You are on page 1of 25

Received: 2 March 2021 Revised: 7 November 2021 Accepted: 9 November 2021

DOI: 10.1002/dac.5055

RESEARCH ARTICLE

ACOMAR: ACO-based MAC-aware routing protocol

Djamila Mechta1 | Oussama Mahgoun1 | Saad Harous2

1
LRSD Laboratory, Computer Science
Department, College of Sciences,
Summary
University of Ferhat ABBAS Sétif-1, Sétif, In this work, we present a new nature-inspired cross-layer routing protocol
Algeria
named ACO-based MAC-aware routing protocol (ACOMAR). An enhance-
2
Department of Computer Science,
ment of time division multiple access (TDMA) algorithm is proposed where
College of Computing and Informatics,
University of Sharjah, Sharjah, slots allocation algorithm considers the geographical distance between nodes
United Arab Emirates and the sink. The nodes that are far away from the sink have better chance to

Correspondence
be scheduled first. This strategy will result in more efficiency than a usual ran-
Djamila Mechta, University of Ferhat dom TDMA schedule. This algorithm uses a metric based on the pheromone's
ABBAS, Sétif-1, Algeria. value of Ant Colony Optimization (ACO) to choose the best next hop with the
Email: mechtadjamila@univ-setif.dz
aim of reducing latency and energy and extending the network lifetime. Com-
parisons are made between ACOMAR, MAC-aware routing (MAR) and the lat-
est cross-layer protocol latency and energy MAC-aware routing (LEMAR). The
experimental results show that ACOMAR outperforms considerably LEMAR
by 80%, 20%, 81%, and 54% in latency, energy, hop count, and throughput,
respectively. Consequently, ACOMAR is a suitable solution for healthcare
applications where WSN-based critical IoT devices interact quickly in emer-
gence cases (high rate in latency optimization 80%).

KEYWORDS
cross-layer routing protocols, nature-inspired methods, optimization, QoS, scheduling,
WSNs

1 | INTRODUCTION

Wireless sensor networks (WSNs) have become in the last few years a highly active application/research domain and
has attracted the attention of researchers due to various civilian and military applications. They consist of a number of
sensor nodes deployed randomly that are interconnected in a wireless manner to collect data of the surrounding envi-
ronment and rely it to the base station (BS) known as the sink.
Since WSN nodes are resource-constrained, reducing latency and conserving energy are prime challenges. In the lit-
erature, a number of protocols have been proposed to attempt to solve these challenges.
All protocols used in the literature can be classified into two classes: event-oriented and monitoring-oriented. In the
first class, the sensor nodes send the data when they receive a request. By contrast, in the second class, the sensor nodes
collect the data and send them periodically. For both classes, there are two ways of communication single-hop and
multi-hop. In the single-hop communication, all the sensors send the data collected directly to the sink. In the multi-
hop communication, the sensors send the data to a neighbor node and those nodes do the same thing until the data
reach the sink.
These protocols are either routing or MAC protocols that are generally implemented independently which leads to
efficient performances in terms of metrics linked to the network or the data link layers, respectively.

Int J Commun Syst. 2021;e5055. wileyonlinelibrary.com/journal/dac © 2021 John Wiley & Sons Ltd. 1 of 25
https://doi.org/10.1002/dac.5055
2 of 25 MECHTA ET AL.

The routing protocols use several QoS mechanisms such as clustering,1-3 routing,4-6 sink mobility,7,8 data
aggregation,9,10 and hybrid11-13 that are based on three different categories, such as classical, optimization, and machine
learning techniques.
For the classical routing, authors in Zhang et al14 proposed an energy-balanced routing method based on forward-
aware factor (FAF-EBRM), which selects the next hop depending on the link weight and forward energy density. Simu-
lations show that FAF-EBRM outperforms LEACH and EEUC in terms of QoS WSN metrics.
Considering optimization category and taking advantages of the bio-inspired approach, Shi et al15 proposed two bio-
inspired clustering routing protocols CNN-LEACH based on hamming network and SMPSO-BP based on neural net-
work. Their hybridization is applied in the WSN data routing process to efficiently enhance the network lifetime and
reduce the redundant data and the energy consumption. Maryam and Naji16 proposed a new decentralized hierarchical
cluster-based routing algorithm for WSNs where clustering and multi-hop routing algorithms are performed during the
same stage to decrease the number of control packets. In their study,17 Zhang et al. proposed an improvement of
dynamic source routing protocol (DSR) for MANETs taking into account the nodes' energy information. This novel
algorithm is a hybridization of genetic (GA) and bacterial foraging optimization (BFO) algorithms to determine the
optimal routing. The GA is performed firstly where its output is used as initialization for BFO algorithm. The simula-
tion shows that the proposed algorithm converges quickly towards the optimal global solution. Authors18 proposed a
new routing protocol named quantum genetic optimized link state routing (QG-OLSR) based on an improved quantum
genetic strategy where a new Q-Learning strategy is embedded. Experiment results show that QG-OLSR selects MPR
efficiently and performs efficiently during transmission phase. In their study,19 Zhang et al. proposed a time duration-
based link model where the link reliability is combined and the heuristic Q-learning algorithm is used to propose a reli-
able self-adaptive routing algorithm (RSAR) for VANET.
Zachariah and Kuppusamy20 proposed HOCK and HECK energy-efficient clustering algorithms for homogeneous
and heterogeneous networks, respectively. Both these algorithms used Krill herd and Cuckoo search for computing the
optimal cluster centroid positions and selecting the optimal cluster heads. The performance of the HOCK algorithm is
evaluated by varying base station locations and node density. To evaluate the HECK algorithm, several levels of hetero-
geneity are considered. The simulation results show that the proposed protocols increase efficiently the network life-
time. Zhang et al21 proposed a novel unequal clustering routing protocol considering energy balancing based on
network Partition & Distance for mobile education (UCNPD). This protocol is based on network partition and distance
considers the energy of CH, the distances to BS and the degrees of node to improve the WSN performance. In their
study,22 Zhang et al. proposed a novel passive multi-hop clustering algorithm (PMC) for VANET. This method aims to
solve the lack of reliability caused by the rapid mobility of vehicles in the cluster. The PMC algorithm is based on the
idea of a multi-hop clustering algorithm that ensures the coverage and stability of cluster. In the cluster head selection
phase, a priority-based neighbor-following strategy is proposed to select the optimal neighbor nodes to join the same
cluster. Comparisons with N-HOP, VMaSC and DMCNF algorithms have been realized to show the efficiency of the
proposed method. In their study,23 Abidoye and Kabaso proposed an ACO-based energy-efficient hierarchical routing
in WSNs based on fog computing (EEHFC). The fog nodes process the received sensed data from WSN' nodes in order
to optimize their power source. The performance of the proposed scheme is compared to P-SEP,24 EDCF,25 and
RABACO26 schemes using real data sets from the wireless environment. Results show that EEHFC performs better than
these three schemes. Chen et al27 proposed a cooperative communication strategy considering both V2I and V2 V com-
munications, mobility of vehicles, and cooperation among vehicles and infrastructure to improve the capacity of vehicu-
lar networks (i.e., the maximum average data rate). Results show that this strategy is more efficient in the case when
the number of vehicles requesting download is low.
Zhang et al28 proposed an efficient comprehensive sensing technology-based data aggregation. Comparisons of
experiments' results show that the proposed method outperforms the relative ones in terms of network total energy con-
sumption either in intra or inter-cluster. Zhang et al29 proposed a new method of data missing estimation with tensor
heterogeneous ensemble learning, for IoV, based on fuzzy neural network (FNN) called FNNTEL. The tensor mode is
adopted to model traffic velocity data. Experiments on real data sensed in Guangzhou and Tianjin, China, show that
FNNTEL outperforms other commonly used technologies and different missing data generation schemes. Zhang et al30
proposed an ad hoc on-demand multi-path distance vector (AOMDV) routing protocol based on link lifetime and
energy consumption prediction (named LLECP-AOMDV) for mobile edge computing. The optimal routing path is
selected according to their links' lifetime and the minimum energy consumption. Experiments results show that
LLECP-AOMDV outperforms AOMDV, AOMR-LV, and FF-AOMDV protocols. Yang et al31 proposed an optimal base
station (BS) antenna downtilt in downlink cellular networks in order to solve area spectral efficiency (ASE) crash
MECHTA ET AL. 3 of 25

