A Comparative Performance Evaluation of Distributed Collision-free

MAC Protocols for Underwater Sensor Networks

Faisal Alfouzan, Alireza Shahrabi, Seyed Mohammad Ghoreyshi and Tuleen Boutaleb

School of Computing, Engineering and Built Environment, Glasgow Caledonian University, Glasgow, U.K.

Keywords:

Underwater Sensor Networks (UWSNs), Medium Access Control (MAC), Collision-free MAC Protocols.

Abstract:

The design of Medium Access Control (MAC) protocols for UWSNs poses many challenges because of their

long propagation delay, high mobility, limited bandwidth, and high bit error rate. Due to these unique acoustic

channel characteristics, most contention-based MAC protocols are costly. Thus, collisions and retransmissions

should be efﬁciently handled at the MAC layer in order to reduce the energy cost and to improve throughput

and fairness across the network. As a consequence, they do not perform as efﬁciently as their achieved perfor-

mance in terrestrial networks. In this paper, we evaluate the performance of three recently reported distributed

collision-free MAC protocols; namely, ED-MAC, DL-MAC, and GC-MAC under various operational condi-

tions. An extensive simulation study is carried out to compare the performance of these MAC protocols in

terms of packet delivery ratio (PDR), throughput, and energy consumption with different scenarios (narrow

and shallow networks) under varying trafﬁc rates and numbers of nodes. Our study results showed that ED-

MAC reaches the best energy efﬁciency in a narrow scenario with a light load than DL-MAC and GC-MAC

protocols. While DL-MAC is a suitable choice for both scenarios among others in terms of ﬂexibility. In terms

of reliability and scalability, GC-MAC achieves the best performance in both scenarios than other protocols.

1 INTRODUCTION

Underwater sensor networks (UWSNs) have attracted

a considerable attention to discover and monitor

aquatic environments. This aims to improve ocean

exploration and underwater applications such as envi-

ronmental monitoring, early warning systems, disas-

ter prevention, intrusion detection, military applica-

tions, and exploration of ocean resource (Ghoreyshi

et al., 2017). In UWSNs, a sink is considered on the

water surface, which is applicable to both an acous-

tic modem for underwater communication and a radio

modem for out-of-the-water communications (from

sink to satellite, and from satellite to monitoring cen-

tre) (Ghoreyshi et al., 2016; Alfouzan et al., 2016).

Anchored nodes are located at the bottom of the ocean

in predetermined locations to collect the information.

That information is delivered to the sink using the

relay nodes which are located between the sink and

the anchored nodes at different depth levels. The

anchored and relay nodes utilise acoustic signals to

transmit the data packets (Ghoreyshi et al., 2018a).

Using acoustic signal differently affects the design

of various services in UWSNs (Akyildiz et al., 2005).

Speciﬁcally, it has completely changed the design of

Medium Access Control (MAC) protocols compared

to that of terrestrial networks (Hsu et al., 2009). Since

radio wave cannot propagate in the underwater envi-

ronment as efﬁciently as it achieved in terrestrial net-

works, currently acoustic communication has exten-

sively been studied (Preisig, 2007; Ghoreyshi et al.,

2018b). However, due to the unique characteristics

of its acoustic channels such as slow signal propaga-

tion speed (about 1500m/s in water), limited channel

capacity, and high dynamics of channel quality, MAC

protocol design for underwater acoustic networks face

several challenges. Particularly, the high propaga-

tion delay which signiﬁcantly affects the MAC design

strategy in UWSNs.

Existing MAC solutions are essentially focused on

TDMA. This is mainly because FDMA is not proper

for UWSNs due to the narrow bandwidth in under-

water acoustic channel as well as the diffuse of lim-

ited band systems to fading and multipath. Moreover,

CDMA is very robust to frequency selective fading

caused by multiple paths. It is therefore unsuitable

for UWSNs due to also its difﬁculties to address the

near-far problem (Xie and Cui, 2007).

Among them, TDMA is the most promising mul-

tiple access technique for UWSNs. This is mainly

Alfouzan, F., Shahrabi, A., Ghoreyshi, S. and Boutaleb, T.

A Comparative Performance Evaluation of Distributed Collision-free MAC Protocols for Underwater Sensor Networks.

DOI: 10.5220/0007379700850093

In Proceedings of the 8th International Conference on Sensor Networks (SENSORNETS 2019), pages 85-93

ISBN: 978-989-758-355-1

because it allows the same frequency channel to be

shared by dividing the signal into different time slots.

Furthermore, it is able to maintain reliable transmis-

sion schedules by performing additional updating and

scheduling phases to also remain all sensor nodes syn-

chronised. TDMA also allows sensors, located out

of each others’ transmission ranges, to transmit data

packets simultaneously without collision by using the

concept of spatial reuse (concurrent sending in dif-

ferent neighbourhoods) (Alfouzan et al., 2018a). For

these reasons, TDMA increases channel reuse and

eliminate packet retransmissions, which results in de-

creased energy consumption and increased network

throughput.

