Sensors 19-03789-v2

sensors
Article
Energy-Efficient Multi-Disjoint Path Opportunistic
Node Connection Routing Protocol in Wireless Sensor
Networks for Smart Grids
Junaid Anees 1,2 , Hao-Chun Zhang 1,* , Sobia Baig 3 and Bachirou Guene Lougou 1
1 School of Energy Science and Engineering, Harbin Institute of Technology, Harbin 150001, China
2 Satellite Control Facility (SCF-L) directorate, SE wing, Space and Upper Atmosphere Research Commission,
Lahore 54000, Pakistan
3 Department of Electrical and Computer Engineering, COMSATS University Islamabad (CUI),
Lahore Campus, Lahore 54000, Pakistan
* Correspondence: hczhang@hit.edu.cn; Tel.: +86-451-86412328
Received: 25 July 2019; Accepted: 30 August 2019; Published: 1 September 2019
Abstract: The gradual increase in the maturity of sensor electronics has resulted in the increasing
demand for wireless sensor networks for many industrial applications. One of the industrial platforms
for efficient usage and deployment of sensor networks is smart grids. The critical network traffic in
smart grids includes both delay-sensitive and delay-tolerant data for real-time and non-real-time usage.
To facilitate these traffic requirements, the asynchronous working–sleeping cycle of sensor nodes
can be used as an opportunity to create a node connection. Efficient use of wireless sensor network
in smart grids depends on various parameters like working–sleeping cycle, energy consumption,
network lifetime, routing protocol, and delay constraints. In this paper, we propose an energy-efficient
multi-disjoint path opportunistic node connection routing protocol (abbreviated as EMOR) for sensor
nodes deployed in neighborhood area network. EMOR utilizes residual energy, availability of
sensor node’s buffer size, working–sleeping cycle of the sensor node and link quality factor to
calculate optimum path connectivity after opportunistic connection random graph and spanning tree
formation. The multi-disjoint path selection in EMOR based on service differentiation of real-time
and non-real-time traffic leads to an improvement in packet delivery rate, network lifetime, end-end
delay and total energy consumption.
Keywords: energy-efficient wireless sensor networks; multipath opportunistic node connection; smart
grids; neighborhood area network; asynchronous working–sleeping cycle strategy; energy-efficient
routing protocol; opportunistic connection random graph; service differentiation in smart grids
1. Introduction
In recent years, the demand for wireless sensor networks has increased for many industrial
applications such as fault diagnosis and monitoring, surveillance monitoring, industrial control
systems, commodity consumption monitoring and plant automation, etc. [1]. In addition, in industrial
applications, wireless sensor networks have many medical, military, civil and environmental
applications as well body area network monitoring [2], civil structure monitoring [3], target tracking in
battlefields [4], earthquake and flood monitoring [5], forest fire monitoring [6], etc.
The technological progression in the field of highly integrated digital sensor electronics, small
scale microprocessors, low power transceivers and radio frequency devices, collectively resulted in the
design of efficient wireless sensors [7]. These wireless sensor devices are responsible for sensing the
change in required physical phenomena of their surrounding environment with the help of a small
microprocessor, radio transceiver, a few transducers and a low capacity battery [7]. The transducers or
Sensors 2019, 19, 3789; doi:10.3390/s19173789 www.mdpi.com/journal/sensors

Sensors 2019, 19, 3789 2 of 32
sensors perform the specific type of sensing in the surrounding environment and pass the sensing
information to a microprocessor for further processing of sensed data [7]. A wireless sensor’s radio
transceiver is used to transmit and receive the sensed data to/from adjacent wireless sensors or to/from
the sink node, depending upon the location of the wireless sensor in the field [7]. Due to low battery
capacity, wireless sensor nodes have to adjust their sleep–awake cycle (or working–sleeping cycle)
according to scenario requirements in order to maximize the network life time and minimize the overall
network delay [8]. The possible existence of these miniaturized wireless sensor devices motivated
researchers to emphasize the significance of wireless sensor node collaboration for data sensing, data
collection and aggregation purposes which resulted in the discovery of an emerging field known as
wireless sensor network (WSN). The purpose of deploying multiple wireless sensor nodes in a field is
to have collaboration between these sensor nodes to achieve a shared goal with the help of sensing,
data aggregation and data sharing between wireless sensor nodes [9].
The frequent transmission failures, electricity theft and congestion problems in traditional
electric grids lead to the consideration that traditional electric grids are insecure and inefficient
in terms of energy management. In order to eradicate these problems, we need to incorporate
state-of-the-art bidirectional communication interfaces, automated control systems and distributed
computing capabilities in our current grid, which will improve energy efficiency, reliability, security
and agility in our electric grid [10]. Furthermore, a highly integrated next generation system is needed
in which electricity service providers, distributors and prosumers are well aware of real-time energy
requirements and capabilities; the system which offers high performance computing, security, scalability,
reliability and security with state-of-the-art communication network infrastructure is our “smart grids”
(SGs) [10,11]. For better realization of SGs, we need to distribute and gather information remotely
and in a timely manner from different phases of our SGs (i.e., generation, transmission, distribution
and consumption) [10–12]. The data acquisition from different devices deployed in the neighborhood
area network (NAN) can be efficiently achieved by incorporating WSN in SGs. The upgradation of
contemporary electric grids to SGs could be supported by self-governing and self-organizing features
of WSN [13]. A traditional electric grid needs installation of costly and inefficient wired monitoring
systems, expensive communication cables and a high-maintenance budget [10]. On the other hand,
SGs with the help of wireless monitoring sensors and proper fault diagnostics, can remarkably reduce
power losses and long-term maintenance budget and does not require expensive communication
cables, thus enhancing the system’s reliability and efficiency [13].
Efficient energy utilization and network lifetime are considered as the main design parameters
in the previous research conducted for WSN [8]. Additionally, the feasibility of mobility usage in
WSN as demonstrated in [14] can improve the network lifetime by replacing the traditional static
wireless sensor nodes with mobile wireless sensor nodes. Similarly, if the center node (or sink) is
mobile, it needs more computation power in comparison to other sensor nodes [14]. The mobility and
more computation power will drain out the battery of that sink node after a few rounds of sensing
and aggregation, and its energy should be replenished in a timely manner. The mobility of a sink
node can be controlled or randomly planned depending upon the particular scenario of the WSN and
in this way, many traditional problems like hot-spot problem [14] can be avoided. With reference
to [15], researchers have proposed a working–sleeping cycle strategy in which free nodes go to sleep to
save their battery power and have a proper node scheduling scheme for efficient data transmission.
These node scheduling schemes were categorized as synchronous and asynchronous working–sleeping
cycle, which means that the network lifetime can be prolonged by changing the node scheduling
schemes in accordance with the scenarios. Although it can affect the link stability between sensor
nodes, it also creates an opportunistic node connection to exist between nodes due to asynchronous
working–sleeping scheduling.

Sensors 2019, 19, 3789 3 of 32
According to [16–20], Opportunistic Routing (OR) is a paradigm for wireless networks which
benefit from broadcast characteristics of a wireless medium by selecting multiple nodes as candidate
forwarders, to improve network performance. In [18–20] a set of nodes are selected as potential
forwarders and the nodes in the selected set forward the packet according to some criteria after they
receive the packet. This group of nodes in OR is called a candidate set (CS). The performance of
OR depends on several key factors such as OR metric, candidate selection algorithm and candidate
coordination method. Boukerche et al. in [20] discussed the basic function of OR by highlighting its
ability to overhear the transmitted packet and to coordinate among relaying nodes. In OR, by using a
dynamic relay node to forward the packet, the transmission reliability and network throughput can be
increased. Thus, the term “Opportunistic Routing” can be defined as the routing scheme in which the
next best forwarder is dynamically selected with respect to OR metric, candidate selection algorithm
and candidate coordinate method as in [18–20].
The reasons why opportunistic node connections exist in WSNs can be summarized as follows:
• WSNs are often deployed in harsh environments (smart grids NAN in our case), where wireless
signals are susceptible to interference, thus causing link instability [21,22], further leading to
opportunistic node connections;
• The sink node mobility usually leads to intermittent links in the network, resulting in opportunistic
node connections [21];
• Due to the limited energy of the nodes, the sensor nodes adopt an asynchronous working–sleeping
cycle strategy to save energy and accordingly, the adjacent nodes may not be able to communicate
with each other continuously as in [22], thus bringing about opportunistic node connections.
By utilizing asynchronous working–sleeping cycle strategy, multiple nodes in WSNs can apply
the concept of OR by overhearing their working neighbor’s transmission. A set of potential forwarders
can be created using opportunistic connection random graph (OCRG) and formation of a spanning
tree can be used to demonstrate that the candidate nodes in OR will forward the packet according to
some criteria (i.e., optimal link and path connectivity calculations in our case).
However, the sensed data being generated as a result of deploying a WSN in a NAN of SGs might
have different attributes like delay tolerance and delay sensitivity [13]. For example, the monitoring
data (e.g., power load) generated by a sensor network which is part of a SG can always be delay
sensitive and should be transmitted to the data processing (or sink) node within certain time limits
thereby increasing the bandwidth requirements, whereas the control data (e.g., changing power load)
generated by the sensor network in SGs can often be considered as delay tolerant data and is not
required to be received by the data processing (or sink) node immediately [13]. Keeping in view the
system requirements of SGs, parameters like energy consumption, mobility, network lifetime, delay
and bandwidth can be treated as performance metrics for effective use of the WSN in SGs.
Many routing techniques specifically designed for WSNs have been proposed in [23–25]. Likewise,
the multipath routing protocols according to [26] have been proposed for WSNs. These multipath
routing protocols use multiple paths for data delivery, and thus improve the network reliability
and robustness. In this paper, we propose an energy efficient and multipath opportunistic node
connection routing protocol for WSNs in NANs to achieve load balancing through splitting up traffic
in terms of real-time and non-real-time across multi-disjoint paths and energy consumption balance
through asynchronous working–sleeping cycle of sensor nodes. Figure 1 demonstrates the wireless
sensor networks deployed in the NAN of SGs in which our NAN gateway is actually the mobile sink.
The opportunistic node connections between mobile sink sensor nodes and between adjacent sensor
nodes can be clearly seen in Figure 1.

Sensors 2019, 19, 3789 4 of 32
As a reflection of opportunistic node connectivity caused by asynchronous working–sleeping
cycle strategy, OCRG is constructed in which a probe message forwarding mechanism is used for each
node to forward the information to the mobile sink node [27]. Additionally, EMOR utilizes the residual
energy, availability of the node’s buffer size, working–sleeping cycle of sensor nodes and link quality
factor to calculate the optimal link and path connectivity for both real-time and non-real-time traffic
and assigns multi-disjoint paths to them accordingly.Sensors 2019, 19, x FOR PEER REVIEW 4 of 33
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGWHGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
HGW
Access GW
Mobile sink (NAN
GW)
Mobile sink (NAN
GW)
SG Control
Center
Backbone Network
Working sensor nodes
Sleeping sensor nodes
Working sensor nodes
outside mobile sink Rl
Mobile sink (NAN Gateway)
Link b/w working nodes and
mobile sink
Low quality link b/w mobile sink
& working nodes outside Rl
Link b/w working sensor nodes
Possible links of sleeping nodes
after wake up
Long radio range of mobile sink
Possible links of mobile sink with
sleeping nodes after transition
Figure 1. Illustration of wireless sensor network (WSN) adopting working–sleeping cycle strategy
with opportunistic node connections in neighborhood area network (NAN).
The rest of the paper is organized as follows: Section 2 describes the related research work
conducted for WSN routing protocols in SGs and various other applications. System modeling is
presented in Section 3. EMOR protocol design for WSNs in NAN is presented in Section 4.
Performance evaluation and simulation results can be depicted from Section 5. Finally, Section 6 and
7 briefly discuss the paper and provide some future research directions.
2. Related Research Work
Keeping in view the need for a more efficient and robust electric grid, a conceptual framework
was proposed by the National Institute of Standards and Technology (NIST) in SGs Interoperability
Standards Roadmap [28,29]. According to this conceptual framework, there is a need to have an inter
domain and intra domain communication between key building blocks like operations, markets,
service providers, generation, transmission, distribution and consumer [28,29]. The NIST framework
states that SGs should support the information flow and electrical flow between all key building
blocks. Information flow includes data acquisition, data processing and data dissemination between
the desired energy subsystems of SGs, while the electrical flow deals with the generation,
transmission and distribution of energy [29]. The main objective of information flow is to monitor
and control the energy whereas the electrical flow is responsible for power delivery, demand and
asset optimization, etc. [29].
Figure 1. Illustration of wireless sensor network (WSN) adopting working–sleeping cycle strategy with
opportunistic node connections in neighborhood area network (NAN).
The rest of the paper is organized as follows: Section 2 describes the related research work
conducted for WSN routing protocols in SGs and various other applications. System modeling is
presented in Section 3. EMOR protocol design for WSNs in NAN is presented in Section 4. Performance
evaluation and simulation results can be depicted from Section 5. Finally, Sections 6 and 7 briefly
discuss the paper and provide some future research directions.
2. Related Research Work
Keeping in view the need for a more efficient and robust electric grid, a conceptual framework
was proposed by the National Institute of Standards and Technology (NIST) in SGs Interoperability
Standards Roadmap [28,29]. According to this conceptual framework, there is a need to have an
inter domain and intra domain communication between key building blocks like operations, markets,
service providers, generation, transmission, distribution and consumer [28,29]. The NIST framework
states that SGs should support the information flow and electrical flow between all key building

Sensors 2019, 19, 3789 5 of 32
blocks. Information flow includes data acquisition, data processing and data dissemination between
the desired energy subsystems of SGs, while the electrical flow deals with the generation, transmission
and distribution of energy [29]. The main objective of information flow is to monitor and control the
energy whereas the electrical flow is responsible for power delivery, demand and asset optimization,
etc. [29].
In [30–32], several key technologies have been identified by the National Energy Technology
Laboratory (NETL) for SGs:
(1) Full-duplex and high-speed communication infrastructure for the information exchange between
different SG entities.
(2) State-of-the-art sensing network infrastructure which will be responsible for the measurement
and relaying of physical data. This sensor network can be used to prevent power theft, improve
demand response, etc.
(3) Fabrication of components related to power electronics, superconductivity and energy storage
should be needed based on the contemporary research being conducted for SGs.
(4) Seamless and real-time decision making at the prosumer end should be possible by improving
the prosumer interface with operations, markets and service providers.
The general overview of SGs concept is presented in [11,12]. The communication architecture and
application needs of SGs are explained in detail in [30]. Moreover, the mechanisms for data collection
and SGs sensing network is surveyed in [13] and strategies for improving information flows in SGs is
mentioned in [33]. Technologies like high speed communication network architecture for SGs have
been studied in [30]. Due to the decentralized and lightweight architecture of WSNs, it can be efficiently
used in SGs and micro-grids [13]. Erol-Kantarci et al. in [33] presented that WSNs can be used in
various SG applications like generation, transmission, distribution and consumption as the WSN can
deliver the information needed by the intelligent algorithms running in the control center of SGs.
An overview of opportunities and challenges of WSNs in SGs is presented in [34]. Likewise, different
researchers have focused on proposing different routing protocols for WSNs based on parameters like
network lifetime, delay, bandwidth, packet to delivery ratio and node mobility [23–26]. As mentioned
in a detailed survey for energy efficient and energy balanced routing protocols for a WSN [35], most of
these protocols emphasize on improving energy consumption during network transmission activities
but very few researches can be found about improvement in energy consumption with node mobility.
In a typical WSN topology, all nodes including the sink nodes are static. This leads to a situation in
which the adjacent or neighboring nodes of the sink node deplete their limited battery powers as they
have to participate in managing more traffic loads for forwarding the sensing data to the sink node as
compared to the nodes which are far from the sink node. This typical scenario is known as hot-spot
problem and adversely affects the network lifetime in any WSN application [14,15]. For this particular
reason, the node mobility concept is introduced by many researchers, which exploits the movement of
the sink node in the WSN to overcome the hot-spot problem [36,37]. As the location of the neighboring
nodes around the sink node are changed due to movement of the sink node, the probability of a
hot-spot problem is reduced and network lifetime is improved due to a more even distribution of
energy consumption in the surrounding nodes [37]. Keeping in view the node mobility, researchers
have worked on the path formation for mobile sinks in [37–39] based on the predefined, controlled and
random path selection procedures. In [40], Alghamdi et al. proposed a new routing aware algorithm
to detect malicious nodes in a concealed data aggregation for WSNs to highlight the significance of
data aggregation in WSNs by reducing communication overhead of sensor nodes.

