Computer Communications 188 (2022) 66–80

Contents lists available at ScienceDirect

Computer Communications
journal homepage: www.elsevier.com/locate/comcom

An improved routing protocol for raw data collection in multihop wireless

sensor networks✩
Yangbin Zhang, Lihua Liu ∗, Mao Wang, Jibing Wu, Hongbin Huang
Science and Technology on Information Systems Engineering Laboratory, National University of Defense Technology, Changsha 410073, China

ARTICLE INFO ABSTRACT

Keywords: Wireless sensor networks (WSNs) are an effective and efficient method for collecting data from a target area,
Wireless sensor networks and prolonging the lifetime of WSNs has been a focus of scientific research due to the limited energy of sensor
Tree-based nodes. However, traditional studies of WSNs are based on the assumption that WSNs collect aggregated data
Cluster-based
with redundant sensor nodes in an ideal radio environment. These assumptions may not be acceptable in
Routing protocol
practical applications. To address this problem, this paper introduces a novel application scenario for WSNs
in which raw data are collected by a multihop network without redundant sensor nodes, and a new hybrid
tree-based and cluster-based routing protocol for raw data collection (HTC-RDC) is proposed to prolong the
lifetime of WSNs that can work in novel application scenarios. The experimental results demonstrate that the
proposed HTC-RDC enables the WSNs to achieve their expected functions and prolongs the network lifetime
by an average of 11.4% compared to the existing protocols in the explored application scenario.

1. Introduction and environmental monitoring [11]. The main limitation of WSNs is

their lifetime, which is constrained by the limited energy of sensor
Due to the tremendous benefits achieved in practical applications, nodes [12]. A WSN with sensor nodes of a small size normally facilitates
Internet of Things (IoT) technologies [1] have become a very popular the network deployment process, but the battery power at each node
research topic in recent years. The operation of the IoT is based on may be limited. Therefore, methods for prolonging the lifetime of WSNs
the perception of the environment, and the focus is the acquisition of have been extensively researched.
data. With the development of technologies such as computer vision, From the perspective of routing protocols, there are two main kinds
high-quantity and high-quality data acquisition has become essential. A of methods used to prolong the lifetime of WSNs. Clustering is a
wireless sensor network (WSN) is an important tool for data acquisition, classic method in which all the nodes are first divided into clusters
and such tools can effectively support the application of the IoT [2].
according to different criteria to prolong the lifetime of a WSN under
A WSN is composed of one base station and a number of sensor
the condition that all sensor nodes can directly communicate with the
nodes. A sensor node is a small unit with sensing and wireless trans-
base station. Then, one node in each cluster is selected as the cluster
mission functionalities. Each sensor node senses and gathers proximal
head to collect the data from the cluster, aggregate [13,14] the data,
environmental information, processes the information, and then for-
and send the aggregated data to the base station. Data aggregation can
wards the data to the base station through a wireless channel. A WSN
composed of all sensor nodes can monitor a specified area, and the greatly prolong the lifetime of a network by reducing the amount of
data that the base station collects can be provided to users for further data transmitted over a wireless channel. However, it is inevitable that
use [3]. The typical structure and function of a WSN are illustrated in this mechanism leads to information loss due to data aggregation by the
Fig. 1. cluster head. For example, the cluster head might use the maximum,
The prominent advantage of a WSN is convenience. The sizes of the minimum, or average value to represent the data collected by all nodes
nodes are small, so they are easy to deploy. A WSN can be deployed in in the cluster in the aggregation stage. Tree-based routing protocols
a target area without any communication infrastructure. In addition, can be used for data collection with multihop methods and prolong
the inexpensive cost of the sensor nodes makes it possible to use the lifetime of a network by employing a hop-spot [15] avoidance
WSNs in business applications. Due to the advantages of WSNs, they mechanism under the condition that some sensor nodes cannot directly
have been widely used in various fields [4,5], including industry [6], communicate with the base station. For example, mountains may pre-
agriculture [7], ocean surveillance [8], medicine and healthcare [9,10], vent direct communication between nodes and the base station, or

✩ This work is supported by NSF of Hunan under Grant 2019JJ50726.

∗ Corresponding author.
E-mail addresses: [email protected] (Y. Zhang), [email protected] (L. Liu).

https://doi.org/10.1016/j.comcom.2022.02.016
Received 19 October 2021; Received in revised form 17 January 2022; Accepted 18 February 2022
Available online 2 March 2022
0140-3664/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Fig. 1. The structure and function of a WSN.