problem. Experimental results guide to set the optimum downtilt angle and consequently maximize the network perfor-
mance. In their study,32 Gao et al. proposed a reliable data routing decision scheme for VANETs based on Manhattan
mobility by analyzing frequently moving vehicles and network connectivity using a probabilistic model. Experiments
show that the new scheme out-performs other baseline protocols in terms of package delivery ratio, time delay and
wireless hops. In their study,33 Huang et al. proposed an allocation scheme for optimization based on user requirements
in a cloud data center. It allocates virtual machines (VMs) to appropriate physical machines (PMs) based on hardware
resources and the throughput. The experiments show that the proposed method is efficient in terms of total energy con-
sumption, CPU utilization, number of PMs used and number of service-level agreement (SLA) violation. A cloud-edge-
based dynamic reconfiguration to service workflow for mobile eCommerce environments is proposed in Gao et al34
where the service neural network-based stability and cost metrics are considered.
Considering machine learning techniques,35 numerous protocols are proposed such as Mehmood et al36 and Huang
et al37 using artificial neural networks and deep learning, respectively.
The MAC protocols use different techniques to access the channel of communication. Farag et al38 proposed an effi-
cient slot stealing medium access control (SS-MAC) for industrial WSNs (IWSNs) designed for mission- and time-
critical applications, which need timely and deterministic data delivery to avoid difficult situations. The protocol was
evaluated mathematically and compared to WirelessHART standard where its performance is proven. A collision-free
time slot scheduling MAC protocol based on multi-level quorum system is proposed for heavy loaded underwater WSNs
(UWSNs).39 This protocol assigns different time slots on demand according to the data transmission amount of nodes
in the same collision area. Simulation results prove that the proposed MAC protocol performed efficient in terms of
avoiding collision, reducing delay, improving system energy-efficiency, and scalable. In their study,40 Zhang et al. pro-
posed an aware-energy game theory-based subcarrier allocation (EESA-EG) that takes into consideration the subcarrier
allocation problem. They modeled the problem as a nonlinear fractional programming problem where the total data
transmission rate limit, total power consumption, and energy consumption on a single subcarrier are constraint condi-
tions. This scheme allocates subcarriers efficiently in terms of energy with better channel state. In their study,41 Zhang
et al. proposed a novel approach to map correlation of ID for RFID anti-collision. Experiments' results show that this
method can greatly improve the recognition efficiency of the system in both cases small/large and static/dynamic num-
ber of tags. Li et al42 optimized the size of the contention window theoretically to reduce both latency and energy con-
sumption. Then, they proposed a global view based adaptive contention window (GV-ACW) MAC protocol to improve
more the network performances. The theoretical analysis and experimental results show that GV-ACW outperforms
the previous MAC protocols in terms of network lifetime and latency.
Huang et al43 proposed a parallel joint optimized relay selection (PORS) protocol to reduce collision, delay as well
as energy consumption for wake-up radio enabled WSNs. The basic consecutive packet transmissions (CPT) method
and maximum waiting queue length and waiting time thresholds are used to define the active nodes (i.e., those that
can transmit data). The selection of the relay considers the number of data packets, waiting time, and remaining
energy. Experiment results show that the proposed strategy outperforms the fixed relaying nodes.
However, the independence of these two categories of protocols (MAC and routing protocols) does not allow the
optimization of the overall performance of the network. The construction of paths to transmit the sensed data may lead
to high latency, because the temporary criterion given by the MAC protocol is not transparent to the network layer.
Conversely, the scheduling of communications (time division multiple access [TDMA]) made by the MAC layer may
disturb the paths chosen by the routing protocol. So these two types of protocols routing and MAC must be correlated
in order to ensure a minimization of communication latency and energy consumption and this by adopting cross-lay-
ering. In this latter approach, either a protocol uses information from another protocol or by combining two or more
protocols to achieve the desired functionalities.

2 | MOTIVATION

There are very few protocols that use the TDMA schedule and the routing tree to reduce the energy consumption con-
sumed and latency. To the best of our knowledge, there are only three classic approaches MAC-aware44-46 (for more
details, see Section 3) in the literature that use the random TDMA schedule information to optimize only energy con-
sumption and latency.
Table 1 shows the characteristics of MAC-aware protocols that use TDMA schedule temporal information when
routing data. ORW44 protocol uses duty cycle principle. Protocols MAR45 and LEMAR46 use, respectively, a two-hop
4 of 25 MECHTA ET AL.

TABLE 1 Protocols of comparison

Protocols ORW44 MAR45 LEMAR46 ACOMAR


Approach Classic Classic Classic Bio-inspired
Neighborhood Node wakes up first Two-hop One-hop One-hop
TDMA Duty cycle Random random Distance-based
Terms used to select the next hop
Distance to the neighbor ✓ ✓ ✓
Distance to the sink ✓
Number of slots ✓ ✓ ✓
Residual energy ✓ ✓
QoS metrics used to measure the performance
Energy consumption ✓ ✓ ✓ ✓
Latency ✓ ✓ ✓ ✓
Packet delivery ratio (PDR) ✓
Hop count ✓
Throughput ✓

(requires several exchanged messages) and one-hop neighborhood knowledge to determine the sender node next hop.
Moreover, MAR exploits the number of slots and geographic distance. LEMAR added a third metric, which is the
energy of nodes.
The related contributions44-46 have limitations in the determination of best paths, which node should be the first
sender (random character in MAC layer does not allow to select carefully the node that starts first the transmission)
and its best neighbor chosen by a classic approach that uses in best cases one-hop neighborhood knowledge and
neglects other necessary metrics. Moreover, for experimental results, they only consider latency and energy consump-
tion to measure their performance.
In this work, we are interested in the monitoring-oriented approach and the multi-hop communication. The main
purpose is to propose a new cross-layer routing protocol ACOMAR taking advantages of previous related works44-46
while overcoming their disadvantages. This protocol contains two phases: During the first phase, it builds a distance-
based TDMA instead of the random one used in the previous protocols. This leads to the selection of the best first
senders (the farthest nodes from the sink) with their optimal neighbors, and therefore, it improves the latency. The sec-
ond phase deals with the determination of the efficient path to transmit the sensed data. This is by using one-hop
neighborhood knowledge instead of two-hop that strongly affect the energy consumed because it requires the exchange
of several messages. The proposed ACOMAR exploits the temporal information of the proposed distance-based TDMA
schedule provided by MAC layer, residual energy of sensor nodes, distance to the direct neighbor, and the number of
time slots. Furthermore, we added a fourth term, which is the distance from the neighbor to the sink, when choosing
the sender best direct neighbor and consequently establish the optimal path to be used for routing the sensed data. The
neighbor that is closest to the sink has a high probability to be the best receiver. All these changes aim to reduce latency
and increase the network performance. This protocol is able to transfer the data collected by the sensor nodes to the
sink and achieves a better energy efficiency and lower communication latency. This proposed solution is based on the
attractive nature-inspired ACO algorithm unlike the classic approach (used in other studies44-46). A TDMA schedule
aims to eliminate the collision during the transmission, the proposed schedule is built based on two metrics. The first
metric is the distance from a node to its neighbors and to the sink. The second one is the one-hop neighborhood knowl-
edge, which minimizes the exchanged messages compared to two-hop neighborhood.
In this paper, the main contributions are summarized as follows:

1. Propose two taxonomies, which classify the different cross-layering protocols based on the QoS and the type of the
approach used (classic or bio-inspired) in developing the solution.
2. Review and compare the protocols, available in the literature, according to their type (MAC-aware/routing-aware),
QoS criteria (energy consumption and latency of communications), objective, method-based (heuristic/meta-heuris-
tic) and path metric.
MECHTA ET AL. 5 of 25

3. Propose a distance-based TDMA schedule (MAC layer) and ACO-based MAC-aware Routing protocol (routing Layer).
4. Performance analysis of proposed solution by including further QoS metrics such as packet delivery ratio (PDR),
hop count, and throughput.

The reminder of this article is organized as follows: In Section 3, we review the literature, analyze various existing
research works in this field and propose a taxonomy of the different cross-layer protocols for scheduling problem used
in WSNs. Section 4 presents preliminaries. Section 5 discusses ACO-based MAC-aware routing protocol (ACOMAR).
Section 6 presents the solution analytical evaluation. In Section 7, we present, compare, and discuss the experimental
results and performance evaluation. Section 8 concludes the paper.

