Received 17 August 2022, accepted 12 October 2022, date of publication 19 October 2022, date of current version 26 October 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3216066

Fault-Tolerant Ad Hoc On-Demand Routing

Protocol for Mobile Ad Hoc Networks
DUC N. M. HOANG 1 , (Graduate Student Member, IEEE),
JONG MYUNG RHEE2 , (Member, IEEE), AND SANG YOON PARK 1, (Member, IEEE)
1 Department of Electronic Engineering, Myongji University, Yongin 17058, South Korea
2 Department of Information and Communications Engineering, Myongji University, Yongin 17058, South Korea

Corresponding author: Sang Yoon Park ([email protected])

This work was supported by the National Research Foundation of Korea (NRF) funded by the Korean Government (MSIT) under Grant
2020R1A2B5B02002201 and Grant 2021R1F1A1050040.

ABSTRACT Mobile ad hoc networks (MANETs) are particularly suited for scenarios that demand rapid
deployment of a communication system without any existing network resources. For instance, a MANET
can facilitate the intercommunication process between members of a rescue party in a natural disaster, where
the underlying routing protocol is crucial to maintaining the dissemination capability of data-critical packets.
However, the backbone of every MANET, i.e., their routing protocol, is limited by the communication
range of nodes, their high-speed mobility, and the capacity constraints of energy. This study proposed a
fault-tolerant ad hoc on-demand routing protocol (FT-AORP) that relies on these characteristics of MANET
nodes to determine reliable paths for data transmission. Subsequently, two of the discovered paths were
used to transmit the duplicates of an original data packet to maximize fault tolerance. Further, using the
OMNeT++ network simulator, the performance of the proposed system was evaluated through extensive
simulation experiments against three simulation parameters: the number of network nodes, node speed, and
data packet sending rate. The simulation results demonstrated that FT-AORP greatly improved the packet
delivery ratio, reduced end-to-end delay, and maintained a higher residual energy level of the transmission
path, compared to other baseline routing protocols.

INDEX TERMS Mobile ad hoc network, fault tolerance, network mobility, on-demand routing protocol.

I. INTRODUCTION Owing to the flexibility in the architecture, MANETs are

A Mobile ad hoc network (MANET) is a wirelessly inter- highly versatile and thus, can be beneficial for a variety of
connected class of networks where nodes can freely move applications [3]. For example, they can be employed by many
and communicate directly with any nearby neighbors. Owing military units [4], [5], [6] because their rapid deployment
to this mobility characteristic, MANET is normally a does not require any previously installed network resources,
self-configuring and infrastructure-less network and thus, thus facilitating the establishment of an information commu-
can be dynamically formed under any topology [1]. How- nication system between the soldiers on duty. In addition,
ever, there should not be any centralized infrastructure in MANETs can be exploited for public safety and disaster
a MANET. In contrast to the infrastructure-based wireless management [7], [8]. For instance, during natural hazards,
network, where the deployment of administering devices, existing telecommunication towers may be dysfunctional and
e.g., base stations or access points, is required for the network an instant installation of a wireless ad hoc communication
operation, each node in a MANET must be able to send, system is extremely critical for the rescue party to com-
receive, and relay network packets [2]. municate with one another while supporting the endangered
victims [9]. Furthermore, the dynamic topology nature of this
The associate editor coordinating the review of this manuscript and network type enables the rescuers to change their positions
approving it for publication was Bijoy Chand Chand Chatterjee . without losing connectivity with the rest of the team [10].

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 10, 2022 111337
 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

Considering these application areas, MANET routing pro- presents the problem formulation through a system model and
tocols have been intensively studied in the literature [11], introduces the proposed routing metrics for the path selec-
[12] owing to their importance in deciding the successful tion issue. In Section IV, the routing protocol is described
operation of a MANET system. However, despite the ease in detail, including the route request, route reply, and node
of the establishment process, MANET routing protocols are discovery schemes. The performance of the proposed rout-
uniquely presented with many challenges [3], [13]. First, ing protocol indicated by simulation results is evaluated
owing to the limited range of radio communication links, in Section V. Finally, the conclusions are presented in
nodes within the network must maintain multi-hop paths Section VI.
between one another [14], [15]. In addition, as the nature
of MANET is typically characterized by numerous proper- II. RELATED WORK
A. FUNDAMENTAL ROUTING PROTOCOLS
ties, MANET routing protocols must be designed to address
Routing schemes for MANETs have been extensively stud-
them upon discovering the paths. A few examples are lim-
ied in the literature. Pioneering routing protocols can be
ited battery capacity [16], security [17], neighboring signal
classified into two main categories: proactive and reactive
strength [18], link reliability [19], or their combinations [20],
protocols [23]. Several well-known proactive protocols are
[21]. Moreover, following the determination of the paths,
destination-sequenced distance vector (DSDV) [24], global
a fault-tolerant transmission strategy is required to guarantee
state routing (GSR) [25], and optimized link state routing
that critical data packets are sent to the destination without
(OLSR) [26]. Meanwhile, certain popular reactive routing
substantial delays. To the best of the authors’ knowledge,
solutions that have been proposed to reduce the overheads
only a few of the proposed routing protocols focus on both
of proactive approaches are ad hoc on-demand distance vec-
an efficient routing scheme and a fault-tolerant capability for
tor [27], dynamic source routing (DSR) [28], and temporally
packet transmissions.
ordered routing algorithm (TORA) [29]. Such protocols have
This study proposed a fault-tolerant ad hoc on-demand
mostly been adapted from traditional routing schemes for
routing protocol, referred to as FT-AORP. It provides reliable
wired networks with minor modifications for wireless net-
transmission for the deployment of MANETs in scenarios
works. Consequently, they were originally designed to select
where safety-critical information must be guaranteed to be
the route with the minimum hop count to transmit the packets
successfully exchanged amongst nodes in the network. The
through. Although this hop count metric is straightforwardly
main contributions of this study are as follows.
useful in cabled networks as the network components are
• A path selection strategy employing three different rout-
often static, MANETs are more dynamic and the number of
ing metrics relevant to MANET characteristics: node
intermediate nodes between the source and destination is thus
mobility, radio signal strength measurement, and energy
highly likely to change over time.
usage rate, was proposed to identify the optimal paths
to the destination for data transmission. These routing B. STATE-OF-THE-ART ROUTING METRICS
metrics can be easily collected without requiring any Considering the limits of the hop count, many studies have
supplemental sensor devices. incorporated different routing metrics into MANET rout-
• We developed a simple multipath discovery scheme that ing protocols. Taha et al. [16] proposed an energy-efficient
returns, if possible, more than one path towards the des- multipath routing protocol that adopted a fitness function
tination for the requesting source node. Based on these as an optimization method to select the best route based
determined paths, the source node can select the two on two criteria: the number of intermediate nodes and their
desired ones using the aforementioned selection strat- remaining energy level. Although this scheme can prolong the
egy. Then, we introduced a fault-tolerant transmission network lifetime, the power consumption rate alone appeared
technique that involves duplicating the important data relatively insufficient when considering aggressive scenarios
packets and transmitting them using two separate node- with more factors affecting the connection links between
disjointed paths. This fault-tolerant design ensured that network nodes. For instance, when an arbitrary node is in
packets were delivered successfully even if transmission an isolated area, it is certainly not an ideal next hop for the
issues occurred in one path. entire route although it might have sufficient energy supply.
• We implemented the proposed protocol as a simulation Therefore, more measures are being exploited to design rout-
framework using OMNeT++ [22] and modeled the ing protocols.
evaluation testbed as close to the real-world environ- For example, Chen et al. [30] proposed a topological
ment as possible with various configurable simulation change adaptive ad hoc on-demand multipath distance vec-
parameters. Moreover, by employing several evaluation tor (TA-AOMDV) routing protocol that can adapt to the
metrics, the better performance of the proposed proto- aggressively dynamic nature of mobile networks. Through
col compared to other revisited routing protocols was the combination of individual node information (i.e., resid-
demonstrated in many testing scenarios. ual energy, queue length, and current bandwidth) as route
The rest of this paper is organized as follows. In Section II, selection metrics with the probability of link stability between
certain relevant state-of-the-art studies that have addressed nodes, the proposed scheme exhibited better performance
the MANET routing problem are reviewed. Section III in many notable metrics such as packet delivery ratio and