Sensors 2019, 19, 3789 6 of 32
All these previous researches were formulated to equalize the energy consumption of nodes,
but they are not able to prolong the network lifetime. Thus, an emerging concept “utilization
of working–sleeping cycle of nodes” was introduced to improve the energy consumption of
nodes and overall network lifetime [22]. The working–sleeping cycle can be segmented into
two categories—synchronous and asynchronous working–sleeping cycles. Authors in [41,42]
revealed that synchronous working–sleeping cycle could help in achieving improvement in energy
consumption. However, the synchronization problem needed significant contributions. Thus in [43],
Ng et al. proposed an energy efficient synchronization algorithm in which adaptive adjustment
of the traffic and wakeup period could improve energy consumption through counter-based and
exponential-smoothing algorithms. In addition to it, some researchers also explored asynchronous
working–sleeping cycle in which all nodes have independent working–sleeping schedules depending
on the network connectivity requirements in terms of traffic coverage area [16,17]. Mukherjee et al.
also proposed an asynchronous working–sleeping technique while focusing on network coverage by
maintaining a minimum number of awake nodes [44]. As a result of working–sleeping cycle strategy,
opportunistic node connections will be established between sensor nodes and the mobile sink which
can further lead to link instability. Therefore, we need a random graph theory to model these kinds of
opportunistic node connections in the WSN. Mostafaei et al. proposed reliable routing with distributed
learning automaton (RRDLA) algorithm, which considers dynamics of links in finding a path from
a source to a destination by considering Quality of Service (QoS) constraints such as end-to-end
reliability and delay [45]. Ben Fradj et al. in [46] presented a new opportunistic routing protocol
called energy-efficient opportunistic routing protocol using a new forward list (EEOR-FL) aiming to
balance energy consumption and maximize the network lifetime by calculating the list of candidates.
Sadatpour et al. proposed a new collision-aware opportunistic routing protocol abbreviated as SCAOR
for highways by utilizing cluster-based scheduling algorithms [47].
Kenniche et al. in [48] indicated that WSN modeling in the case of working–sleeping cycle strategy
and opportunistic node connections can be modeled using a random geometric graph in which a set of
vertices represent sensor nodes and a set of edges represent links between those vertices or sensor
nodes. In reference [49], Norman et al. proposed a novel random graph modeling for heterogeneous
sensor networks based on different transmission ranges and a new routing metric. Referring to [50],
in which Ren et al. focused on the challenges and weaknesses of using random graph theory for
modeling WSN, they thus proposed a new weighted topology model for WSN based on the random
geometric theory. In [51], Fu et al. developed an optimal policy for connection between source and
destination nodes in random WSN topology using random graph. For this policy to be implemented
correctly, an energy efficient probe message forwarding mechanism was proposed. After analyzing
the source information in the received probe message, the mobile sink calculates the link connectivity
between any two adjacent nodes after determining their neighbor relationships. Thus, an opportunistic
connection random graph can be constructed.
As SGs require real-time (delay sensitive) and non-real-time (delay tolerant) traffic being routed
through the WSN while minimizing the energy consumption and increasing the network lifetime,
we need to utilize the WSN routing protocol which can offer energy efficiency and multi-disjoint
path routing. Keeping in view these SGs requirements, Dulman et al. in [52] discussed the trade-offs
between traffic overhead and reliability in multi-path routing for WSN. In [35], the authors discussed
the taxonomy of cluster-based routing protocols for WSN, with respect to energy efficient and energy
balance routing protocols. Ben-Othman et al. discussed the energy efficient and QoS based routing
protocol in [53]. Mostafaei et al. in [54] investigated the problem of self-protection in WSNs and devised
the Self Protection Learning Automaton (SPLA) algorithm in which sensing graph of the network plays
the main role in finding the minimum number of nodes to protect the nodes. The proposed solution
takes advantage of irregular cellular learning automaton (ICLA) to properly schedule the sensors into
either an active or idle state. In [53], researchers proposed an energy efficient and QoS-based multi-path
routing protocol with disjoint paths for real-time and non-real-time traffic management. In [55],

Sensors 2019, 19, 3789 7 of 32
Liang et al. presented a novel routing optimization technique based on improvement in Low Energy
Adaptive Clustering Hierarchy (LEACH) in WSNs. According to this novel routing optimization
technique, energy efficiency per unit node per round can be achieved and network lifetime can be
prolonged. In our proposal, we utilized certain ideas from the previous routing protocols in WSN and
proposed a detailed solution for optimally tackling the problems like energy consumption and network
lifetime enhancement through multi-disjoint path opportunistic node modeling using random graph
theory for delay-tolerant and delay-sensitive data in SGs NAN.
3. System Modeling
In this paper, we assumed that all sensor nodes are being deployed in the NAN randomly and
independently and the mobile sink is considered as the NAN gateway for collection of monitoring
and control data from the sensor nodes. Also, we assumed that all sensor nodes follow asynchronous
working–sleeping cycle strategy with their working time as WV and their sleeping time as SV.
The residual energy of the sensor nodes is based on the simplified energy consumption model [18].
Figure 2 depicts the asynchronous working–sleeping cycle strategy of nodes along with their residual
energy slope. Both of these features are used to visualize the possible link connectivity between
adjacent sensor nodes, Vi, Vj, and Vk, which are within their radio range r. With reference to [18],
the energy consumed during transmission and reception for adjacent nodes Vi and Vj is given by:
ET(Vi, Vj) = (E + εampd2
(Vi,Vj)
)B (1)
ER(Vj) = EB (2)
where E is the basic energy consumed during transmission and reception per bit, B is the number
of bits transmitted or received, εamp is the energy consumed by transmission amplifier, d(Vi,Vj) is the
distance between node Vi and Vj, ET and ER are the energy consumed by nodes for transmission and
reception, respectively. Figure 2 shows that the link connectivity between adjacent nodes depends on
asynchronous working–sleeping cycle as well as residual energy of nodes. When the node is in working
mode, its residual energy starts depleting and follows the residual energy slope. The residual energy at
the start of the working mode is high and is low at the end of the working mode. So, the opportunistic
link connection if established at the start of the working mode will lead to higher link connectivity
value as compared to the connection established at the center or end of the working mode.
Sensors 2019, 19, x FOR PEER REVIEW 7 of 33
sensor nodes, Vi, Vj, and Vk, which are within their radio range r. With reference to [18], the energy
consumed during transmission and reception for adjacent nodes Vi and Vj is given by:
2
( , )
( , ) ( )  i jT i j amp V V
E V V E d B (1)
( )R j
E V EB (2)
where E is the basic energy consumed during transmission and reception per bit, B is the number
of bits transmitted or received, amp is the energy consumed by transmission amplifier, ( , )i jV Vd is the
distance between node Vi and Vj, ET and ER are the energy consumed by nodes for transmission and
reception, respectively. Figure 2 shows that the link connectivity between adjacent nodes depends on
asynchronous working–sleeping cycle as well as residual energy of nodes. When the node is in
working mode, its residual energy starts depleting and follows the residual energy slope. The
residual energy at the start of the working mode is high and is low at the end of the working mode.
So, the opportunistic link connection if established at the start of the working mode will lead to higher
link connectivity value as compared to the connection established at the center or end of the working
mode.
Node Vi
Node Vj
Node Vk
WVi SVi
WVj SVj
LViVk> LViVj LVjVk> LViVk> LVjVk > LViVj (LVjVk ≈ LViVk)> LViVj
RE
High
RE
Low
RE
Low
RE
High
RE
High
RE
Low
t1 t2 t3 t4 t5 t6
Time
NodeStatus
Transitions
Residual Energy
Slope
Nodes deployed
in SGs NAN
Status Transitions
Timeslot for possible
communication b/w
nodes
WVk
SVk
WVi WVj WVk : working time
of node Vi Vj Vk
SVi SVj SVk : working time of
node Vi Vj Vk
1
0
1
0
1
0
Figure 2. Illustration of link connectivity based on asynchronous working–sleeping cycle strategy and
residual energy of nodes deployed in NAN. SGs—smart grids.
4. EMOR Protocol Design for WSN in NAN
The following five phases reflect the proposed design of EMOR for WSN deployed in NAN.
4.1. Initialization Phase
Figure 2. Illustration of link connectivity based on asynchronous working–sleeping cycle strategy and
residual energy of nodes deployed in NAN. SGs—smart grids.

Sensors 2019, 19, 3789 8 of 32
4. EMOR Protocol Design for WSN in NAN
The following five phases reflect the proposed design of EMOR for WSN deployed in NAN.
4.1. Initialization Phase
It is assumed that the mobile sink node can start the data collection anywhere and anytime in the
network by broadcasting a message which contains the sink ID (SID) and data collection duration.
Upon receiving the message from the mobile sink, the sensor nodes within long radio range Rl
obtain the data collection duration the sink and start calculating their working schedule (based on
the working–sleeping cycle) and status transition frequency (i.e., switching from working to sleeping
and vice versa), within this particular data collection duration. In case the sensor node is not working
(sleeping state), then it can participate in the next data collection period when the sink node broadcasts
the same message again to facilitate the sleeping sensor nodes and acquisition of probe messages
from the new sensor nodes. Here, we have assumed that the relative signal strength indication (RSSI)
parameter can be used to estimate the distance between the mobile sink and sensor nodes. In order to
minimize the energy consumption, we employed the expected optimal hops (EOH) approach. EOH
can be defined as the number of hops needed to forward a probe message from any node to the mobile
sink node with minimum energy consumption [22]. According to [22], EOH can be expressed as:
EOHVi
=
εamp
2E
d(Vi,Vsin k) (3)
where E is the basic energy consumed during transmission and reception per bit, εamp is the energy
consumed by the transmission amplifier, d(Vi,Vsin k) is the distance between node Vi and VSINK and
EOHVi
is the expected optimal hops of node Vi needed to select proper forwarders to the mobile sink
with minimum energy consumed.
4.2. Probe Message Forwarding Mechanism in Data Collection Scope
All sensor nodes within long radio range Rl which received the broadcast message from the mobile
sink, need to send a data set comprising of information such as source ID, status transition frequency,
working–sleeping schedule and all neighbor IDs to the mobile sink node during the data collection
period. The format of probe message being forwarded to the mobile sink can be seen from Table 1.
Table 1. Format of probe message in data collection scope.
Fields
Source
ID
Working–Sleeping
Schedule
Status
Transition
Frequency
Neighbor
IDs
Sink
ID
Forwarders
ID
Expected
Optimal Hops
(EOH)
Length (bits) 10 15 10 10 10 60 5
The lengthof the probemessage is 120 bits in which thesource IDoccupies 10 bits, working–sleeping
schedule occupies 15 bits, status transition frequency and neighbor IDs have 10 bits each, sink ID
has 10 bits, forwarders ID has 60 bits and EOH contains 5 bits. The “Source ID” represents the ID
of the node sending the message to the mobile sink. The “Working–Sleeping Schedule” represents
the information about whole working time and whole sleeping time of the respective node. “Status
Transition Frequency” provides information about the transitions from working mode to sleeping
mode and vice versa of the sensor node. “Neighbor IDs” includes the ID list of neighboring nodes to
the source node. If there is no neighbor or all neighboring nodes are in sleeping mode, the “Neighbor
IDs” field remains empty. Moreover, the “Sink ID” is related to mobile sink’s SID and is used to indicate
that the probe message being forwarded from the sensor node is specifically targeted for the desired
sink node in case of multiple mobile sinks present in the network. “Forwarders ID” stores the IDs of
nodes which have already forwarded this probe message to the mobile sink. This field is updated

Sensors 2019, 19, 3789 9 of 32
when a new node forwards this probe message to the mobile sink and adds its ID in the forwarders ID
list. Likewise, “EOH” field stores the value of expected optimal hops needed for the source ID message
to reach the sink node. As this probe message can be generated by any sensor node within the radio
range Rl of the mobile sink, so it’s quite probable that different working–sleeping schedules and status
transition frequencies of sensor nodes can lead to inefficient probe message forwarding. Therefore,
only the nodes that have received the probe message can participate in this forwarding mechanism.
For example, for an intermediate node VS receiving the probe message from one of its neighbors Vr as
shown in Figure 3, the first step would be to check its own ID in the “Forwarders ID” field. If the probe
message was not forwarded by this node VS, it will add its ID to “Forwarders ID” field, determine
the total number of forwarders (TNF) for this probe message and then calculate (EOH-TNF). Now it
will check the “Neighbor IDs” field and see which neighbor’s EOH is closest to (EOH-TNF) as an
optimal forwarder.
Since the asynchronous working–sleeping cycle strategy is used in sensor nodes, if any sensor
node VS receives the probe message from its surrounding node Vr, it will forward the probe message
to one of its neighbors only if that neighbor node is in working mode. Likewise, it will not forward the
probe message if there is no neighboring node or the neighbor is in sleeping mode. The probe message
receiving and forwarding mechanism based on asynchronous working–sleeping cycle of nodes is
depicted in Figure 3. Algorithm 1 provides the details about probe message forwarding mechanism.
message receiving and forwarding mechanism based on asynchronous working–sleeping cycle of
nodes is depicted in Figure 3. Algorithm 1 provides the details about probe message forwarding
mechanism.
r
r
r
rr
(a)
(d)
(b)
(c)
Vr
VS
Vr
VS
Vt
Vr
VS VS
Figure 3. Probe message receiving and forwarding mechanism in EMOR. (a) Vs receives probe
message from Vr; (b) Vs forwards the received probe message to working neighbor Vt; (c) Vt has only
one neighbor Vs to forward the probe message; (d) Vs has no neighbor.
Algorithm 1: Initialization and Probe Message Forwarding Mechanism in EMOR
Input:
1. Mobile sink broadcasting SID and data collection duration to all nodes within Rl
2. Sensor nodes within Rl receive the data collection message
Output:
3. Select neighbor which is optimal forwarder
Begin:
4. Calculate the working–sleeping cycle and status transition frequencies
5. Estimate the distance between sensor node and mobile sink using RSSI
6. Calculate EOH based on distance estimation
sin( , )i kV V
d
7. Determine the optimal forwarders based on the following criteria:
8. When an intermediate working node Vs receives the mobile sink message
9. if (Vs has not forwarded the probe message to any of its neighbors), then
10. Add Vs source ID in the forwarders ID list and calculate total number of
Figure 3. Probe message receiving and forwarding mechanism in EMOR. (a) Vs receives probe message
from Vr; (b) Vs forwards the received probe message to working neighbor Vt; (c) Vt has only one
neighbor Vs to forward the probe message; (d) Vs has no neighbor.