Recently, three new TDMA-based distributed

collision-free MAC protocols, ED-MAC (Alfouzan

et al., 2017; Alfouzan et al., 2018a), DL-MAC (Al-

fouzan et al., 2018b), and GC-MAC (Alfouzan et al.,

2018c), have been reported in the literature with var-

ied performance under various operational conditions

and different assumptions. The aim of this paper is

to compare the performance of these protocols under

two different practical scenarios, namely shallow and

narrow scenarios, and to investigate their reliability,

scalability and ﬂexibility under a wide range of ex-

perimental conditions.

The remainder of this paper is organised as fol-

lows. Section 2 describes and classiﬁes the protocols

that we have investigated. Section 3 investigates and

compares the performance evaluation. Section 4 con-

cludes the paper.

2 DESCRIPTION OF THE

PROTOCOLS

This section presents a description of each collision-

free MAC protocol. It is then followed by a classiﬁca-

tion that includes all the requirements and properties

of each, as illustrated in Table 1.

2.1 ED-MAC

An Efﬁcient Depth-based MAC protocol (ED-MAC)

has been proposed in (Alfouzan et al., 2017; Alfouzan

et al., 2018a) with the aim of improving the energy ef-

ﬁciency, throughput, and fairness. It is a reservation-

based MAC protocol in which a duty cycle mecha-

nism is used by assigning time slots to every indi-

vidual node in the network in a distributed manner.

This is used to reduce the energy consumption by us-

ing a wake-up scheduling scheme; nodes are awake in

some slots to transmit or receive data and are asleep

over the remaining slots.

ED-MAC operates in three phases: initial,

scheduling, and normal operational phase. In the ﬁrst

phase, all the sensors randomly transmit a number of

small beacons to detect their one-hop neighbouring

nodes. The length of the initial phase is a predeﬁned

ﬁxed value for all sensors.

The primary goal of the scheduling phase is to al-

locate a unique time-slot to every sensor in the net-

work. A timer is utilised at every sensor to prioritise

slot reservation depending on the sensor depth; the

deeper the sensor, the higher priority to reserve a slot.

This timer lets a sensor which is located in a deeper

area to reserve a slot sooner than its above neighbour-

hoods. The value of this timer at each sensor is given

by (Alfouzan et al., 2017; Alfouzan et al., 2018a):

sch



Depth

+ N

Depth

× T

Delay



− T

Delay

, (1)

where M

Depth

is the depth of the network area and

Depth

is a sensor depth in the network. T

Delay

is the

length of the scheduling phase that is a predeﬁned

ﬁxed value, set during the deployment process based

on the application requirements. The value of T

Delay

depends also on the density of the nodes in an under-

water area. This phase is signiﬁcantly shorter than

that of the normal operational phase.

The third phase, the normal operational phase, is

divided into a number of rounds. Each of them in-

cludes a number of slots. Sensors in each round are

aware of their reserved slots, and other slots that re-

served by their neighbourhoods. Thus, all sensors

can schedule to wake-up either to send their own data

packets during the reserved slots or to receive a data

packet from a neighbouring sensor. They switch to a

sleeping mode in the remaining slots when there is no

data transmission or reception.

2.2 DL-MAC

A Depth-based Layering MAC protocol (DL-MAC)

has been proposed in (Alfouzan et al., 2018b) to

improve energy efﬁciency, reliability, and ﬂexibility.

This is achieved by dividing the aquatic network area

into a number of horizontal layers to avoid any chance

of vertical collision. A simple clustering approach

for one-hop neighbouring sensors are utilised to also

avoid any possibility of collision between sensors in

the same layer. Therefore, DL-MAC is able to ad-

dress spatial-temporal uncertainty, the near-far effect,

and any hidden/exposed terminal problems.

In DL-MAC, sensors in the network operate in

three phases: updating, scheduling, and operational.

The goal of the updating phase is to gather informa-

tion about one-hop neighbouring nodes. This is per-

formed by exchanging some updating messages. The

SENSORNETS 2019 - 8th International Conference on Sensor Networks

length of this phase is a constant value set during the

deployment time for all sensors.

The purpose of the second phase, the schedul-

ing phase, is to assign a time-slot to every individ-

ual sensor in the network to access the medium with

no chance of collision. This is achieved by divid-

ing the network into multi layers, grouped into three

frames, and every frame is also divided into a num-

ber of sub-frames. Each sub-frame consists of a num-

ber of time-slots. A distributed clustering approach is

utilised to allow cluster heads, CHs, selecting unique

sub-frames, which should be different from the adja-

cent clusters. By using a simple clustering approach,

every cluster head is eventually able to assign all its

cluster members, i.e., those are located within a one-

hop neighbourhood, different time slots. Through this

principle, every sensor in the network has a different

time-slot in any two-hop neighbourhood; hence no

collisions can occur. To determine a CH, this model

gives higher priority to a sensor that can cover more

sensors in its 1-hop range at the same layer. This is ap-

plied via a timer-based approach, called degree timer,

which can be given by (Alfouzan et al., 2018b):

=(d

max

−d

)×(T

sch

max

)±λ (2)

where d

denotes the sensor degree and d

max

indicates

the maximum node degree in the network topology.