111338 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

end-to-end delay. However, while the protocol is a prospec- TABLE 1. Commonly used notation.
tive choice for applications in high-speed MANETs with
quality-of-service (QoS) constraints, it may not be suitable
for life-critical systems as its packet delivery rate perfor-
mance is not consistently satisfactory as the network nodes
move chaotically fast (e.g., 30 m/s and higher).
In a similar manner, Dhananjayan and Subbiah [31]
proposed a trust-aware ad hoc routing protocol using
stability factors such as energy level, mobility, and
RSSI-based distance measurement from each node’s
accumulated logging information to select the optimal
next hop from surrounding neighbors. Simulation results
showed that this reputation-based protocol achieved high network nodes to change their movement patterns, which may
throughput, improved packet delivery rate, and low end- not always be plausible owing to the highly unpredictable
to-end delay. However, the log collection and examination MANET mobility [35].
process performed by each node might demand enormous Recently, Srilakshmi et al. [36] proposed a secure opti-
computational costs and subsequently high-power consump- mization routing protocol that simultaneously addresses the
tion, which is normally limited for resource-constrained energy and communication issues of MANETs. In detail,
MANETs. there are certain cluster heads in the network that are in charge
More recently, Sirmollo and Bitew [32] proposed a of routing the packets, and they are chosen based on their
mobility-aware routing scheme for MANETs. This protocol calculated trust values. However, the trust values are mainly
takes multiple factors into consideration, such as the speed derived from the transmission duration, which is partially
of the nodes, the relative distance between them, and their susceptible to the dynamic nature of MANETs. Interestingly,
residual energy level, to choose the best nodes to forward Pattnaik et al. [37] presented a multipath routing scheme for
the packets to during the route request and route discov- MANETs, which makes use of De Casteljau’s algorithm with
ery phases. Simulation results also demonstrated the supe- the Bezier curve for the mobility awareness of multiple speed
rior performance over other routing protocols. However, the levels. The optimal routing path for data transmission is then
proposed scheme mainly relies on the calculation of the selected based on various factors. Routing in obstacle-ridden
node distance, which requires an accurate positioning system. environments is certainly a promising research topic. How-
In reality, this might not always be possible, especially when ever, it is outside of the scope of this work and can be of great
it comes to indoor contexts. potential for our future research directions.
In addition, Naseem et al. [38] proposed a novel
C. FAULT-TOLERANT ROUTING PROTOCOLS energy-efficient routing protocol for MANETs that also
Regarding fault tolerance in MANETs, Nsaif et al. [33] intro- employs manifold routing criteria, such as the number of
duced an approach referred to as seamless routing for wireless hops, the round-trip time, and the remaining energy level,
ad hoc networks (SRAD). Here, each pair of nodes had two to discover the optimal data transmission path. Additionally,
link-disjointed paths, which were used to transmit two redun- it is also a multipath-enabled routing scheme that aims to
dant frame copies to the destination such that zero-recovery offload the data transmission among various paths for load
time was achieved in the case where one of the operating balancing. In contrast, in our work, the discovered paths
paths fails. Instead of using the shortest paths, the path are used to simultaneously send data packets to the desired
selection is based on its bit error rate (BER) measurement. destination node. As a result, we can guarantee the maxi-
However, the process to calculate the BER for the links is mization of fault tolerance for MANETs. Likewise, Sarhan
unclear and the proposed protocol has been evaluated for and Sarhan [39] employed the elephant herding optimization
a benign scenario only (e.g., the network containing only algorithm to choose the routing paths with optimal residual
21 nodes moving at 10 m/s). energy in an on-demand multipath routing protocol called
Song et al. [34] proposed several topology control algo- EHO-AOMDV. Then, various paths are also used for the
rithms employing the Kalman filter, which can anticipate data transmission’s load balancing between the source and
the movement tendency of other surrounding nodes and con- destination nodes. The actual number of packets assigned to
sequently adjust its movement to maintain wireless links a path will be determined using the minimum energy level
with other nodes. In addition, they are localized algo- of its intermediate nodes, which ensures that the path with
rithms wherein no more than two hops’ information is higher residual energy will be responsible for relaying more
required. Simulation results against node velocity and the data packets.
number of cluster members showed that one of the pro-
posed schemes can effectively restore connectivity while III. PROBLEM STATEMENT
maintaining a minimal deviation from the original task-based In this section, we describe the general system model, formu-
direction. However, these proposed algorithms require the late the routing problem, and introduce our proposed routing

VOLUME 10, 2022 111339

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

Definition 1: From many possible paths between nodes vi

and vj , path p is the optimal path if its routing measure 2(p)
is the largest one.
Therefore, in the proposed study, the act of selecting the
two optimal paths was equal to selecting the two paths with
the two largest routing measures 2. In the subsequent
sections, we will provide details for the computation process
FIGURE 1. Comparison between (a) node-disjointed and
of the routing measure value for nodes and paths, i.e., 8 and
(b) link-disjointed paths. 2 respectively.

metrics as part of the routing protocol. The commonly used B. NODE ROUTING MEASURE
notation in this paper is shown in Table 1. In the proposed scheme, when the network nodes are oper-
ating, each one computes its own node routing measure 8
A. SYSTEM MODEL AND PROBLEM FORMULATION at every selected interval. This 8 value is calculated from
• A path pij consisting of l nodes is denoted as pij = the information of the current state of the node and its sur-
[vi , . . . , vj ], where kpij k = l with k·k being the cardi- rounding environment. A node routing measure is composed
nality of a set; of three routing metrics.
• For two adjacent nodes, e.g., nodes vi and vi+1 , a bidi- 1) Mobility indicator 8m ;
rectional link is denoted as (vi , vi+1 ) ∈ E; 2) Neighboring signal strength 8s ;
• There should be k possible paths between vi and vj , 3) Energy level 8e .
h i
(1) (2) (k)
they collectively form a route rij = pij , pij , . . . , pij , The proposed routing metrics can be easily obtained from
where krij k = k. commercially available wireless products without any addi-
tional designated sensors or peripherals. In particular, the
To provide reliability for in-network data communication,
mobility indicator suggests the tendency of a node to change
a node should maintain information on alternative paths in
its neighbors, neighboring signal strength indicates if a node
addition to the main path [40]. Therefore, in this work, we aim
is within an area of strong wireless signal coverage, and
to develop a multipath routing protocol that discovers more
energy level reveals the current energy level of a node and
than one path (i.e., k > 1) from the source to the destination
its power consumption rate. Finally, the routing measure 8
node. This design strategy can cope with potential link break-
of node v is expressed as
ages due to network mobility and unstable wireless medium.
Thus, k = 2 being set to a larger value may cause excessive 8(v) = 8m + 8s + 8e . (2)
protocol overhead but show no significant improvement [41].
Next, to maximize the fault-tolerant capability, discovered 1) MOBILITY INDICATOR
paths should be node-disjointed [42], [43] instead of sim- Mobility indicator metric was first proposed in [44] to incor-
ply link-disjointed. This is because, for two node-disjointed porate the mobility pattern of a node considering changes in
paths, there is no common node, whereas link-disjointed its neighboring nodes. A node is deemed more reliable to be
paths can share one or more intermediate nodes, as depicted in part of a transmission link if it tends to stay with a higher
Fig. 1. In this example, following the node-disjointed design number of other surrounding nodes. The mobility indicator
scheme, the routing protocol can avoid the situation where 8m (t) for a node v at time t is calculated using
a single point of failure occurs at the single common node  (v) (v)
of two paths, which is denoted as node F in Fig. 1b. In the 
 0, if kNt k = kNt−1t k = 0,
(1) (2)
proposed system model, this implies that (pij , pij ) ∈ rij :
v u (v) (v)
8m (t) = u t kNt ∩ Nt−1t k , otherwise,
(1) (2)
pij ∩ pij = {vi , vj } as krij k = k = 2 for every pair of source 
 kN (v) ∪ N (v) k

vi and destination node vj . t t−1t
In addition, as this study considered k = 2 paths to be (3)
used to transmit the data packets between the source and (v) (v)
where Nt (resp. Nt−1t ) denotes the set of neighboring nodes
destination nodes, an algorithm to select two paths from many of node v at time t (resp. t −1t), and 1t is a sufficiently large
possibly discovered paths is required. Therefore, a routing sampling period.
measure was introduced for each operating node in the net- Lemma 1: 0 ≤ 8m ≤ 1.
work. In particular, the routing measure for node v at time t is Proof: First, it is evident from (3) that, in the case of
defined as 8(v) (t). Recall that as a path p is formed between (v) (v)
kNt k = kNt−1t k = 0, 8m = 0. In addition, 8m is a
nodes vi and vj , it consists of l intermediate nodes. Thus, the square root and therefore, 8m > 0. Next, it is apparent that
routing measure 2(p) for the entire path p is expressed as (v) (v) (v) (v) (v) (v) (v)
kNt ∩ Nt−1t k ≤ kNt ∪ Nt−1t k or kNt ∩ Nt−1t k/kNt ∪
1 X (vi ) (v) (v) (v)
2(p) = 8 . (1) Nt−1t k ≤ 1, and the equality holds when Nt = Nt−1t .
l v ∈p
i Hence, the lemma is proven. 