Sensors 2019, 19, 3789 10 of 32
Algorithm 1: Initialization and Probe Message Forwarding Mechanism in EMOR
Input:
1. Mobile sink broadcasting SID and data collection duration to all nodes within Rl
2. Sensor nodes within Rl receive the data collection message
Output:
3. Select neighbor which is optimal forwarder
Begin:
4. Calculate the working–sleeping cycle and status transition frequencies
5. Estimate the distance between sensor node and mobile sink using RSSI
6. Calculate EOH based on distance estimation d(Vi,Vsin k)
7. Determine the optimal forwarders based on the following criteria:
8. When an intermediate working node Vs receives the mobile sink message
9. if (Vs has not forwarded the probe message to any of its neighbors), then
10. Add Vs source ID in the forwarders ID list and calculate total number of forwarders (TNF)
11. Compute EOHVS
− TNF
12. if (Vs has working neighbors), then
13. Check EOH of all of its working neighboring nodes
14. if (EOHNE ≈ EOHVS
− TNF), then
15. Select the neighbor as an optimal forwarder
16. end if
17. endif
18. else if (Vs has already forwarded probe message to neighbor node Vt but receives it again), then
19. Explore the working neighbors other than Vt that have not forwarded the message yet
20. Remove the recently added source IDs of Vs, Vt from forwarders ID list
21. Add Vs source ID in the forwarders ID list and calculate total number of forwarders (TNF)
22. Compute EOHVS
− TNF
23. if (Vs has working neighbors other than Vt), then
24. Check EOH of working neighboring nodes other than Vt
25. if (EOHNE ≈ EOHVS
− TNF), then
26. Select the neighbor as an optimal forwarder
27. end if
28. end if
29. endif
30. if (all Vs neighbors are sleeping), then
31. Stop forwarding the probe message
32. Wait for the next data collection by sink node.
33. end if
4.3. Construction of Opportunistic Connection Random Graph
The construction of OCRG is dependent on the asynchronous working–sleeping cycle of sensor
nodes and additional source information received within probe message forwarding during the
initialization phase. In order to construct an OCRG, we assumed that the mobile sink is always in
working mode and any sensor node within the short radio range RS of the mobile sink can communicate
with the mobile sink at any time. Furthermore, we proposed a data set D(S,Ns,W/S,FST,RE,BS,S/N)
consisting of source (S) and its neighboring nodes (NS) information, working–sleeping cycle schedule
(W/S), status transition frequencies (FST) for every sensor node, residual energy of every sensor node after
every round of communication (RE, buﬀer size (BS) and signal-to-noise ratio (S/N)) of every sensor node
with its adjacent nodes. Using this data set D(S,Ns,W/S,FST,RE,BS,S/N), we analyzed the opportunistic

Sensors 2019, 19, 3789 11 of 32
connection between any adjacent sensor nodes which are in working mode. Additionally, for EMOR,
the link connectivity between adjacent nodes is dependent on asynchronous working–sleeping cycle
W/S, status transition frequencies FST of the adjacent nodes, residual energies of the adjacent nodes RE,
remaining buffer size of the adjacent nodes to cache the sensory data and link quality factor between
adjacent nodes in terms of signal-to-noise ratio S/N. More status transitions of a node will lead to an
improvement in its link connectivity with the adjacent node as the probability of establishing a link
connection will increase. Keeping in view the data set D(S,Ns,W/S,FST,RE,BS,S/N), the time-frequency
parameter TFViVj of link connectivity LViVj can be calculated as:
TFViVj =
FSTVi
FSTmax
×
WVi
TCP
FSTVj
FSTmax
×
WVj
TCP
(4)
TFViVSINK
=
FSTVi
FSTmax
×
WVi
TCP
WVSINK
(5)
where WVi and WVj are the whole working time of the adjacent nodes Vi and Vj, TCP is the data
collection period during probe message forwarding mechanism, FSTVi and FSTVj are the status transition
frequencies of adjacent nodes Vi and Vj, TFViVj is the time-frequency parameter of link connectivity
LViVj, and FSTmax is the max status transition frequency value obtained during TCP which is used here
for normalization of the sensor node’s FST. The mobile sink is always in working mode, so WSINK = 1
and Equation (5) will be become,
TFViVSINK
=
FSTVi
FSTmax
×
WVi
TCP
(6)
Hence, the time-frequency parameter TFViVSINK
only depends upon the whole working time
and status transition frequency of the sensor nodes. With time-frequency parameter TF, we have to
calculate the residual energy, remaining buffer size and link quality factor between adjacent nodes.
In order to determine the next best hop, we assume that there are N nodes deployed in the NAN,
so our link connectivity function LViVj in terms of TF, RE, BS, S/N will be:
LViVj = max
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σS/NViVj} (7)
LVSINKVi
= max
Vi∈N
{αTFVSINKVi
+ βRE,Vi
+ γBS,Vi
+ σS/NVSINKVi
} (8)
where RE,Vj
, RE,Vi
are the contemporary residual energies of node Vj and Vi, BS,Vj
, BS,Vi
are the available
buffer size of nodes Vj and Vi, S/NViVj
, S/NVSINKVi
are the link quality factors in terms of signal-to-noise
ratio between Vi and Vj and VSINK and Vi, respectively. TFViVj is the time-frequency parameter of link
connectivity LViVj, α, β, γ, σ are the appropriate weights assigned to time-frequency parameter, residual
energy, buffer size and link quality factor, respectively. Also, we have considered the residual energy
and buffer size of node Vj only in Equation (7) because node Vj consumes energy and buffer capacity
for both data reception and transmission in accordance with a simplified energy model [18]. Moreover,
illustration of opportunistic connection random graph to understand the link and path connectivity
based on asynchronous working–sleeping cycle of adjacent nodes can be seen in Figure 4.
4.4. Optimal Paths Calculation and Spanning Tree Design
After the construction of OCRG, we need to find the optimal path in EMOR for our sensor
node’s data to reach the sink node successfully. In order to establish the optimal path, we need to
design a spanning tree algorithm which will help us in determining the maximum value of path
connectivity from any sensor node to mobile sink node. Using this algorithm, we will select the optimal
multi-disjoint paths for real-time and non-real-time traffic towards mobile sink (NAN gateway) in
Section 4.5.

Sensors 2019, 19, 3789 12 of 32
Path connectivity (PC) depends on the link connectivity of adjacent sensor nodes and is calculated
as the product of individual link connectivity values between sensor nodes along the path towards the
sink node. The reliability of the data delivery depends on the link connectivity between adjacent sensor
nodes and optimal path selection. PC for a path (Vi, Vi+t, t) according to Figure 4 can be formulated in
Equation (9) as:
PCViVj
= PCViVi+t =
t−1
k=0
LVi+k,Vi+(k+1)
(9)
Similarly, the PC for neighboring nodes Vi of mobile sink node VSINK will be:
PCVSINKVi
= LVSINK,Vi
(10)
Figure 4. Representation of opportunistic connection random graph (OCRG) based on the link
connectivity.
( 1)
1
,
0
  



  i j i k i k
t
VV ViVi t V V
k
PC PC L (9)
Similarly, the PC for neighboring nodes Vi of mobile sink node VSINK will be:
,SINK i SINK iV V V VPC L (10)
During the construction of the spanning tree, it is pertinent to mention that the nodes whose
optimal path are formed already, since they were the neighbors of mobile sink, can help the mobile
sink node in getting the optimal paths of intermediate and far-end nodes. Equations (8), (9) and (10)
help us in determining the PC of node Vj to mobile sink. The PC from mobile sink node to an unknown
sensor node Vj depends on 2 factors: (i) PC between mobile sink and its neighboring nodes and (ii)
link connectivity between mobile sink’s neighboring node Vi and unknown sensor node Vj.
( 1)
1
,
0
  


 SINK j SINK i k i k
t
V V V Vi V V
k
PC PC L (11)
1( , ,*) { , , 1} { , , , }   SINK j SINK j SINK i i t jPATH V V V V t V V V V (12)
where SINKV ViPC is the path connectivity from the mobile sink to neighboring node Vi. Here we
have assumed that Vj = Vi+t for which t is the number of hops on the path from Vi to Vj and it satisfies
the criteria1 1t N   . In every round of communication (iteration), the mobile sink acquires the
Figure 4. Representation of opportunistic connection random graph (OCRG) based on the
link connectivity.
During the construction of the spanning tree, it is pertinent to mention that the nodes whose
optimal path are formed already, since they were the neighbors of mobile sink, can help the mobile sink
node in getting the optimal paths of intermediate and far-end nodes. Equations (8), (9) and (10) help
us in determining the PC of node Vj to mobile sink. The PC from mobile sink node to an unknown
sensor node Vj depends on 2 factors: (i) PC between mobile sink and its neighboring nodes and (ii)
link connectivity between mobile sink’s neighboring node Vi and unknown sensor node Vj.
PCVSINKVj
= PCVSINKVi ×
t−1
k=0
LVi+k,Vi+(k+1)
(11)
PATH(VSINK, Vj, ∗) = VSINK, Vj, t + 1 = VSINK, Vi, . . . Vi+t−1, Vj (12)
where PCVSINKVi is the path connectivity from the mobile sink to neighboring node Vi. Here we have
assumed that Vj = Vi+t for which t is the number of hops on the path from Vi to Vj and it satisﬁes
the criteria 1 ≤ t ≤ N − 1. In every round of communication (iteration), the mobile sink acquires the
updated path connectivity values of intermediate and far-end sensor nodes deployed in the NAN,
compares it with the previous PC values, and selects the max PC value. The same process is repeated

Sensors 2019, 19, 3789 13 of 32
until mobile sink node receives the updated path and max PC information of all the nodes present in
the network.
PC
updated
VSINKVj
= max{PCVSINKVj
, PCVSINKVi
×
t−1
k=0
L
updated
Vi+k,Vi+(k+1)
} (13)
PATH(VSINK, Vj, ∗) = VSINK, Vj, t + 1updated = VSINK, Vi, . . . Vi+t−1(updated), Vj (14)
This spanning tree formation is useful in reducing the overall delay in the network as if the
mobile sink needed to explore all the possible paths before selecting the optimal path and determining
which one has maximum PC value would have resulted in more delay and a less eﬃcient spanning
tree design.
The formation of a spanning tree and optimal path calculation after construction of OCRG can
be seen in Figure 5 and Table 2. It can be depicted from Figure 5b that after construction of OCRG,
the mobile sink starts the initialization phase by connecting to its immediate neighbors. Figure 5c
shows the path connectivity between the mobile sink and neighbors of the neighbors of mobile sink
(i.e., node V4 and V5). The process of spanning tree formation continues from Figure 5d–i based on
the optimal path connectivity rule in Algorithm 2. According to the optimal path connectivity rule,
we have to choose the maximum values between the current path connectivity value and updated
path connectivity value. The optimal path connectivity values for Figure 5 are given in Table 2.
The initialization step of Table 2 is synchronized with Figure 5b where only immediate neighbors are
connected to the mobile sink. From steps 1–7, the optimal path connectivity rule is followed to reduce
the overall delay in the network.
26. while (PC from VSINK to entire network is not yet established)
27. Select the optimal path using following expression
28. , ( 1)
1
0
max{ , }SINK j SINK iV V V VSINK j i k i k
t
updated updated
V V V V
k
PC PC PC L
  


 
29. Update the path from VSINK to Vj
30. ( ) 1( )( , , 1) { , , 1 } { , , , }SINK j SINK j updated SINK i i t updated jPATH V V t V V t V V V V    
31. end while
Figure 5. Formation of spanning tree based on OCRG. (a) OCRG with VSINK and 10 sensor nodes; (b)
initialization and path connectivity of immediate neighbors of VSINK; (c) path connectivity of V4 and
V5; (d) path connectivity of V6, V7 and V8; (e) path connectivity of V9 and V10; (f) alternate paths for
immediate nodes of VSINK; (g) alternate paths for intermediate and far end nodes; (i) spanning tree
completed.
Table 2. Path connectivity from VSINK to each sensor node in the spanning tree.
Step [v1] v[2] v[3] v[4] v[5] v[6] v[7] v[8] v[9] v[10]
Initializat
ion
PCV1 PCV2 PCV3 0 0 0 0 0 0 0
1 PCV1 PCV2 PCV3
PCV1
*LV1V4
PCV3
*LV3V5
0 0 0 0 0
V2
V3
V1
VSINK
V2
V1
VSINK
V5
V3
V4
V2
V4
V6
V7
V8
V9
V10
VSINK
V3
V5
V1
V2
V1
VSINK
V4
V5
V3
V6
V7
V8
V2
V1 VSINKV4
V6
V7
V8
V3
V5
V10
V9
V2
V1 VSINKV4
V7
V3
V10
V9
V5
V8
V6
V2
V1 VSINKV4
V7
V3
V10
V9
V5
V8
V6
V2
V1 VSINKV4
V7
V3
V10
V9
V5
V8
V6
V2
V1 VSINKV4
V7
V3
V10
V9
V5
V8
V6
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
Figure 5. Formation of spanning tree based on OCRG. (a) OCRG with VSINK and 10 sensor nodes;
(b) initialization and path connectivity of immediate neighbors of VSINK; (c) path connectivity of V4
and V5; (d) path connectivity of V6, V7 and V8; (e) path connectivity of V9 and V10; (f) alternate paths
for immediate nodes of VSINK; (g) alternate paths for intermediate and far end nodes; (i) spanning
tree completed.