This can be estimated based on the sensor deployment

process, number of sensor nodes, and network dimen-

sions and it can also be known to all nodes during the

deployment time. T

sch

is the scheduling interval time

and λ is a short random time duration to differentiate

the underwater sensor nodes with the same d

The operational phase is divided into a number of

cycles, each consisting of three frames. Each frame

is composed of k sub-frames. Frames and sub-frames

are used to avoid vertical layers and horizontal clus-

ter interference respectively. Each sub-frame also en-

compasses a number of slots. At each cycle, every

sensor is aware of its frame, sub-frame, and its own

reserved slots, as well as the slots reserved by its

neighbouring sensors. They can therefore be sched-

uled to wake-up either to transmit their own data

packet during the reserved slots or to possibly receive

a data packet from a neighbouring sensor. They are

asleep in the remaining slots when there is no data

transmission or reception.

2.3 GC-MAC

A collision-free Graph Colouring MAC protocol

(GC-MAC) has been proposed in (Alfouzan et al.,

2018c) which aims to achieve better performance

in terms of throughput, energy efﬁciency, and fair-

ness than other collision-free MAC protocols. GC-

MAC uses the concept of graph colouring to de-

velop a reservation-based contention-free MAC pro-

tocol. This has been achieved by using a distributed

clustering approach for up to two-hop neighbouring

sensors and then to address the hidden and expose

node problems by removing the possible colour con-

ﬂict in two-hop neighbouring graph. Using a TDMA-

like approach, GC-MAC is able to assign a time-slot,

colour, to every individual node in the network in a

distributed manner. Nodes with the same colours can

thus transmit concurrently without collision. Nodes

are awake in some slots to transmit or receive data

packets and asleep over the remaining slots.

GC-MAC includes three phases to operate, which

are initial, scheduling, and operational phase. In the

initial phase, two rounds of beaconing are conducted

to discover two-hop neighbouring sensors. This is

performed by exchanging some beacons between sen-

sors. The length of this phase is set as a constant value

for all sensors during the deployment time.

The primary goal of the scheduling phase is to as-

sign different colours to all nodes which are located

within any two-hop neighbourhood using a simple a

clustering approach. This is achieved by allowing a

cluster head (CH), which is determined as the closest

sensor to a reference point during the ﬁrst phase, to

assign a different colour for all its one-hop neighbour-

ing (inner) sensors. Afterwards, the outer sensors,

those located outside the cluster, decide about their

own colours individually. By the end of the schedul-

ing phase, every sensor node has a various colour in

any two-hop neighbouring graph and hence no colli-

sion can occur.

The operational phase is divided into a number of

rounds. Each round consists of a number of time-

slots. These time-slots are reserved by assigning a

various colour to each. Nodes with the same colour

can transmit data packets at the same time without any

collision while the hidden and exposed node prob-

lems are properly addressed. In this phase, the sen-

sor nodes wake-up and sleep periodically. In other

words, sensors can schedule to wake-up to send their

own data packets during the reserved slots or to re-

ceive a data packet from a neighbourhood, while they

are asleep in other remaining slots when there is no

data transmission or reception.

2.4 Qualitative Comparison

In this section, we classify each protocol (ED-MAC,

DL-MAC, and GC-MAC) requirements and proper-

ties, and compare them qualitatively.

Table 1 lists all assumptions in which every pro-

A Comparative Performance Evaluation of Distributed Collision-free MAC Protocols for Underwater Sensor Networks

Table 1: Comparisons of the collision-free MAC protocols

for UWSNs.

ED-MAC DL-MAC GC-MAC

Category TDMA-based TDMA-based TDMA-based

Year 2017 2018 2018

Schedule Distributed Distributed Distributed

TDMA status Adaptive slotted Adaptive slotted Adaptive slotted

Network division No Divided to layers Divided to cubes

Clustered No Yes Yes

neighbourhood info One-hop neighbours One-hop neighbours Two-hop neighbours

Priority Depth-based timer Degree timer Node ID

Random time No Yes No

Synchronised Yes Yes Yes

GPS No No Yes

Number of slots 2 × N

max

Equal to d

max

Fixed

Conﬂict Avoidance No No Yes

tocol is based on. In fact, they are all classiﬁed as

a TDMA-based duty-cycle MAC protocols while the

basic information required for their operation are dif-

ferent.

During the deployment process, DL-MAC and

GC-MAC require the network area to be divided into

a number of layers and cubes, respectively, to im-

prove the efﬁciency of their distributed scheduling.

ED-MAC, however, does not require any network di-

visions. At the same time, its function is not based

on any kind of clustering, whereas this is one of the

requirements of DL-MAC and GC-MAC.

During the ﬁrst phase, ED-MAC and DL-MAC

require to exchange one-hop neighbouring informa-

tion to be obtained before the scheduling phase. In

their scheduling phases, both protocols set their prior-

ity timers differently. More speciﬁcally, ED-MAC’s

timer is used at each node to prioritise slot reserva-

tions depending on the node depth; the deeper the

node, the higher the priority to reserve a slot. Equa-

tion 1 demonstrates how the priority of each sensor

node is calculated. In DL-MAC, the degree timer is

used in each sensor node to start the scheduling phase;

a node with higher d-hop neighbouring nodes, which

can cover more nodes in its 1-hop range at the same

layer, has a higher priority to become a cluster head

(CH) than other nodes. Equation 2 gives the degree

timer of each node in the network, considering a short,

random time duration, λ, to differentiate the sensor

nodes that have the same sensor degree, d

In the ﬁrst phase of GC-MAC, however, two-hop

neighbouring information is required to be obtained

before the scheduling phase. The goal of exchanging

the two-hop neighbouring information during the ﬁrst

phase is to detect the hidden terminal nodes which are

located outside the two-hop neighbouring district of

each other. After creating the neighbouring graph, N

by each sensor node, the nearest node’s distance to a

reference point, r p, becomes a CH, which is respon-

sible for independently choosing a colour for itself

and different colours for its cluster members (CMs).