111340 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

From Lemma 1, it is apparent that the fewer the changes by applying the log sum inequality [47] for the set β of kN (v) k
in the neighbor set at the two different time instances, the positive numbers given in (4) and a set {1, 1, . . . , 1} of the
larger the value of 8m , implying a more reliable node. In other same size, we obtain
studies, e.g., [45], global positioning system (GPS) devices γ 
vi
X X
may be used to compute the mobility of a node based on γvi ln(γvi ) = γvi ln
1
the relative distance between nodes. However, the GPS fix vi ∈N (v) v ∈N (v)
i  P
process may result in large power consumption [46], which is
v ∈N (v) γvi
X
not suitable for mobile nodes powered by on-board batteries. ≥ γvi  ln P i
(v)
vi ∈N vi ∈N (v) 1
2) NEIGHBORING SIGNAL STRENGTH = − lnkN (v) k, (8)
Mobility indicator suggests the change rate in the number of
γvi = 1 according to (6). Therefore, it is obvious
P
neighboring nodes within a certain amount of time. However, for vi ∈N (v)
(v)
this information alone does not indicate whether the node that 8s ≤ ln(|lnkN k
ᾱ|+kN (v) k)
≤ 1 and the proof is completed. 
is currently within a sufficiently close distance to properly In particular, this study exploited the characteristics of
receive the radio signal from other transmitting nodes. There- the entropy concept [48] to compute the routing metric
fore, this study introduced the neighboring signal strength 8s . According to the entropy concept, given P a set X =
metric that employs the received signal strength indicator {x1 , x2 , . . . , xn } satisfying 0 ≤ xi ≤ 1 and xi = 1,
(RSSI) to show the proximity of a node within the transmis- thePentropy of this set can be computed using H (X ) =
sion ranges of its neighbors. − xi log xi . Regarding the computation of the proposed
As a node receives a packet from one surrounding node, routing metric, a node obtains a set β of RSSI coefficients
it can retrieve information on the strength level of the signal, from its neighbors’ RSSI values, as given in (4). Thus, from
from which it successfully receives that packet. In particular, conditions (5) and (6), it is obvious that the entropy of β can
for 0(v, vi ) < 0 in dBm as the RSSI value of node vi ’s signal be calculated, as provided in the numerator of (7).
measured by the node v, it is known that the closer to 0 the Remark 1: The entropy value of a set has the following
value of 0, the better the signal. Subsequently, node v builds relevant properties, which can be used to describe the rela-
up a collection of the measured signal levels of its neighbors tionship between the RSSI values γvi collected by a node and
(v) (v)
at time t, denoted as α = {(0(v, vi )) | vi ∈ Nt } with Nt 6 = its metric 8s .
∅. Then a set β of RSSI coefficients is computed using
( ) • 8s should be proportional to γvi , implying that a small
(0(v, vi ))−1 (v) increase or decrease in γvi values result in a small
β = P −1
| vi ∈ Nt increase or decrease in 8s , respectively. According
(v) (0(v, vi ))
vi ∈Nt
n o to this property, a node obtaining higher RSSI values
(v)
= γvi | vi ∈ Nt , (4) exhibits a higher 8s value, thus, it is more likely to be
chosen to be part of a routing path;
(0(v, vi ))−1 • 8s should be maximal if γvi values are equally likely.
where γvi = P −1
. Because of the computa-
(v) (0(v, vi )) This implies that if all RSSI values are equally high,
vi ∈Nt
tion of γvi , we obtain the node is in an area with strong signal coverage and
therefore becomes an optimal candidate to be included
0 < γvi ≤ 1, (5) in a routing path. However, there is a case when γvi
and values are equally low and 8s still has its maximum
X value. To overcome this, the denominator ln (|ᾱ| + kN k)
γvi = 1. (6) as given in (7) was introduced. With |ᾱ| being the abso-
vi ∈N (v) lute value of the mean of the RSSI set, this implies that if
Then, the neighboring signal strength metric 8s at time t is all the RSSI values are equally small, the 8s of the node
expressed as will be divided by a large value as the absolute value of a
 non-positive RSSI value is being used. In contrast, if all
0, P if kN (v) k = 0, the RSSI values are equally high, 8s will be larger.
8s (t) = (v) γvi ln γvi (7)
− vi ∈N , otherwise,
ln(|ᾱ| + kN (v) k) 3) ENERGY LEVEL
where ᾱ is the mean value of set α and |·| denotes the absolute Assume that a node is able to acquire its battery level at
(v)
value. time t, denoted by Et in Joule (J). The energy level metric
Lemma 2: 0 ≤ 8s ≤ 1. is computed as
Proof: From (7), for kN (v) k = 0, 8s = 0. Then, from (v)
if Et < ζ,

(5), it is clear that ln γvi < 0 or − ln γvi > 0. In addition, the 0,

(v)
denominator given in (7) is always positive for kN (v) k ≥ 1. 8e (t) = Et (9)
 (v) , otherwise,