Sensors 2019, 19, 3789 14 of 32
Algorithm 2: Optimal Paths Calculation and Spanning Tree Design for Non-Distinguishable Service
Input:
1. Formation of data set D(S,Ns,W/S,FST,RE,BS,S/N)
2. Construction of opportunistic connection random graph (OCRG)
Output:
3. Optimal path connectivity and spanning tree formation
Begin:
4. Determine link connectivity LViVj in terms of TF, RE,BS, S/N of adjacent nodes
5. LViVj = max
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σS/NViVj}
6. Set the path connectivity of VSINK, i.e., PCVSINKVSINK
= 1;
7. Determine the path connectivity for neighboring nodes Vi of mobile sink node VSINK
8. PCVSINKVi
= LVSINK,Vi
9. for each awake node Vj in network, do //subroutine_wakeup
10. if (Vj is directly connected to neighboring node Vi of mobile sink node VSINK), then
11. PCVSINKVj
= LVSINK,Vi
× LVi,Vj
12. PATH(VSINK, Vj, ∗) = VSINK, Vj, 2 = VSINK, Vi, Vj
13. elseif (Vj is not directly connected to neighboring nodes Vi of mobile sink node VSINK), then
14. PCVSINKVj
= PCVSINKVi ×
t−1
k=0
LVi+k,Vi+(k+1)
whereVi+t = Vj
15. PATH(VSINK, Vj, ∗) = VSINK, Vj, t + 1 = VSINK, Vi, . . . Vi+t−1, Vj
16. else PCVSINKVj
= 0;
17. PATH(VSINK, Vj, ∗) = ∅;
18. end if
19. end for
20. for each sleeping node Vj in network, do
21. if (Vj wakes up); it should send the probe message to mobile sink with updated information
22. VSINK should reconstruct OCRG in the next iteration
23. subroutine_wakeup: go to 9 //execute steps 9–20
24. end if
25. end for
26. while (PC from VSINK to entire network is not yet established)
27. Select the optimal path using following expression
28. PC
updated
VSINKVj
= max{PCVSINKVj
, PCVSINKVi
×
t−1
k=0
L
updated
Vi+k,Vi+(k+1)
}
29. Update the path from VSINK to Vj
30. PATH(VSINK, Vj, t + 1) = VSINK, Vj, t + 1(updated) = VSINK, Vi, . . . Vi+t−1(updated), Vj
31. end while
After the formation of spanning tree, the mobile sink should broadcast it to the sensor nodes
within long radio range Rl in every round of communication. The sensor nodes send the sensing
data to mobile sink using this spanning tree information. If a sensor node does not ﬁnd itself on the
spanning tree, it can wait for the next broadcast message from the mobile sink during data collection
period and resend its probe message to mobile sink. Upon receiving this probe message, the mobile
sink will reconstruct OCRG, re-formulate the spanning tree by re-calculating the optimal path from
itself to that node.

Sensors 2019, 19, 3789 15 of 32
Table 2. Path connectivity from VSINK to each sensor node in the spanning tree.
Step [v1] [v2] [v3] [v4] [v5] [v6] [v7] [v8] [v9] [v10]
Initialization PCV1 PCV2 PCV3 0 0 0 0 0 0 0
1 PCV1 PCV2 PCV3 PCV1 *LV1V4 PCV3 *LV3V5 0 0 0 0 0
2
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3}
PCV1 *
LV1V4
PCV3 *LV3V5
PCV4 *
LV4V6
PCV2 *
LV2V7
PCV5 *
LV5V8
0 0
3
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3}
Max{
PCV1 *
LV1V4,
PCV2 *
LV2V4}
Max{
PCV3 *
LV3V5,
PCV2 *
LV2V5}
PCV4 *
LV4V6
PCV2 *
LV2V7
PCV5 *
LV5V8
PCV7 *
LV7V9
PCV8 *
LV8V10
4
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3}
Max{
PCV1 *
LV1V4,
PCV2 *
LV2V4}
Max{
PCV3 *
LV3V5,
PCV2 *
LV2V5}
Max{
PCV4 *
LV4V6,
PCV2 *
LV2V6}
PCV2 *
LV2V7
Max{
PCV5 *
LV5V8,
PCV2 *
LV2V8}
PCV7 *
LV7V9
PCV8 *
LV8V10
5
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3,
PCV1 *
LV1V3}
Max{
PCV1 *
LV1V4,
PCV2 *
LV2V4}
Max{
PCV3 *
LV3V5,
PCV2 *
LV2V5}
Max{
PCV4 *
LV4V6,
PCV2 *
LV2V6,
PCV1 *
LV1V6}
PCV2 *
LV2V7
Max{
PCV5 *
LV5V8,
PCV2 *
LV2V8}
Max{
PCV7 *
LV7V9,
PCV6 *
LV6V9}
PCV8 *
LV8V10
6
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3,
PCV1 *
LV1V3}
Max{
PCV1 *
LV1V4,
PCV2 *
LV2V4}
Max{
PCV3 *
LV3V5,
PCV2 *
LV2V5,
PCV7 *
LV7V5}
Max{
PCV4 *
LV4V6,
PCV2 *
LV2V6,
PCV1 *
LV1V6,
PCV7 *
LV7V6}
Max{
PCV2 *
LV2V7,
PCV5 *
LV5V7,
PCV6 *
LV6V7}
Max{
PCV5 *
LV5V8,
PCV2 *
LV2V8}
Max{
PCV7 *
LV7V9,
PCV6 *
LV6V9}
Max{
PCV8 *
LV8V10,
PCV9 *
LV9V10}
7
Max{
PCV1,
PCV2 *
LV2V1}
Max{
PCV2,
PCV1 *
LV1V2,
PCV3 *
LV3V2}
Max{
PCV3,
PCV2 *
LV2V3,
PCV1 *
LV1V3}
Max{
PCV1 *
LV1V4,
PCV2 *
LV2V4}
Max{
PCV3 *
LV3V5,
PCV2 *
LV2V5,
PCV7 *
LV7V5}
Max{
PCV4 *
LV4V6,
PCV2 *
LV2V6,
PCV1 *
LV1V6}
Max{
PCV2 *
LV2V7,
PCV5 *
LV5V7,
PCV6 *
LV6V7}
Max{
PCV5 *
LV5V8,
PCV2 *
LV2V8,
PCV7 *
LV7V8}
Max{
PCV7 *
LV7V9,
PCV6 *
LV6V9,
PCV10*LV10V9}
Max{
PCV8 *
LV8V10,
PCV9 *
LV9V10,
PCV7 *
LV7V10}
4.5. Optimal Multi-Disjoint Path Selection for Real-Time and Non-Real-Time Traffic
We know that the link connectivity function LViVj
depends on several factors such as time-frequency
parameter, residual energy of each neighboring sensor node, remaining buffer size and link quality
factor in terms of signal-to-noise ratio. Based on the link connectivity function, we select our next best
hop to create an optimal link between sensor node and one of its neighboring nodes, which further
results in the optimal path connectivity towards mobile sink node. But this approach just includes the
same optimal path for all kinds of data delivery, which will not suit delay-sensitive and delay-tolerant
traffic requirements in SGs. Keeping in view the SGs control and monitoring data requirements in
NAN, we need to split up our real-time and non-real-time data in such a way that instant priority and
an optimal multi-path connectivity should be provided to real-time traffic, whereas for non-real-time
traffic, secondary priority should be assigned, and an alternate multi-disjoint path connectivity should
be provided.
Therefore, we need to find the next most preferred neighboring node (second best hop) for an
alternative multi-disjoint path connectivity of non-real-time traffic. In this way, we can have two-path
connectivity:

Sensors 2019, 19, 3789 16 of 32
(a) Primary multi-path connectivity for real-time traffic based on first best hop decision criteria in
link connectivity function
(b) Alternate multi-path connectivity for non-real-time based on second best hop decision criteria in
link connectivity function



LRT
ViVj
= max
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σ(S/N)ViVj} . . . . . . . . . H1 : real − time
LNT
ViVj
= 2ndmax
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σ(S/N)ViVj} . . . H2 : non − real − time
(15)



PCRT
Vi,Vi+t
= LRT
Vi,Vi+1
×
t−2
k=1
LRT
Vi+kVi+(k+1)
× LRT
Vi+t−1Vi+t
. . . . . . H1 : real − time
PCNT
Vi,Vi+t
= LNT
Vi,Vi+1
×
t−2
k=1
LNT
Vi+kVi+(k+1)
× LNT
Vi+t−1Vi+t
. . . . . . H2 : non − real − time
(16)
where LRT
ViVj
is the link connectivity for real-time traffic, LNT
ViVj
is the link connectivity for non-real-time
traffic, PCRT
Vi,Vi+t
is the path connectivity for real-time traffic from node Vi to any other node Vi+t and
PCNT
Vi,Vi+t
is the path connectivity for non-real-time traffic from node Vi to any other node Vi+t in
Equations (15) and (16).
Foregoing in view, the constructed paths are node-disjoint paths which have no rendezvous point
except source and destination. Node-disjoint paths are also preferred because they utilize most available
network resources while avoiding the bottle necks by keeping energy balance. If an intermediate node
fails in a node-disjoint path, only the path containing that failed node will be affected, thus maintaining
the diversity of the routes intact with minimum impact. To avoid misuse of energy resources, we limit
each sensor node to involve either in first-best hop decision or second-best hop decision, so that no
sensor node is involved in constructing paths for both real-time and non-real-time traffic. Algorithm 3
provides details about the multi-disjoint path selection for real-time and non-real-time traffic.
After the construction of multi-disjoint paths, we need to divide our total paths for real-time and
non-real-time traffic (i.e., out of N available paths, let us assume that there are µ paths that correspond
to a probability of successfully delivering data to a destination) [46]. For real-time traffic, we need τ
paths and for non-real-time traffic, we need paths, where the total traffic µ = τ + ε. Assuming that
the traffic size of real-time and non-real-time data is known, we can easily calculate τ and ε. If RT
represents the real-time traffic size and NT represents non-real-time traffic size, we can have:
τ = (
RT
RT + NT
)µ (17)
ε = (
NT
RT + NT
)µ (18)
Moreover, the path connectivity from mobile sink VSINK to Vj for real-time and non-real-time
traffic can be expressed in Equations (19) and (20). The path information from mobile sink VSINK to Vj
for real-time and non-real-time traffic can be seen in Equations (21) and (22).
PCRT
VSINKVj
= PCRT
VSINKVi
×
t−2
k=0
LRT
Vi+kVi+(k+1)
× LRT
Vi+t−1Vi+t
(19)
PCNT
VSINKVj
= PCNT
VSINKVi
×
t−2
k=0
LNT
Vi+kVi+(k+1)
× LNT
Vi+t−1Vi+t
(20)
PATH(VSINK, Vj, ∗) = VSINK, Vj, t + 1 = VSINK, VRT
i , . . . VRT
i+t−1, Vj (21)
PATH(VSINK, Vj, ∗) = VSINK, Vj, t + 1 = VSINK, VNT
i , . . . VNT
i+t−1, Vj (22)

Sensors 2019, 19, 3789 17 of 32
where t is the hops on the path between node Vi to Vj and t + 1 hops on the path between VSINK to Vj.
Algorithm 3: Optimal Multi-Disjoint Paths Selection for Real-Time and Non-Real-Time Traffic
(Distinguishable Service)
Input:
1. Formation of data set D(S,Ns,W/S,FST,RE,BS,S/N)
Output:
2. Optimal multi-disjoint paths for delay-sensitive and delay-tolerant traffic
Begin:
3. for each awake node Vi and Vj in network, do
4. Determine LViVj in terms of TF, RE, BS, S/N for first and second best hop decision criteria
5. if (delay sensitive traffic is needed), then //real-time data
6. LRT
ViVj
= max
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σ(S/N)ViVj}
7. elseif (delay tolerant traffic is needed), then //non-real-time data
8. LNT
ViVj
= 2ndmax
Vj∈N
{αTFViVj + βRE,Vj + γBS,Vj + σ(S/N)ViVj}
9. else Link connectivity cannot be established due to status transition (i.e., node sleeping)
10. end if
11. Determine the path connectivity for both delay-sensitive and delay-tolerant traffic
12. if (delay-sensitive traffic is needed), then //real-time data
13. PCRT
VSINKVj
= PCRT
VSINKVi
×
t−2
k=0
LRT
Vi+kVi+(k+1)
× LRT
Vi+t−1Vi+t
14. elseif (delay tolerant traffic is needed), then //non-real-time data
15. PCNT
VSINKVj
= PCNT
VSINKVi
×
t−2
k=0
LNT
Vi+kVi+(k+1)
× LNT
Vi+t−1Vi+t
16. else link and path connectivity failed (i.e., path is no longer available)
17. end if
18. Formulate real-time and non-real-time paths from total paths µ
19. if (delay sensitive traffic is needed), then //real-time data
20. total real-time paths will be τ ∗ PATH(VSINK, Vj, ∗) = τ ∗ VSINK, VRT
i
, . . . VRT
i+t−1
, Vj
21. else total non-real-time paths will be ε ∗ PATH(VSINK, Vj, ∗) = ε ∗ VSINK, VNT
i
, . . . VNT
i+t−1
, Vj ;
22. end if
23. Select the optimal path using following expressions
24. max{PCRT
VSINKVj
, PCRT
VSINKVi
×
t−2
k=0
L
RT_updated
Vi+kVi+(k+1)
× L
RT_updated
Vi+t−1Vi+t
}
25. max{PCNT
VSINKVj
, PCNT
VSINKVi
×
t−2
k=0
L
NT_updated
Vi+kVi+(k+1)
× L
NT_updated
Vi+t−1Vi+t
}
26. end for
After the selection of optimal paths for real-time and non-real-time traffic, we need to define our
routing strategy for data transmission and reception. The traditional opportunistic node connection
routing only includes asynchronous working–sleeping cycle of sensor nodes which could lead to
failed link connection sometimes, due to the sleeping mode of any forwarder node. Also, it does not
support different type of data requirements in SGs, so in order to resolve this ambiguity, we need
energy-efficient multi-disjoint path supporting opportunistic connection routing protocol. Algorithm 4
provide details about our routing strategy for delay sensitive and delay tolerant traffic in EMOR.