3 | R E LA T E D WOR K

In the literature, there are several approaches that focus on the reduction of energy consumed and latency in WSN
using different strategies. In this section, we focus on scheduled-based protocols, which use the strategy of TDMA to
schedule the communication between nodes.
In scheduled-based protocols, the contention between nodes on the transmission medium is completely eliminated
and the collision is less by comparison with contention protocols. The idea of these scheduling protocols is to assign to
each node in the network a specific time slot for the transmission. During this time slot, the node has a full access to
the channel and can use it until the end of the slot. For the reception, a number of slots which equal to the number of
node's neighbors is given. In the remaining slots that are not used neither for transmission nor for reception, the node
switches to sleep mode to save its energy.
In the literature, several MAC-aware protocols have been proposed. These protocols aim to improve sensors perfor-
mances by decreasing the energy consumption and increasing the lifetime of the whole network. However, the vast
majority of them has some shortcoming.
Opportunistic routing for WSN (ORW)44 is a cross-layer mac-aware routing protocol. It uses the duty-cycle principle
to reduce energy consumption and the latency. The idea of ORW is to use the neighbor that wakes up first and provides
a routing progress as a forwarder. Hence, the sender sends a stream of packets until the intended receiver wakes up
and acknowledges it. ORW uses a routing metric called expected duty cycled (EDC), which describes the number of
wake-ups until a packet reaches the intended destination. When a node sends a packet, the neighbor that wakes up first
and has a smaller EDC value than the sender receives the packet and acknowledges it. This neighbor becomes then the
forwarder. ORW is considered a cross-layer protocol because it uses the information of MAC protocols. This informa-
tion consists of the knowledge of node's neighbor that wakes up first.
Among major strengths of ORW is that it can be implemented in two versions centralized and distributed. In the
centralized implementation, the sink needs a global view of the network to calculate the EDC value of all nodes in the
network. In the distributed implementation an iterative process is required to stabilize the computation of the EDC
metric. Indeed, it is clear that the node, who woke up first and acknowledged the packet, is the forwarder. Therefore,
this is the way to select the forwarder without taking into account the spatial information such as the distance between
nodes and the residual energy.
MAC-aware routing for WSN (MAR-WSN)45 is a cross-layer MAC-aware protocol. It uses the information of a ran-
dom TDMA schedule and the nodes spatial information in order to reduce the energy consumption and the latency in
the network. MAR-WSN uses the time slot information of a random TDMA schedule by exploitation of one-hop and
two-hop neighbors sets, when looking for the next hop for each node to transmit the data. The forwarding sets ensure
that the data is forwarded towards the sink. The role of TDMA schedule is to ensure that each node sends and receives
the data from or to the other nodes without any collision. In Louail et al,45 it is assumed that each node in the network
knows its position, the position of its one-hop neighbors and two-hop neighbors and the position of the sink.
A weighted distance delay (WDD) selection function based on delay and distance metrics is defined. Each node uses
this function to get its best optimal next-hop among the forwarding sets. The Forwarding node for node i is node j that
belongs to one-hop neighbors set that has a neighbor k in two-hop neighbors set for which WDD has the lowest value.
This process is repeated until the data reaches the sink.
The use of TDMA schedule improves the performance of MAR-WSN as shown in its results. Also, the metric WDD
ensures that the next hop is the best progress in terms of delay and distance. It provides a better performance than
ORW in terms of average energy consumption and average latency. However, MAR-WSN has some limitations and
6 of 25 MECHTA ET AL.

disadvantages. MAR-WSN uses a random TDMA schedule which leads to a random order in transmission. Some nodes
that are far away from the sink might need to take the first slot to transmit their data and this gives a better result in
term of latency. Moreover, MAR-WSN uses two-hop neighboring knowledge, which requires several messages
exchanged in the network. This large number of messages requires from nodes a significant energy for the transmission
and the reception to and from the other nodes. Also, the sender uses the same node as a forwarder in each transmission
and it does not take into account the decrease of its residual energy.
Latency and energy MAC-aware routing (LEMAR-WSN )46 is a cross-layer MAC-aware protocol that uses the tem-
poral information of a random TDMA schedule, the geographical position of nodes (one-hop neighbors) and the energy
consumed by each node in order to choose the next hop. Each node knows its own geographical position, the position
of its one-hop neighbors and the position of the sink. Each node in the network chooses the next hop using a metric
called slot distance energy (SDE). This metric helps the node to choose the nearest and the cheapest next hop in terms
of energy consumption and latency.
Unlike MAR-WSN that uses two-hop neighborhood knowledge, LEMAR-WSN uses one-hop neighborhood knowl-
edge. This reduces the number of messages exchanged in the network by half. The node chooses the best hop in terms
of energy and latency. A different next hop might be chosen by the node at each transmission due to the variation of
energy consumption that affects the SDE value too. It means that the sender chooses the best next hop in terms of both
energy and latency. This way of selecting the next hop has proved to be more efficient than the technique used by
MAR-WSN (the same node as a forwarder in each transmission).
Although LEMAR-WSN provides an efficient way to reduce energy consumption and latency by choosing the best next
hop in terms of energy consumed, distance, and latency, it has some limitations. We notice that LEMAR-WSN uses a random
TDMA schedule like MAR-WSN to order the nodes. It can be enhanced using some kind of metrics to assign priority to nodes
based on the distance of the nodes from the sink. Also, it uses the geographical distance between the nodes and the sink
which is not always accurate, it depends on the path length to the sink that can be longer than the geographical distance.
The length of the path in red color, shown in Figure 1, is longer than the one in blue despite that the geographical
distance between Node 1 and its one hop neighbors (Nodes 2 and 3) is considered in such way that the closer neighbor
is chosen.
Effective time division multiplication access based on genetic algorithm (ETDMA-GA)47 is a TDMA scheduling
based on genetic algorithms. It uses latency as a fitness function in order to reduce the communication latency in the
WSN. The scheduling algorithm is divided into two principal phases: initialization and TDMA schedule construction.
In the initialization phase, the sink constructs a routing tree like min hop or depth first trees using the information of
sensor nodes. In the second phase, the sink constructs a routing tree traversal, calculates the TDMA schedule, and
broadcasts both of routes and TDMA schedule to all sensor nodes in WSN.
ETDMA-GA finds the best routing tree ordering and the corresponding TDMA schedule. The TDMA schedule is
designed using a two-dimensional chromosomes matrix with R rows and W columns. R is the number of time slots, and W
is the number of sensor nodes. The number of time slots is equal to the maximum interference degree in the network. A
gene in a chromosome represents one time slot whereas the value of gene indicates the number inside the slot. The TDMA
schedule is constructed in several iterations. The fitness function used is based on latency of each sensor node in WSN.

FIGURE 1 Network topology example


MECHTA ET AL. 7 of 25

The usage of GA to enhance the TDMA schedule leads to efficient results in term of latency in WSN. The authors
claim that ETDMA-GA outperforms existing state-of-the-art approaches in terms of average latency, average normal-
ized latency, and the average length of schedule. ETDMA-GA is a centralized contention-free approach. The TDMA
schedule and the routes information are generated by the sink. Thus, a significant energy is consumed by the sink to
create the TDMA schedule, construct the routing tree, and broadcast this information to the nodes. The enhancement
of the TDMA schedule is not enough to reduce the latency where the energy is important factor that must be taken into
consideration when optimizing WSN performances.
IDeg-LO and IDegRe-LO48 both are a routing-aware approach that exploits the information from the routing layer
to construct a TDMA schedule. IDeg-LO uses the routing tree and the interference degree to schedule the sensor nodes.
The algorithm begins by scheduling the leaves based on the interference degree in descending order. The authors define
the interference degree to be the number of one hop and two hop neighbors (a node is counted only once because it
could happen that a node is one-hop and two-hop neighbors at the same time). After scheduling the leaves, the internal
nodes are scheduled in the same order as the leaves. The internal nodes are scheduled as close as possible to their chil-
dren. If there is no slot that a node can use for its transmission, a new slot is added to the schedule. IDegRe-LO follows
a very similar process as IDeg-LO. The main difference is that IDegRe-LO removes from the routing tree the leaves that
have been already scheduled and the process continues with the internal nodes, which gives a chance to the nodes that
have high degree of interference.
Both IDeg-LO and IDegRe-LO perform efficiently in terms of latency. Both schemes proposed an efficient way to
reduce the latency in WSN. IDeg-LO and IDegRe-LO improve the average latency by up to 44.14% and 53.33%, respec-
tively. However, it is still not enough to extend the network lifetime because energy consumption is not taken into con-
sideration. IDeg-LO may schedule a parent before some of its children, which may lead to the parent wasting some time
waiting for its children that are scheduled after its sending slot. Moreover, the internal nodes that have a high interfer-
ence degree do not have a chance to be scheduled soon, because they are scheduled in the same order as their leaves.
We propose in this paper two taxonomies for these protocols, shown in Figure 2, where the first one is method nature-
based (classic or bio-inspired) and the second is QoS-based (latency-aware, energy-aware, or latency/energy-aware).
We have noticed that most of the mentioned protocols are based on heuristics. Only one of them uses an inspired
method to enhance the TDMA schedule, which performed efficiently results show better performance in terms of
latency. The effectiveness of meta-heuristics has been proven in different fields.
Table 2 summarizes the studied protocols corresponding to the QoS in terms of energy and latency, their involved
layers (MAC or routing), algorithm nature (distributed or centralized) and the method used (classic or nature-inspired).
All these protocols are centralized except ORW is hybrid (centralized and distributed).