Nodes which are located between two adjacent cluster

heads within more than two-hop neighbouring nodes

decide their own colours individually; the lower the

node ID, the higher the priority to select the ﬁrst avail-

able colour among others.

According to the number of slots, each protocol

has its own algorithm and assumptions which it uses

to divide its operational window differently. In ED-

MAC, for instance, the number of slots is double the

maximum number of nodes in a neighbourhood, N

max

in order to exclude the possibility of concurrent data

transmission from a sensor located outside a one-hop

neighbourhood and the node within the neighbour-

hood. Meanwhile, in DL-MAC, the number of slots

is proportional to the maximum node degree found

in the one-hop neighbourhood graphs, d

max

. In GC-

MAC, the number of slots depends on the duration of

the operational phase as well as the slot length, which

is equal to the propagation delay plus a small guard

time to ensure that a packet is entirely received at the

destination before data transmission by another node

can begin.

In addition, GC-MAC the only protocol among

the selection discussed here to introduce the concept

of conﬂict detection (CD). The primary goal is to de-

tect and resolve any conﬂicts that may occur between

sensor nodes during the scheduling phase.

3 PERFORMANCE EVALUATION

In this section, we ﬁrst discuss the simulation scenar-

ios and settings used in our comparison study in the

Aqua-Sim underwater simulation (Xie et al., 2009).

We then deﬁne the metrics used in our performance

study. Finally, we compare the design trade-off be-

tween ED-MAC, DL-MAC, and GC-MAC within the

various given scenarios and networks.

3.1 Simulation Scenarios and Settings

We implement our ED-MAC, DL-MAC, and GC-

MAC protocols in Aqua-Sim, which is an NS-2 based

simulator for UWSNs. In our simulations, we con-

sider two scenarios, each of which is evaluated with

reference to two parameters: the trafﬁc rate, and the

number of nodes. In the ﬁrst scenario, all the under-

water sensor nodes are randomly distributed in a 3D

shallow region of 500m × 500m × 250m. In the sec-

ond scenario, the sensor nodes are randomly deployed

in 3D narrow region of 300m × 300m × 600m. The

following parameters are used in all scenarios, unless

noted otherwise: the trafﬁc rate increases from 0.05

up to 0.5 packets per second. In this set, we randomly

distributed 100 sensor nodes in the given two scenar-

ios. The node density is also increased from 50 to

SENSORNETS 2019 - 8th International Conference on Sensor Networks

(a) PDR

(b) Throughput

Figure 1: Shallow region scenario: PDR, Throughput, and Energy consumption vs. Trafﬁc rate.

500 sensor nodes within the same two scenarios. In

this case, the trafﬁc rate is ﬁxed at 0.1 packets per

second. In both scenarios, the power consumption on

transmission mode is 2 Watts, the power consumption

on receive mode is 0.75 Watts, and the power on sleep

mode is 8 mW. The data packet size is set to 1000 bits,

and all other control packets are set at 100 bits. The

channel bit rate (i.e., bandwidth) is set at 10 kbps and

the maximum transmission range is 100 meters.

It should be noted that all the results are averaged

over 20 runs, and that each is obtained with a ran-

domly generated topology in each scenario. The to-

tal simulation time for each run is set to 3600 sec-

onds. In our simulation setup, the length of the ﬁrst

phase is a predeﬁned ﬁxed value for all sensor nodes,

which is set at each sensor before deployment. Thus,

we consider this phase to be 30 seconds in all three

MAC protocols. The second phase, the scheduling

phase, also uses a predeﬁned constant interval set for

all sensors at the deployment time based on the appli-

cation requirements. Hence, we consider the schedul-

ing phase, T

sch

, to be 90 seconds in these reported

MAC protocols, except GC-MAC which includes the

conﬂict detection interval, T

added to its scheduling

phase. The T

is set at 30 seconds.

3.2 Performance Metrics

We deﬁne the most important metrics in medium ac-

cess control protocols during this comparison study to

evaluate the performance of ED-MAC, DL-MAC, and

GC-MAC protocols as packet delivery ratio (PDR),

throughput, and energy consumption per packet suc-

cessfully received in joules. These metrics are deﬁned

as follows:

The packet delivery ratio, PDR, is deﬁned as the

ratio of the packets successfully received to the to-

tal packets generated in the network. The network

throughput is deﬁned as the number of successful re-

ceived packets divided by the simulation time. The

energy consumption per packet is obtained by divid-

ing the overall energy usage in the network by the

successfully delivered data packets, where the energy

consumption here is measured in joules per packet.