Sensors 2019, 19, 3789 18 of 32
Algorithm 4: Routing Strategy for Delay Sensitive and Delay Tolerant Traffic in EMOR
Input:
1. OCRG, spanning tree formation, optimal path connectivity for distinguishable service
Output:
2. Energy efficient routing protocol based on optimal path connectivity for distinguishable service
Begin:
3. Assuming that NAN sensing data from node Vm is forwarded to VSINK using optimal path
4. When an intermediate node Vr receives the sensing data in NAN:
5. if (the successor node Vs of node Vr is in working mode), then
6. Forward the sensing data to Vs
7. if (Vs knows optimal path towards VSINK in its routing table), then
8. Vs receives the sensing data from Vr and forwards the data on optimal path;
9. elseif (Vs does not know optimal path towards VSINK in its routing table), then
10. if (Vs has neighbors on the spanning tree), then
11. for (each working neighbor Vi of Vs node), do
12. if (delay sensitive traffic is needed), then
13. Calculate LRT
VsVi
for real-time data
14. Determine PCRT
VsVi
for real-time data based on LRT
VsVi
15. elseif (delay tolerant traffic is needed), then
16. Calculate LNT
VsVi
for non-real-time data
17. Determine PCNT
VsVi
for non-real-time data based on LNT
VsVi
18. end if
19. end for
20. Select the optimal path using Algorithm 2 and 3
21. elseif (Vs has no neighbors on the spanning tree), then
22. Stop forwarding the data
23. end if
24. end if
25. elseif (the successor node Vs of node Vr is sleeping), then
26. Search for other neighbors or wait until Vr wakes up
27. end if
Based on the routing strategy algorithm for delay-sensitive and delay-tolerant traffic in EMOR,
the detail process of a source node forwarding its sensed data to the mobile sink can be depicted from
Figure 6. When a random node Vs receives the data from one of its neighboring nodes, it follows the
optimal multi-disjoint path selection for delay sensitive and delay tolerant traffic. Node V1 receives the
delay sensitive traffic as it has the maximum link connectivity with Vs and node V2 receives the delay
tolerant traffic as it has the second maximum link connectivity value with Vs. During the transmission
of delay-sensitive and delay-tolerant traffic in the network, we have to ensure that the nodes can be
part of the optimal path towards mobile sink if they offer higher link connectivity. Although the nodes
appearing as green in Figure 6 are working, they do not offer max or second max link connectivity,
so they are not part of the delay-sensitive or delay-tolerant traffic paths. Moreover, the nodes appearing
as pink in Figure 6 are sleeping nodes, so they could not be involved in the processes of OCRG,
spanning tree, and optimal paths towards mobile sink.
After the path connectivity for delay-sensitive and delay-tolerant traffic is established, we have to
update the path connectivity value by comparing the path connectivity value in previous rounds with
the path connectivity value in the current round of communication and select the maximum of the two

Sensors 2019, 19, 3789 19 of 32
values. In this way, we will be able to acquire the optimal path from any working sensor node to the
mobile sink.
the nodes appearing as pink in Figure 6 are sleeping nodes, so they could not be involved in the
processes of OCRG, spanning tree, and optimal paths towards mobile sink.
After the path connectivity for delay-sensitive and delay-tolerant traffic is established, we have
to update the path connectivity value by comparing the path connectivity value in previous rounds
with the path connectivity value in the current round of communication and select the maximum of
the two values. In this way, we will be able to acquire the optimal path from any working sensor
node to the mobile sink.
Figure 6. Multi-disjoint path selection and routing strategy for real-time and non-real-time traffic.
5. Performance Evaluation of EMOR
In this section, we describe the simulation environment and simulation results with reference to
performance metrics used for EMOR.
5.1. Simulation Environment
In order to validate the effectiveness of EMOR in NAN, we evaluate the performance of EMOR
in MATLAB 2018 simulator [56] and the parameters used in this simulation are defined in Table 3 as:
Table 3. Simulation parameters.
Parameters Values
Network Size 500 × 500 m2
Number of the Mobile Sink 1
Number of Sensors 200
Mobility Pattern Randomly
Duration for a Data Collection Period 600 s
Communication Range for Sensor Nodes 20 m
(Data + Overhead) Packet Size 1024 bits
Probe Message Size 120 bits
Transmit Power 15 mW
Receive Power 13 mW
Medium Access Control (MAC) Layer IEEE 802.11
Max Buffer size 512 K-bytes
Figure 6. Multi-disjoint path selection and routing strategy for real-time and non-real-time traffic.
5. Performance Evaluation of EMOR
In this section, we describe the simulation environment and simulation results with reference to
performance metrics used for EMOR.
5.1. Simulation Environment
In order to validate the effectiveness of EMOR in NAN, we evaluate the performance of EMOR in
MATLAB 2018 simulator [56] and the parameters used in this simulation are defined in Table 3 as:
Table 3. Simulation parameters.
Parameters Values
Network Size 500 × 500 m2
Number of the Mobile Sink 1
Number of Sensors 200
Mobility Pattern Randomly
Duration for a Data Collection Period 600 s
Communication Range for Sensor Nodes 20 m
(Data + Overhead) Packet Size 1024 bits
Probe Message Size 120 bits
Transmit Power 15 mW
Receive Power 13 mW
Medium Access Control (MAC) Layer IEEE 802.11
Max Buffer size 512 K-bytes
Target Reliability in Probabilistic and Opportunistic Flooding
Algorithm (POFA)
0.6
Initial Energy of Nodes 2.5 J
Buffer Threshold 1024 bits
Eelec 20 × 10−7 J/bit
εamp 10 × 10−9 J/bit/m2
Weights (α, β, γ and σ) respectively 0.4, 0.3, 0.1, 0.2

Sensors 2019, 19, 3789 20 of 32
We conducted a set of simulation experiments for EMOR protocol and performed a comparative
analysis of EMOR with several other protocols used in sensor networks. Our simulation environment
includes a network field of 500 m × 500 m in which 200 sensors were deployed randomly. We have
assumed that all sensor nodes are independent and identical with a communication range of 20 m.
We have fixed the duration of the data collection period in our experiments to check the impact of several
other parameters on EMOR’s performance and to compare that performance with various algorithms
used in sensor networks. The classical data transmission algorithms like Extremely Opportunistic
Routing (ExOR) [19] improves the reliability of data forwarding by exploiting the broadcasting nature
of wireless medium in such a way that allows multiple neighboring nodes to overhear the ongoing
transmission and participate in data forwarding, whereas probabilistic and opportunistic flooding
algorithm (POFA) [57] achieves the target reliability on each hop of data forwarding by adopting the
controlled transmissions but at the cost of higher energy consumption and increased number of hopes.
Venkatesha et al. in [58] presented reliable proliferation routing with low duty cycle (RPRDC) by
integration of concepts like randomized dispersity, forwarding, and reliable path finder to improve
the packet delivery ratio. Yang et al. in [22] presented a data collection model based on opportunistic
node connection (DCBONC) in wireless sensor networks. DCBONC calculates the optimal path from
the mobile sink to every sensor node in the graph by formulation of a spanning tree. DCBONC also
utilizes a routing protocol which adapts to different sensor nodes statutes to improve the reliability of
data transmission. DCBONC depicts improvement in performance in parameters like packet delivery
ratio (PDR), energy consumption, and network lifetime. Foregoing in view, most of the classical
algorithms used in sensor networks, consider unreliable wireless links due to lack of link quality
parameter. Most of the algorithms do not include multipath links for data forwarding except ExOR
and POFA. In addition to it, no sensor network algorithm developed so far has contributed towards
SGs NAN while focusing on real-time and non-real-time traffic requirements and efficiently managing
the energy resources.
Moreover, the traditional data forwarding algorithms and schemes do not incorporate OCRG in
calculating the optimal path towards the sensor nodes, so an algorithm is needed which utilizes the
concept of OCRG, highlights the link quality aspect, and also demonstrates adherence to accomplish
the real-time and non-real-time requirements of SGs NAN while efficiently managing the energy
resources. Therefore, we propose an energy efficient multipath opportunistic routing protocol for smart
grids NAN. The performance comparison of EMOR with that of ExOR, POFA, RPRDC, and DCBONC
is completed with reference to performance metrics like link connectivity, PDR, energy consumption,
network lifetime, and average end-end delay.
5.2. Results
5.2.1. Link Connectivity
As from the Section 4.3, “Construction of Opportunistic Connection Random Graph”, we know
that link connectivity between a mobile sink node and any sensor node or between sensor nodes depend
on several parameters like time-frequency parameter, residual energies of both nodes, remaining buffer
size of both nodes, and link quality factor. Two scenarios are designed to conduct simulations to
evaluate link connectivity. In the first scenario, the link connectivity is calculated between mobile sink
node and sensor node. In this scenario, we assume that mobile sink node is always in working status
but the sensor node, depending on its working–sleeping cycle, performs certain status transitions,
so the link connectivity if calculated between the mobile sink node and any sensor node, depends on
the time frequency (TF) parameter of that sensor node, residual energy of that sensor node, remaining
buffer size of that sensor node, and signal-to-noise ratio between that sensor node and mobile sink
node. In this scenario, we consider multiple sensor nodes with different values of TF, residual energy,
and link quality factor to evaluate their link connectivity against mobile sink node. Also, as we have
assumed that the mobile sink node always remains in the working state, so its energy is always

Sensors 2019, 19, 3789 21 of 32
100% and the total working time of sensor nodes is 75% of data collection duration TCP. Figure 7
depicts that the overall link connectivity decreases with the increase in max status transition frequency.
The main reason will be the reduction in residual energies of nodes and the decreased status transition
frequencies of individual nodes. For nodes with higher status transition frequency, the link connectivity
is higher. Foregoing in view, the higher status transition of a node will create increased opportunities
of sensing data sharing between that node and mobile sink node, thus improving the link connectivity
between them but at the same time, for higher status transitions of a node, there will be a corresponding
decrement in the residual energy as the power is being consumed every time (waking up and sleeping)
a node undergoes a status transition as shown in Figure 2.
sensor node, depends on the time frequency (TF) parameter of that sensor node, residual energy of
that sensor node, remaining buffer size of that sensor node, and signal-to-noise ratio between that
sensor node and mobile sink node. In this scenario, we consider multiple sensor nodes with different
values of TF, residual energy, and link quality factor to evaluate their link connectivity against mobile
sink node. Also, as we have assumed that the mobile sink node always remains in the working state,
so its energy is always 100% and the total working time of sensor nodes is 75% of data collection
duration TCP. Figure 7 depicts that the overall link connectivity decreases with the increase in max
status transition frequency. The main reason will be the reduction in residual energies of nodes and
the decreased status transition frequencies of individual nodes. For nodes with higher status
transition frequency, the link connectivity is higher. Foregoing in view, the higher status transition
of a node will create increased opportunities of sensing data sharing between that node and mobile
sink node, thus improving the link connectivity between them but at the same time, for higher status
transitions of a node, there will be a corresponding decrement in the residual energy as the power is
being consumed every time (waking up and sleeping) a node undergoes a status transition as shown
in Figure 2.
Figure 7. Link connectivity between different sensors and mobile sink (based on time frequency (TF)
parameter, residual energy, buffer size, and link quality factor) against maximum status transition
frequency of nodes in a spanning tree.
In the second scenario, the link connectivity is calculated between different adjacent sensor node
pairs. In this scenario, we assume that all the sensor nodes, depending on their working–sleeping
cycle, perform certain status transitions, so the link connectivity if calculated between adjacent sensor
node pairs, depending on the TF parameter of those sensor nodes, the residual energy of those sensor
nodes, the remaining buffer size of those sensor nodes, and signal-to-noise ratio between that node
pair. In this scenario, we consider multiple sensor nodes with different values of TF, residual energies,
and link quality factors to evaluate their link connectivity between them.
Figure 8 depicts that the overall link connectivity decreases with the increase in maximum status
transition frequency. The main reason will be the decreasing status transition frequencies of different
node pairs and cumulative reduction in residual energies used for simulation. We can see from Table
4 that when node pair status transition frequency decreases from 10 to 4, there is a corresponding
effect on the overall link connectivity between each node pair. The difference between the first
scenario and second scenario is the mobile sink. When we use a mobile sink and a sensor node for
data sharing, we have to consider only the effect of sensor node parameters on the link connectivity
whereas in the case of adjacent node pairs, we have to consider the effect of parameters related to the
node pair. Keeping in view the weights used in Table 3, we can analyze the decrease in overall
connectivity in Figure 8, as compared to link connectivity calculated in the first scenario (i.e., for fv1
= fv2 = 10, we can see a reduction of 25% in the impact of TF parameter for link connectivity in the
second scenario, for fv3 = fv4 = 8, we have 40% decrease in the impact of TF parameter for link
Figure 7. Link connectivity between different sensors and mobile sink (based on time frequency (TF)
parameter, residual energy, buffer size, and link quality factor) against maximum status transition
frequency of nodes in a spanning tree.
In the second scenario, the link connectivity is calculated between different adjacent sensor node
pairs. In this scenario, we assume that all the sensor nodes, depending on their working–sleeping
cycle, perform certain status transitions, so the link connectivity if calculated between adjacent sensor
node pairs, depending on the TF parameter of those sensor nodes, the residual energy of those sensor
nodes, the remaining buffer size of those sensor nodes, and signal-to-noise ratio between that node
pair. In this scenario, we consider multiple sensor nodes with different values of TF, residual energies,
and link quality factors to evaluate their link connectivity between them.
Figure 8 depicts that the overall link connectivity decreases with the increase in maximum status
transition frequency. The main reason will be the decreasing status transition frequencies of different
node pairs and cumulative reduction in residual energies used for simulation. We can see from Table 4
that when node pair status transition frequency decreases from 10 to 4, there is a corresponding effect
on the overall link connectivity between each node pair. The difference between the first scenario and
second scenario is the mobile sink. When we use a mobile sink and a sensor node for data sharing,
we have to consider only the effect of sensor node parameters on the link connectivity whereas in
the case of adjacent node pairs, we have to consider the effect of parameters related to the node pair.
Keeping in view the weights used in Table 3, we can analyze the decrease in overall connectivity in
Figure 8, as compared to link connectivity calculated in the first scenario (i.e., for fv1 = fv2 = 10, we can
see a reduction of 25% in the impact of TF parameter for link connectivity in the second scenario,
for fv3 = fv4 = 8, we have 40% decrease in the impact of TF parameter for link connectivity). In case of
fv5 = fv6 and fv7 = fv8, the effect of TF-parameter in the link connectivity calculation is reduced by
55% and 70%, respectively. Using Equations (4) and (6) and the assumption that total working time of