Owing to these conditions and (5), we obtain 8s ≥ 0. Next, Et−1t

VOLUME 10, 2022 111341

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

where ζ is a sufficiently small safety threshold to indicate that 8s , and 8e ) of each node. Based on this maximization, the
this node is no longer suited to be involved in the network chosen path will be the optimal one and have the following
activity owing to a shortage in power storage. characteristics.
Lemma 3: 0 ≤ 8e ≤ 1. Theorem 1: For a path with the maximum 2 value, its
Proof: First, because the battery level is always non- intermediate nodes are less likely to change their neighbors.
negative, 8e ≥ 0 and the equality holds when Et value Proof: From Lemmas 1 and 4, it is evident that one
is below the given ζ threshold. Then, as it is clear that the of the conditions for achieving the maximum value of path
battery level at a particular time will always be less than or measure 2(p) is that nodes in the path p appear to have few
(v) (v) (v) (v)
equal to the previous time, Et ≤ Et−1t or Et /Et−1t ≤ 1. changes in the set of their neighboring nodes. Consequently,
The equality holds when the node consumes minimal power, the selected path will remain stable over time and there should
resulting in relatively no change between the two acquisition be a reduction in the number of intermediate link breaks as
(v) (v) nodes do not tend to flee from one another. 
attempts of battery level, or Et = Et−1t . 
Through the introduction of this metric, this study aimed Theorem 2: For a path with the maximum 2 value, its
to calculate the power consumption rate of a node and make intermediate nodes tend to stay in a region with equally
decisions accordingly. For example, if the node is quickly distributed strong signal coverage from the neighbor nodes.
drawing power from its battery for network activities such Proof: From Lemmas 2 and 4, it can be observed that
(v) (v)
as packet transmission or reception (i.e., Et  Et−1t ), 8e one of the conditions for achieving the maximum value of
will have a smaller value and this node stands a small chance path measure 2(p) is that nodes in the path p can obtain
of being an intermediate hop in a path. Moreover, from (9), equally high RSSI values of the radio signal from all the
if the battery level is less than ζ , this node will be certainly neighbor nodes. This is thoroughly justified in Remark 1. 
excluded from the path formation, thereby preventing later Theorem 3: For a path with the maximum 2 value, its
possible link interruptions owing to power outages. intermediate nodes are higher in their current energy level
and have slower rates of power consumption.
C. PATH ROUTING MEASURE Proof: From Lemmas 3 and 4, it can be observed that
Now, recall that path p between two nodes v1 and vl compris- one of the conditions for achieving the maximum value of
ing l nodes is denoted as p = [v1 , v2 , . . . , vl ], where kpk = l. path measure 2(p) is that nodes in the path p appear to be
The routing measure for node vi and path p is given by 8(vi ) maximized in the battery level. In addition, if nodes appear
and 2(p) , respectively, and by substituting (2) into (1), the as consuming energy quickly (i.e., Et  Et−1t ), it results in
following is obtained reduced 8e and subsequently 2 value. 
From (10), it is observed that the routing measure for
1 X (vi ) 1 X  (vi ) 
2(p) = 8 = 8m + 8s(vi ) + 8e(vi ) . (10) a path is the summation of the routing metrics of all the
l v ∈p l v ∈p nodes in that path, including the source and the destination
i i
nodes. However, a node is not required to know the entire
Lemma 4: 0 ≤ 2(p) ≤ 3. network topology to calculate the routing measure for every
Proof: First, from Lemmas 1 to 3, it isPclear that for possible path between itself and the desired destination node.
every vi ∈ p, 8(vi ) ≥ 0, thus 2(p) = 1l vi ∈p 8(vi ) ≥ In fact, this computation is distributedly performed by each
0 because l is also positive. Then, intermediate hop during the route discovery procedure, which
1 X  (vi )  is addressed in the next section.
max 2(p) = max 8m + 8s(vi ) + 8e(vi )
l v ∈p
i IV. PROPOSED ROUTING PROTOCOL
1X  
In this section, a fault-tolerant ad hoc on-demand routing pro-
≤ max 8m (vi )
+ 8s(vi ) + 8e(vi )
l v ∈p tocol (FT-AORP) is presented. First, the manner in which the
i
1 X  conduction of the route discovery procedure as a source node
leq max 8m (vi )
+ max 8s(vi ) + max 8(v
e
i)
requires the transmission of a data packet to a specific des-
l v ∈p
i tination (route information is not yet available) is discussed.
1X Next, after the source node acquires the route information of
= (1 + 1 + 1)
l v ∈p the destination node inside its routing table, a fault-tolerance
i
scheme of data packet transmission was proposed, wherein
3kpk
= two duplicates of the original data packet were sent over the
l network to guarantee reliable transmission for critical data
= 3, (11)
packets.
also from Lemmas 1 to 3 and for kpk = l 6 = 0. Hence, the
lemma is proven.  A. ROUTE DISCOVERY
From Definition 1, the act of selecting the best path is The route discovery strategy of FT-AORP is partially based
equal to selecting the route with the maximum path measure on the well-known ad hoc on-demand distance vector
2, i.e., maximizing the constituent routing metrics (i.e., 8m , (AODV) routing protocol [27] with certain modifications for

111342 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

current node. The routing measure of a node is computed

using (2).
Before eventually sending the RREQ packet, the source
node must encapsulate its RREQ-identifying information
inside. Thus, it increments its own RREQ ID number by one
and embeds this value inside the RREQ ID field of the packet.
FIGURE 2. Structure of the RREQ packet with the newly added ‘‘Routing
Later, as other nodes receive this broadcasted RREQ packet,
Measure’’ field. they can check whether it has received the earlier copy before
by examining both the RREQ ID and Originator IP Address
the multipath routing purpose. Similar to the AODV protocol, fields. In addition, each node maintains a sequence number
there are two phases from the point the source node initiates such that other nodes can check for this information by exam-
the route discovery until it is responded to with the requested ining the Originator Sequence Number field to ensure that
route: route request (RREQ) and route reply (RREP). In gen- it is updated with the originator node regarding the routing
eral, the route request process floods the RREQ packets into control packets. The route request process performed by the
the network and as they reach the desired destination node, source node is described in Fig. 3.
it sends back the RREP packets to the source node via unicast. If a node receives an RREQ packet, it (with the exception of
the destination node), determines whether it has received an
1) ROUTE REQUEST earlier RREQ packet with the same RREQ ID and Originator
RREQ is a broadcast packet that is sent by a source node if it IP Address or not. If it has, the node simply discards the
does not know a route to a destination. Thus, the source node newly arrived RREQ packet to avoid unnecessary dissemi-
wants to send a data packet to a certain node and it searches its nation of the RREQ packet all over the network. However,
routing table for an entry that contains relevant information if the node cannot find the RREQ packet’s identity inside its
regarding the destination node. As the node finds nothing, record, it increases the Hop Count field by one, adds its own
it starts the route request process by broadcasting the RREQ routing measure value to the accumulated Routing Measure,
packet. The format of the RREQ packet is shown in Fig. 2. and rebroadcasts the RREQ packet. In contrast to the origi-
It should be noted that all the packet structures proposed in nal AODV protocol, if the destination node receives RREQ
this paper are for illustrative purposes only and are not byte packets, it does not check for any duplications. Instead,
aligned. it maintains a record of every arrived RREQ packet inside
As evident from Fig. 2, the RREQ packet defines the its routing table to enable multipath routing. Subsequently,
following fields: the destination node responds to each RREQ packet with an
RREP packet, which begins the route reply phase.
• Type is the type of FT-AORP control packets and RREQ
packet is type number 1;
2) ROUTE REPLY
• U is the unknown sequence number flag, indicating that
the destination sequence number is unknown; As the destination node receives an RREQ packet, it imme-
• Hop Count is the number of intermediate hops from the diately generates an RREP packet; thus, the number of the
originator node of an RREQ packet to the node currently RREQ packets possibly reaching the destination is equal
handling it. This value is incremented for every node that to the number of paths between the source and destination
the RREQ packet passes; nodes. In contrast to the broadcasted RREQs, the RREP
• RREQ ID is a monotonically increasing number that, responses are performed via unicast as the destination and
coupled with the originator IP address, uniquely identi- intermediate nodes already learn the reverse route from the
fies an RREQ packet. RREQ ID number is incremented destination back to the source node during the route request
by one whenever an RREQ packet is created; period. The format of the RREP packet is shown in Fig. 4.
• Destination IP Address is the IP address of the destina- Specifically, the RREP packet defines the following fields:
tion node of the route that is being discovered; • Type is the type of FT-AORP control packets and RREP
• Destination Sequence Number is the most recently is type number 2;
received sequence number by the originator for any route • Hop Count is similar to that defined in the RREQ
to the destination; packet’s structure;
• Originator IP Address is the IP address of the node • Destination IP Address is the IP address of the destina-
that started the route discovery and initiated the route tion node of the route. As the RREP packet is always
request; created by the destination node, this field should contain
• Originator Sequence Number is the current sequence the IP address of the destination itself;
number to be used for the route entry to the initiator of • Destination Sequence Number is similar to that defined
the route request; in the RREQ packet, and its purpose is to ensure
• Routing Measure is the accumulated routing measure that other nodes are up-to-date regarding its routing
value 8 of all the nodes from the originator node to the activities;

VOLUME 10, 2022 111343

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

Because the RREP packets are sent via unicast, an inter-

mediate hop need not check if it has received the same packet
earlier or not. Instead, it increments the Hop Count value by
one, adds its own routing measure to the Routing Measure
field, and simply forwards the RREP packet to the next hop
according to its routing table. However, as the source node
receives the RREP packets through which it has requested
the route, it updates its routing table with the obtained route
information from such RREP packets and becomes ready to
transmit the data packet to the desired destination.
An example of the entire route request and route reply
procedures is shown in Fig. 5. In this scenario, first, the
source node S broadcasts an RREQ packet as it desires to
find paths to the destination node D. The solid red arrows
are the first copies of the RREQ packet that unprecedent-
edly reach the neighbor nodes. As the later arrived RREQ
packets (illustrated by the dotted black arrows) are discarded,
there is only one single red arrow towards a node except for
the destination node D because it accepts multiple arriving
RREQ packets. In addition, as the destination D receives an
FIGURE 3. Flowchart of creating and sending an RREQ packet.
RREQ packet, it immediately responds with an RREP packet
via unicast, which is illustrated by the solid blue arrows.
Following the entire route request and reply process, there
are two discovered paths between nodes S and D, namely
p(1) = [S, I1 , I4 , I5 , I8 , D] and p(2) = [S, I2 , I7 , I10 , I9 , D].
Given the found paths p(1) and p(2) , the S node is ready to
transmit the data packet using them.

