void region-, power control-, and mobility assisted-based solutions. We discuss the general design principles, advantages and

&

12:

20:

2.1. Geographic-based routing protocol

 m: an incoming beacon message if m.seq_num > seq_num or (m.seq_num = seq_num m.hop_count + 1 < hop_count) then NDF _dir (sender ) ← m.DF _dir hop_count ← m.hop_count + 1 next _hopv ← m(sender ) if depth(v ) > depth(sender ) then DF _dir ← UP else DF _dir ← DOW N end if end if

22:

if DF _dir = UP then next _hopv ← undefined depthnh ← depth(v ) for u ∈ Ne (e ) do if depth(u ) < depthnh and NDF _dir (u ) = UP then next _hopv ← u depthnh ← depth(u ) end if end for end if

Herein, we discuss UWSNs’ routing protocols employing bypass void region-based recovery algorithm. In Vector-based Routing (VBF) protocol [8], data packets are routed along a virtual “routing pipe”. The routing pipe is formed by the vector, known as forwarding vector, calculated from the sender and destination location information. A node that has successfully received a packet continues to forward it, if its distance to the forwarding vector is less than a predefined threshold. Otherwise, the packet is discarded. The disadvantages of VBF arise from the node density inside the routing pipe. High densities cause redundant packet transmissions. Low densities lead to packet losses, due to the problem of void region. To mitigate the effects of node density in VBF, [21] proposed the Hop-by-Hop Vector-based Forwarding (HH-VBF) routing protocol. HH-VBF uses a different virtual pipe per-hop from each forwarder to the sink, to cope with the absence of nodes lying within the virtual forwarding pipe of VBF in sparse network scenarios. VBVA [18] routing protocol, also designed to improve VBF, attempts

4

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

Table 1 Proposed classification of void-handling procedures in UWSNs. Methodology

Basic principle

Advantages

Disadvantages

Power control

Void nodes increase their communication range to find a next-hop node. An alternative path circumventing the void region is determined.

Simple, no changes in the sensing coverage area. No changes in the sensing coverage area.

Void nodes are vertically moved to new locations.

No increment of the number hops and interference zone.

Increase energy consumption, packet error and collisions. Overhead to find alternative paths, long routing paths, which increase packet collisions and end-to-end delay. Change the initial sensing coverage.

Bypassing void region

Mobility assisted

to bypass the boundary of the void region by shifting the forwarding vector of the data packet, and by a back-pressure method when the vector-shifting method fails (convex voids). Hydrocast routing protocol [14] uses the pressure (depth) information of the nodes to greedily route data packets towards the surface of the sea. In each hop, the forwarder node selects a subset of its neighbors, named as candidates, with shallower depths than its own. The candidates continue the multi-hop routing in a prioritized way, such that a node will forward the packet if all the high priority candidates fail to do so. Priority is given according to both the progress of each node to the surface of the sea, and the probability of packet delivery from the source to each node. Sinks deployed at the surface are responsible for receiving packets. The communication void region problem occurs when a node lacks a neighbor closer to the surface. In this case, the void node uses controlled flooding to search for a node at a depth lower than its own, and explicitly maintains a path to that node to route the data packets along the void region. Noh et al. [10] proposed the Void-Aware Pressure Routing (VAPR) protocol. In VAPR, a data packet is routed along a directional path to a surface sonobuoy. This directional path is determined from beacon messages disseminated from the sonobuoys. When a node receives a beacon message, it updates its forwarding direction (upward or downward) according to the depth location of the sender (shallow or deep). Thus, each node becomes aware of whether it is in a communication void region. Void nodes route data packets bypassing the boundary of void regions by matching the neighbors’ forwarding direction with their own. 2.3. Power control-based approaches A simple way to cope with the communication void region problem is through the increment of the void node communication range, as depicted in Fig. 1c. In this strategy, the void node increases its communication range by means of the power control, to find a neighbor node that can continue forwarding the packet. Void handling algorithms, using this strategy, are quite simple. However, a high energy cost is incurred as the transmission power is increased. Moreover, packet collisions and, consequently, more retransmissions will take place, since the increment of the interference zone. Algorithm 3 shows a naive power control-based void handling algorithm. Each node starts to find a next-hop forwarder using the lowest transmission power level (Lines 4–9). The node will increase to the subsequent transmission power level, regardless of whether it is in a communication void region (Lines 10–12). This algorithm is repeated until either the node determines its next-hop forwarder, or the maximum transmission power level is reached. Finally, the neighbor should also be instructed to adjust its communication range, in order to avoid asymmetric links. Some routing protocols for underwater sensor networks adjust the power transmission to improve link reliability, as in CARP [22]. Herein, we are interested in those works in which the power control is used to find a next-hop forwarder. In Focused Beam Routing (FBR) protocol [17], the forwarder node multicasts a request to

Algorithm 3 Power control–based void node recovery algorithm. 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14:

Rc (v) ← next_level next_hopv ← undefined depthnh ← ∞ for u ∈ Ne (v ) do if depth(u ) < depthnh then next _hopv ← u depthnh ← depth(u ) end if end for if next _hopv = undefined then Rc (next _hopv ) ← Rc (v ) else Go to Line 1 end if

send (RTS) message, using the lowest power level. This message contains the location of both the forwarder and the destination. Each receiver node that lies within a cone of angle ± θ /2 emanating from the transmitter towards the final destination is a candidate to forward the packet, and responds to the RTS message. If there is no node within the transmission cone that can be reached at the current power level, the forwarder node increases its transmission power to the next level and multicasts the RTS packet. Al-Bzoor et al. [23] proposed a routing schema in which nodes are assigned to concentric layers centered on the sink. Data packets are forwarded to the nodes located in the next closest layer to the sink, until the packet reaches it. Adaptive power control is used by the nodes to increase their power level to ensure connectivity when the network is sparse, and decrease it to reduce the energy consumption when the network is dense and static. 2.4. Mobility assisted-based approaches A more recent strategy proposed to deal with the communication void region problem in underwater sensor networks is through mobility assistance (depth adjustment) of the underwater sensor nodes (Fig. 1d). Void nodes or auxiliary nodes are moved to new depth locations in order to resume the greedy upward forwarding strategy. The main disadvantage of the mobilityassisted methodology is the energy cost of moving void nodes. Algorithm 4 presents a void handling algorithm based on the proposals [11,24,25]. Taking advantage of node depth adjustment capabilities, O’Rourke et al. [26] proposed a multi-modal communication. In their proposal, sensor nodes equipped with acoustic and radio modems compute the trade-off between energy cost and data latency, based on the amount of data needing to be sent and the cost of surfacing; they then decide what technology should be used for transmitting the data. Depth-Controlled Routing (DCR) protocol [24] uses the mobility of the nodes by means of the depth adjustment to organize the network topology. The idea is to move some nodes to new depth