4 | PRELIMINARIES

4.1 | Network topology or system model

We model the WSN as an undirected graph. The sensor nodes are represented by numbered vertices and the edge repre-
sents the communication between two sensor nodes. The nodes of the network are numbered from 1 to n and each node

FIGURE 2 Proposed taxonomies for existing cross-layer protocols


8 of 25 MECHTA ET AL.

TABLE 2 Comparison between TDMA scheduling protocols

Protocols Type Objective Energy Latency Method based Path metric


44
ORW Mac-aware routing Reduces the energy Low Low Classic (heuristic) First awoken
consumed & latency neighbor
MAR-WSN45 Mac-aware routing Reduces the energy Low Low Classic (heuristic) Neighbor with
consumed & latency low WDD
LEMAR-WSN46 Mac-aware routing Reduces the energy Low Low Classic (heuristic) Neighbor with
consumed & latency low SDE
IDeg-LO48 Route-aware Mac Reduces Latency — Medium Classic (heuristic) Parent (routing tree)
IDegRe-LO48 Route-aware Mac Reduces Latency — Low Classic (heuristic) Parent (routing tree)
ETDMA-GA47 Route-aware Mac Reduces Latency — Low Inspired (meta-heuristic) Parent (routing tree)

FIGURE 3 Sensor network example

has a unique identifier i different from the other nodes where i  [1, n] and n is the number of nodes in the network.
Also, each node is connected to at least one neighbor, which means the graph connected. The nodes are deployed in
random way in the coverage area. All nodes have the same transmission range. Only one node is designated as the sink.
Figure 3 shows a connected graph that models a network of 14 nodes where node number 1 is the sink.
In this example, the neighbors of node 10 (i.e., one-hop neighbors) are {3, 7, 9, 12}, and the neighbors of the neigh-
bors of node 10 (i.e., two-hop neighbors) are {1, 2, 4, 6, 14}.
All data gathered by sensor node must be sent to the sink, either directly without passing by any other node or by
using multi-hop technique. The direct technique requires an enormous amount of energy to send the data because the
distance between some sensor nodes and the sink could be longer than the transmission range. So it is necessary for
these distant sensor nodes to use the multi-hop technique to send the data to the sink.

4.2 | Energy model

In WSN, the node has three different modes (standby, transmission, and reception). The energy consumption is differ-
ent from one mode to another. The transmission consumes more than reception or standby mode; energy consumption
is at a low level in standby mode where the node saves its energy.
In this work, we assumed that all sensor nodes have initially equal energy that is decreased over time by different rate
depending on the node's mode. The time is divided into slots. For example, Table 3 shows the time division for the network in
MECHTA ET AL. 9 of 25

Figure 3. Whenever the transmission slot of any node starts, the node uses some of its power to transmit its data to the suitable
next hop. Then, we compute the energy consumed during the transmission and the reception using LEACH model as follows:
The energy consumed by the node in the transmission is calculated according to formula (1).

E T i,j ðk, di,j Þ ¼ E elec ∗ k þ ϵamp ∗ k ∗ d2i,j : ð1Þ

The energy used by any node to receive a packet is computed according to Equation (2)

ERi ðkÞ ¼ E elec ∗ k, ð2Þ


where:

• Eelec is the energy consumed by node to start transmitter or receiver circuit,


• k is the length of the data in bits,
• ϵamp represents transmit amplifier, and it equals 100 pJ/bit/m2,
• di, j represents the distance between the sender i and receiver j.

5 | P R O P O S ED A C O - B A S E D M A C - A W AR E R O U T I N G P R O T O C O L : A C O M A R

In this section, we describe in detail the QoS-aware cross-layer solution for the data routing in WSN networks (ACOMAR).
The solution is composed of the distance-based TDMA and the ACO-based routing phases as summarized in Figure 4.
Figure 4A.(a) describes the Distance-based TDMA schedule algorithm 1. The first statement (V is empty) is to make
sure that all nodes in the vector V got one transmission slot. If the statement is false, the farthest node in the vector V is
picked up and a loop starts to find the appropriate slot for it. Once there is no slot that can be given to the node, the
statement (allocate) is false and the slot counter S is increased by 1. After each node gets its slot, the first statement
(V is empty) becomes true, which means the vector of nodes V is empty. Table 3 represents a distance-based TDMA
schedule for the network in Figure 3 generated using Algorithm 1.
Figure 4B represents the flowchart of ACOMAR protocol. After the pheromone is initialized to the inverse of the dis-
tance between each node and its neighbors, each node saves its residual energy, the residual energy of its neighbors, and
the distance between the neighbors and the sink. When a node “i” has a data to send which, is described in the state-
ment (node “i” is ready to transmit), it calculates the probability of the next hop. Once the packet reaches the destination
node, the pheromone on the path used by the packet is updated. Each node in the network follows the same process.

5.1 | Distance-based TDMA schedule (MAC layer)

TDMA is a commonly used technique in network to avoid collision when nodes send their data. Time is divided into
equal intervals called slots; each node has a dedicated slot to transmit its data. During other slots, the node could either
in sleeping mode to save energy, or in receiving mode to receive data from other nodes. In the previous related works,
the random TDMA is used to avoid collisions and coordinate the access to the channel. The slot's allocation method
runs randomly which means a slot is assigned to a node for transmission with only one constraint, the neighborhood
constraint where node cannot be scheduled in the same slot with its one hop or two hop neighbors because collision
may be caused when the data is sent at the same time by nodes that are close to each other.
The use of random TDMA schedule may lead to inefficient results in terms of latency. It means that some nodes
could waste a considerable time because the establishment process of the schedule does not take into consideration any
information about the geographical position of nodes (except the neighborhood constraint) or the distance between the
sender and the sink. This information is an important factor that could give a better result if it is used during the crea-
tion of the schedule. In this work, we use the distance between the node and the sink in addition to the neighborhood
criterion to create the TDMA schedule.
We build a TDMA schedule based on the spatial information of nodes (see Algorithm 1), where the nodes far away
from the sink are more likely to be scheduled first. The sink is scheduled to use the slot allocation process because its
main task is to collect the data from all nodes and then send it to the BS. In addition, the nodes that are so close to the
sink can take the same slot the farthest nodes took but the neighborhood constraint must be respected.
10 of 25 MECHTA ET AL.

FIGURE 4 Proposed solution flowchart


MECHTA ET AL. 11 of 25

T A B L E 3 An example of a distance-based TDMA


schedule for the network of Figure 3

We represent the TDMA schedule by a matrix MNxL where N is the number of rows (number of nodes in the net-
work) and L represents the number of columns (number of slots). To assure collision free transmission, the sender and
its one-hop and two-hop neighbors cannot use the same slot. The minimum value that can be chosen as a number of
slots is equal to the maximum degree in the graph plus 1.
The elements of the TDMA schedule in Table 3 are integers that have the following specific meanings:

1. TDMA[i, j] = i, means that node i uses the slot j for the transmission, for example in Table 3:
- Distance-based TDMA½13, 3 ¼ 13, node 13 uses slot number 3 as a transmission slot.
2. TDMA[i, j] = k, means that node i in the slot j is in receiving mode, where it receives the data sent by node k (node
k is a direct neighbor of node i ), for example,
- Distance-based TDMA½14, 3 ¼ 13, Node 14 receives data from Node 13 in slot number 2, that is, the slot for receiv-
ing data from Node 13 by Node 14 is slot number 3.
3. TDMA½i, j ¼ 0, means the node i in the slot j is in sleep mode, it does not send or receive data to or from any node.

The nodes that have the same color can transmit their data in parallel way without collisions.

5.2 | ACO-based routing model (routing layer)

The sensor nodes in WSN collect data from the surrounding area to be send after to the sink using a routing technique.
Most of routing protocols establish routes between each node in the network and the sink. Those routes form a routing
tree. Each node selects one node to forward data to and this node is called parent.
Figure 5 illustrates an example of a routing tree of the network shown in Figure 3.
Ant colony optimization (ACO) algorithm is one of the most efficient and well-regarded algorithms. ACO is member
of ant colony algorithms family and has been applied to a range of problems in the literature. It aims to find approxi-
mate solutions to difficult optimization problems. It can be used, in the context of WSN, to find the best path in terms
of energy consumption and transmission latency. The ACO algorithm uses the pheromone concentration to decide
which path costs less than the others. In real world when ants are looking for food, they lay down a chemical on the
path to direct other ants to the food, the ants choose the path with the highest concentration of pheromone.
In this work, we propose a new cross-layer routing approach based on ACO. It uses the information coming from
the TDMA schedule, the spatial information where each node knows the distances to its direct neighbors and the resid-
ual energy at each node. The aim is to minimize nodes' energy consumption by nodes, extend the network lifetime and
reduce the communication latency between the nodes.
12 of 25 MECHTA ET AL.