B. DATA PACKET TRANSMISSION

After a successful route discovery, the source node can finally
FIGURE 4. Structure of the RREP packet with the newly added ‘‘Routing utilize the paths found to send the data packets. However,
Measure’’ field.
as now there can be more than two possible paths to the desti-
nation (there is no limit to the number of RREQ packets that
• Originator IP Address is the IP address of the node the destination node can reply to), the source node is required
that started the route discovery and initiated the route the decision on which paths to use. As introduced earlier
request; in Definition 1, the optimal path has the largest calculated
• Routing Measure is similar to that defined in the RREQ routing measure 2 given in (10). Based on this, the source can
packet’s structure, except that it is added up from the select the best two paths to the destination from its routing
destination back to the source node. This is because table. The implementation of the path selection strategy is
the nodes should have the latest possible information described in Algorithm 1.
regarding routing measure values of the path. From Algorithm 1, it is evident that if there is no path
Analogous to the creation of RREQ packets, the desti- available even after the route discovery scheme, the source is
nation node also attaches the necessary information to the still unable to transmit the data packet. In the case of only one
RREP packet. In fact, the RREP packet’s structure has fewer path being found, the two output paths p1 and p2 are identical
fields as it does not need to carry any identifiable information, and the data packet is duplicated and transmitted using this
such as the RREQ ID field, as required in the case of the single path. However, if there are two or more paths, the
RREQ packet. However, the RREP packets still must con- source simply selects the two paths with the largest routing
tain their latest sequence number to facilitate other nodes measure computed using (10). In this case, each path carries
in determining the freshness level of the received RREP a duplicate of the original data packet.
packets via the Destination Sequence Number field. Thus, the Finally, as the copies of the data packet reach the desti-
destination node increments its own sequence number by one nation node, the sequence number attached to a packet is
every time it sends an RREP packet. In addition, the RREP checked to determine whether it has received the same one
packet defines the new Routing Measure compared to the earlier. If the sequence number of the packet is not in the
conventional AODV protocol, which is initialized with the destination node’s record, it accepts the packet and processes
destination node’s routing measure 8 as it prepares the RREP the encapsulated payload. In contrast, if this packet has
packet. been already received, the destination node silently ignores

111344 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

• Routing measure is the associated cost of the path and is

used to determine the optimal path to the destination;
• Hop count is the number of intermediate hops that this
path includes;
• Lifetime is the point of time at which a path in the routing
table is considered obsolete. Here, the entry associated
with this path is removed from the routing table.
At predefined intervals, a node checks for expired entries in
its routing table. If the node needs to send data packets to
the destination via this outdated path, it must initiate a route
discovery scheme again and waits for at least one available
path before being eventually able to transmit any data packets.

FIGURE 5. Example of route request and reply procedures. D. COMPLEXITY ANALYSIS

Because the route discovery process of the proposed
FT-AORP is mainly based on the original AODV protocol, the
Algorithm 1 Path Selection Algorithm complexity analysis of FT-AORP can be straightforwardly
Input: List r of found paths to the destination, krk ≥ 0 inferred from the AODV protocol [49], [50]. We describe
Output: Two selected paths p1 and p2 two classes of complexity that can be used to characterize
if krk = 0 then the proposed algorithm: time complexity and communication
p1 ← p2 ← ∅ complexity.
else if krk = 1 then In detail, time complexity is defined as the amount of time
p1 ← p2 ← p ∈ r that is required for the entire network to perform a com-
else plete protocol operation (in this case, we consider a success-
p1 ← p ∈ r | ∀pi ∈ r : 2(p) ≥ 2(pi ) ful route discovery). Meanwhile, communication complexity
p2 ← p ∈ r | ∀pi ∈ r, pi 6 = p1 : 2(p) ≥ 2(pi ) denotes the number of protocol messages that are used to do
end if the same task. Without a loss of generality, we assume that
the whole network operates in a synchronous manner. This
means that all the nodes collectively execute at fixed time
instants; in other words, both the processing and propagation
it. Thus, by using multipath routing, the reliability of packet
delays of packets are negligible for all nodes. Subsequently,
transmission is maximized even if one path is not operational.
FT-AORP has the time complexity of O(2d), where d is
the diameter of the network. In the worst case, where the
C. NODE DISCOVERY AND ROUTING TABLE source and destination nodes are on two opposite sides of the
Along with the route discovery scheme and data packet trans- network, recalling that l is the number of intermediate nodes
mission, the network activities also include node discovery. between two nodes, we obtain l = d. It is then intuitively
A node’s discovery of its neighbors enables it to beware of the obvious that a route discovery process must traverse 2l or 2d
existence of the other surrounding nodes. Hence, it should be nodes, including both the route request from the source to the
able to compute its own routing measures 8m and 8s using destination node and vice versa (i.e., the route reply phase).
this information coupled with the measured RSSI values. Similarly, the communication complexity of FT-AORP is
Moreover, node discovery also allows a node to update its O(2n), where n is the number of nodes in the network. This
routing table with direct paths to the nearby nodes, thereby is due to the fact that the operation of FT-AORP is mainly
avoiding the route discovery overhead if the destination is based on the broadcasting mechanism, which means that all
already within a direct communication range with the source network nodes are able to send and receive the protocol’s
node itself. To realize the node discovery scheme, each node control packets in both the route request and route reply
in the network broadcasts a HELLO packet at every prede- phase. Consequently, FT-AORP has a reasonably acceptable
fined interval. convergence rate of the route discovery process given small
Another important component of the routing protocol is the to medium-sized networks.
routing table maintained by every node in the network. More
specifically, Table 2 presents an example of the routing table V. PERFORMANCE EVALUATION
with two entries. The routing table has the following fields: In this section, the proposed routing protocol is imple-
• Destination is the IP address of the destination node; mented in the OMNeT++ discrete event simulator [22]
• Next hop is the IP address of the next hop that also and thus evaluated by conducting various extensive simu-
contains the path information towards the destination lation experiments. The performance was assessed in terms
node; of packet delivery ratio, end-to-end delay, and path energy

VOLUME 10, 2022 111345

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

TABLE 2. Example of the routing table.