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

Algorithm 4 Mobility assisted–based void node recovery algorithm. 1:   set of void nodes 2:  D set of depth candidates to the void node 3: for u ∈  do D←∅ 4: for v ∈ V \  do 5: d ← D(u, v )  Euclidean distance considering only X,Y co6: ordinates if d ≤ Rc (u ) then 7: determine Z (u )new , such that 8: (X (u ) − X (v ))2 + (Y (u ) − Y (v ))2 + (Z (u )new − Z (v ))2 ≤ 9: Rc ( u ) if Z (v ) < Z (u )new then 10: D ← D ∪ {Z (u )new } 11: end if 12: 13: end if end for 14: if |D| > 0 then 15: 16: Z (u ) ← Z  ∈ D : ∀Z ∈ D, |Z (u ) − Z  | < |Z (u ) − Z |  ←  \ {u} 17: end if 18: 19: end for

locations in order to reduce the number of both void and disconnected nodes in scenarios of attached underwater sensor networks. Using a similar premise, in our previous work [12,25] we proposed a centralized (CTC) and distributed (DTC) topology control algorithm to move void nodes to new depths, with the goal of resuming the greedy forwarding routing in long-term monitoring of underwater sensor networks. GEDAR [11,13] routing protocol uses the depth adjustment capability of the nodes to move void nodes to new depth locations, in order to resume greedy forwarding in the scenario of short-term underwater mobile sensor networks. 3. Preliminaries Before we proceed further and present our proposed analytical framework, we wish to provide some preliminary background on the UWSN model used in our framework. 3.1. Network model We concentrate on static underwater sensor network scenarios, where self-buoyant sensor nodes are deployed with moorings (e.g., AquaNode [26] or Telesonar SM-975 SMART modems from Teledyne Benthos [27]). Sensor nodes are randomly deployed with moorings at the bottom of the sea. In those scenarios where mobility assisted is used for void-handling, the nodes can adjust their depth by means of inflatable buoys or winch-based apparatus by adjusting the tether, incurring in an energy cost of Em joules/meter, such as in the AquaNode [26]. As in the related work [8,11,22], we consider that each sensor node generates data packets of Ld bytes according to a Poisson process with the same parameter of λ pkts/min. Acknowledgment packets of size La bytes are used to confirm the successful reception of the data packet in each hop. Sonobuoys (or sinks) can be deployed at the surface of the sea in a planned manner. They are responsible for delivering the data packets collected from the sensor network to the monitoring center. As in [9,10], we assume that a packet is correctly received and can be delivered to the monitoring center if it arrives at a sonobuoy, since the sound propagates (speed of 1.5 × 103 m/s in water) five orders of magnitudes slower than radio (3 × 108 m/s in air).

5

The network topology is modeled by a graph G = (V, E ), where V is the set of nodes, and E is the finite set of links between them. The set V = {N ∪ S } of nodes is composed by the union of the sets of underwater sensor nodes, N = {N1 , . . . , N|N | } and surface sonobuoys S = {S1 , . . . , S|S | }. Each node v has a communication range Rc (v) and a carrier-sensing range Rh (v), Rh (v) ≥ Rc (v). The carrier-sensing range is the area, in addition to the communication range area, in which a transmission is heard, but the signal is so weak it may not be decoded correctly. In those scenarios where power control is used for void-handling, each node v can adjust its transmission power to the corresponding communication range values of the finite set  = {Rc1 , Rc2 , . . . , Rcn }. We define Ne (v) as the neighborhood set of v formed by the nodes with a distance less than or equal to Rc (v), and the Nh (v) formed by nodes in the region Rc (v ) − Rh (v ). When a node v transmits a packet, all nodes inside the sphere with the radius Rh (v), Ne (v)∪Nh (v), can hear the transmission. However, only the nodes with a distance less than or equal to Rc (v), Ne (v), will receive the signal with sufficient power strength to be correctly decoded with high probability. Thus, all nodes inside the carrier-sensing range of v will interpret the channel as busy, and will not be capable of accessing the medium during the v’s transmission [28]. Note that considering the node u ∈ Ne (v) as the next-hop of v, the set of nodes H (v ) = Ne (v ) \ {u} ∪ Nh (v ) will consume energy for the reception of the unintended v’s packets, even discarding them. Finally, let p(v, s) be the routing path from the node v to a surface sonobuoy s. We define the set P = { p(v, s ) : ∀v ∈ N } composed by the routing paths of all sensor nodes. 3.2. A review of the underwater acoustic propagation model We use the passive sonar equation to determine the transmission power in W/m2 for a target signal to noise ratio (SNR). The passive sonar equation characterizes the signal to noise (SNR) of an emitted underwater signal at the receiver as [29,30]:

SNR = SL − A(d, f ) − N ( f ) + DI ≥ SINRth ,

(1)

where SL is the source level, A(d, f) is the transmission loss, N(f) is the noise level, DI is the directivity index, and SINRth is a decoding threshold. In Eq. (1), all quantities are in dB re μ Pa. In this work, we use the Urick model [31] to capture the underwater acoustic signal attenuation. Whereas more realistic models for predicting acoustic attenuation can be found in the literature, such as the Rogers model [32] and Bellhop software [33], Urick model is simple to evaluate and can provide a useful approximation when parameters are properly chosen, as discussed in [34]. This model is based on empirical equations for acoustic power spreading and absorption loss. Accordingly, the path loss, which describes the attenuation on a single, unobstructed propagation path, over a distance d for a signal of frequency f due to large scale fading, is given as:

10 log A(d, f )/A0 = k · 10 log d + d · 10 log a( f )d ,

(2)

where k is the spreading factor and a(f) is the absorption coefficient. The absorption coefficient a(f), in dB/km for f in kHz, is described by the Thorp’s formula [35] given by:

10 log a( f ) =

0.11 × f 2 44 × f 2 + + 2.75 × 10−4 f 2 + 0.003. 1 + f2 4100 + f 2 (3)

The geometry of propagation is described using the spreading factor k. Its commonly used values are k = 2 for spherical spreading, k = 1 for cylindrical spreading, and for a practical scenario, k is given as 1.5. To properly fit the behaviors of the Roger and Bellhop models, we have used k = 1.7, as pointed out in [34].

6

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

The noise level N(f) that affects the underwater channel can be modeled using four sources: turbulence (Nt ), shipping (Ns ), waves (Nw ), and thermal noise (Nth ). The overall ambient noise is N ( f ) = Nt ( f ) + Ns ( f ) + Nw ( f ) + Nth ( f ). A detailed mathematical formulation to calculate the intensity of each source of noise for a given frequency f can be found in [29,36]. In Eq. (1), the source level (SL) is also related to the transmitted signal intensity It (μ Pa) at 1 m from the source, expressed as:

SL = 10 log

It . 1 μP a

It = 10SL/10 × 0.67 × 10−18 .