FIGURE 5 Fig A routing tree for the network in Figure 3

TABLE 4 Similarity between ACO and WSN

ACO WSN
Food source Sink
Food placement Sink position
Ants colony Sensor nodes
Ants nest Sender node i
Pheromone initial value Invert of the distance between node i and its neighbor
Probability Pi, j Fitness function
Best value of fitness function Optimal next hop

We assume that each node knows its distance to the sink, its one hop neighbors, the distance to each one them, and
the distance of each one of them to the sink. In addition to the spatial information, each node knows the residual
energy of all its direct neighbors.
Each sender node runs ACOMAR to determine its best optimal next hop until it reaches the sink. The sequence of
optimal next hops constitutes the efficient-routing path from the furthest node (sender) to the sink (destination).
The ACO-based efficient-routing technique consists of four steps that are carried out, iteratively, for each sender
node as shown in Algorithm 2. The sender nodes represent the ant nest while the sink is the food source as shown in
Table 4. Ants are used to determine the best next hop among the sender's direct neighbors.

- Step 1 (Initialization): First, each link (i,j) between two nodes i and j in the network is initialized by a certain con-
centration of pheromone during the initialization phase of the algorithm. The initial pheromone value between nodes
i and j by ant k is computed using Equation (3).
8
< 1 k th ant travels on the link ði, jÞ
Δτi,j ¼ di,j
k
ð3Þ
:
0 otherwise,

where di, j is the distance between nodes i and j.


When the next hop j is chosen by an ant k, the pheromone of the link (i,j) connecting nodes i and j is then updated
using Equation (4).
MECHTA ET AL. 13 of 25

τi,j ¼ Σ m
k¼1 Δτi,j ,
k
ð4Þ

where m represents the total number of ants that selected the link (i, j).
- Step 2 (Best next hop selection): The efficient-path selection from the sender node i to the sink is correlated to the
nodes' next hop selection. First, the ant k at node i selects the next hop j from the direct neighbors list by calculating
a probability Pki,j according to Equation (5). Then, the roulette wheel principle is used to decide finally the best next
hop based on the computed probabilities to which the data will be sent. The probability value is defined according to
the amount of the pheromone, number of slots, residual energy percentage (as given by Equations (4), 6, and 8
respectively), and the spatial distance.

½τi,j α ∗ ½δi,j β ∗ ½σ j γ ∗ ½φj,s λ


Pki,j ¼ P α
, ð5Þ
k  F i ð½τ i,k  ∗ ½δi,k β ∗ ½σ k γ ∗ ½φk,s λ Þ

where:
P i, j is the probability of selecting node “j” as a next hop by node “i” to transmit its data;
F i is the set of one-hop neighbors of node “i” that are closer to the sink than node i, where F i ¼ k  N i ; dk,s < di,s ; di, S
is the distance between node i and the sink S;
τ i, j represents the concentration of pheromone on the link (i, j);
δi, j is the invert of the number of slots between the sending slot of node i and the sending slot of node j. It is com-
puted as follows: 1
δi,j ¼ ð6Þ
Δt i,j

where Δti, j is the number of slots between the sending slot of node i and the sending slot of node j computed using
the Distance-based TDMA schedule as follows:

ðslot j  slot i Þ mod l if slot j ≥ slot i
Δt i,j ¼ ð7Þ
ðl  slot i þ slot j Þ mod l if slot i > slot j ,

where:
l is the length of TDMA schedule
sloti represents the slot of node i.
For example, in the Distance-based TDMA schedule:

Δt 2,13 ¼ ð3  1Þ mod 7 ¼ 2, because slot 13 ¼ 3 > slot 2 ¼ 1

Δt 13,14 ¼ ð7  3 þ 2Þ mod 7 ¼ 6, because slot 14 ¼ 2 < slot 13 ¼ 3

σ j is the percentage of residual energy of node j computed as:

RE j
σj ¼ : ð8Þ
100

φj, s represents the invert of the distance between node j and the sink s where
8
< 1 if dj,s > 0
φj,s ¼ dj,s ð9Þ
:
1 Otherwise,

α, β, γ, and λ are coefficients that should be chosen properly to control the importance of measured metrics.
Table 4 shows ACOMAR's relevant modeling parameters.
14 of 25 MECHTA ET AL.

- Step 3 (Pheromone update): The pheromone value on each link (i, j) in the network is modeled as a square matrix
where the number of rows and columns equals the number of nodes in the network. Nodes that do not have links
between each other, which means they are not neighbors, the pheromone value is undefined.
The pheromone value on links between nodes that are neighbors is initialized using Equation (4). After choosing the
next hop, the pheromone value on the path between sender and receiver is updated and increased using Equation (11)
while the other unused paths the pheromone value is decreased using Equation (10).
- Step 4 (Pheromone evaporation): After the ant chooses the link with the highest pheromone concentration, it
deposits a certain quantity of pheromone, which increases the pheromone concentration in this link. In case the
pheromone did not evaporate, the chosen next hop would not be the optimal next hop for the ants because the phero-
mone accumulated will stack and the ant will not be able to take another link, it will choose the same link over and
over again. In order to avoid this scenario, we propose an evaporation mechanism to decrease the pheromone con-
centration of the chosen link between the sender and its non-selected neighbors. Equation (10) shows how the phero-
mone decreases between node i and node j:

τi,j ¼ ð1  ρÞ ∗ τi,j j  Ni ð10Þ

where ρ is pheromone evaporation rate ρ  [0, 1] and Ni represents the set of one-hop neighbors of node i. After the
ant chooses the path and arrives to the destination, it deposits a certain quantity of pheromone on the path and the
pheromone concentration on that path is updated as follows:

τi,j ¼ ð1  ρÞ ∗ τi,j þ Δτi,j ð11Þ

where Δτi, j represents the quantity of pheromone deposits by the ant on its way to the next hop as defined in
Equation (4).
All ACO-based routing stages are summarized in Algorithm 2.

6 | APPR OACH ANALYS IS

6.1 | Solution performance

• Energy consumption Since the energy is proportional to the distance, the decrease of energy consumption is
ensured by choosing the best next hop using both distance-based TDMA and ACO system. This results in the closest
node to the sender that verifies the neighborhood constraints.
• Latency Latency in WSN indicates the time needed for a packet to reach the sink from any node. In this work, the
time is expressed by the number of slots and is calculated using the TDMA schedule. The path that the packet takes
to reach the sink is calculated from the routing tree.
MECHTA ET AL. 15 of 25

The packet sent by any node in the network travels through several nodes to reach the sink. The packet sent by node
i goes through multiple nodes to reach the sink as follows:

nodei ! node1 ! node2 !   ! nodex ! sink

The latency is computed as follows:

LT i ¼ slot i þ Δt i,1 þ Δt 1,2 þ   þ Δt x1,x ð12Þ

For example, to calculate the latency of a packet sent by Node 5 in the network shown in Figure 3, we need to find the
routing path of the packet, which is
5 ! 4 ! 11 ! 6 ! 8 ! 1

- Using random TDMA schedule,

LT 5 ¼ slot 5 þ Δt 5,4 þ Δt 4,11 þ Δt 11,6 þ Δt 6,8


LT 5 ¼ 1 þ 5 þ 3 þ 4 þ 1 ¼ 14:

- Using distance-based TDMA schedule in Table 3,

LT 5 ¼ slot 5 þ Δt 5,4 þ Δt 4,11 þ Δt 11,6 þ Δt 6,8


LT 5 ¼ 1 þ 1 þ 2 þ 4 þ 2 ¼ 10:

Both TDMA schedule can be used to avoid collision that might happen if the transmission process of the sensor
nodes is not scheduled. Random TDMA schedule is established using only the neighborhood knowledge. In this case a
node, and a node from the set of its one-hop or two-hop neighbors cannot be scheduled in the same slot.
This constraint is used to establish the distance based TDMA as well in addition to the use of the geographical
distance. We used the distance between each node and the sink to assign a priority in the slot allocation mecha-
nism, in which the farthest node in the network takes the first slot. This process is repeated until all nodes are
assigned a transmission slot. If they could not find an appropriate slot, a new one will be added to the schedule
and the process is repeated until all nodes are allocated a slot. The sink node will be the last node in the schedule
process.

6.2 | Algorithms' complexity

In this section, we analyze the algorithms temporal complexity where the time is the resource quantity required to run
an algorithm. In our proposed approach, we have used distance-based scheduling and ACO-based routing algorithms to
get the best optimal solution (paths from the sender to the BS). Distance-based scheduling algorithm assures that the
time scheduling is respected during the data transmission phase in order to avoid collisions in the network. ACO-based
routing is exploited to find the best next hops among the direct nodes' neighbors when exploiting on temporal informa-
tion provided by Distance-based scheduling.
Since distance-based scheduling algorithm is lunched at the start of execution, let us compute its complexity first.
The distance-based scheduling algorithm has two nested iteration loops in addition to several elementary instruc-
tions as shown in Table 5. Assume N is the size of V (the vector of distances between nodes and sink) and S is the num-
ber of slots.
16 of 25 MECHTA ET AL.