Sensors 2019, 19, 3789 22 of 32
sensor nodes is 75% of the data collection duration, the effect of the TF parameter on link connectivity
calculation can be determined by the following expression,
TFViVj =
FSTVi
FSTmax
×
(0.75 ∗ TCP)
TCP
FSTVj
FSTmax
×
(0.75 ∗ TCP)
TCP
(23)
TFViVSINK =
FSTVi
FSTmax
×
(0.75 ∗ TCP)
TCP
(24)
TFViVj
TFViVSINK
=
FSTVi
FSTmax
× 0.75 (25)
where the ratio of the TF parameter for adjacent node pairs to the TF parameter between mobile sink
and single sensor node in Equation (25), is used to study the impact or influence of the TF parameter on
the link connectivity calculation in the second scenario. In addition to it, the effect of residual energy
on the link connectivity calculation can be determined using Equations (26) and (27):
RE(1stscenario) = βREi; RE(2ndscenario) = β(REi ∗ REj) (26)
RE(2ndscenario)
RE(1stscenario)
= REj (27)
STmax CPSTmax CPF T   
(0.75* ) 
  
 
ViVSINK
STVi CP
STmax CP
TF
F T
F T
(24)
0.75
 
 
 
ViVj
ViVSINK
STVi
STmax
TF
TF
F
F
(25)
where the ratio of the TF parameter for adjacent node pairs to the TF parameter between mobile
sink and single sensor node in Equation (25), is used to study the impact or influence of the TF
parameter on the link connectivity calculation in the second scenario. In addition to it, the effect of
residual energy on the link connectivity calculation can be determined using Equations (26) and (27):
(1 ) (2 )
; ( * )st ndEi Ei EjE scenario E scenario
R R R RR    (26)
(2 )
(1 )
nd
st
Ej
E scenario
E scenario
R
R
R
 (27)
The residual energy ratio in Equation (27) can be used to study the impact of residual energy on
the link connectivity calculation in the second scenario. It is quite obvious that the only difference in
residual energy impact on link connectivity is residual energy of node j in the node pair.
For example, if the residual energy of a node is 0.94 in the first scenario and the residual energies
of the adjacent node pair are 0.94 and 0.74, subject to the condition that the node used in the first
scenario belongs to the node pair in the second scenario, then the percentage change in the impact of
RE on overall link connectivity will be 26% decreased according to Equation (27) and Table 5. It should
also be noted that during our simulation, the effect of buffer size for all nodes is not considered as the
percentage change in overall link connectivity because the buffer size was negligible.
Figure 8. Link connectivity between different adjacent sensors (based on the TF parameter, residual
energy, buffer size, and link quality factor) against maximum status transition frequency of nodes in
the spanning tree.
Table 4. Effect of TF parameter on link connectivity.
Figure 8. Link connectivity between different adjacent sensors (based on the TF parameter, residual
energy, buffer size, and link quality factor) against maximum status transition frequency of nodes in
the spanning tree.
Table 4. Effect of TF parameter on link connectivity.
Parameter
Status Transition
Frequency in First
Scenario
Status Transition
Frequency in
Second Scenario
Percentage
Change
(%)
Overall Link
Connectivity
Increase/Decrease
TF Parameter
Fv1 = 10 Fv1 = Fv2 = 10 25% Decrease
Fv3 = 8 Fv3 = Fv4 = 8 40% Decrease
The residual energy ratio in Equation (27) can be used to study the impact of residual energy on
the link connectivity calculation in the second scenario. It is quite obvious that the only difference in
residual energy impact on link connectivity is residual energy of node j in the node pair.

Sensors 2019, 19, 3789 23 of 32
For example, if the residual energy of a node is 0.94 in the first scenario and the residual energies
of the adjacent node pair are 0.94 and 0.74, subject to the condition that the node used in the first
scenario belongs to the node pair in the second scenario, then the percentage change in the impact of
RE on overall link connectivity will be 26% decreased according to Equation (27) and Table 5. It should
also be noted that during our simulation, the effect of buffer size for all nodes is not considered as the
percentage change in overall link connectivity because the buffer size was negligible.
Table 5. Effect of residual energy on link connectivity.
Parameter
Initial
Residual
Energies
in First
Scenario
Initial
Residual
Energies
in Second
Scenario
Change in
Initial
Residual
Energies
(%)
Final
Residual
Energies
in First
Scenario
Final
Residual
Energies
in Second
Scenario
Change in
Final
Residual
Energies
(%)
Overall
Link
Connectivity
Increase/
Decrease
Residual
Energy
RE1 = 0.99
RE1 = 0.99,
RE2 = 0.96
4% RE1 = 0.72
RE1 = 0.72,
RE2 = 0.65
35% Decrease
RE3 = 0.94
RE3 = 0.94,
RE4 = 0.99
1% RE3 = 0.60
RE3 = 0.60,
RE4 = 0.74
26% Decrease
RE5 = 0.84
RE5 = 0.84,
RE6 = 0.94
6% RE5 = 0.60
RE5 = 0.60,
RE6 = 0.78
22% Decrease
RE7 = 0.94
RE7 = 0.94,
RE8 = 0.74
26% RE7 = 0.78
RE7 = 0.78,
RE8 = 0.48
52% Decrease
5.2.2. Packet Delivery Ratio (PDR)
The metric PDR can be defined as the ratio of the packets received by the mobile sink node to the
packet sent by the sensor nodes. We have designed two scenarios to perform simulation for the evaluation
of PDR. In the first scenario, PDR is evaluated against long radio range Rl of the mobile sink node whereas
in the second scenario, the performance of the PDR was measured against network density. In order
to set up the simulation environment for PDR, we need to set a benchmark for the packet successfully
transmitted and packet dropped due to certain reasons. If the node finds one of its neighboring nodes
is in working mode or if the data packet is forwarded between adjacent sensor nodes for which both
of the sensor nodes are in working mode, we can consider it as the benchmark of a packet successfully
transmitted; whereas if the node cannot find the forwarder due to asynchronous working–sleeping cycle
strategy or if the data packet is forwarded between adjacent sensor nodes and one of the node experiences
status transition (i.e., working to sleeping mode, thus causing transmission failure), then we will consider
it as the benchmark for a packet drop. Figure 9 shows that PDR for POFA [57] increases with the increase
in long radio range Rl but for schemes like DCBONC [22], RPRDC [58], ExOR [19] and EMOR, the PDR
decreases with the increase in long radio range Rl. The reason for the increase in PDR for POFA can be
explained in terms of unchanged network density. When Rl increases, more nodes will be part of the data
collection process but as the total number of nodes remains the same, it results in the increase of hops for
all the sensor nodes in the data collection scope. This increase in number of hops will eventually decrease
the success rate of transmission and delivering the sensory data to mobile sink node correspondingly.
However, the sensory data will be forwarded to mobile sink node through multiple paths in POFA,
thus increasing the possible paths towards mobile sink node and improving the success rate of data
transmission and reception.
It is evident from Figure 9 that our proposed scheme EMOR incorporates real-time and
non-real-time disjoint paths and the performance of both is better than POFA (for long radio range
Rl up to 71%). For real-time EMOR, the performance is better than ExOR, RPRDC, and DCBONC
throughout the graph as Rl expands due to the reason that the proposed scheme always forwards the
sensing data along its optimal path which is based on maximum value of path connectivity. In order to
strengthen our basis for improvement in PDR, the link connectivity (in EMOR) depends on several
factors like time-frequency parameter, residual energies of sensor nodes, buffer capacity of sensor
nodes, and link quality factor between sensor nodes and mobile sink node. Furthermore, the proposed

Sensors 2019, 19, 3789 24 of 32
scheme also utilizes the routing protocol which is based on the concept of multi-disjoint path selection
for real-time and non-real-time data, in which real-time EMOR follows the criteria of the ﬁrst best hop
and non-real-time EMOR follows the criteria of the second best hop, thus improving the PDR.
factors like time-frequency parameter, residual energies of sensor nodes, buffer capacity of sensor
nodes, and link quality factor between sensor nodes and mobile sink node. Furthermore, the
proposed scheme also utilizes the routing protocol which is based on the concept of multi-disjoint
path selection for real-time and non-real-time data, in which real-time EMOR follows the criteria of
the first best hop and non-real-time EMOR follows the criteria of the second best hop, thus improving
the PDR.
Figure 9. Packet delivery ratio against long radio range of mobile sink where number of nodes (N) =
1000. DCBONC: Data Collection Algorithm based on Opportunistic Node Connection; EXOR:
Extremely Opportunistic Routing; POFA: Probabilistic and Opportunistic Flooding Algorithm;
RPRDC: Reliable Proliferation Routing with Low Duty Cycle; EMOR: Energy Efficient Multi-Disjoint
Path Opportunistic Node Connection Routing Protocol
The PDR against network density graph can be seen in Figure 10, in which PDR increases with
an increase in the number of nodes in the network for all of the schemes used in the simulation. The
number of nodes in the working mode will increase due to increased network density, therefore we
can see an increase in the overall PDR. The proposed scheme achieves 138%, 110%, 10.5%, and 36.39%
improvement in PDR over ExOR, RPRDC, DCBONC, and POFA (up to 925 nodes), respectively. The
improvement in PDR for POFA beyond 900 nodes is better than our proposed scheme due to the fact
that there are multiple forwarders in POFA to achieve the target reliability value. When the number
of connections increases, the possibility of achieving that target reliability value also increases, thus
improving the PDR. In RPRDC, the routing is segmented into random dispersion and reliability path
exploration with the consideration of variables such as residual energy, packet reception rate, and
link quality but it does not provide enough information between sink node and sensor nodes for
achieving better PDR. ExOR [19] works on the metric expected transmission times (ETX) in such a
way that the node with a minimum value of ETX will be selected for forwarding data towards the
sink node and does not include information related to path connectivity, thus leading to decreased
PDR.
Figure 9. Packet delivery ratio against long radio range of mobile sink where number of nodes (N)
= 1000. DCBONC: Data Collection Algorithm based on Opportunistic Node Connection; EXOR:
Extremely Opportunistic Routing; POFA: Probabilistic and Opportunistic Flooding Algorithm; RPRDC:
Reliable Proliferation Routing with Low Duty Cycle; EMOR: Energy Eﬃcient Multi-Disjoint Path
Opportunistic Node Connection Routing Protocol.
The PDR against network density graph can be seen in Figure 10, in which PDR increases with
an increase in the number of nodes in the network for all of the schemes used in the simulation.
The number of nodes in the working mode will increase due to increased network density, therefore we
can see an increase in the overall PDR. The proposed scheme achieves 138%, 110%, 10.5%, and 36.39%
improvement in PDR over ExOR, RPRDC, DCBONC, and POFA (up to 925 nodes), respectively.
The improvement in PDR for POFA beyond 900 nodes is better than our proposed scheme due to
the fact that there are multiple forwarders in POFA to achieve the target reliability value. When the
number of connections increases, the possibility of achieving that target reliability value also increases,
thus improving the PDR. In RPRDC, the routing is segmented into random dispersion and reliability
path exploration with the consideration of variables such as residual energy, packet reception rate,
and link quality but it does not provide enough information between sink node and sensor nodes for
achieving better PDR. ExOR [19] works on the metric expected transmission times (ETX) in such a way
that the node with a minimum value of ETX will be selected for forwarding data towards the sink
node and does not include information related to path connectivity, thus leading to decreased PDR.
Figure 10. Packet delivery ratio (PDR) against network density with long radio range of mobile sink
Rl as 100 m.
5.2.3. Energy Consumption
In order to evaluate the performance of energy consumption, we have designed two scenarios
using our simulation parameters. In the first scenario, we consider that the source (sensor) node is
forwarding its sensing data to the mobile sink, so we evaluate the energy consumption after
Figure 10. Packet delivery ratio (PDR) against network density with long radio range of mobile sink Rl
as 100 m.

Sensors 2019, 19, 3789 25 of 32
5.2.3. Energy Consumption
In order to evaluate the performance of energy consumption, we have designed two scenarios using
our simulation parameters. In the ﬁrst scenario, we consider that the source (sensor) node is forwarding
its sensing data to the mobile sink, so we evaluate the energy consumption after contemplating this
consideration. In the second scenario, we evaluate the total energy consumption against the simulation
time. Figure 11 depicts that energy consumed during the sensing data forwarding by a sensor node to
mobile sink changes by network density. We have assumed that the maximum distance between the
sensor node and the mobile sink can be 3Rl/4 just for the sake of this experiment.
Figure 11. Energy consumption against network density, with long radio range of mobile sink Rl as
100 m.
Figure 12 shows the increase in number of hops when the network density is increased. It is
evident that the increase in number of hops in the POFA scheme are higher than in any other scheme.
The number of hops for non-real-time EMOR are more than that of real-time EMOR because of the
fact that non-real-time path selection is based on the second best hop decision in which the second-
to-highest priority is given to the non-real-time path. The probability of a node having a certain status
transition is higher in the non-real-time path than in the real-time path (depending on the TF
parameter). Moreover, the immediate nodes present near the mobile sink are already loaded as the
traffic convergence nodes, so in order to avoid them, we have to select alternative paths (resulting in
increased hops), which means that as the network density increases, the number of hops for non-real-
time path selection are increased more than the real-time path.
Figure 12. The variation in number of hops against network density, with long radio range of mobile
sink Rl as 100 m.
In the second scenario, we conducted the total energy consumption against the network density
simulation. It is quite obvious from Figure 13 that total energy consumed increases as the simulation
Figure 11. Energy consumption against network density, with long radio range of mobile sink Rl as
100 m.
We can observe from the Figure 11 that energy consumption increases when the number of
nodes increases in the network or when network density increases. The energy consumption by the
POFA scheme is more obvious than in any other scheme including our proposed scheme. The energy
consumption in POFA can be explained in a way that when network density increases, the number of
hops used for data delivery between the sensor node and mobile sink increases too, so the compilation
of energy consumed by every hop results in the overall increase of energy consumption. Furthermore,
the energy consumed by the proposed scheme is slightly higher than DCBONC and ExOR schemes.
The reason is when a node sleeps in the neighborhood of a source node due to asynchronous
working–sleeping cycle strategy, which was used as a data forwarding node towards mobile sink in the
past, several processes like optimal path calculation, optimal path selection for real-time/non-real-time
data and multipath opportunistic routing protocol are repeated for neighbors which are awake, thus
resulting in a slight increase in energy consumption but still far better than POFA scheme. The energy
consumption for non-real-time EMOR is slightly higher than real-time EMOR as we select the second
best hop for our path connectivity in non-real-time EMOR in which it becomes more probable that the
PDR for the second best hop path will be slightly less, and energy consumption will be slightly more
than that of real-time EMOR.
Figure 12 shows the increase in number of hops when the network density is increased. It is
evident that the increase in number of hops in the POFA scheme are higher than in any other scheme.
The number of hops for non-real-time EMOR are more than that of real-time EMOR because of the fact
that non-real-time path selection is based on the second best hop decision in which the second-to-highest
priority is given to the non-real-time path. The probability of a node having a certain status transition is
higher in the non-real-time path than in the real-time path (depending on the TF parameter). Moreover,
the immediate nodes present near the mobile sink are already loaded as the traﬃc convergence nodes,
so in order to avoid them, we have to select alternative paths (resulting in increased hops), which