(5)

Finally, the transmission power Pt in Watts needed to achieve intensity It at a distance of 1 m from the source in direction to the receiver is:

Pt = 2π × H × It ,

(6)

where H is the depth of the water in meters. 3.3. Underwater packet delivery probability estimation In this section, we estimate the underwater packet delivery probability p(d, m) for a packet of m bits transmitted between any pair of nodes with a distance of d meters from each other. By doing so, we use Rayleigh fading to model small scale where the signalto-noise (SNR) over distance d, given by Eq. (1), has the following probability distribution [37]:



∞ 0



∞

0

(8)

Here, pe (X) is the probability of error for an arbitrary modulation at a specific value of SNR X. In this paper, we use BPSK (Binary Phase Shift Keying) modulation, which is widely used in state-ofthe-art acoustic modems [27,38]. In BPSK, each symbol carries a bit. In [39], the probability of bit error over distance d is given as:





where ρ = λTslot , is the system utilization factor. From the n nodes affecting one another, a successful transmission occurs when only one node has a packet to transmit for a given slot; that is, the queue of the other nodes is empty. This probability is given according to:

 

Ps =

n Pne (1 − Pne )n−1 . 1

(12)

From Eqs. (11) and(12), the collision probability Pc is [42]:

Pc = 1 − Ps − (1 − Pne )n .

(13)

In the remaining of our proposed model, the above defined collision probability Pc is used together with the packet error probability, 1 − p(d, m ), devised in Eq. (10), to estimate the expected number of transmission until a successful packet delivery between two any given nodes. We use the concept of coverage of an edge [43] to determine the number of concurrent nodes n during a transmission from the node u to v. Let D(u, r) denote the sphere centered at u with radius r, the coverage of the edge e = (u, v ) is defined to be the cardinality of the set of nodes covered by the sphere, induced by u and v as:

n = |{w ∈ N

| w is covered by D(u, Rc (u ))}∪ {w ∈ N | w is covered by D(v, Rc (v ))}|. (14)

(7)

pe (X ) pd (X )dX.

1 pe ( d ) = 1− 2

(11)

4. The proposed analytical framework

X 1 e− SNR(d ) . SNR(d )

The probability of error can be evaluated as:

pe ( d ) =

Pne = min{ρ , 1},

(4)

Solving Eq. (4) the transmitted signal intensity is given by Eq. (5), where the constant converts μ Pa to W/m2 .

pd ( X ) =

1/Tslot . For a more complete description see [42]. The probability that the node’s queue is not empty is:



SNR(d ) . 1 + SNR(d )

(9)

From Eq. (9), the delivery probability of a packet of m bits for any pair of nodes with distance d is:

p(d, m ) = (1 − pe (d ))m .

(10)

3.4. Packet collision probability estimation We consider that all nodes use slotted-ALOHA as the MAC layer protocol. The choice of this MAC protocol was made in light of its superior performance to random access and schedule-based MAC protocols in heavier traffic load scenarios [40,41]. In slottedALOHA, the time is divided into slots of fixed duration, Tslot , equal to the transmission time of the maximum frame plus the guard time. Thus, each node with a packet to send starts its transmission at the beginning of the slot it has selected. If a conflict occurs, the transmission is rescheduled to a random slot in the future. The MAC layer of each node can be modeled as a M/G/1 queue with an arrival rate of λ, as the packet generation rate in each node follows a Poisson process with the same parameter λ, and service rate of

Recall that the main contribution of this paper is to propose an analytical framework to study and investigate the methodologies used for the design of void-handling algorithms in UWSNs. To this end, we devise two important metrics to study the trade-offs of each methodology: sensing coverage rate and energy consumption. 4.1. Sensing coverage rate modeling The main objective of a sensor network is to collect data from variables and about events of interest. In this context, the sensing coverage area is an important metric, since it measures the fraction of the area of interest that is covered by sensor nodes that can deliver data packets from monitored events. In the following, we devise the sensing coverage rate metric to (i) evaluate the performance of void-handling algorithms in delivery data from sensor nodes in different locations of the environment, and (ii) measure the impact of nodes’ movement in the sensing task, when mobility assisted-based void-handling algorithms are employed. The sensing coverage rate of a sensor network is defined as the portion of the monitored area from the total area of interest. A point of the area of interest is said to be covered when its Euclidean distance to a sensor node is less than or equal to the sensing coverage radius [44]. Let nu be an underwater sensor node with a sphere sensing coverage of radius Rs , centered at ξ u (xnu , ynu , znu ). The sensing coverage area of nu is defined by the set of points:

Conu (Rs ) = {ξ ∈ R3 : ξu − ξ

≤ R s } ,

(15)

where  ξu − ξ  is the Euclidean distance between the locations ξ u and ξ . In order to determine the set of covered points according to Eq. (15), the area of interest D is divided into (Lx × Ly × Lz )/l3 cubes of side l. We define an indicator function f(ξ i , u) to determine whether a cube whose center point is ξ i (xi , yi , zi ) is covered

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

by the sensor nu ∈ Nn , as:



f ( ξi , n u ) =

1,

ξi ∈

0,

Conu

( Rs )

(16)

otherwise.

The points of the area monitored by a sensor node are considered to be covered only if the node covering them has a routing path to a surface sonobuoy, i.e., if it can deliver the generated packet to the destination. To model that, we define the indicator function to determine the reachable nodes nu ∈ Nn by:



g( n u , s ) =

1, 0,

∃s ∈ Ns | nu  s

(17)

otherwise.

Finally, the coverage rate defined as the fraction of the cubes covered by at least one sensor node with a path to a sonobuoy is given as follows:

(Lx ×Ly ×Lz )/l 3 Co =

i=0

maxnu ∈Nn { f (ξi , nu ) × g(nu , s )} Lx ×Ly ×Lz l3

.