TABLE 5 Distance-based scheduling

TABLE 6 ACO-based routing

Complex  scheduling ¼ 1 þ OðNÞ þ N ∗ ð1 þ 1 þ OðSÞ þ S ∗ ð1 þ 1 þ 1 þ 1ÞÞ þ 1 ¼ N ∗ S


because 2 þ N þ 2 ∗ N þ 4 ∗ S is less than N ∗ S:

Therefore, the time complexity of the algorithm is O(N ∗ S).


We note that the complexity of this algorithm is O(N) when S is small and it is O(N2) when S converges towards N.
ACO-based routing algorithm is lunched by each sender node. It has two nested loops where the first one is for iter-
ations and the second for ants that are looking for the sender optimal next hop as shown in Table 6. So, the ACO com-
plexity for the sender node in its transmission slot is computed as

Complex  ACO ¼ 1 þ maxiter ∗ m ∗ ð1 þ 1Þ þ 1 þ 1 þ 1 ¼ 4 þ maxiter ∗ 2 ∗ m


¼ maxiter ∗ m other values are neglected

where m is the number of ants and maxiter is the maximum number of iterations.
Since the network topology contains n sensor nodes (all of them might be sender at one time or another), the com-
plexity will be

Complex  ACO ¼ maxiter ∗ m ∗ n:

Taking into account that the sender nodes' direct neighbors (nodes  F1(sender) do not have the same transmission slots
(to avoid collision), so their number is subtracted from the total number of nodes, and then the formula will be

Complex  ACO ¼ maxiter ∗ m ∗ s


ð13Þ
where s ¼ n  jF 1 ðall sendersÞj and s < n:
MECHTA ET AL. 17 of 25

6.3 | Convergence of the solution

ACOMAR runs in two phases (distance-based scheduling and ACO-based data routing) each one in its appropriate
layer (MAC and routing layers).
In the first phase, the establishment of time scheduling is ensured by the condition of the while loop of Algorithm 1
where we use a not empty vector V that contains distances between nodes and the sink. We make sure that the vector
will be empty after several iterations and thus the algorithm converges towards the required optimal solution.
The convergence of any meta-heuristic based approach is proven in the literature including ACO system, which is
used in our solution. This ACO system guarantees the convergence to the optimal solution by choosing carefully its
parameters' values.

Theorem 1. The proposed Mac-aware routing protocol ACOMAR assures the existence of an elementary path
between any node of the network and the sink.

Proof. Let, N be the total number of nodes and S be a limited number of time slots where S is strictly less
than N (S < N). We associate to each time slot a number of nodes that can transmit data (sender nodes).
From the definition, each node n  Si sends its data to a node m  Sj such that i < j.

We suppose that there is a node, which cannot transmit its data to the sink. We assume that there exists a node
n  Si such that there is no path between node n and the sink. As n  Si, node n sends its data to another node f  Sj
such that i < j.
If the direct neighbor of node f is the sink, so the transmitted data from n is received by the sink. Consequently, the
path between node n and the sink exists. Otherwise, there is at most (S  i) time slots where one of these neighbors has
the sink as the next hop. Then, there exists at least one path between node n and the sink with at most (S  i) slots or
hops. Accordingly, any node can transmit its data to the sink, and therefore, the hypothesis is wrong.

7 | PERFORMANCE EVALUATION

7.1 | Experiment environment

To evaluate the proposed ACOMAR cross-layer protocol performance, we carried out series of experiments using NS3
network simulator. The network consists of homogeneous sensor nodes, equal initially in energy, which is set at 2 J,
scattered randomly within a targeted area of 100  100 m2. All sensors including the sink are stationary. The sensor
nodes are energy-constraint; however, we assume that the sink has unlimited power and it is fixed outside the target
area (top left side). The packets are sent from the sender nodes with size 512 Bytes where the transmission range was
20 m. For ACO, experimental parameters are chosen carefully to guarantee that ACOMAR achieves its best
performances.
Five metrics are used to evaluate the performance of the proposed scheme ACOMAR and the schemes to which it is
compared. These metrics are the energy consumption, latency, hop count, PDR, and throughput detailed in Section 7.2.
We consider scenario where we varied the number of nodes from 20 to 200 in increment of 20 nodes as shown in
Figures 6, 7, 8, 9, and 10.
A comparative study is performed between ACOMAR, LEMAR, and MAR protocols to analyze their effectiveness.
In the following subsections, we present and discuss the simulation results of ACOMAR, LEMAR, and MAR
protocols.
Table 7 summarizes the parameters used in the simulation. Table 8 shows the parameters' values for which ACO
algorithm performed best in terms of latency and energy consumption.

7.2 | Experimental results and comparisons with other schemes

To minimize variances and be consistent we run each experiment 100 times, then we took the average of the obtained
results. We evaluated the impact of the network density. In addition, we compared the proposed scheme with well-
18 of 25 MECHTA ET AL.

FIGURE 6 Energy Consumption by MAR, LEMAR, and ACOMAR

FIGURE 7 Latency in ACOMAR versus LEMAR and MAR

known cross-layer routing protocols (MAR45 and LEMAR46). LEMAR outperforms MAR, which has better perfor-
mances than ORW. We compare ACOMAR performance to MAR and LEMAR protocols because (to the best of our
knowledge) are the best existing protocols in the literature. Next, we discuss the obtained results.
MECHTA ET AL. 19 of 25

FIGURE 8 Hop counts in ACOMAR versus LEMAR and MAR

FIGURE 9 Packet delivery ratio

(A) The average energy consumption: It refers to the average quantity of power consumed by each node
in the network. It computed using the same technique as the one used by LEACH protocol (see
Equation 14).
20 of 25 MECHTA ET AL.

FIGURE 10 Throughput

TABLE 7 Simulation parameters

Parameters values
Area 100  100 m
Number of sensor nodes 20–200 nodes
Nodes' distribution Random
Initial Energy 2J (Joules)
Packet Size 512B (Bytes)
Transmission Range 20 m
Position of the sink (100,0)
Mobility model Static
Slot duration 20 ms

TABLE 8 ACO parameters

Parameters Symbols values


Pheromone variable α 0.01
Heuristic of latency β 0.1
Heuristic of energy γ 1
Heuristic of distance λ 2
Pheromone evaporation coefficient ρ 0.7
MECHTA ET AL. 21 of 25

X N 
AvgECenergyconsumption ¼ E
i¼1 initialenergyðiÞ
 E residualenergyðiÞ =N: ð14Þ

Figure 6 presents the average of energy consumption of three protocols MAR, LEMAR, and ACOMAR in small- and
large-scale networks. Only the energy consumption of data transmission is considered when computing the network
energy consumption. The results show that ACOMAR outperforms both MAR and LEMAR in terms of energy con-
sumption. Hence, it saves the sensor nodes' battery power and then prolongs the network lifetime. This is due to the
best next hop chosen by ACOMAR that consumes less energy in data transmission phase.

(B) The average latency: It is the average amount of time that the packet takes to reach the sink from any node in the
network. It is expressed in number of slots as shown in Equation (12). Figure 7 shows the average latency of MAR,
LEMAR, and ACOMAR protocols when the number of sensor nodes is varied from 20 to 200. We notice that the aver-
age latency of ACOMAR protocol does not exceed 40 even for the case when we have 200 nodes in the network. By
contrast, when LEMAR and MAR protocols are used, the average latency increases very fast to 440 and 480, respec-
tively, when the number of nodes is 200. This is because ACOMAR chooses the next hop that allocates the nearest
sending slot, which decreases the communication latency in a significant way. This means that ACOMAR protocol is
able to deliver data faster than LEMAR and MAR protocols, which is an important factor in WSNs.
(C) The average hop count: It indicates the average number of nodes that the packet passes through to reach the
sink. Figure 8 displays the average hop counts of ACOMAR protocol compared to MAR and LEMAR protocols.
The hop count is a critical factor for the energy efficiency in WSN. Therefore, if the path from the sender to the
sink has optimal hop count it consumes less energy for the transmission and receiving process which prolongs the
network lifetime. The obtained results show, when ACOMAR protocol is used, the average number of hops that
one node needs to deliver its data to the sink is less than seven hops even if the network has more than 180 nodes.
Since the energy consumption and latency are both low, this indicates that ACOMAR provides the next hop that is
close to the sink and it consumes less energy in the transmission process and low latency.