3.3 Simulation Evaluation

In this section, we present the results of the com-

parative performance evaluation of the protocols de-

scribed in Section 2. We ﬁrst display the simulation

results of the shallow region scenario, before going on

to demonstrate the simulation results of the narrow re-

gion scenario.

3.3.1 Shallow Region Scenario

In this scenario, we ﬁrst evaluate the PDR, through-

put, and energy consumption under varying trafﬁc

rates in all three MAC protocols, as illustrated in Fig.

(1). In this case, the number of nodes is set to 100

sensor nodes.

Fig. (1a) shows the PDR as a function of data gen-

eration rate. With a low trafﬁc load, the PDR for all

protocols is 100%. When the trafﬁc load further in-

creases, the PDR of ED-MAC and DL-MAC (with

4 and 3 sub-frame conﬁgurations) signiﬁcantly de-

creases. Due to the fact that the scheduling design of

ED-MAC is based on depth criteria, it is more suitable

to a narrow region rather than a shallow one. With a

medium trafﬁc rate (0.25 packets per second), GC-

MAC performs best in terms of packet delivery ratio

by delivering all the data packets. As the trafﬁc rate

further increases, GC-MAC still has the ability to de-

liver most of the packets, followed by DL-MAC with

only 2 sub-frames. This is mainly because GC-MAC

employs the conﬂict detection mechanism during its

scheduling phase. Regarding the DL-MAC, because

the trafﬁc rate has a reverse relationship with the du-

ration of the operational window, DL-MAC with 2

sub-frames performs better than DL-MAC with 3 and

4 sub-frame conﬂagrations. DL-MAC with 3 sub-

frames also outperforms DL-MAC with 4 sub-frames.

This is due to the fact that the higher the trafﬁc rate,

the lower the duration of the operational cycle; hence,

no more available slots can be reserved.

A Comparative Performance Evaluation of Distributed Collision-free MAC Protocols for Underwater Sensor Networks

(a) PDR

(b) Throughput

Figure 2: Shallow region scenario: PDR, Throughput, and Energy consumption vs. Number of nodes.

As can be seen in Fig. (1b), the network through-

put of all protocols is proportional to the trafﬁc rate.

When the trafﬁc load increases, the throughput of all

protocols increases correspondingly up to their satu-

ration points. ED-MAC and DL-MAC (with 3 and 4

sub-frames) achieve their maximum capacities sooner

than GC-MAC and DL-MAC do, with 2 sub-frames,

within 0.35 packets per second. DL-MAC with 2

sub-frames and GC-MAC reach their capability re-

garding handling data packets within 0.40 and 0.50

packets per second respectively. We can observe that

GC-MAC outperforms all the other protocols in terms

of throughput with the same trafﬁc rates. This phe-

nomenon is particularly evident in the case of GC-

MAC, because it employs an effective conﬂict detec-

tion algorithm to avoid collisions, thus considerably

improving network throughput.

Fig. (1c) conﬁrms what we have already seen in

both Fig. (1a) and Fig. (1b). All the protocols are us-

ing a reliable and efﬁcient mechanism for channel ac-

cess and a very limited number of small packets dur-

ing the initial and scheduling phases in order to elim-

inate collisions and achieve high throughput. How-

ever, Fig. (1c) also shows the energy usage per data

packet successfully received. On the one hand, ED-

MAC consumes less energy than any other protocol

during all trafﬁc rates. This is mainly because it has

fewer requirements than the others during the deploy-

ment process as well as the ﬁrst and second phases,

such as multi layers in DL-MAC and reference points

in GC-MAC. On the other hand, ED-MAC spends a

higher amount of energy per data packet successfully

received than the others, because it delivers a lower

amount of packets than GC-MAC and DL-MAC with

both 2 and 3 sub-frame conﬁgurations, as depicted in

Fig. (1a). GC-MAC consumes lower energy in total,

therefore it is the best protocol in terms of energy con-

sumption per packet. It is more interesting to observe

that DL-MAC with 4 sub-frames consumes more en-

ergy per packet at a high trafﬁc rate. This happens

because the higher the trafﬁc rate, the lower the cycle

duration; therefore, there is a short number of slots,

causing high competition between nodes to reserve

those limited slots, and thus greater energy consump-

tion.

In this scenario, we also evaluate the performance

of all the protocols in terms of PDR, throughput, and

energy consumption under varying number of nodes,

as shown in Fig. (2). This ﬁgure illustrates how sparse

and dense sensor nodes in a shallow network can af-

fect the protocols, and examines the scalability and

ﬂexibility among them.

Fig. (2a) illustrates the PDR as a function of num-

ber of nodes. In this ﬁgure, GC-MAC clearly outper-

forms all other protocols by handling all data packets

up until 250 nodes. When the node density further in-

creases, the PDR of GC-MAC slightly decreases by

almost 20%. This is because of its efﬁcient schedul-

ing and the conﬂict detection mechanism that used

during the scheduling phase. However, when the node

density is low, ED-MAC delivers most of the packets

(almost 98%). When the node density is increased,

the PDR of ED-MAC decreases correspondingly to

only deliver 31% with a high dense network. This is

mainly because the ED-MAC policy depends on the

nodes’ depth, i.e., the deeper the node, the higher the

priority to reserve a slot. ED-MAC does not consider

horizontal hidden and exposed nodes located in shal-

lower areas.