(18)

t (u )), computed as and relaying of the neighbors’ data packets (Erel t (u ) + E t (u ). Let u denote a source node and u + 1 Etrans (u ) = Egen rel its next-hop forwarder. The u’s energy consumption per minute due to transmissions is given as: t Egen (u ) = λ · Pt · (N¯ u,u+1 + 1 )



4.2.1. Expected network energy cost Each sensor node v will route its data packets to a sonobuoy s through the path p(v, s ) = {v, v + 1, . . . , s}. The nodes belonging to the path p(v, s) are determined from the greedy upward forwarding (GUF) strategy or from the void handling algorithm when the node is a void node. Let u be a sensor node belonging to the path p(v, s), the node u + 1 (u − 1) corresponds to the next (predecessor) from the node u. Data packets will be discarded if their nodes do not have a next-hop forwarder, even after the run of the communication void recovery algorithms. In order to keep our model simple and tractable, we devise the lower bound of the expected network energy cost, for a network operation time of T minutes, as:

Etot =



 T Etrans (u ) + Erec (u ) + Eo (u ) + Ea (u ) ,

(19)

u∈N

where Etrans is the expected transmission energy cost; Erec is the expected reception energy cost; Eo is the expected overhearing energy cost; and Ea is the expected depth adjustment energy cost. In the following, we devise each expected energy cost considered in Eq. (19). 4.2.2. Expected transmission energy cost The energy spent for each node u due to data transmission is t ( u )) a result of both transmissions of its generated packets (Egen

L d

αB

,

(20)

where Pt is the transmission power given by Eq. (6), B is the bandwidth, α is the bandwidth efficiency, and N¯ u,u+1 is the number of unsuccessful attempts to deliver the packet from u to its nexthop forwarder u + 1. Assuming that the number of retransmissions is not limited, i.e., underwater monitoring applications requiring high-fidelity data collection, the expected number of transmissions (ETX) might be used to compute the number of attempts to successfully deliver the packet between the hops u and u + 1. As an acknowledgment packet is short compared to data packets, we assume their expected number of transmissions is equal to 1. The expected number of transmissions for data packets sent from node u to its next-hop forwarder u + 1 is given from Eqs. (10) and(13), as:

4.2. Energy consumption modeling Herein, we focus upon on the network energy consumption, as it is critical in UWSNs [45]. We model the most relevant sources of energy consumption in underwater sensor networks: communication and node depth adjustment when mobility assisted-based void-handling procedure is used. Regarding the energy consumption of communication, we approach only the cost relative to data and acknowledge packet transmissions and receptions, although the overhead to neighborhood discovery and routing path configuration is an important aspect to be considered in the design of routing protocols. This is because we are interested in evaluating resulting routing paths after running the next-hop forwarder selection algorithms derived from the power control-, bypassing void regions- and mobility assisted-based strategies; that is, we investigate how the peculiarities in the routing paths built from each void handling strategy affects the network performance. Moreover, independently of the void handling algorithm, the routing protocol will perform the neighbor discovery, which will be basically the same throughout our analysis.

7



1 , RT , p(d, m ) · (1 − Pc )

N¯ u,u+1 = min 1 −

(21)

where RT is the maximum number of tries. Unless otherwise specified, we set RT equals to 5 attempts. Besides its own packet transmission, the node u also will spend energy in relaying data packets from the neighbors where it is the next-hop forwarder. The relay energy cost is due to the transmission of the acknowledgment to confirm to the node u − 1 the successful data packet reception, and to the relaying of the data packet to the next-hop forwarder u + 1, given as:





t Erel (u ) = λ · Pt ·

(N¯ u,u+1 + 1 ) ×

p(v,s )∈P v=u

L L  a d + ×1 , αB αB u∈ p(v,s)

(22) where 1|u ∈ p(v, s) is the indicator function of the condition that the node u belongs to the path p(v, s), and La is the size of the acknowledgment packet. 4.2.3. Expected reception energy cost The energy consumption for data reception is a combination of the energy cost for the reception of acknowledgment packets and r (u ) + E r (u ). First, data packets sent by neighbors, as Erec (u ) = Egen rel the node will consume energy for the reception of the acknowledgment of its own transmitted data packets, given as: r Egen (u ) = λ · Pr

L a

αB

,

(23)

where Pr is the reception power, the value of which depends on the acoustic modem interface. Besides, for each path P where u is part and acts as a next-hop forwarder for the node u − 1, u will spend energy to receive the data packets that should be forwarded. This cost is given as follows: r Erel (u ) =

p(v,s )∈P v=u



(N¯ u−1,u + 1 ) ×

L d

αB

+

L  a

αB

× 1

u∈ p(v,s)

.

(24) 4.2.4. Expected overhearing energy cost Sensor nodes spend energy when they overhear unintended data transmissions. For each node u, this cost corresponds to the u’s unintended data transmissions performed by sensor nodes inside of u’s carrier-sensing range. The energy consumption relative

8

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

Fig. 2. Planned deployment of surface sonobuoys.

to overhearing is given as:

Eo (u ) = λ · Pr ·

(v,s )| 

| p

(N¯ vi ,vi+1 + 1 ) ×

p(v,s )∈P i=0

L L  a d + ×1 . αB αB u∈H (vi ) (25)

4.2.5. Expected depth adjustment energy cost Finally, sensor nodes spend energy during the depth adjustment regardless of whether they are using a mobility assistedbased void handling algorithm. Let u be a void node initially deployed at zu . If it is a void node, the recovery algorithm will move it to a new depth location, for instance zu . From u’s displacement δu = |zu − zu | and the energy cost in Joules per meter for the adjustment Em , the energy consumption of the node relative to the depth adjustment is given as

Eadj (u ) = δu × Em .

(26)

4.2.6. Significance of the proposed energy model While the above devised model of the energy consumption is simple, it is useful for quickly providing insights from the lower bound of the energy consumption of an underwater sensor network. Moreover, this simplicity is needed to make the model tractable. Most importantly, the proposed energy consumption model incorporates well the main features of void-handling methodologies that result in energy expenditure. For instance, the energy consumption due to the packet transmission (Eq. (20)) considers the different communication range of the nodes when power controlbased void-handling algorithm is used, through the transmission power Pt . The energy consumption, due to an increased number of hops, is accounted in individual nodes receiving data packets from the neighbors and sending to them ACK packet. Finally, the energy consumption of the depth adjustment of nodes, due to the mobility assisted void-handling technique, is considered in Eq. (26). Based on these models, it is possible to evaluate the energy cost of different approaches for void-handling in UWSNs. 5. Numerical results In this section, we instantiate our proposed analytical framework (Section 4) to evaluate the performance of the discussed void handling strategies for the design of communication void recovery protocols. By doing so, we implement Algorithms 1–4, discussed in Section 2, which are a non-void-handling, power control-, bypassing void regions-, and mobility assisted-based void-handling procedures, respectively. Herein, the main goal is to investigate the strengths and weaknesses of each void-handling paradigm according to several UWSN density scenarios and configurations. We use the R package1 to implement and instantiate the proposed analytical framework, as well as the considered voidhandling algorithms. In our numerical evaluation, we implement 1

https://www.r-project.org/.