FIGURE 11 Gain of ACOMAR compared to LEMAR


22 of 25 MECHTA ET AL.

(D) Packet delivery ratio: PDR measured as ratio of the number of packets successfully received by the sink to the
total number of packets sent by sensor nodes in the network.
Packets received
PDR ¼  100%: ð15Þ
Total number of packets sent

Figure 9 shows the packet PDR for MAR, LEMAR, and ACOMAR protocols. ACOMAR outperforms LEMAR and
MAR for different network sizes, especially when the number of nodes is greater than or equal 160. LEMAR the
PDR for 60 nodes is by up to 80% and for ACOMAR the PDR for 60 nodes is by up to 90%. The comparison shows
that ACOMAR achieves also a PDR higher than LEMAR in large-scale networks (more than 160 nodes).
(E) Throughput: It is the number of packets successfully transmitted from the source to the destination per second.
In our study throughput is expressed in kilo bytes per second (kbps). Figure 10 shows the average throughput when
the number of nodes is varied from 20 to 200 in increment of 20. The results show that the throughput is clearly
higher when ACOMAR protocol is used. The ACOMAR average throughput is about 200 KB/s, however when
LEMAR and MAR protocols are used, the average throughput is around 125 KB/s and 90 KB/s, respectively.
(F) Discussion: Corresponding to the previous comparisons where ACOMAR performs better than LEMAR and is
more efficient than MAR in terms of all evaluated metrics, we only considered ACOMAR and LEMAR for the gain
calculation. Figure 11 represents the gain of protocol ACOMAR compared to LEMAR. We notice that the gain of
latency, energy, hop count and throughput reaches more than 80%, 20%, 81%, and 54%, respectively.

The details of these gains are summarized in Tables 9,10, and 11.

TABLE 9 Results considering different metrics (energy consumption, latency, and hop count) in LEMAR and ACOMAR

Energy consumption (J) Latency (No. of slots) Hop count

Nodes LEMAR ACOMAR LEMAR ACOMAR LEMAR ACOMAR


20 1.93473 0.622637 30.7243 11.718 5.42553 1.84872
40 5.34034 1.42246 59.5348 22.6878 5.40494 1.85848
60 12.4349 3.05754 136.27 34.17 8.99954 2.65008
80 18.8313 9.34434 104.15 44.7482 12.0807 5.99466
100 29.8883 12.4447 130.866 51.5149 15.3006 6.37082
120 36.7557 13.34 188.494 48.518 15.654 5.68142
140 53.7973 13.7399 267.859 50.9408 19.6155 5.00975
160 61.676 15.3413 301.908 49.5942 19.3328 4.86
180 79.6514 17.9501 369.186 64.2929 21.7715 5.02538
200 101.733 19.3914 454.336 64.6515 25.2701 4.74132

TABLE 10 Results considering different metrics (PDR and throughput) in LEMAR and ACOMAR

Packet Delivery Ratio Throughput


Nodes
LEMAR ACOMAR LEMAR ACOMAR
20 91.4806 97.6608 151.357 202.273
40 65.5087 90.3744 138.3 213.7
60 81.5167 90.6629 103.2 212.6
80 100 100 134 162
100 100 100 143 164
120 100 100 158 196
140 100 100 139 199
160 68.5071 96.5339 118 206
180 71.2814 97.3309 103 225
200 78.1166 86.8118 100 200
MECHTA ET AL. 23 of 25

TABLE 11 GAIN values in different metrics in LEMAR and ACOMAR

Nodes Energy (gain) Joules Latency (gain) Hop count (gain) Throughput (gain)
20 3.2802325 61.8608072 65.9255409 25.1719211
40 4.89735 61.8915323 65.6151595 35.2831072
60 7.81446667 74.9247817 70.5531616 5.14581373
80 5.92935 57.0348536 50.3782066 17.2839506
100 8.7218 60.6353828 58.3622864 12.804878
120 9.75654167 74.2601887 63.7062732 19.3877551
140 14.3062143 80.9822332 74.4602483 30.1507538
160 14.4795938 83.5730752 74.8613755 42.7184466
180 17.13925 82.585228 76.9176217 54.2222222
200 20.5854 85.7701129 81.2374308 50

8 | C ON C L U S I ON

In this paper, we have proposed an ACO-based MAC-aware routing protocol known as ACOMAR to ensure QoS in
WSN networks and collision-free transmission. This new cross-layer approach exploits the TDMA schedule information
and considers residual energy of sensor nodes, distance to the direct neighbor, distance to the sink and the number of
time slots to select the best next hop when routing data. ACOMAR aims to enhance the network performances in terms
of energy efficiency, network lifetime extension, and latency minimization. ACOMAR is validated with simulation and
analytical results. Experiment results of three protocols MAR, LEMAR, and ACOMAR reveal that ACOMAR outper-
forms MAR and LEMAR for both small and large-scale networks. This is due to the use of the bio-inspired algorithm
ACO, which helps ACOMAR to quickly converge to the optimal neighbor of a sensor node. In addition, our study
shows that the ACOMAR protocol is the most suitable for delay-sensitive applications and huge WSN-based IoT
networks.
In the future, we plan to improve ACOMAR protocol by combining nodes' energy consumption and spatial informa-
tion in the TDMA schedule.

DATA AVAILABILITY STATEMENT


Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

ORCID
Djamila Mechta https://orcid.org/0000-0003-3643-7979

R EF E RE N C E S
1. Mechta D, Harous S. QGCA: quantum genetic based-clustering algorithm for WSNS. In: 14th International Wireless Communications &
Mobile Computing Conference (IWCMC). Limassol, Cyprus; 2018:82-88.
2. Gupta SK, Jana PK. Energy-efficient clustering and routing algorithms for wireless sensor networks: GA-based approach. Wirel Pers
Commun. 2015;83(3):2403-2423. https://doi.org/10.1007/s11277-015-2535-7
3. Hatamian M, Barati H, Movaghar A, Naghizadeh A. CGC: centralized genetic-based clustering protocol for wireless sensor networks
using onion approach. Telecommun Syst. 2016;62(4):657-674. https://doi.org/10.1007/s11235-015-0102-x
4. Keerthika A, Hency VB. A survey of routing protocols of wireless sensor network with mobile sinks. J Eng Appl Sci. 2016;11:6951-6963.
5. Kromer P, Musilek P. Bio-inspired routing strategies for wireless sensor networks. Propagation Phenomena in Real World Networks. Intel-
ligent Systems Reference Library. Switzerland: Springer International Publishing; 2015:657-674. https://doi.org/10.1007/978-3-319-15916-
4_7
6. Kaddi M, Banana A, Omari M. ECO-BAT: a new routing protocol for energy consumption optimization based on BAT algorithm in
WSN. Comput Mater Continua. 2021;66(2):1497-1510. http://www.techscience.com/cmc/v66n2/40628
7. Mechta D, Harous S, Alem I. Improving wireless sensor networks durability through efficient sink motion strategy: TMSRP. Wirel Pers
Commun. 2018;99(4):1661-1682. https://doi.org/10.1007/s11277-018-5329-x
8. Mechta D, Harous S. LBSNN: neural networks-based moving sink. 9th IEEE Annual Ubiquitous Computing, Electronics & Mobile Com-
munication conference (UEMCON). Columbia University, New York, USA: IEEE; 2018:475-481.
24 of 25 MECHTA ET AL.