level. In addition, for each evaluation metric above, three and operated at 2.4 GHz. In addition, each node was powered
network configurations were altered to observe their impact by energy storage with a nominal capacity of 5328 J. During
on the overall performance, including the node density or the the start phase of the simulation, the battery capacity of each
number of network nodes, the node speed, and the data packet node was initialized with a random value between 70% and
sending rate as follows. 100% of the nominal capacity. Moreover, when a node’s
• For the node density scenario, node speeds were fixed at energy storage is below a threshold ζ of 5%, it was considered
15 m/s and the source node’s sending rate was fixed at energy-depleted. Finally, each entry in the routing table had
0.5 Mbps; an active route timeout of 1.5 s and unless it was updated;
• Regarding the node speed scenario, there were a total of this entry was unusable and removed. The key simulation
20 nodes and the source node’s sending rate was also parameters are summarized in Table 3.
0.5 Mbps;
• For the sending rate scenario, 20 nodes were used, while B. PACKET DELIVERY RATIO
the node speed was fixed at 15 m/s. Packet delivery ratio (PDR) is computed as the ratio of the
In addition, to obtain reliable results, 10 simulation runs were number of successfully delivered packets from the source to
conducted for each configuration with different random seed the destination to the number of generated packets by the
numbers. Ultimately the mean results were computed. source node in a simulation run. This metric is crucial to
For comparative analysis, four representative routing data-critical packet transmissions because it demonstrates the
protocols, namely ad hoc on-demand distance vector reliability and fault-tolerant capability of a protocol. There-
(AODV) [27], ad hoc on-demand multipath distance vec- fore, given ng and nd being the number of generated and
tor (AOMDV) [51], AOMDV with the fitness function successfully delivered packets, respectively, the PDR is cal-
(FF-AOMDV) [16], and load-balanced multi-path routing culated using
protocol with energy constraints (EE-LB-AOMDV) [38] nd
PDR = . (12)
were employed. It is noted that for the EE-LB-AOMDV ng
protocol, we did not implement the load balancing function
Fig. 6 shows the evaluation of the PDR performance of
and we instead use the discovered paths to transmit multiple
five protocols under three different scenarios. First, it was
copies of the same packet to enable fault tolerance. These pro-
measured as the number of nodes increased from 10 to 80,
tocols were re-implemented in the simulation environment
as shown in Fig. 6a. It is evident that, with only 10 nodes, all
and were provided with the exact same configurations for
the protocols succeeded in delivering more than 90% of the
each simulation run in all scenarios.
data packets to the destination node. However, as the number
of nodes increased to 80, the other four protocols, except for
A. SIMULATION MODEL
FT-AORP, failed to provide reliable transmission and their
In the simulation testbed, every simulation run shared the
PDRs were as low as 0.8. In contrast, FT-AORP guaranteed a
following configurations. First, various numbers of nodes
PDR of more than 80%, even in the most aggressive scenario
were initially randomly scattered within a 1000 × 1000 (m2 )
of the network density. This may be justified by the fact that
obstacle-free area. The simulation model employed the ran-
FT-AORP employs various relevant routing metrics to avoid
dom waypoint mobility model [52], implying every node
the occurrence of potential link failures on one path.
moved to a uniformly distributed position with a varied num-
In addition, Fig. 6b shows the PDR performance against
ber of speeds and a zero-pause time. During the simulation
various node speeds, from 5 to 60 m/s. It can be observed
time, the source node generated fixed-size data packets with
that as node speed increases, PDR decreases owing to the
different rates. In addition, the ratio medium of the simulation
high likelihood of link breaks as the nodes constantly change
environment had a background noise power of −110 dBm.
their positions. However, EE-LB-AOMDV and FT-AORP
Each node was equipped with a wireless interface with the
outperformed the other three baseline protocols in nearly
following settings:
every scenario. Even at a very high speed of 60 m/s, the two
• transmitter power: 1.2 mW; protocols were still able to deliver 70% of the data pack-
• receiver sensitivity: −87 dBm; ets to the destination, whereas the recorded numbers of the
• signal-to-noise-plus-interference ratio (SNIR) thresh- remaining protocols were all 10% lower. Because FT-AORP
old: 3 dB. considers the node mobility and its location with respect to
Regarding the MAC protocol, the wireless interfaces used other nodes when selecting the optimal paths, its results were
IEEE 802.11n standard with the ad hoc management mode marginally better than EE-LB-AOMDV in general.

111346 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

TABLE 3. Key simulation parameters. However, the FT-AORP protocol exhibited smaller delays
overall than the other four protocols for all network sizes.
This may be because the intermediate nodes move away from
one another and potentially cause link breaks, while the four
protocols mainly transmit packets using the path with the
smallest number of hop counts
Fig. 7b depicts the delay performance with regard to var-
ious node speed values. It is evident that as the nodes sped
up, the network topology kept changing, possibly causing the
nodes to lose their current established links. At 5 and 15 m/s,
all the protocols showed approximately the same delay in
packet transmission of 0.013 and 0.008 s, respectively. As the
speed increased to 30 m/s and higher, FF-AOMDV and
EE-LB-AOMDV exhibited the best delay results, while the
performance of FT-AORP was also very similar. This is
because they could choose more stable paths compared to
AODV and AOMDV.
Fig. 6c further demonstrates the fault tolerance of
Regarding the end-to-end delay results against various
FT-AORP against different packet sending rates between
sending rates, as shown in Fig. 7c, as the data packets
0.1 and 5 Mbps. In general, the PDR performance of the
being injected into the network in a time instant increased,
three protocols was less materially affected by this factor
the transmission delay increased. This may be attributed to
than by the network density and node speed. At 0.1 Mbps,
lower-layer channel conflicts, as the nodes compete for chan-
the reduced packet delivery efficiency is possibly due to the
nel access. On the whole, the four multipath protocols outper-
low data rate compared to the predetermined node speed and
formed AODV in all scenarios because they could transmit
density. To be more specific, a certain number of packets
the packets using two paths and thus avoid conflicts happen-
might be undelivered due to the fast changes in the network
ing on one path. In fact, EE-LB-AOMDV performed slightly
topology. In the scenario that required the largest throughput
better than FT-AORP because it considered round-trip time
(i.e., 5 Mbps), FT-AORP successfully transmitted nearly 85%
as a routing metric. Overall, the applicability of FT-AORP
of the data packets, which was just 3% fewer compared to
to time-critical systems where the packet’s transmission time
the other less throughput-demanding scenarios of 0.1, 0.5,
should be stably minimal in a variety of contexts is proven.
and 1 Mbps. Meanwhile, EE-LB-AOMDV had a very com-
parable simulation result, and AODV did not provide such
D. PATH ENERGY
a satisfactory performance. Thus, FT-AORP is also an ideal
With this metric, the average energy level of the immediate
routing protocol for bandwidth-demanding applications, such
nodes in a path that is used to transmit a data packet was
as voice and video calls.
computed. It indicates that the FT-AORP protocol prefers
to choose the path with high-energy intermediate nodes.
C. END-TO-END DELAY
Because there may be two discovered paths for data trans-
End-to-end delay is the elapsed time from when a data packet
mission between the source and the destination, the path that
is generated by the source node to when it is successfully
delivers the data packet to the destination first, was consid-
received by the destination node. This latency includes all
ered. The path energy level for a simulation run is computed
potential delays such as queuing at the wireless interface
as
and re-transmission attempts by the Medium Access Control.  
For this metric, the average delay of all the successfully nd
1 X 1 X
transmitted packets in a simulation run was calculated as Path energy =  E j ,
(v )
(14)
nd kp(i) k (i)
i=1 vj ∈p
Pn d g

i=1 τi − τi
d
End-to-end delay = , (13) where nd is the number of successfully transmitted packets,
nd
p(i) is the path used to transmit i-th packet, vj is the j-th node
g g
where τi (resp. τid ), τid > τi , is the time instant where the in the path, and E (vj ) is the energy level (percent) of node vj .
i-th packet is generated by the source node (resp. successfully Fig. 8 presents the average energy level of the trans-
received by the destination node). mitting paths with three different simulation parameters.
Fig. 7 shows the end-to-end delay performance under three First, the performance of the protocols was compared in the
different simulation parameters. In particular, Fig. 7a com- testbed including from 10 to 80 nodes, as shown in Fig. 8a.
pares end-to-end latency with different levels of the network It is evident that both AODV and AOMDV shared very
density measured in the number of nodes. In most cases, the similar simulation results, as there was not much discrep-
latency increased as the network grew larger because there ancy in their operating principles. Regarding FF-AOMDV,
were possibly more nodes involved in a transmission path. EE-LB-AOMDV, and FT-AORP, because these protocols can

VOLUME 10, 2022 111347

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

FIGURE 6. Packet delivery ratio performance against (a) node density, (b) node speed, and (c) sending rate.

FIGURE 7. End-to-end delay performance against (a) node density, (b) node speed, and (c) sending rate.

FIGURE 8. Path energy performance against (a) node density, (b) node speed, and (c) sending rate.

learn and select paths of higher in the energy levels, they paths with higher energy nodes. Consequently, the network
certainly exhibited better performance regardless of the node lifetime was significantly improved, as the energy-depleted
density. nodes did not participate in the data transmission activities.
Next, Fig. 8b shows the measured path energy against vari- Finally, Fig. 8c depicts the path energy performance with
ations in node mobility. It is evident that the overall trend is an multiple data-sending rates. It is obvious that higher the send-
increase in the energy level as the node speed increases for all ing rates result in more power being consumed by the nodes
the routing protocols. For certain speed values, specifically 5, to transmit the data packets. In detail, AODV and AOMDV
30, and 60 m/s, FT-AORP exhibited slightly lower energy had the two smallest path energy levels overall because they
level results compared to the other protocols. However, for did not consider energy as a routing metric. Meanwhile,
the remaining scenarios, FT-AORP was capable of selecting FF-AOMDV, EE-LB-AOMDV, and FT-AORP exhibited the