Sensors 2019, 19, 3789 26 of 32
means that as the network density increases, the number of hops for non-real-time path selection are
increased more than the real-time path.
The number of hops for non-real-time EMOR are more than that of real-time EMOR because of the
fact that non-real-time path selection is based on the second best hop decision in which the second-
to-highest priority is given to the non-real-time path. The probability of a node having a certain status
transition is higher in the non-real-time path than in the real-time path (depending on the TF
parameter). Moreover, the immediate nodes present near the mobile sink are already loaded as the
traffic convergence nodes, so in order to avoid them, we have to select alternative paths (resulting in
increased hops), which means that as the network density increases, the number of hops for non-real-
time path selection are increased more than the real-time path.
sink Rl as 100 m.
time increases. Here, we have considered only the total energy consumed during the data collection
phase by the mobile sink and energy consumed during the sensing data delivery by source node,
regardless of the condition that the sensing data is successfully delivered to the mobile sink or not.
In comparison to proposed scheme, ExOR consumes 84% more energy, POFA consumes 307% more
energy and RPRDC consumes 107% more energy, whereas in the case of DCBONC, the energy
consumption of the proposed scheme EMOR is slightly higher (i.e., 18% more). The main reason is
sink Rl as 100 m.
time increases. Here, we have considered only the total energy consumed during the data collection
phase by the mobile sink and energy consumed during the sensing data delivery by source node,
regardless of the condition that the sensing data is successfully delivered to the mobile sink or
not. In comparison to proposed scheme, ExOR consumes 84% more energy, POFA consumes 307%
more energy and RPRDC consumes 107% more energy, whereas in the case of DCBONC, the energy
consumption of the proposed scheme EMOR is slightly higher (i.e., 18% more). The main reason is
that in DCBONC, the probe message forwarding mechanism in the data collection period consumes
less energy due to the incorporation of EOH [22]. Furthermore, the spanning tree is constructed and
broadcasted by mobile sink so that the source node could obtain the max path connectivity information
beforehand, thus saving a lot of energy consumed by selecting the forwarders. However, the additional
energy consumed during the selection of optimal paths for real-time and non-real-time traffic is the
difference in total energy of EMOR and DCBONC.
that in DCBONC, the probe message forwarding mechanism in the data collection period consumes
less energy due to the incorporation of EOH [22]. Furthermore, the spanning tree is constructed and
broadcasted by mobile sink so that the source node could obtain the max path connectivity
information beforehand, thus saving a lot of energy consumed by selecting the forwarders. However,
the additional energy consumed during the selection of optimal paths for real-time and non-real-time
traffic is the difference in total energy of EMOR and DCBONC.
Figure 13. Total energy consumption against network simulation time, with Rl as 100 m and N =
1000.
5.2.4. Average End-End Delay
Average end-end delay is an important metric in evaluating the performance of energy efficient
QoS-based routing protocols. Here, we have conducted the simulation in terms of average packet
delay. Figure 14 depicts the average packet delay against packet arrival rate for several different
schemes. In this experiment, we can observe the change in delay with respect to change in packet
arrival rate. We measured the average packet delay for both real-time and non-real-time traffic in our
Figure 13. Total energy consumption against network simulation time, with Rl as 100 m and N = 1000.
5.2.4. Average End-End Delay
Average end-end delay is an important metric in evaluating the performance of energy efficient
QoS-based routing protocols. Here, we have conducted the simulation in terms of average packet

Sensors 2019, 19, 3789 27 of 32
delay. Figure 14 depicts the average packet delay against packet arrival rate for several different
schemes. In this experiment, we can observe the change in delay with respect to change in packet
arrival rate. We measured the average packet delay for both real-time and non-real-time traffic in
our proposed scheme to validate whether EMOR can successfully differentiate between them or not.
From the graph, it is quite clear that high priority is given to real-time traffic due to the reason that
real-time traffic is given absolute preferential treatment over low priority non-real-time traffic in EMOR
design, thus resulting in low end-end delay. Real-time EMOR performs much better than non-real-time
EMOR in comparison to other schemes because of the reason that we use the queueing model for SGs
NAN environment. The average packet delay in POFA is much less as it incorporates the multipath
forwarding strategy depending on its target reliability. Additionally, DCBONC performance is slightly
better than real-time EMOR due to several reasons (i.e., decision variables for calculating the link
connectivity and path connectivity in DCBONC are very few). Furthermore, there is no differentiation
between the traffic types in DCBONC, thus resulting in lower average packet delay. Non-real-time
EMOR performs better than RPRDC up to 48 packets received per second and then achieves a higher
delay due to the queueing model being employed in our design.
Figure 14. Average end-end delay against packets received.
5.2.5. Network Lifetime
Here we conducted the simulation with the condition that our mobile sink can launch multiple
data collection tasks anywhere within the network. The performance of network lifetime can be
evaluated in terms of number of dead nodes for which according to [59] Tunca et al. have highlighted
that network lifetime for WSNs can be defined in terms of the time when 25% of the nodes present in
the network have no residual energy left to continue their data delivery tasks. Figure 15 demonstrates
the network lifetime performance of different schemes against their simulation time. It can be seen
that the number of dead nodes increases as the simulation time increases. In this simulation, we have
assumed that the data collection period for mobile sink is 600 seconds and our mobile sink can start
the data collection task randomly within the network.
In POFA, the node dies faster than any other scheme due to the reason that POFA focuses more
on the reliability of data transmission at the expense of energy consumption, so it reaches the 25%
dead node mark faster than any other scheme. It is evident from the graph that the performance of
EMOR is better than all schemes including DCBONC. The main reason is the calculation of optimal
paths for nodes in advance and employment of flexible routing protocol. The slight difference
between DCBONC and real-time EMOR can be explained by the fact that decision variables for
calculation of optimal paths in EMOR are more effective than that of DCBONC. In DCBONC, only
TF-parameter is used in calculation of optimal paths whereas in EMOR, we use the TF parameter,
residual energy, buffer size, and link quality factor to achieve better network lifetime. Moreover, the
energy consumed per node is less in EMOR in comparison to other schemes due to the fact the EMOR
can change the data forwarding paths adaptively, especially in the low connectivity areas of the
network.
Figure 14. Average end-end delay against packets received.
5.2.5. Network Lifetime
Here we conducted the simulation with the condition that our mobile sink can launch multiple
data collection tasks anywhere within the network. The performance of network lifetime can be
evaluated in terms of number of dead nodes for which according to [59] Tunca et al. have highlighted
that network lifetime for WSNs can be defined in terms of the time when 25% of the nodes present in
the network have no residual energy left to continue their data delivery tasks. Figure 15 demonstrates
the network lifetime performance of different schemes against their simulation time. It can be seen
that the number of dead nodes increases as the simulation time increases. In this simulation, we have
assumed that the data collection period for mobile sink is 600 seconds and our mobile sink can start
the data collection task randomly within the network.
In POFA, the node dies faster than any other scheme due to the reason that POFA focuses more
on the reliability of data transmission at the expense of energy consumption, so it reaches the 25%
dead node mark faster than any other scheme. It is evident from the graph that the performance of
EMOR is better than all schemes including DCBONC. The main reason is the calculation of optimal
paths for nodes in advance and employment of flexible routing protocol. The slight difference between
DCBONC and real-time EMOR can be explained by the fact that decision variables for calculation of
optimal paths in EMOR are more effective than that of DCBONC. In DCBONC, only TF-parameter
is used in calculation of optimal paths whereas in EMOR, we use the TF parameter, residual energy,
buffer size, and link quality factor to achieve better network lifetime. Moreover, the energy consumed

Sensors 2019, 19, 3789 28 of 32
per node is less in EMOR in comparison to other schemes due to the fact the EMOR can change the
data forwarding paths adaptively, especially in the low connectivity areas of the network.
Figure 15. Network life time (number of dead nodes against simulation time) with Rl = 90 m and N =
850.
6. Discussion
We have introduced the energy efficient multipath opportunistic routing protocol for SGs NAN
environment. The previous opportunistic routing schemes were designed without consideration of
service differentiation and multiple decision variables for calculation of optimal path connectivity. In
EMOR, we utilized the concepts of multiple decision variables in optimal path connectivity and
service differentiation in optimal path selection for real-time and non-real-time traffic. The proposed
design of EMOR could be applicable to the following different approaches in future:
1) Incorporation of relay nodes and finding the optimum number of relay nodes to support
network lifetime.
2) Design an energy efficient multipath opportunistic routing protocol for wireless cognitive
sensor networks.
3) Keeping in view the variable design of sensors and network architectures, heterogeneous
clustering could be used with EMOR for wide-range connectivity.
4) How to increase the network lifetime of WSNs having sensor nodes which follow
asynchronous working–sleeping cycle strategy by achieving a tradeoff between EMOR and
energy balancing routing protocols.
5) How to reduce the network end-end delay of WSNs having sensor nodes which follow
asynchronous working–sleeping cycle strategy by achieving a tradeoff between POFA and
real-time EMOR.
6) Unmanned Aerial Vehicle (UAV) sensor networks, Internet-of-Things (IOT) enabled home
energy systems, cyber-physical systems, etc.
7) Energy harvesting wireless sensor networks [60].
7. Conclusion
In this paper, we proposed an energy-efficient multipath opportunistic routing protocol for
wireless sensor networks, which can be used in neighborhood area network of smart grids. In this
proposed scheme, the mobile sink launches the data collection task anywhere and anytime in the
network by broadcasting its SID in the tag message. Upon receiving the tag message, the sensor nodes
start formulating their working–sleeping schedule and then forward a probe message to the mobile
sink as a response to the tag message. The probe message contains the source information, working–
sleeping schedule, and status transition frequency of that sensor node. When the mobile sink receives
this probe message, it constructs an opportunistic connection random graph and calculates the
optimal path from itself to each sensor node. The optimal path formation is based on the dataset of
link connectivity and path connectivity, which includes time-frequency parameter, residual energy,
Figure 15. Network life time (number of dead nodes against simulation time) with Rl = 90 m and
N = 850.
6. Discussion
We have introduced the energy efficient multipath opportunistic routing protocol for SGs NAN
environment. The previous opportunistic routing schemes were designed without consideration of
service differentiation and multiple decision variables for calculation of optimal path connectivity.
In EMOR, we utilized the concepts of multiple decision variables in optimal path connectivity and
service differentiation in optimal path selection for real-time and non-real-time traffic. The proposed
design of EMOR could be applicable to the following different approaches in future:
(1) Incorporation of relay nodes and finding the optimum number of relay nodes to support
network lifetime.
(2) Design an energy efficient multipath opportunistic routing protocol for wireless cognitive
sensor networks.
(3) Keeping in view the variable design of sensors and network architectures, heterogeneous
clustering could be used with EMOR for wide-range connectivity.
(4) How to increase the network lifetime of WSNs having sensor nodes which follow asynchronous
working–sleeping cycle strategy by achieving a tradeoff between EMOR and energy balancing
routing protocols.
(5) How to reduce the network end-end delay of WSNs having sensor nodes which follow
asynchronous working–sleeping cycle strategy by achieving a tradeoff between POFA and
real-time EMOR.
(6) Unmanned Aerial Vehicle (UAV) sensor networks, Internet-of-Things (IOT) enabled home energy
systems, cyber-physical systems, etc.
(7) Energy harvesting wireless sensor networks [60].
7. Conclusions
In this paper, we proposed an energy-efficient multipath opportunistic routing protocol for
wireless sensor networks, which can be used in neighborhood area network of smart grids. In this
proposed scheme, the mobile sink launches the data collection task anywhere and anytime in the
network by broadcasting its SID in the tag message. Upon receiving the tag message, the sensor
nodes start formulating their working–sleeping schedule and then forward a probe message to the
mobile sink as a response to the tag message. The probe message contains the source information,

Sensors 2019, 19, 3789 29 of 32
working–sleeping schedule, and status transition frequency of that sensor node. When the mobile sink
receives this probe message, it constructs an opportunistic connection random graph and calculates
the optimal path from itself to each sensor node. The optimal path formation is based on the dataset of
link connectivity and path connectivity, which includes time-frequency parameter, residual energy,
buffer capacity, and link quality factor. After calculation of the optimal path, the first spanning tree is
generated in which mobile sink is considered as the root node. Keeping in view the nature of NAN
traffic in smart grids, we performed multi-path selection for real-time and non-real-time traffic based
on the first and second best possible decisions for link connectivity and path connectivity. The second
spanning tree after the selection of real-time and non-real-time path was generated and optimal paths
for real-time and non-real-time traffic were calculated. Consequently, we designed the routing protocol
based on the optimal paths for real-time and non-real-time traffic of sensor nodes in the spanning tree
while focusing on different working–sleeping cycle strategies of sensor nodes.
The possible future work could be incorporation of networks like cloud computing and fog
computing with EMOR in which the high computation needs of the mobile sink in the sensor network
can be fulfilled using fog computing nodes bridged with cloud. Additionally, the novel concept
could be applied to cognitive radio sensor networks to deal with problems like predictive channel
assignments and opportunistic spectrum access. Moreover, supervised machine learning techniques
can also be used with our proposed scheme to investigate the performance improvements in OCRG and
spanning tree formation, predictive changes in the real-time and non-real-time traffic trends (including
volumes, faults, power surges, etc.).
Author Contributions: Conceptualization, J.A.; methodology, J.A.; formal analysis, J.A. and S.B.; investigation,
J.A. and S.B.; resources, H.-C.Z.; supervision, H.-C.Z.; project administration, S.B. and H.-C.Z.; software, J.A.;
validation, H.-C.Z. and B.G.L; writing—review and editing, J.A., H.-C.Z. and B.G.L.
Funding: The research was funded by the National Natural Science Foundation of China, grant number 51776050.
Acknowledgments: The authors would like to thank all the reviewers for their insightful comments and deliberate
suggestions for improving the quality of this paper.
Conflicts of Interest: The authors declare no conflicts of interest.
References
1. Gungor, V.C.; Hancke, G.P. Industrial Wireless Sensor Networks: Challenges, Design Principles, and Technical
Approaches. IEEE Trans. Ind. Electron. 2009, 56, 4258–4265. [CrossRef]
2. Movassaghi, S.; Abolhasan, M.; Lipman, J.; Smith, D.; Jamalipour, A. Wireless Body Area Networks: A
Survey. IEEE Commun. Surv. Tutor. 2014, 16, 1658–1686. [CrossRef]
3. Xu, N.; Rangwala, S.; Chintalapudi, K.K.; Ganesan, D.; Broad, A.; Govindan, R.; Estrin, D. A Wireless Sensor
Network for Structural Monitoring. In Proceedings of the 2nd International Conference on Embedded
Networked Sensor Systems (SenSys), Baltimore, MD, USA, 3–5 November 2014; pp. 13–24. [CrossRef]
4. Bokareva, T.; Hu, W.; Kanhere, S.; Ristic, B.; Gordon, N.; Bessell, T.; Rutten, M.; Jha, S. Wireless Sensor
Networks for Battlefield Surveillance. In Proceedings of the Land Warfare Conference (LWC), Brisbane,
Australia, 24–27 October 2006.
5. Yawut, C.; Kilaso, S. A Wireless Sensor Network for Weather and Disaster Alarm Systems. In Proceedings of
the International Conference on Information and Electronics Engineering, Bangkok, Thailand, 28–29 May
2011; pp. 155–159.
6. Hefeeda, M.; Bagheri, M. Wireless Sensor Networks for Early Detection of Forest Fires. In Proceedings of the
IEEE International Conference on Mobile Adhoc and Sensor Systems MASS-2007, Pisa, Italy, 8–11 October
2007; pp. 1–6.
7. Manshahia, M.S. Wireless Sensor Networks: A Survey. IJSER 2016, 74, 710–716.
8. Luo, J.; Panchard, J.; Piórkowski, M.; Grossglauser, M.; Hubaux, J.P. MobiRoute: Routing Towards a
Mobile Sink for Improving Lifetime in Sensor Networks. In Proceedings of the International Conference on
Distributed Computing in Sensor Systems, San Francisco, CO, USA, 18–20 June 2006. [CrossRef]