It is noteworthy that DL-MAC with higher sub-

frame conﬁgurations performs better than DL-MAC

with lower sub-frames. In this ﬁgure, DL-MAC with

4 sub-frames outperforms DL-MAC with 2 and 3 sub-

frames. This is because a higher number of sub-

frames implies the ability to handle more nodes. DL-

MAC, with a higher number of sub-frames as a func-

tion of the number of nodes, gives the opposite results

when it is as a function of the trafﬁc rate as depicted in

Fig. (1a). This is because the trafﬁc rate has a reverse

relationship with the duration cycle, as mentioned ear-

lier, while the number of nodes does not, meaning that

when the node density increases, the trafﬁc rate is kept

ﬁxed at 0.1 packets per second.

Changing the node density from 50 to 500 nodes

generates the results shown in Fig. (2b) which shows

that all the throughput of the protocols begin with

SENSORNETS 2019 - 8th International Conference on Sensor Networks

(a) PDR

(b) Throughput

Figure 3: Narrow region scenario: PDR, Throughput, and Energy consumption vs. Trafﬁc rate.

5 packets per second when the node density is low.

As the node density further increases, the through-

put of the protocols rises signiﬁcantly and eventu-

ally reaches saturation point. In contrast, GC-MAC

has a higher throughput than the others by reaching

nearly 40 packets per second within 500 nodes. DL-

MAC, with three different conﬁgurations shows bet-

ter throughput than ED-MAC during all numbers of

nodes. In particular, DL-MAC with 4 sub-frames is

able to handle more data packets than DL-MAC with

2 and 3 sub-frames by almost 21.4% and 6.3% respec-

tively, with a high node density. ED-MAC achieves

almost 20 packets per second within 350 nodes, and

it then degrades as the node density increases. This

is because the ED-MAC’s scheduling policy does not

take the horizontal two-hop neighbourhood into ac-

count. It is therefore not a suitable choice for a shal-

low network.

Fig. (2c) illustrates the relationship between Fig.

(2a) and Fig. (3b). This ﬁgure conﬁrms the best

protocol performance in terms of energy consump-

tion. It can clearly be seen that GC-MAC displays

the best performance in terms of energy consump-

tion per data packet over the other protocols during

the whole node density. It is followed by DL-MAC

with 4 sub-frames, which consumes less energy than

DL-MAC with 3 and 2 sub-frame conﬁgurations. ED-

MAC, however, spends less energy within 50 and 100

nodes, before increasingly consuming higher energy

per packet up to almost double the others with a high

dense network. This is because, as shown in Fig. (2a),

it delivers fewer packets than the other protocols and

at the same time, it consumes more energy in joules

than the others.

3.3.2 Narrow Region Scenario

In the second scenario, we assess and investigate the

performance of the protocols within a narrow under-

water network in terms of PDR, throughput, and en-

ergy consumption under varying trafﬁc rates, as de-

picted in Fig. (3).

Fig. (3a) shows that the PDR of the three pro-

tocols is proportional to the trafﬁc rate, and that the

PDR of GC-MAC reduces to almost 93% and 80%

corresponding with a trafﬁc rate of 0.4 and 0.5 packets

per second respectively. This is because a higher traf-

ﬁc rate leads to a lower duration of operational round,

resulting in the channel capacity being achieved. Sim-

ilar phenomena are also observed with ED-MAC and

DL-MAC in three different conﬁgurations. How-

ever, by carefully considering collision avoidance,

GC-MAC achieves fairly good results for packet de-

livery ratio compared to the others. The result of DL-

MAC with 4 sub-frames is very low with a high trafﬁc

rate because the duration cycle becomes low; there-

fore, nodes are unable to ﬁnd a free time-slot, i.e.,

a medium channel is divided into three frames, each

frame consists of 4 sub-frames, and each sub-frame

includes a number of slots which is insufﬁcient for

the one-hop neighbouring nodes.

Fig. (3b) presents the network throughput of all

protocols as a function of the trafﬁc rate. The through-

put of the protocols is observed to increase as the

trafﬁc rate increases. This is because the trafﬁc rate

does not exceed the network capacity. Within a traf-

ﬁc rate of (0.3 and 0.4 packets per second), the net-

work congested, resulting in a decreasing throughput

for ED-MAC and DL-MAC with all sub-frame con-

ﬁgurations with a growth in trafﬁc rate, except the

GC-MAC trends of 40 packets per second at 0.45 and

0.50 trafﬁc rates. This is mainly because any possi-

bility of collisions is removed by applying the con-

ﬂict detection interval at the end of the scheduling

phase. Therefore, the GC-MAC always achieves the

best throughput compared to ED-MAC and DL-MAC

with different sub-frame conﬁgurations.