111348 VOLUME 10, 2022

D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

highest remaining energy levels overall, especially at 5 Mbps, [12] R. A. Nazib and S. Moh, ‘‘Reinforcement learning-based routing protocols
where FT-AORP outperformed all other routing protocols. for vehicular ad hoc networks: A comparative survey,’’ IEEE Access, vol. 9,
pp. 27552–27587, 2021.
[13] M. Ayyash, Y. Alsbou, and M. Anan, ‘‘Introduction to mobile ad-hoc and
VI. CONCLUSION vehicular networks,’’ in Wireless Sensor and Mobile Ad-Hoc Networks:
This study proposed a fault-tolerant ad hoc on-demand rout- Vehicular and Space Applications, D. Benhaddou and A. Al-Fuqaha, Eds.
New York, NY, USA: Springer, 2015, pp. 33–46.
ing protocol for MANET systems, referred to as FT-AORP. [14] M. Conti and S. Giordano, ‘‘Multihop ad hoc networking: The theory,’’
Considering the mobility tendency of network nodes, the IEEE Commun. Mag., vol. 45, no. 4, pp. 78–86, Apr. 2007.
radio signal strength levels they obtain from their neighbors, [15] M. Conti and S. Giordano, ‘‘Multihop ad hoc networking: The evolutionary
path,’’ in Mobile Ad Hoc Networking. Hoboken, NJ, USA: Wiley, 2013,
and their energy consumption rate, the proposed routing pro- pp. 1–33.
tocol can make rational decisions on the prospective paths [16] A. Taha, R. Alsaqour, M. Uddin, M. Abdelhaq, and T. Saba, ‘‘Energy
to be used to transmit the data packets. Simulation results efficient multipath routing protocol for mobile ad-hoc network using the
fitness function,’’ IEEE Access, vol. 5, pp. 10369–10381, 2017.
using the OMNeT++ environment indicated that FT-AORP [17] U. Srilakshmi, N. Veeraiah, Y. Alotaibi, S. A. Alghamdi, O. I. Khalaf,
can improve the packet delivery ratio, reduce the end-to-end and B. V. Subbayamma, ‘‘An improved hybrid secure multipath routing
transmission delay, and select high-energy paths. Conse- protocol for MANET,’’ IEEE Access, vol. 9, pp. 163043–163053, 2021.
quently, the proposed protocol is helpful for applications [18] J. Park, S. Moh, and I. Chung, ‘‘A multipath AODV routing protocol in
mobile ad hoc networks with SINR-based route selection,’’ in Proc. IEEE
where data-critical packets are required to be successfully Int. Symp. Wireless Commun. Syst., Oct. 2008, pp. 682–686.
delivered to the destination with low latency, and a rapid [19] C. T. Hieu and C.-S. Hong, ‘‘A connection entropy-based multi-rate routing
installation of an ad hoc network is strongly favored. protocol for mobile ad hoc networks,’’ J. Comput. Sci. Eng., vol. 4, no. 3,
pp. 225–239, Sep. 2010.
In future work, other routing metrics will be developed [20] S. S. Solapure and H. H. Kenchannavar, ‘‘Design and analysis of RPL
to guarantee the quality of service for specific applications objective functions using variant routing metrics for IoT applications,’’
where a traffic prioritization scheme is demanded. In addi- Wireless Netw., vol. 26, no. 6, pp. 4637–4656, Aug. 2020.
[21] H. H. El-Sayed, A. Younes, and F. A. Alghamdi, ‘‘Multiobjective multicast
tion, the route discovery strategy can be improved to reduce DSR algorithm for routing in mobile networks with cost, delay, and hop
the protocol control overhead. Finally, because radio signal count,’’ Complexity, vol. 2021, pp. 1–8, Jun. 2021.
characteristics could not be entirely modeled by simulations [22] OMNeT++ Discrete Event Simulator. Accessed: Oct. 20, 2022. [Online].
Available: https://omnetpp.org/
conducted, the proposed protocol will be implemented on [23] M. Abolhasan, T. Wysocki, and E. Dutkiewicz, ‘‘A review of routing pro-
real-world mobile devices to further experiment and prove the tocols for mobile ad hoc networks,’’ Ad hoc Netw., vol. 2, no. 1, pp. 1–22,
applicability potential of FT-AORP. Jan. 2004.
[24] C. E. Perkins and P. Bhagwat, ‘‘Highly dynamic destination-sequenced
distance-vector routing (DSDV) for mobile computers,’’ in Proc. SIG-
REFERENCES COMM, New York, NY, USA, Oct. 1994, pp. 234–244.
[1] R. Bruzgiene, L. Narbutaite, and T. Adomkus, MANET Network in Internet [25] T.-W. Chen and M. Gerla, ‘‘Global state routing: A new routing scheme
of Things System. Rijeka, Croatia: IntechOpen, May 2017. for ad-hoc wireless networks,’’ in Proc. IEEE Int. Conf. Commun., vol. 1,
[2] U. Venkanna, J. K. Agarwal, and R. L. Velusamy, ‘‘A cooperative routing Jun. 1998, pp. 171–175.
for MANET based on distributed trust and energy management,’’ Wireless [26] P. Jacquet, P. Muhlethaler, T. Clausen, A. Laouiti, A. Qayyum, and
Pers. Commun., vol. 81, no. 3, pp. 961–979, Apr. 2015. L. Viennot, ‘‘Optimized link state routing protocol for ad hoc networks,’’ in
[3] D. Ramphull, A. Mungur, S. Armoogum, and S. Pudaruth, ‘‘A review of Proc. IEEE Int. Multi Topic Conf., Lahore, Pakistan, Dec. 2001, pp. 62–68.
mobile ad hoc network (MANET) protocols and their applications,’’ in [27] C. E. Perkins and E. M. Royer, ‘‘Ad-hoc on-demand distance vector rout-
Proc. 5th Int. Conf. Intell. Comput. Control Syst. (ICICCS), May 2021, ing,’’ in Proc. 2nd IEEE Workshop Mobile Comput. Syst. Appl., Jan. 1999,
pp. 204–211. pp. 90–100.
[4] L. Kant, K. Young, O. Younis, D. Shallcross, K. Sinkar, A. Mcauley, [28] D. B. Johnson, D. A. Maltz, and J. Broach, ‘‘DSR: The dynamic source
K. Manousakis, K. Chang, and C. Graff, ‘‘Network science based routing protocol for multi-hop wireless ad hoc networks,’’ Ad Hoc Netw.,
approaches to design and analyze MANETs for military applications,’’ vol. 5, pp. 139–172, Jan. 2001.
IEEE Commun. Mag., vol. 46, no. 11, pp. 55–61, Nov. 2008. [29] V. Park and M. Corson, ‘‘A highly adaptive distributed routing algorithm
[5] J. Sandeep and J. S. Kumar, ‘‘Efficient packet transmission and energy for mobile wireless networks,’’ in Proc. IEEE Int. Conf. Comput. Commun.
optimization in military operation scenarios of MANET,’’ Proc. Comput. (INFOCOM), vol. 3, Apr. 1997, pp. 1405–1413.
Sci., vol. 47, pp. 400–407, Jan. 2015. [30] Z. Chen, W. Zhou, S. Wu, and L. Cheng, ‘‘An adaptive on-demand mul-
[6] I.-R. Chen, R. Mitchell, and J.-H. Cho, ‘‘On modeling of adversary behav- tipath routing protocol with QoS support for high-speed MANET,’’ IEEE
ior and defense for survivability of military MANET applications,’’ in Access, vol. 8, pp. 44760–44773, 2020.
Proc. IEEE Mil. Commun. Conf., Oct. 2015, pp. 629–634. [31] G. Dhananjayan and J. Subbiah, ‘‘T2AR: Trust-aware ad-hoc routing
[7] M. H. Hassan, S. A. Mostafa, H. Mahdin, A. Mustapha, A. A. Ramli, protocol for MANET,’’ SpringerPlus, vol. 5, no. 1, p. 995, Jul. 2016.
M. H. Hassan, and M. A. Jubair, ‘‘Mobile ad-hoc network routing proto- [32] C. Z. Sirmollo and M. A. Bitew, ‘‘Mobility-aware routing algorithm for
cols of time-critical events for search and rescue missions,’’ Bull. Electr. mobile ad hoc networks,’’ Wireless Commun. Mobile Comput., vol. 2021,
Eng. Informat., vol. 10, no. 1, pp. 192–199, Feb. 2021. pp. 1–12, May 2021.
[8] Y. Jin, X. Tan, W. Feng, J. Lv, A. Tuerxun, and K. Wang, ‘‘MANET [33] S. A. Nsaif, S. Y. Park, and J. M. Rhee, ‘‘SRAD: A novel approach to
for disaster relief based on NDN,’’ in Proc. 1st IEEE Int. Conf. Hot seamless routing for wireless ad hoc networks,’’ in Proc. 23rd Int. Conf.
Information-Centric Netw. (HotICN), Aug. 2018, pp. 147–153. Adv. Commun. Technol. (ICACT), Feb. 2021, pp. 172–175.
[9] P. K. Garg, ‘‘Potentials of network-based unmanned aerial vehicles,’’ in [34] X. Song, L. Zhou, H. Zhao, X. Hu, and J. Wei, ‘‘Localized fault tolerant and
Cloud IoT-Based Vehicular Ad Hoc Networks. Hoboken, NJ, USA: Wiley, connectivity restoration algorithms in mobile wireless ad hoc network,’’
2021, pp. 369–397. IEEE Access, vol. 6, pp. 36469–36478, 2018.
[10] T. Ohta, M. Nishi, T. Terami, and Y. Kakuda, ‘‘Information dissemination [35] H. Nishiyama, T. Ngo, N. Ansari, and N. Kato, ‘‘On minimizing the impact
using MANET for disaster evacuation support,’’ IEICE Trans. Commun., of mobility on topology control in mobile ad hoc networks,’’ IEEE Trans.
vol. 102, no. 4, pp. 670–678, 2018. Wireless Commun., vol. 11, no. 3, pp. 1158–1166, Mar. 2012.
[11] Y. Jahir, M. Atiquzzaman, H. Refai, A. Paranjothi, and P. G. LoPresti, [36] U. Srilakshmi, S. A. Alghamdi, V. A. Vuyyuru, N. Veeraiah, and
‘‘Routing protocols and architecture for disaster area network: A survey,’’ Y. Alotaibi, ‘‘A secure optimization routing algorithm for mobile ad hoc
Ad Hoc Netw., vol. 82, pp. 1–14, Jan. 2019. networks,’’ IEEE Access, vol. 10, pp. 14260–14269, 2022.