Sensors 2019, 19, 3789 30 of 32
9. Xujing, L.; Wei, L.; Mande, X.; Anfeng, L.; Ming, Z.; Neal, N.X.; Miao, Z.; Wan, D. Differentiated Data
Aggregation Routing Scheme for Energy Conserving and Delay Sensitive Wireless Sensor Networks. Sensors
2018, 18, 2349.
10. Available online: https://www.energy.gov/sites/prod/files/oeprod/DocumentsandMedia/TransmissionGrid.pdf
(accessed on 18 September 2018).
11. Available online: https://www.energy.gov/oe/downloads/smart-grid-introduction-0 (accessed on 20
March 2019).
12. Ma, R.; Chen, H.-H.; Huang, Y.-R.; Meng, W. Smart Grid Communication: Its Challenges and Opportunities.
IEEE Trans. Smart Grid 2013, 4, 36–46. [CrossRef]
13. Fadel, E.; Gungor, V.C.; Nassef, L.; Akkari, N.; Malik, M.G.A.; Almasri, S.; Akyildiz, I.F. A Survey on Wireless
Sensor Networks for Smart Grid. Comput. Commun. 2015, 71, 22–33. [CrossRef]
14. Faheem, Y.; Boudjit, S.; Chen, K. Data Dissemination Strategies in Mobile Sink Wireless Sensor Networks: A
Survey. In Proceedings of the 2nd IFIP Conference on Wireless Days, Paris, France, 15–17 December 2009.
[CrossRef]
15. Alfayez, F.; Hammoudeh, M.; Abuarqoub, A. A Survey on MAC Protocols for Duty-Cycled Wireless Sensor
Networks. Procedia Comput. Sci. 2015, 73, 482–489. [CrossRef]
16. Ghadimi, E.; Landsiedel, O.; Soldati, P.; Duquennoy, S.; Johansson, M. Opportunistic Routing in Low Duty
Cycle Wireless Sensor Networks. ACM Trans. Sens. Netw. 2014, 10, 1–39. [CrossRef]
17. Lai, Y.; Lin, Z. Data Gathering in Opportunistic Wireless Sensor Networks. Int. J. Distrib. Sens. Netw. 2012,
2012, 1319–1322. [CrossRef]
18. Luo, J.; Hu, J.; Wu, D.; Li, R. Opportunistic Routing Algorithm for Relay Node Selection in Wireless Sensor
Networks. IEEE Trans. Ind. Inform. 2015, 11, 112–121. [CrossRef]
19. Morris, R.; Morris, R. ExOR: Opportunistic multi-hop routing for wireless networks. In Proceedings of the
Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications,
Philadelphia, PA, USA, 22–26 August 2005.
20. Boukerche, A.; Darehshoorzadeh, A. Opportunistic Routing in Wireless Networks: Models, Algorithms, and
Classifications. ACM Comput. Surv. 2014, 47, 1–36. [CrossRef]
21. Priya, D.; Haripriya; Kulothungan. An Energy Efficient Link Stability Based Routing Scheme for Wireless
Sensor Networks. In Proceedings of the International Conference on Communication and Signal Processing
(ICCSP), Chennai, India, 6–8 April 2017. [CrossRef]
22. Yang, G.; Peng, Z.; He, X. Data Collection Based on Opportunistic Node Connections in Wireless Sensor
Networks. Sensors 2018, 18, 3697. [CrossRef] [PubMed]
23. Akkaya, K.; Younis, M. A Survey on Routing Protocols for Wireless Sensor Networks. J. Ad Hoc Netw. 2005,
3, 325–349. [CrossRef]
24. Al-Karaki, J.N.; Kamal, A.E. Routing Techniques in Wireless Sensor Networks: A Survey. IEEE J.
Wirel. Commun. 2004, 11, 6–28. [CrossRef]
25. Akkaya, K.; Younis, M. An energy aware QoS routing protocol for wireless sensor networks. In Proceedings
of the 23rd International Conference on Distributed Computing Systems Workshops, Providence, RI, USA,
19–22 May 2003; pp. 710–715.
26. Quynh, T.N.; Vinh, T.T.; Quynh, M.B.T. Multipath routing for cluster-based and event-based protocols in
wireless sensor networks. In Proceedings of the Third ACM Symposium on Information and Communication
Technology SoICT, Halong, Vietnam, 23–24 August 2012; pp. 172–179. [CrossRef]
27. Ren, Y.; Qin, Y.; Wang, B. A random geometric graph coverage model of wireless sensor networks. In
Proceedings of the IET International Conference on Wireless, Mobile and Multimedia Networks, Hangzhou,
China, 6–9 November 2006.
28. National Institute of Standards and Technology, Standards Identified for Inclusion in the Smart Grid
Interoperability Framework 2009, 1.0. Available online: https://www.nist.gov/sites/default/files/documents/
smartgrid/FinalSGDoc2010019-corr010411-2.pdf (accessed on 20 March 2019).
29. Fang, X.; Misra, S.; Xue, G.; Yang, D. Smart Grid New and Improved Power Grid: A Survey. IEEE Commun.
Surv. Tutor. 2012, 14, 944–980. [CrossRef]
30. Gao, J.; Xiao, Y.; Liu, J.; Liang, W.; Chen, C.L.P. A survey of communication/networking in smart grids. Future
Gener. Comput. Syst. 2012, 28, 391–404. [CrossRef]

Sensors 2019, 19, 3789 31 of 32
31. NETL, A systems view of the modern grid. Available online: http://www.netl.doe.gov (accessed on 17
March 2019).
32. NETL, Environmental Impacts of Smart Grids. Available online: https://www.netl.doe.gov/projects/files/
FY11_EnvironmentalImpactsofSmartGrid_010111.pdf (accessed on 17 March 2019).
33. Erol-Kantarci, M.; Mouftah, H.T. Energy-Efficient Information and Communication Infrastructures in the
Smart Grid: A Survey on Interactions and Open Issues. IEEE Commun. Surv. Tutor. 2015, 17, 179–197.
[CrossRef]
34. Gungor, V.C.; Lu, B.; Hancke, G.P. Opportunities and Challenges of Wireless Sensor Networks in Smart Grid.
IEEE Trans. Ind. Electron. 2010, 57, 3557–3564. [CrossRef]
35. Ogundile, O.; Alfa, A. A Survey on an Energy-Efficient and Energy-Balanced Routing Protocol for Wireless
Sensor Networks. Sensors 2017, 17, 1084. [CrossRef]
36. Han, G.; Yang, X.; Liu, L.; Zhang, W. A Joint Energy Replenishment and Data Collection Algorithm in
Wireless Rechargeable Sensor Networks. IEEE Internet Things J. 2017, 99, 1. [CrossRef]
37. Yang, G.; Liu, S.; He, X.; Xiong, N.; Wu, C. Adjustable Trajectory Design Based on Node Density for Mobile
Sinking WSNs. Sensors 2016, 16, 2091. [CrossRef] [PubMed]
38. Kaswan, A.; Singh, V.; Jana, P.K. A novel multi-objective particle swarm optimization based energy efficient
path design for mobile sink in wireless sensor networks. Pervasive Mob. Comput. 2018, 46, 122–136. [CrossRef]
39. Yang, G.; Wang, L.; Jiang, L.; He, X. Biased trajectory dissemination of uncontrolled mobile sinks for event
collection in wireless sensor networks. Int. J. Distrib. Sens. Netw. 2016, 12, 1–14. [CrossRef]
40. Alghamdi, W.; Rezvani, M.; Wu, H.; Kanhere, S.S. Routing-Aware and Malicious Node Detection in a
Concealed Data Aggregation for WSNs. ACM Trans. Sens. Netw. 2019, 15, 1–20. [CrossRef]
41. Guntupalli, L.; Martinez-Bauset, J.; Li, F.Y. Performance of Frame Transmissions and Event-Triggered Sleeping
in Duty-Cycled WSNs with Error-Prone Wireless Links. Comput. Netw. 2018, 134, 215–227. [CrossRef]
42. Montieri, A.; Montieri, A.; Persico, V.; Pescap, A. A sleep scheduling approach based on learning automata
for WSN partial coverage. J. Netw. Comput. Appl. 2017, 80, 67–78.
43. Ng, K.P.; Tsimenidis, C.; Woo, W.L. C-Sync: Counter-based synchronization for duty-cycled wireless sensor
networks. Ad Hoc Netw. 2017, 61, 51–64. [CrossRef]
44. Mukherjee, M.; Shu, L.; Hu, L.; Hancke, G.P.; Zhu, C. Sleep Scheduling in Industrial Wireless Sensor Networks
for Toxic Gas Monitoring. IEEE Wirel. Commun. 2017, 99, 2–8. [CrossRef]
45. Mostafaei, H. Energy-Efficient Algorithm for Reliable Routing of Wireless Sensor Networks. IEEE Trans.
Ind. Electron. 2019, 66, 5567–5575. [CrossRef]
46. Ben Fradj, H.; Anane, R.; Bouallegue, R. Opportunistic Routing Protocols in Wireless Sensor Networks.
Wirel. Pers. Commun. 2018. [CrossRef]
47. Sadatpour, V.; Zargari, F.; Ghanbari, M. A Collision Aware Opportunistic Routing Protocol for VANETs in
Highways. Wirel. Pers. Commun. 2019. [CrossRef]
48. Kenniche, H.; Ravelomananana, V. Random Geometric Graphs as model of Wireless Sensor Networks. In
Proceedings of the International Conference on Computer and Automation Engineering (ICCAE), Singapore,
26–28 February 2010.
49. Norman, J.; Joseph, P. A New Graph Model for Heterogeneous WSN. Commun. Comput. Inf. Sci. 2011, 197,
187–196.
50. Ren, Y.; Zhang, Z. A Weighted Network Topology Model of WSN Based on Random Geometric Graph
Method. Adv. Mater. Res. 2014, 962–965, 2898–2902. [CrossRef]
51. Fu, L.; Wang, X.; Kumar, P.R. Are We Connected? Optimal Determination of Source-Destination Connectivity
in Random Networks. IEEE/ACM Trans. Netw. 2017, 25, 751–764. [CrossRef]
52. Dulman, S.; Nieberg, T.; Wu, J.; Havinga, P. Trade-off between traffic overhead and reliability in multipath
routing for wireless sensor networks. In Proceedings of the IEEE Wireless Communications and Networking
Conference WCNC, New Orleans, LA, USA, 16–20 March 2003; pp. 1918–1922. [CrossRef]
53. Ben-Othman, J.; Yahya, B. Energy efficient and QoS based routing protocol for wireless sensor networks.
J. Parallel Distrib. Comput. 2010, 70, 849–857. [CrossRef]
54. Mostafaei, H.; Obaidat, M.S. Learning automaton-based self-protection algorithm for wireless sensor
networks. IET Netw. 2018, 7, 353–361. [CrossRef]
55. Liang, H.; Yang, S.; Li, L.; Gao, J. Research on routing optimization of WSNs based on improved LEACH
protocol. EURASIP J. Wirel. Commun. Netw. 2019, 194. [CrossRef]

Sensors 2019, 19, 3789 32 of 32
56. Matlab. R2014b Simulator; MathWorks: Natick, MA, USA, 2014.
57. Chang, D.; Cho, K.; Choi, N.; Kwon, T.T.; Choi, Y. A probabilistic and opportunistic ﬂooding algorithm in
wireless sensor networks. Comput. Commun. 2012, 35, 500–506. [CrossRef]
58. Venkatesha; Ashwini, T.N.; Akshay, A.L.; Tejaswi, V.; Venugopal, K.R.; Iyengar, S.S.; Patnaik, L.M. RPRDC:
Reliable Proliferation Routing with low Duty-cycle in Wireless Sensor Networks. Procedia Comput. Sci. 2015,
54, 37–46. [CrossRef]
59. Tunca, C.; Isik, S.; Donmez, M.Y.; Ersoy, C. Ring Routing: An Energy-Eﬃcient Routing Protocol for Wireless
Sensor Networks with a Mobile Sink. IEEE Trans. Mob. Comput. 2015, 14, 1947–1960. [CrossRef]
60. Adu-Manu, K.S.; Adam, N.; Tapparello, C.; Ayatollahi, H.; Heinzelman, W. Energy-Harvesting Wireless
Sensor Networks (EH-WSNs). ACM Trans. Sens. Netw. 2018, 14, 1–50. [CrossRef]
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Sensors 19-03789-v2

More Related Content

What's hot (18)

Similar to Sensors 19-03789-v2 (20)

Recently uploaded (20)

Sensors 19-03789-v2