Table 2 Model configuration. Parameters

Values

Sensing cov. radius (Rs ) Carrier sensing threshold Node battery Packet gen. rate (λ) Mac layer protocol Frequency (f) Packet payload size Bandwidth (B) Power reception (Pr ) Bandwidth efficiency (α ) Energy to move nodes (Ed ) SNRth Monitoring mission

250 m 10 dB 1122 Wh 1 pkts/min slotted-ALOHA 24 kHz 50 bytes 20 0 0 Hz 0.62 W 1 bps/Hz 15 J/m 10 dB 150 days

the Urick’s model, described in Section 3.2, to simulate the underwater environment and physical layer acoustic communication. The Urick’s model has been widely used in several UWSN simulators (e.g., Aqua-Sim [46] and SUNSET [47]). The transmission power at each node is calculated according to Eq. (6), based on the acoustic modem configuration given in Table 2. As an important input of our model, we generate several network topologies for each considered density. Thus, the obtained results for each configuration correspond to an average value of 25 runs of our model, with a confidence interval of 95%. 5.1. Model setup It is important to mention that in the current testbeds of UWSNs, only a few units of underwater sensor nodes are utilized and carefully deployed at specific locations [48,49]. This is due to the high cost of these devices. In our study, we tried to simulate underwater network scenarios closely to the real deployments. In addition, we used system configurations that were extensively considered in the literature when simulating well-known geographic routing protocol proposed for underwater sensor networks [10,11,19,24–26,42,50,51]. Despite low density scenarios, we consider random deployments of moderate network density to investigate the void-handling methodologies in future deployments of low cost underwater sensor nodes [52], as well as to be consistent with related work. Accordingly, the default simulation parameters are presented in Table 2. For the numerical analysis, we randomly deploy a varying numbers of underwater sensor nodes with moorings, i.e., non-mobile underwater nodes, ranging from 80 to 200 in a 3 D region measuring 2500 m × 2500 m × 1500 m. Surface sonobuoys are deployed in a predetermined manner. We considered different scenarios with 4, 9 and 16 surface sonobuoys deployed into equally divided surface grids, as showed in Fig. 2. The goal of varying the number of sonobuoys is to assess the upward greedily forwarding strategy that is proposed in the majority of geographic-based routing protocol designed for UWSNs (e.g., [9,10,19]).

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

9

Fig. 3. Percentage of void nodes.

In our numeric analysis, each node v has a communication range Rc (v) of 250 m. For the power control-based strategy, each node can set its transmission power to the equivalent communication range of  = {250, 350, 450} m. The transmission power of each desired communication range is calculated from Eq. (6). The carrier sensing range Rh (v) is determined from the transmission power and the carrier sensing threshold that was set to 10 dB. In the plots, the acronym GUF designates Algorithm 1, which is the greedy upward forwarding routing protocol without a recovery mode; GUF+PA designates Algorithm 3, which is the greedy forwarding routing protocol with power control-based void handling algorithm; GUF+VA designates Algorithm 2, which is the routing protocol with the bypassing void region-based void handling algorithm; and GUF+DA designates Algorithm 4, which is the routing protocol with the mobility assisted-based void handling algorithm. 5.2. Greedy forwarding analysis In this section, we evaluate the performance of the voidhandling methodologies in terms of reducing the fraction of void nodes. The main goal is to observe how the void-handling algorithms can outperform the traditional geographic-based routing without any recovery procedure in terms of nodes having a greedy geographic path to a sonobuoy. Fig. 3 depicts the fraction of nodes located into communication void regions, for different UWSN densities. First, it is important to note that void-handling methodologies do not perform well in very low density scenarios. This aims that for these cases, a planned deployment could be considered. Interestingly, the use of mobility assisted (GUF+DA) and power control (GUF+PA) outperformed the widely used approach of circumventing void regions (GUF+VA). For instance, GUF+PA and GUF+DA reduced in 20% and 25% the fraction of void nodes, when compared to the GUF and GUF+VA, in the low density scenario of 4 sonobuoys and 80 underwater nodes, respectively. The same trend can be observed as the network density increases. The reason for this trend is that bypassing void regionbased void-handling algorithm can determine an alternative path only if it exists on the network graph model, while power controland mobility assisted-based void-handling strategies can create additional paths to the previously disconnected nodes, as both strategies change the network topology. The aforementioned conclusion provides the following useful insight. For future design of geographic-based routing protocol in low to moderate UWSN density, topology control-based methodologies are more suitable to be used as void-handling procedures.

This is because they can organize the network topology in a such way that new routing paths will be created. Bypassing void region is suitable for scenarios of high density with few void nodes. In this case, it is highly probable that a void node will have a neighbor that can continue forwarding the packet through a path circumventing the void region. 5.3. Covered area of interest In this section, we evaluate how the use of void-handling approaches can improve the fraction of covered area. It is important to remember that in our analytical framework, a point of the area is covered if it is inside of the sensing range of at least one node that can deliver collected data to surface sonobuoys, as modeled in Section 4.1. Fig. 4 shows the fraction of the area of interest that is covered by at least one sensor node. As expected and confirmed in the plot, the fraction of covered points increases in accordance with network density. When the number of surface sonobuoys increases, the fraction of covered area increases. This is because more nodes become connected with a routing path to a sonobuoy. Moreover and more importantly, it is possible to observe that the topology control-based void-handling algorithms (GUF+DA and GUF+PA) perform better that GUF+VA, significantly increasing the covered area. For instance, when using GUF or GUF+VA the covered area is less than 1%. When topology control-based void-handling (GUF+DA and GUF+PA) is used, the covered area increases from 10% to up to 60% when network density increases. This is because these approaches increased the fraction of nodes that have a path to a given destination, as corroborated by Fig. 3. Thus, more nodes are capable of delivering collected data. To summarize the aforementioned result, the investigated topology control-based void-handling strategies for underwater sensor networks proved efficient in coping with the void node problem, from the perspective of improving path determinations to the destinations. As an immediate consequence of this result, the sensing coverage can be improved as more nodes can deliver the gathered data packets. However, the widely used bypassing voidhandling approach will not perform well especially in scenarios of low density. 5.4. Energy consumption analysis In this section, we evaluate how the void-handling approaches will impact the network lifetime. Here, we compute the network

10

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

Fig. 4. Percentage of the covered area of interest.