Fig. (3c) shows the results in terms of energy ef-

ﬁciency, i.e., the energy consumed per successfully

delivered packet. We can ﬁrst observe that ED-MAC

is more energy efﬁcient than the others at low traf-

ﬁc rates, because of the lower requirements it faced

during the scheduling phase than other protocols, re-

sulting in a high delivery rate and low energy con-

sumed. It is more interesting to observe that although

GC-MAC consumed more energy than the others at

A Comparative Performance Evaluation of Distributed Collision-free MAC Protocols for Underwater Sensor Networks

(a) PDR

(b) Throughput

Figure 4: Narrow region scenario: PDR, Throughput, and Energy consumption vs. Number of nodes.

high trafﬁc rates, it can still be considered to have

the best energy efﬁciency at medium and high traf-

ﬁc rates. This is because its delivery rate outperforms

that of all the other protocols at medium and high traf-

ﬁc rates, as shown previously in Fig. (3a). DL-MAC

with 4 sub-frames, however, is considered inefﬁcient

because it spends double the energy of GC-MAC at

a medium trafﬁc rate. With a high trafﬁc rate, DL-

MAC with 4 sub-frames also consumes nearly three

times more energy than GC-MAC.

This scenario also allows a comparative evaluation

to be carried out of the performance of ED-MAC, DL-

MAC, and GC-MAC in terms of PDR, throughput,

and energy consumption under different sets of node

density, as presented in Fig. (4). This ﬁgure shows

how sparse and dense nodes in a narrow network can

affect the performance of the three MAC protocols,

and how scalable and ﬂexible their designs are.

As can be seen in Fig. (4a), the PDR of GC-MAC

remains constant at 100%, as it delivers all the data

packets successfully up to a density of 250 nodes,

then the delivery rate reduces signiﬁcantly as the node

density further increases. For DL-MAC with three

different sub-frame conﬁgurations, the delivery rate

with 2 sub-frames reduces sooner than DL-MAC with

3 and 4 sub-frames. As the node density increases,

the PDR of DL-MAC with 2, 3, and 4 sun-frames is

signiﬁcantly reduced to deliver approximately 61%,

66%, and 70% data packets respectively in a high

density network (with 500 nodes). This is because a

higher number of sub-frames under a ﬁxed trafﬁc rate

implies an ability to handle more nodes. However, the

PDR of ED-MAC delivers most of the data packets

up to 150 nodes, then it dramatically decreases as the

number of nodes further increases. We can observe

that the delivery rate of ED-MAC is good with an

increased number of nodes, but compared with GC-

MAC and DL-MAC, it has the lowest delivery rate

with all numbers of nodes.

Fig. (4b) presents the results in terms of through-

put. Within 50 and 100 nodes, the network through-

put of all the protocols signiﬁcantly increases while

achieving 5 and 10 packets per second respectively.

When the number of nodes further increases, the

throughput correspondingly goes up and eventually

reaches the channel threshold. In contrast, GC-MAC

outperforms all the other protocols by handling more

than 35 packets per second with high node density

(within 450 and 500 nodes). Meanwhile, the through-

put of ED-MAC handles lower data packets than the

others. This is because ED-MAC has some limita-

tions during its scheduling phase which do not apply

to GC-MAC and DL-MAC. More speciﬁcally, ED-

MAC’s scheduling policy does not take the horizontal

two-hop neighbouring nodes into account, meaning

that a few collisions might occur among them when

the number of nodes increases. DL-MAC is more

scalable and ﬁxable than other protocols in employ-

ing sub-frame conﬁgurations. For instance, DL-MAC

with a lower number of sub-frames performs better

than DL-MAC with higher sub-frames in terms of in-

creasing the trafﬁc rate and keeping the node den-

sity ﬁxed. Conversely, when the number of nodes

increases under a ﬁxed trafﬁc rate, DL-MAC with

higher sub-frame conﬁgurations performs better than

with lower sub-frames. This is mainly because the

trafﬁc rate has a reverse relationship over the duration

of the operational cycle which, as well as a higher

number of sub-frames, implies the ability to handle

more nodes.

Fig. (4c) illustrates the energy consumption. This

ﬁgure shows the best performance in terms of packet

delivery rate along with the energy consumption as-

sociated with sparse and dense nodes. ED-MAC

can be observed to have consumed less energy per

packet than the others with 50 and 100 nodes. As

the node density increases, its energy consumption

rises sharply, reaching just over 4.6 joules per packet

with 500 nodes. However, GC-MAC starts off con-

stant by consuming 1 joule per packet until the num-

ber of nodes exceeds 200 nodes, then its energy con-

sumption increases as the number of nodes increases

too by consuming nearly 2.8 joules per packet with a

high node density. Overall, GC-MAC performs well

from 150 to 500 nodes, followed by DL-MAC with 4

sub-frames. This is because both protocols have more

SENSORNETS 2019 - 8th International Conference on Sensor Networks

beneﬁts than the others depending on their scheduling

strategies, as has been explained above.

4 CONCLUSIONS

This paper has presented a comparative performance

evaluation of three collision-free MAC protocols for

channel access in underwater sensor networks. We

have investigated several scenarios that are typical of

the current underwater channel access research. The

ﬁrst scenario evaluated a shallow network area with

low, medium, and high trafﬁc rates as well as sparse

and dense nodes under a ﬁxed trafﬁc rate. The second

scenario examined the same parameters within a nar-

row network area. These scenarios are mainly used to

study the three protocols’ overall ability, scalability,

and ﬂexibility under sparse and dense sensor nodes as

well as under light to heavy trafﬁc. Our study points

out that every protocol has its own advantages and dis-

advantages, which means that no protocol can ﬁt all

needs in all scenarios.