9. Mechta D, Harous S. HC-LEACH: Huffman coding-based energy-efficient LEACH protocol for WSN. In: 11th IEEE annual ubiquitous
computing, electronics mobile communication conference (UEMCON); 2020:932-938. https://doi.org/10.1109/UEMCON51285.2020.
9298061
10. Nagesh R, Raga S, Mishra S. Elimination of redundant data to enhance wireless sensor network performance using multi level data
aggregation technique. In: 2019 10th International Conference on Computing, Communication and Networking Technologies
(ICCCNT); 2019; Kanpur, India:1-5. https://doi.org/10.1109/ICCCNT45670.2019.8944647
11. Sheta AF, Solaiman B. Evolving clustering algorithms for wireless sensor networks with various radiation patterns to reduce energy con-
sumption. In: Science and information conference (SAI); 2015:1037-1045. https://doi.org/10.1109/SAI.2015.7237270
12. Wang J, Yang X, Zhang Z, Bin L, Kim J. A survey about routing protocols with mobile sink for wireless sensor network. Int J Futur
Gener Commun Netw. 2014;7:221-228. https://doi.org/10.14257/ijfgcn.2014.7.5.18
13. Hadjadj A, Aribi B, M.Amad. Energy-efficient and degree-distance clustering based hierarchical routing protocol for WSNs. Recent Adv
Comput Sci Commun. 2021;14(6):1811-1826. https://doi.org/10.2174/2666255813666191204161724
14. Zhang D, Li G, Zheng K, Ming X, Pan Z-H. An energy-balanced routing method based on forward-aware factor for wireless sensor net-
works. IEEE Trans Ind Inf. 2014;10(1):766-773. https://doi.org/10.1109/TII.2013.2250910
15. Shi L, Mengyao L, Li X. WSN data fusion approach based on improved BP algorithm and clustering protocol. In: The 27th Chinese Con-
trol and Decision Conference (2015 CCDC); 2015; Qingdao, China:1450-1454. https://doi.org/10.1109/CCDC.2015.7162147
16. Maryam S, Naji H. A decentralized energy efficient hierarchical cluster-based routing algorithm for wireless sensor networks. AEU - Int
J Electron Commun. 2015;69. https://doi.org/10.1016/j.aeue.2015.01.002
17. Zhang D, Liu S, Liu X, Zhang T, Cui Y. Novel dynamic source routing protocol (DSR) based on genetic algorithm-bacterial foraging opti-
mization (GA-BFO). Int Journal Commun Syst. 2018;31(18):1-20. https://doi.org/10.1002/dac.3824
18. Zhang D, Zhang T, Dong Y, Liu X, Cui Y, Zhao D. Novel optimized link state routing protocol based on quantum genetic strategy for
mobile learning. J Netw Comput Appl. 2018;122:37-49. https://doi.org/10.1016/j.jnca.2018.07.018
19. Zhang D, Zhang T, Liu X. Novel self-adaptive routing service algorithm for application in VANET. Appl Intell. 2019;49(5):1866-1879.
https://doi.org/10.1007/s10489-018-1368-y
20. Zachariah U, Kuppusamy L. A hybrid approach to energy efficient clustering and routing in wireless sensor networks. Evol Intel. 2021:
1-13. https://doi.org/10.1007/s12065-020-00535-0
21. Zhang D, Liu S, Zhang T, Liang Z. Novel unequal clustering routing protocol considering energy balancing based on network partition &
distance for mobile education. J Netw Comput Appl. 2017;88:1-9. https://doi.org/10.1016/j.jnca.2017.03.025
22. Zhang D, Ge H, Zhang T, Cui Y-Y, Liu X, Mao G. New multi-hop clustering algorithm for vehicular ad hoc networks. IEEE Trans Intell
Transport Syst. 2019;20(4):1517-1530. https://doi.org/10.1109/TITS.2018.2853165
23. Abidoye AP, Kabaso B. Energy-efficient hierarchical routing in wireless sensor networks based on fog computing. J Wirel Commun Netw.
2021;8(1687-1499):1497-1510. https://doi.org/10.1186/s13638-020-01835-w
24. Borujeni EM, Rahbari D, Nickray M. Fog-based energy-efficient routing protocol for wireless sensor networks. J Supercomput. 2018;
74(12):6831-6858.
25. Sun Y, Dong W, Chen Y. An improved routing algorithm based on ant colony optimization in wireless sensor networks. IEEE Commun
Lett. 2017;21(6):1317-1320.
26. Priyadarshi R, Rawat P, Nath V. Energy dependent cluster formation in heterogeneous wireless sensor network. Microsystem Technolo-
gies. 2019;25(6):2313-2321.
27. Chen J, Mao G, Li C, Liang W, Zhang D. Capacity of cooperative vehicular networks with infrastructure support: multiuser case. IEEE
Trans Veh Technol. 2018;67(2):1546-1560. https://doi.org/10.1109/TVT.2017.2753772
28. Zhang D, Zhang T, Zhang J, Dong Y, Zhang X. A kind of effective data aggregating method based on compressive sensing for wireless
sensor network. J Wireless Com Network. 2018;2018(1):159. https://doi.org/10.1186/s13638-018-1176-4
29. Zhang T, Zhang D, Yan H, Qiu J, Gao J. A new method of data missing estimation with FNN-based tensor heterogeneous ensemble
learning for internet of vehicle. Neurocomputing. 2021;420:98-110. https://doi.org/10.1016/j.neucom.2020.09.042
30. Zhang D, Chen L, Zhang J, Chen J, Zhang T, Tang Y-M, Qiu J-N. A multi-path routing protocol based on link lifetime and energy con-
sumption prediction for mobile edge computing. IEEE Access. 2020;8:69,058-69,071. https://doi.org/10.1109/ACCESS.2020.2986078
31. Yang J, Ding M, Mao G, Lin Z, Zhang D-G, Luan TH. Optimal base station antenna downtilt in downlink cellular networks. IEEE Trans
Wireless Commun. 2019;18(3):1779-1791. https://doi.org/10.1109/TWC.2019.2897296
32. Gao H, Liu C, Li Y, Yang X. V2VR: reliable hybrid-network-oriented V2V data transmission and routing considering RSUs and connec-
tivity probability. IEEE Trans Intell Transp Syst. 2021;22(6):3533-3546. https://doi.org/10.1109/TITS.2020.2983835
33. Huang Y, Xu H, Gao H, Ma X, Hussain W. SSUR: an approach to optimizing virtual machine allocation strategy based on user require-
ments for cloud data center. IEEE Trans Green Commun Netw. 2021;5(2):670-681. https://doi.org/10.1109/TGCN.2021.3067374
34. Gao H, Huang W, Duan Y. The cloud-edge-based dynamic reconfiguration to service workflow for mobile ecommerce environments: a
QoS prediction perspective. ACM Trans Internet Technol. 2021;21(1). https://doi.org/10.1145/3391198
35. Nayak P, Swetha GK, Gupta S, Madhavi K. Routing in wireless sensor networks using machine learning techniques: challenges and
opportunities. Measurement. 2021:108974. https://doi.org/10.1016/j.measurement.2021.108974
36. A.Mehmood, Lv Z, Lloret J, Umar MM. ELDC: an artificial neural network based energy-efficient and robust routing scheme for pollu-
tion monitoring in WSNs. IEEE Trans Emerg Top Comput. 2017;8(1):106-114.
MECHTA ET AL. 25 of 25

37. Huang R, Ma L, Zhai G, He J, Chu X, Yan H. Resilient routing mechanism for wireless sensor networks with deep learning link reliabil-
ity prediction. IEEE Access. 2020;8:64,857-64,872. https://doi.org/10.1109/ACCESS.2020.2984593
38. Farag H, Gidlund M, Österberg P. A delay-bounded MAC protocol for mission- and time-critical applications in industrial wireless sen-
sor. IEEE Sensors J. 2018;18(6):2607-2616. https://doi.org/10.1109/JSEN.2018.2793946
39. Sun N, Wang X, Han G, Peng Y, Jiang J. Collision-free and low delay MAC protocol based on multi-level quorum system in underwater
wireless sensor networks. Computer Commun. 2021;173:56-69. https://doi.org/10.1016/j.com.2021.03.020
40. Zhang D, Chen C, Cui Y, Zhang T. New method of energy efficient subcarrier allocation based on evolutionary game theory. Mobile
Netw Appl. 2021;26(2):523-536. https://doi.org/10.1007/s11036-018-1123-y
41. Zhang D, Wang X, Song X, Zhao D. A novel approach to mapped correlation of ID for RFID anti-collision. IEEE Trans Serv Comput.
2014;7(4):741-748. https://doi.org/10.1109/TSC.2014.2370642
42. Li F, Huang G, Yang Q, Xie M. Adaptive contention window MAC protocol in a global view for emerging trends networks. IEEE Access.
2021;9:18402-18423. https://doi.org/10.1109/ACCESS.2021.3054015
43. Huang C, Huang G, Liu W, Wang R, Xie M. A parallel joint optimized relay selection protocol for wake-up radio enabled WSNs. Phys
Commun. 2021;47:101320. https://doi.org/10.1016/j.phycom.2021.101320
44. Landsiedel O, Ghadimi E, Duquennoy S, Johansson M. Low power, low delay: opportunistic routing meets duty cycling; 2012.
45. Louail L, Felea V, Bernard J, Guyennet H. MAC-aware routing in wireless sensor networks. 225–229; 2015.
46. Hebal S, Louail L, Harous S. Latency and energy optimization using MAC-aware routing for WSNs. Int J Business Data Commun Netw
(IJBDCN). 2020;16:19-27.
47. Osamy W, El-Sawy AA, Khedr AM. Effective TDMA scheduling for tree-based data collection using genetic algorithm in wireless sensor
networks. Peer-to-Peer Networking and Applications. 2020;13:796-815.
48. Louail L, Felea V. Routing-aware TDMA scheduling for wireless sensor networks. 1–8; 2016.

How to cite this article: Mechta D, Mahgoun O, Harous S. ACOMAR: ACO-based MAC-aware routing
protocol. Int J Commun Syst. 2021;e5055. doi:10.1002/dac.5055

You might also like