Fig. 5. Network lifetime.

lifetime as the time, in days, of the battery depletion of the first sensor node. It is important to mention that this metric has been widely considered to measure energy-efficient networking protocol for wireless sensor networks [53]. This metric is still more suitable for UWSNs characterized by low-density deployments. In this case, whenever a central node dies [54], the network may partition from the routing perspective. In addition, the choice of this metric is because the pure and direct analysis of the energy consumption values might not be suitable for observing the drawbacks of each void-handling strategy. The energy consumption is sensitive to several aspects of both the network and the application besides the topology. For instance, a packet transmission among dozens of hops may waste more energy than a single packet transmission at a high transmission power. This is due to the effects of packet reception and overhearing on the neighboring nodes. However, if we consider a high traffic load, the node using a high transmission power will deplete its battery firstly. Fig. 5 depicts the network lifetime. As shown in the plot, GUF and GUF+VA have a prolonged network lifetime as compared to the topology control void-handling approaches (GUF+DA and GUF+PA). This is due its poor performance in reducing the percentage of nodes in void areas, as showed in Fig. 3. GUF+DA outperforms GUF+PA mainly in medium- to high-density network scenarios of 16 sonobuoys. The reason for this is that in high density scenarios of increased sonobuoys, void nodes are moved to among short

distances (please refer to Fig. 7b), whereas GUF+PA still result in a significant number of nodes using more transmission power value, as shown in Fig. 6. Moreover, the use of high communication range in the GUF+PA approach, more collisions and retransmissions will take place because of the longer carrier sensing range, and an increasing number of nodes will be within the interference zone of the transmitters, increasing the energy consumption relative to the overhearing packets operation. To summarize the aforementioned results, GUF+VA could be suitable for scenarios of low network density. This is because the three approaches have similar results in terms of reducing void nodes, but GUF+VA spends less energy than GUF+DA and GUF+PA in those scenarios. However, GUF+PA and GUF+DA are suitable for scenarios of medium to high density, in which the fraction of void nodes ranges from moderate to low. In these scenarios, GUF+PA is adequate for very low traffic loads, whereas GUF+DA suits moderate to high loads, as the energy cost to move the nodes is diluted over time.

5.5. Network topology analysis In this section, we analyze various peculiarities related to the resulting network topology after each void-handling strategy, which impact on the network performance and energy consumption.

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

11

Fig. 6. Fraction of nodes per communication range value.

Fig. 7. Moved nodes: a) Percentage of void nodes that were moved. b) Average node displacement.

Fig. 6 shows the fraction of nodes by each transmission power level on the power control-based void handling strategy. This plot shows that when the network density increases in term of number of surface sonobuoys, the fraction of nodes using 350 m or 450 m as their communication range decreases. This occurs because isolated nodes that select a large communication range in low-density scenarios to have a path to a sonobuoy will decrease with the increment of the number of sensor nodes, as confirmed by the results of GUF performance in Fig. 3. Fig. 7a shows the percentage of nodes that were moved to new depths when the mobility assisted-based void-handling algorithm was used. As expected and confirmed in the plot, the increment of the surface sonobuoys decreases the percentage of moved nodes. In the hard scenario of low-density, a low number of void nodes where moved. This is because the displacement of some nodes do not guarantee that routing paths will be created (please refer to Fig. 3). When the number of sensor nodes increases, the fraction of moved nodes increases. The reason for this is that some isolated nodes in low density scenarios cannot be moved, as they cannot communicate with sonobuoys, regardless of proximity to the surface of the sea. Fig. 7b depicts the average displacement of void nodes. We can note similar behavior to the above discussed. For low density scenarios, void nodes tends to move along short distances. Basically, these are the nodes closer to the sonobuoys that move to communicate directly with them. As the network density increases, the average displacement increases. However, a better surface cover-

age in terms of an increment on the number of sonobuoys will limit the displacement of underwater nodes. Fig. 8 shows the complementary cumulative distribution function (CCDF) of the number of hops for a different number of sensor nodes and surface sonobuoys. In general, it is claimed that the use of bypassing void region void-handling methodology results in long paths. However, for the considered scenarios, results in Fig. 8 show that GUF+DA and GUF+PA have long paths as compared to the GUF+VA. This is explained by the fact that high fraction of nodes using GUF+VA approach are void nodes and are not able to deliver data packet to a sonobuoy (please refer to Fig. 3). 6. Conclusion We proposed an analytic framework to evaluate the performance of the void handling algorithms in underwater sensor networks. The proposed model considered the unique characteristics of the underwater acoustic communication and the peculiarities of power control-, bypassing void region-, and mobility assistedbased void node recovery strategies. Numerical results showed that the widely employed bypassing void region methodology is not the better strategy to cope with void nodes in low density UWSN scenarios. However, topology control-based methodologies are preferable since they change the topology creating new communication links. More specifically, the proposed mathematical framework provided useful insights for the future design of void-handling proce-

12

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14

Fig. 8. CCDF of the # of hops for different network densities. (a)–(c) 4 sonobuoys. (d)–(f) 9 sonobuoys. (g)–(i) 16 sonobuoys.

dures. For instance, power control-based strategy proved to be impractical for low-density scenarios, given the energy cost for data communication along long-range links. The bypassing void regionbased strategy showed to be an attractive solution for low-density scenarios, as a high energy cost is incurred to use power control and mobility assisted-based strategies due to the transmission between far nodes and the cost to move void nodes in the depth adjustment strategy. The mobility assisted-based strategy with the depth adjustment of some nodes was determined as a viable solution for moderate- to high-density and traffic load scenarios. In these scenarios, nodes will node among short distances, and the energy cost to do this is amortized as long as more data packets are transmitted. In future work, we plan to extend this methodology to evaluate the performance of void-handling algorithms in mobile underwater sensor network scenarios. This further evaluation should lead to insights for the design of adaptive void handling algorithms, in which the neighborhood density of the nodes change over time.

Acknowledgment This work is partially supported by NSERC TRANSIT Project, NSERC Strategic DIVA Network Research Program, Canada Research Chairs Program, and MRI/ORF Funds. We also would like to thank the Brazilian CNPq, CAPES and FAPEMIG agencies.

References [1] I.F. Akyildiz, D. Pompili, T. Melodia, Underwater acoustic sensor networks: research challenges, Ad Hoc Netw. 3 (3) (2005) 257–279. [2] I. Vasilescu, K. Kotay, D. Rus, M. Dunbabin, P. Corke, Data collection, storage, and retrieval with an underwater sensor network, in: Proc. of the 3rd ACM Int’l Conf. on Embedded Networked Sensor Systems (SenSys), 2005, pp. 154–165. [3] M.C. Domingo, R. Prior, Energy analysis of routing protocols for underwater wireless sensor networks, Comput. Commun. 31 (6) (2008) 1227–1238. [4] J. Heidemann, M. Stojanovic, M. Zorzi, Underwater sensor networks: applications, advances and challenges, Philosoph. Trans. R. Soc. A 370 (2012) 158–175. [5] M. Stojanovic, Underwater acoustic communications: design considerations on the physical layer, in: Proc. of the 5th Annual Conf. on Wireless on Demand Network Systems and Services (WONS), 2008, pp. 1–10.