VOLUME 10, 2022 111349

 D. N. M. Hoang et al.: Fault-Tolerant Ad Hoc On-Demand Routing Protocol for Mobile Ad Hoc Networks

[37] P. K. Pattnaik, B. K. Panda, and M. Sain, ‘‘Design of novel mobility and DUC N. M. HOANG (Graduate Student Member,
obstacle-aware algorithm for optimal MANET routing,’’ IEEE Access, IEEE) received the B.E. degree in computer engi-
vol. 9, pp. 110648–110657, 2021. neering from the Ho Chi Minh City University of
[38] M. Naseem, G. Ahamad, S. Sharma, and E. Abbasi, ‘‘EE-LB-AOMDV: Technology (HCMUT), VNU—HCM, Vietnam,
An efficient energy constraints-based load-balanced multipath routing in 2020. He is currently pursuing the M.S. degree
protocol for MANETs,’’ Int. J. Commun. Syst., vol. 34, no. 16, p. e4946, with the Department of Electronics Engineering,
Nov. 2021. Myongji University, South Korea. His research
[39] S. Sarhan and S. Sarhan, ‘‘Elephant herding optimization ad hoc on-
interests include ad-hoc networking, wireless net-
demand multipath distance vector routing protocol for MANET,’’ IEEE
works, fault-tolerant networks, and the Internet of
Access, vol. 9, pp. 39489–39499, 2021.
[40] M. Tarique, K. E. Tepe, S. Adibi, and S. Erfani, ‘‘Survey of multipath Things.
routing protocols for mobile ad hoc networks,’’ J. Netw. Comput. Appl.,
vol. 32, no. 6, pp. 1125–1143, 2009.
[41] S. Mao, S. Lin, Y. Wang, S. S. Panwar, and Y. Li, ‘‘Multipath video JONG MYUNG RHEE (Member, IEEE) received
transport over ad hoc networks,’’ IEEE Wireless Commun., vol. 12, no. 4, the Ph.D. degree from North Carolina State Uni-
pp. 42–49, Aug. 2005. versity, USA, in 1987.
[42] X. Li and L. Cuthbert, ‘‘Stable node-disjoint multipath routing with low After 20 years at the Agency for Defense
overhead in mobile ad hoc networks,’’ in Proc. IEEE Comput. Soc. 12th Development in South Korea, where he made
Annu. Int. Symp. Modeling, Anal., Simulation Comput. Telecommun. Syst., noteworthy contributions to C4I and military
Oct. 2004, pp. 184–191. satellite communications, he joined DACOM and
[43] Y. H. Robinson, E. G. Julie, K. Saravanan, R. Kumar, and L. H. Son, ‘‘FD- Hanaro Telecom, in 1997 and 1999, respectively.
AOMDV: Fault-tolerant disjoint ad-hoc on-demand multipath distance At Hanaro Telecom, which was the second-largest
vector routing algorithm in mobile ad-hoc networks,’’ J. Ambient Intell. local carrier in South Korea, he served as the Chief
Humanized Comput., vol. 10, no. 11, pp. 4455–4472, Nov. 2019. Technology Officer (CTO) with a senior executive vice-president position.
[44] C. Wu, K. Kumekawa, and T. Kato, ‘‘A MANET protocol considering His main duty at Hanaro Telecom was a combination of management
link stability and bandwidth efficiency,’’ in Proc. Int. Conf. Ultra Modern
and new technology development for high-speed Internet, VoIP, and IPTV.
Telecommun. Workshops, Oct. 2009, pp. 1–8.
In 2006, he joined Myongji University, and currently, he is a Special
[45] D. Macone, G. Oddi, and A. Pietrabissa, ‘‘MQ-routing: Mobility-, GPS-
and energy-aware routing protocol in MANETs for disaster relief scenar- Mission Professor with the Information and Communications Engineering
ios,’’ Ad Hoc Netw., vol. 11, no. 3, pp. 861–878, May 2013. Department. His current research interests include centered on military com-
[46] R. Jurdak, P. Corke, D. Dharman, and G. Salagnac, ‘‘Adaptive GPS duty munications and smart grids, including ad-hoc and fault-tolerant networks.
cycling and radio ranging for energy-efficient localization,’’ in Proc. 8th
ACM Conf. Embedded Netw. Sensor Syst. (SenSys), Zürich, Switzerland,
2010, pp. 57–70. SANG YOON PARK (Member, IEEE) received
[47] T. M. Cover and J. A. Thomas, Elements of Information Theory. Hoboken, the B.S. degree in electrical engineering and the
NJ, USA: Wiley, 2006. M.S. and Ph.D. degrees in electrical engineer-
[48] C. E. Shannon, ‘‘A mathematical theory of communication,’’ Bell Syst. ing and computer science from Seoul National
Tech. J., vol. 27, no. 3, pp. 379–423, Jan. 1948. University, Seoul, Republic of Korea, in 2000,
[49] M. Srivastava and P. Yadav, ‘‘A performance analysis of routing protocols 2002, and 2006, respectively. He joined the
in mobile ad-hoc networks,’’ International J. Eng. Res. Technol., vol. 1,
School of Electrical and Electronic Engineering,
no. 8, p. 11, Oct. 2012.
Nanyang Technological University, Singapore,
[50] A. Lee, I. Ra, and H. Kim, ‘‘Performance study of ad hoc routing proto-
cols with gossip-based approach,’’ in Proc. Simulation Multiconference, as a Research Fellow, in 2007. From 2008 to 2014,
Mar. 2009, pp. 1–8. he was a Research Scientist with the Institute for
[51] M. K. Marina and S. R. Das, ‘‘On-demand multipath distance vector Infocomm Research, Singapore. Since 2014, he has been with the Depart-
routing in ad hoc networks,’’ in Proc. IEEE 9th Int. Conf. Netw. Protocols ment of Electronic Engineering, Myongji University, Yongin, Republic of
(ICNP), Nov. 2001, pp. 14–23. Korea, where he is currently an Associate Professor. His research interests
[52] C. Bettstetter, H. Hartenstein, and X. Pérez-Costa, ‘‘Stochastic properties include design of dedicated and reconfigurable architectures for low-power
of the random waypoint mobility model,’’ Wireless Netw., vol. 10, no. 5, and high-performance digital communication systems.
pp. 555–567, 2004.

111350 VOLUME 10, 2022