This study has concluded that ED-MAC achieves

the best performance in a narrow scenario with a light

trafﬁc rate. It is able to handle 99% of the data pack-

ets at a lower energy cost than other protocols. How-

ever, it is not a suitable choice in a shallow scenario

as it does not consider two-hop neighbouring nodes

horizontally. To improve its performance in this re-

gion, more slots are needed to reduce the probabil-

ity of collisions that may occur between nodes lo-

cated horizontally. DL-MAC has the ability to per-

form better than ED-MAC in narrow and shallow re-

gions by scheduling all nodes vertically and horizon-

tally. This is because of its ability to address most

of the MAC problems such as spatial-temporal un-

certainty, the near-far effect, and any hidden/exposed

terminal problems. In terms of ﬂexibility, DL-MAC is

more ﬂexible than other protocols by dividing the net-

work area into frames, and the sensors in each frame

into sub-frames. It is, therefore, suitable for both sce-

narios. While GC-MAC has achieved the best perfor-

mance than others in both narrow and shallow scenar-

ios in terms of reliability and scalability. However, it

has consumed more energy than other MAC protocols

because of the conﬂict detection interval.

REFERENCES

Akyildiz, I. F., Pompili, D., and Melodia, T. (2005). Under-

water acoustic sensor networks: research challenges.

Ad hoc networks, 3(3):257–279.

Alfouzan, F., Shahrabi, A., Ghoreyshi, S., and Boutaleb,

T. (2018a). An efﬁcient scalable scheduling mac

protocol for underwater sensor networks. Sensors,

18(9):2806.

Alfouzan, F., Shahrabi, A., Ghoreyshi, S. M., and Boutaleb,

T. (2016). Performance comparison of sender-based

and receiver-based scheduling mac protocols for un-

derwater sensor networks. In Network-Based Infor-

mation Systems (NBiS), 2016 19th International Con-

ference on, pages 99–106. IEEE.

Alfouzan, F., Shahrabi, A., Ghoreyshi, S. M., and Boutaleb,

T. (2017). Efﬁcient depth-based scheduling mac pro-

tocol for underwater sensor networks. In Ubiquitous

and Future Networks (ICUFN), 2017 Ninth Interna-

tional Conference on, pages 827–832. IEEE.

Alfouzan, F., Shahrabi, A., Ghoreyshi, S. M., and Boutaleb,

T. (2018b). An energy-conserving depth-based lay-

ering mac protocol for underwater sensor networks.

In 2018 IEEE 88th Vehicular Technology Conference

(VTC), pages 1–6. IEEE.

Alfouzan, F., Shahrabi, A., Ghoreyshi, S. M., and Boutaleb,

T. (2018c). Graph colouring mac protocol for under-

water sensor networks. In 2018 IEEE 32nd Interna-

tional Conference on Advanced Information Network-

ing and Applications (AINA), pages 120–127. IEEE.

Ghoreyshi, S. M., Shahrabi, A., and Boutaleb, T. (2016).

A novel cooperative opportunistic routing scheme for

underwater sensor networks. Sensors, 16(3):297.

Ghoreyshi, S. M., Shahrabi, A., and Boutaleb, T. (2017).

Void-handling techniques for routing protocols in

underwater sensor networks: Survey and chal-

lenges. IEEE Communications Surveys & Tutorials,

19(2):800–827.

Ghoreyshi, S. M., Shahrabi, A., and Boutaleb, T. (2018a).

An efﬁcient auv-aided data collection in underwater

sensor networks. In 2018 IEEE 32nd International

Conference on Advanced Information Networking and

Applications (AINA), pages 281–288. IEEE.

Ghoreyshi, S. M., Shahrabi, A., and Boutaleb, T. (2018b).

A stateless opportunistic routing protocol for under-

water sensor networks. Wireless Communications and

Mobile Computing, 2018.

Hsu, C.-C., Lai, K.-F., Chou, C.-F., and Lin, K.-J. (2009).

St-mac: Spatial-temporal mac scheduling for under-

water sensor networks. In INFOCOM 2009, IEEE,

pages 1827–1835. IEEE.

Preisig, J. (2007). Acoustic propagation considerations for

underwater acoustic communications network devel-

opment. ACM SIGMOBILE Mobile Computing and

Communications Review, 11(4):2–10.

Xie, P. and Cui, J.-H. (2007). R-mac: An energy-efﬁcient

mac protocol for underwater sensor networks. In

2007 International Conference on Wireless Algo-

rithms, Systems and Applications, pages 187–198.

IEEE.

Xie, P., Zhou, Z., Peng, Z., Yan, H., Hu, T., Cui, J.-H., Shi,

Z., Fei, Y., and Zhou, S. (2009). Aqua-sim: An ns-2

based simulator for underwater sensor networks. In

OCEA. 2009, MTS biloxi-marine tec. for our fut.: glo.

and local chall., pages 1–7. IEEE.

A Comparative Performance Evaluation of Distributed Collision-free MAC Protocols for Underwater Sensor Networks