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14 [6] H. Kulhandjian, T. Melodia, Modeling underwater acoustic channels in short-range shallow water environments, in: Proc. of the ACM Int’l Conference on Underwater Networks & Systems (WUWNet), 2014, pp. 26:1–26:5. [7] J. Partan, J. Kurose, B.N. Levine, A survey of practical issues in underwater networks, in: Proc. of the 1st ACM international workshop on Underwater networks (WUWNet), 2006, pp. 17–24. [8] P. Xie, J.-H. Cui, L. Lao, VBF: vector-based forwarding protocol for underwater sensor networks, in: Proc. 5th Int’l IFIP-TC6, in: NETWORKING’06, 2006, pp. 1216–1221. [9] H. Yan, Z.J. Shi, J.-H. Cui, DBR: depth-based routing for underwater sensor networks, in: NETWORKING 2008, vol. 4982, Springer Berlin Heidelberg, 2008, pp. 72–86. [10] Y. Noh, U. Lee, P. Wang, B.S.C. Choi, M. Gerla, VAPR: void-aware pressure routing for underwater sensor networks, IEEE Trans. Mob. Comput. 12 (5) (2013) 895–908. [11] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, GEDAR: geographic and opportunistic routing protocol with depth adjustment for mobile underwater sensor networks, in: Proc. of the IEEE International Conference on Communications (ICC), 2014, pp. 251–256. [12] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, A novel void node recovery paradigm for long-term underwater sensor networks, Ad Hoc Netw. 34 (2015) 144–156. [13] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, Geographic and opportunistic routing for underwater sensor networks, IEEE Trans. Comput. 65 (2) (2015) 548–561. [14] Y. Noh, U. Lee, S. Lee, P. Wang, L.F.M. Vieira, J.H. Cui, M. Gerla, K. Kim, Hydrocast: pressure routing for underwater sensor networks, IEEE Trans. Vehic. Technol. 65 (1) (2016) 333–347. [15] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, Design guidelines for opportunistic routing in underwater networks, IEEE Comm. Mag. 54 (2) (2016) 40–48. [16] D. Chen, P. Varshney, A survey of void handling techniques for geographic routing in wireless networks, IEEE Comm. Surv. Tuts. (2007) 50–67. [17] J.M. Jornet, M. Stojanovic, M. Zorzi, Focused beam routing protocol for underwater acoustic networks, in: Proc. of the 3rd ACM international workshop on Underwater Networks (WUWNet), 2008, pp. 75–82. [18] P. Xie, Z. Zhou, Z. Peng, J.-C. Cui, Z. Shi, Void avoidance in three-dimensional mobile underwater sensor networks, in: Wireless Algorithms, Systems, and Applications, vol. 5682, Springer Berlin Heidelberg, 2009, pp. 305–314. [19] U. Lee, P. Wang, Y. Noh, L.F.M. Vieira, M. Gerla, J.-H. Cui, Pressure routing for underwater sensor networks, in: Proc. IEEE INFOCOM, 2010, pp. 1–9. [20] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, Local maximum routing recovery in underwater sensor networks: performance and trade-offs, in: Proc. of the IEEE 22nd International Symposium on Modelling, Analysis & Simulation of Computer and Telecommunication Systems (MASCOTS), 2014, pp. 112–119. [21] N. Nicolaou, A. See, P. Xie, J.-H. Cui, D. Maggiorini, Improving the robustness of location-based routing for underwater sensor networks, in: Proc. of the MTS/IEEE OCEANS, 2007, pp. 1–6. [22] S. Basagni, C. Petrioli, R. Petroccia, D. Spaccini, Carp: a channel-aware routing protocol for underwater acoustic wireless networks, Ad Hoc Netw. 34 (2014) 92–104. [23] M. Al-Bzoor, Y. Zhu, J. Liu, A. Reda, J.-H. Cui, S. Rajasekaran, Adaptive power controlled routing for underwater sensor networks, in: Wireless Algorithms, Systems, and Applications, vol. 7405, Springer Berlin Heidelberg, 2012, pp. 549–560. [24] R.W.L. Coutinho, L.F.M. Vieira, A.A.F. Loureiro, DCR: depth-controlled routing protocol for underwater sensor networks, in: Proc. of the IEEE Symposium on Computers and Communications (ISCC), 2013, pp. 453–458. [25] R.W.L. Coutinho, L.F.M. Vieira, A.A.F. Loureir, Movement assisted-topology control and geographic routing protocol for underwater sensor networks, in: Proc. of the 16th ACM Int’l Conf. on Modeling, Analysis & Simulation of Wireless and Mobile Systems (MSWiM), 2013, pp. 189–196. [26] M. O’Rourke, E. Basha, C. Detweiler, Multi-modal communications in underwater sensor networks using depth adjustment, in: Proc. of the 7th ACM International Conference on Underwater Networks and Systems (WUWNet), 2012, pp. 31:1–31:5. [27] Teledyne-Benthos, Acoustic Modems, 2017, (http://teledynebenthos.com/ product/smart_products/sm-975). [Online]. [28] F. Bouabdallah, N. Bouabdallah, R. Boutaba, On balancing energy consumption in wireless sensor networks, IEEE Trans. Vehic. Technol. 58 (6) (2009) 2909–2924. [29] R. Jurdak, C.V. Lopes, P. Baldi, Battery lifetime estimation and optimization for underwater sensor networks, IEEE Sensor Netw. Oper. 2006 (2004) 397–420.

13

[30] H. Yang, B. Liu, F. Ren, H. Wen, C. Lin, Optimization of energy efficient transmission in underwater sensor networks, in: Proc. IEEE Global Telecommunications Conference (GLOBECOM), 2009, pp. 1–6. [31] J.R. Urick, Principles of Underwater Sound, McGraw-Hill, 1983. [32] P.H. Rogers, Onboard Prediction of Propagation Loss in Shallow Water, Naval Research Lab Defense Technical Information Center, Washington DC, 1981. [33] Bellhop - Ocean Acoustics Library, 2015, (http://oalib.hlsresearch.com/Rays/). [Online]. [34] G. Toso, P. Casari, M. Zorzi, The effect of different attenuation models on the performance of routing in shallow-water networks, in: Proc. of the Underwater Communications and Networking (UComms), 2014, pp. 1–5. [35] L.M. Brekhovskikh, Y. Lysanov, Fundamentals of Ocean Acoustics, Springer, 2003. [36] M. Stojanovic, On the relationship between capacity and distance in an underwater acoustic communication channel, in: Proc. of the 1st ACM Isnternational Workshop on Underwater Networks (WUWNet), 2006, pp. 41–47. [37] M. Stojanovic, Recent advances in high-speed underwater acoustic communications, IEEE J. Oceanic Eng. (1996) 125–136. [38] L. Freitag, M. Grund, S. Singh, J. Partan, P. Koski, K. Ball, The whoi micro– modem: an acoustic communcations and navigation system for multiple platforms, in: Proc. MTS/IEEE Oceans, 2005, pp. 1086–1092. [39] T. Rappaport, Wireless Communications: Principles and Practice, Prentice Hall, 2002. [40] P. Xie, J.-H. Cui, Exploring random access and handshaking techniques in large-scale underwater wireless acoustic sensor networks, in: Proc. of the OCEANS, 2006, pp. 1–6. [41] Y. Su, Y. Zhu, H. Mo, J.-H. Cui, Z. Jin, A joint power control and rate adaptation mac protocol for underwater sensor networks, Ad Hoc Netw. 26 (0) (2015) 36–49. [42] Y. Zhu, Z. Jiang, Z. Peng, M. Zuba, J.-H. Cui, H. Chen, Toward practical mac design for underwater acoustic networks, in: Proceedings IEEE INFOCOM, 2013, pp. 683–691. [43] M. Burkhart, P. von Rickenbach, R. Wattenhofer, A. Zollinger, Does topology control reduce interference? in: Proc. 5th ACM Int’l Symposium on Mobile ad hoc Networking and Computing (MobiHoc), 2004, pp. 9–19. [44] B. Liu, F. Ren, C. Lin, Y. Yang, R. Zeng, H. Wen, The redeployment issue in underwater sensor networks, in: Proc. of the IEEE Global Telecommunications Conference (GLOBECOM), 2008, pp. 1–6. [45] R.W.L. Coutinho, A. Boukerche, L.F.M. Vieira, A.A.F. Loureiro, On the design of green protocols for underwater sensor networks, IEEE Commun. Mag. 54 (10) (2016) 67–73. [46] P. Xie, Z. Zhou, Z. Peng, H. Yan, T. Hu, J.H. Cui, Z. Shi, Y. Fei, S. Zhou, Aqua-sim: an ns-2 based simulator for underwater sensor networks, in: Proc. of the OCEANS, 2009, pp. 1–7. [47] C. Petrioli, R. Petroccia, J.R. Potter, D. Spaccini, The sunset framework for simulation, emulation and at-sea testing of underwater wireless sensor networks, Ad Hoc Netw. 34 (2015) 224–238. [48] L. Pu, Y. Luo, H. Mo, S. Le, Z. Peng, J.-H. Cui, Z. Jiang, Comparing underwater MAC protocols in real sea experiments, Comput. Commun. 56 (2015) 47–59. [49] H. Zhang, S. Xiong, Z. Yue, Z. Wang, Sea trials of an underwater acoustic network in the east china sea 2015, in: IEEE/OES China Ocean Acoustics (COA), 2016, pp. 1–5. [50] E. Demirors, B.G. Shankar, G.E. Santagati, T. Melodia, Seanet: a software-defined acoustic networking framework for reconfigurable underwater networking, in: Proc. of the 10th International Conference on Underwater Networks & Systems (WUWNET), 2015, pp. 11:1–11:8. [51] Y. Han, Y. Fei, A. Ding, SMARP: a stochastic mac protocol with randomized power control for underwater sensor networks, in: Proc. of the 13th Annual IEEE International Conference on Sensing, Communication, and Networking (SECON), 2016, pp. 1–9. [52] L.F.M. Vieira, M.A.M. Vieira, D. Pinto, J.A.M. Nacif, S.S. Viana, A.B. Vieira, Hydronode: an underwater sensor node prototype for monitoring hydroelectric reservoirs, in: Proc. of the ACM Int’l Conference on Underwater Networks & Systems (WUWNet), 2012, pp. 43:1–43:2. [53] A.A. Aziz, Y.A. Sekercioglu, P. Fitzpatrick, M. Ivanovich, A survey on distributed topology control techniques for extending the lifetime of battery powered wireless sensor networks, IEEE Commun. Surv. Tut. 15 (1) (2013) 121–144. [54] R.W. Coutinho, A.F. Boukerche, L. Vieira, A. Loureiro, A novel centrality metric for topology control in underwater sensor networks, in: Proc. of the 19th ACM Int’l Conf. on Modeling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM), 2016, pp. 205–212.

14

R.W.L. Coutinho et al. / Computer Networks 126 (2017) 1–14 Rodolfo W.L. Coutinho is currently a Postdoctoral Fellow at the PARADISE Research Laboratory, University of Ottawa, Canada. He received the Joint Ph.D. degree from the University of Ottawa, Canada, and Federal University of Minas Gerais (UFMG), Brazil in 2017. He received his Masters and Bachelors degree in 2010 and 2009, both from the Federal University of Para (UFPA), Brazil. His research interests include underwater sensor networks, wireless networking, and mobile computing.

Azzedine Boukerche (FIEEE, FEiC, FCAE, FAAAS) is a full professor and holds a Canada Research Chair Tier-1 position at the University of Ottawa. He is founding director of the PARADISE Research Laboratory and the DIVA Strategic Research Centre at the University of Ottawa. He has received the C. Gotlieb Computer Medal Award, Ontario Distinguished Researcher Award, Premier of Ontario Research Excellence Award, G. S. Glinski Award for Excellence in Research, IEEE Computer Society Golden Core Award, IEEE CS-Meritorious Award, IEEE TCPP Leaderships Award, IEEE ComSoc ASHN Leaderships and Contribution Award, and University of Ottawa Award for Excellence in Research. He serves as an Associate Editor for several IEEE transactions and ACM journals, and is also a Steering Committee Chair for several IEEE and ACM international conferences. His current research interests include wireless ad hoc and sensor networks, wireless networking and mobile computing, wireless multimedia, QoS service provisioning, performance evaluation and modeling of large-scale distributed and mobile systems, and large scale distributed and parallel discrete event simulation. He has published extensively in these areas and received several best research paper awards for his work. He is a Fellow of the Engineering Institute of Canada, a Fellow of the Canadian Academy of Engineering, and a Fellow of the American Association for the Advancement of Science.

Luiz F.M. Vieira received the Computer Science degree from Universidade Federal de Minas Gerais (UFMG), Brazil, and the Ph.D. degree from the University of California at Los Angeles (UCLA). He is currently a professor at the Computer Science Department at UFMG. His current research interests include network coding, wireless networks and underwater sensor networks. He has published extensively in the areas of wireless networks and underwater sensor networks, including ACM WUWNET and IEEE Journal on Selected Areas in Communications. He is currently a TPC member for IEEE INFOCOM and ACM WUWNET and reviewer for IEEE/ACM Transactions on Networking and Ad Hoc Networks. He is a member of the IEEE (see http://www.dcc.ufmg.br/∼lfvieira for recent publications).

Antonio A.F. Loureiro is a full professor at UFMG, where he leads the research group on mobile ad hoc networks. He received his B.Sc. and M.Sc. degrees in computer science from UFMG, and his Ph.D. degree in computer science from the University of British Columbia, Canada. He was the recipient of the 2015 IEEE Ad Hoc and Sensor (AHSN) Technical Achievement Award. He is a regular visiting professor and researcher at the PARADISE Research Laboratory at the University of Ottawa and is an international research partner of DIVA Strategic Research Networks. His main research areas include wireless sensor networks, mobile computing, and distributed algorithms. In the last 10 years, he has published regularly in international conferences and journals related to those areas, and has also presented tutorials at international conferences.