Fig. 2. Diagram of the proposed scenario.

the distance between nodes and the base station may be beyond the
(4) The experimental results indicate that HTC-RDC outperforms the
corresponding communication radius in a vast target area. However,
compared methods.
this kind of method is commonly used in dynamic networks and might
This paper is organized as follows.
lack effective means to prolong the lifetime of a static network.
Section 2 briefly surveys the related works on methods for prolong-
The above traditional methods assume that there are redundant
ing the lifetime of networks. Section 3 describes the network model.
sensor nodes and that the data can be aggregated in the collection
Section 4 summarizes and analyzes the principles used to prolong the
process. However, in practical applications WSNs are expected to be
lifetime of networks. Section 5 presents the HTC-RDC protocol. In
deployed without redundancy to control cost and facilitate the process
Section 6, the effectiveness of the HTC-RDC protocol is verified based
of deployment for users, especially when sensor nodes are expensive,
on experiments, and the proposed approach is compared with other
or the cost of deploying nodes is high. For example, when deploying
techniques. Finally, Section 7 concludes the paper.
sensor nodes on the mountain for early warning of landslides, the
monitoring data is expected to be collected by deploying a minimum
number of nodes. That means each sensor node is crucial. In a situation 2. Related works
without redundant sensor nodes, the whole network works only if all
sensors are working. Therefore, the lifetime of the network is defined as Since the energy consumed by data transmission in the wireless
the lifetime of the node that first stops working. In addition, aggregated channel accounts for the highest proportion of the energy consumption
data reduce the energy consumption of the network; however, raw of the entire network, routing protocols are focused on to prolong
data that provide valuable information could yield important benefits the network lifetime. Most routing protocols in the literature can be
for real-world applications. For example, computer vision technologies divided into two categories: cluster-based routing protocols and tree-
require image data as inputs, and if data aggregation is applied in image based routing protocols. Both of these two categories of protocols use
collection with a WSN, the result of data aggregation will be textual or a variety of different strategies for different application scenarios to
numerical data but not image data. prolong the network lifetime.
Based on the above considerations, a new application scenario is
illustrated in Fig. 2, and its specifications are proposed as follows. (1) 2.1. Cluster-based routing protocols
The raw data from a target area must be collected without aggregation.
(2) Each sensor node is crucial in a WSN. (3) Some sensor nodes cannot Cluster-based routing protocols are typically used in static WSNs.
directly communicate with the base station. In this application scenario, With cluster-based routing protocols, sensor nodes are divided into
a new protocol, named the hybrid tree-based and cluster-based routing different clusters, and there is one cluster head in each cluster. The
protocol for raw data collection (HTC-RDC), is proposed to prolong cluster-head node collects the data from all the nodes in the cluster,
the lifetime of WSNs. In the HTC-RDC protocol, a multihop network aggregates the data [16] and sends the aggregated data to the base
is built with a tree structure to establish communication among all station or a cluster head at a higher level in the node hierarchy. The
the sensor nodes and the base station. In addition, a cluster structure main purpose of clustering is to prolong the lifetime of the network,
is incorporated into the tree structure to enhance the network energy and several methods can be employed to achieve this goal. The main
efficiency. The proposed hybrid structure is further reinforced to pro- difference among these methods is related to the selection criteria for
long the lifetime of the network based on the mechanisms exploited cluster-head nodes.
in tree-based or cluster-based protocols, such as load balancing and The low-energy adaptive clustering hierarchy (LEACH) protocol
hot-spot avoidance. The hybrid structure is established to collect raw [17] was the first dynamic cluster-based routing protocol. In the LEACH
data according to the above principle and combinatorial optimization. protocol, each node has a probability 𝑝 of being selected as a cluster-
Experiments are conducted under the proposed scenario. The results head for certain rounds in one period. The probability 𝑝 is also a
indicate that HTC-RDC outperforms the compared methods. reflection of the number of cluster head nodes in the population of
The contributions of this paper are summarized as follows: all nodes. If a node is a cluster head, in the rest of the period, it will
(1) A new scenario without redundant sensor nodes is defined to not be a cluster head again. After certain rounds, some other nodes are
collect raw data in practical situations. selected as new cluster heads. The other nodes join the cluster for which
(2) A mathematical model based on optimization theory is formu- the cluster head is the nearest. Other protocols that are improvements
lated for the proposed scenario. of LEACH, such as the centralized LEACH (LEACH-C) protocol [18], the
(3) A new HTC-RDC protocol is proposed to prolong the lifetime of two-level hierarchy LEACH (TL-LEACH) protocol [19], the threshold-
networks in the proposed scenario. based LEACH (T-LEACH) protocol [20], the density of sensor LEACH

67
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

(DS-LEACH) protocol [21], the LEACH energy-based protocol (LEACH- station. In tree-based routing protocols, the sensor nodes and the base
EP) [22] and the distance-based threshold LEACH (LEACH-DT) pro- station form a tree structure topology with the base station as the root.
tocol [23], were discussed in [24] and deemed classical approaches. A node transmits data to the base station through its parent nodes.
The advantage of these classical approaches is that they are easy to Studies of tree-based networks have also focused on how to prolong
implement because they are distributed methods with low computa- the lifetime of networks.
tional complexity and overhead, but they have some limitations in load If the coordinates of the nodes and the base station are avail-
balancing and coping with personalized node distributions. able based on positioning system (GPS) or other localization methods,
Fuzzy logic is introduced to improve the classical approach to load geographic information can be utilized to help prolong the lifetime
balancing. The main method involves using fuzzy logic to structure of a network. The geographic adaptive fidelity (GAF) protocol [38]
unequal clusters so that hot spots can be avoided as much as possible. builds virtual grids based on geographical information. All nodes in
For example, in the multiobjective fuzzy clustering algorithm (MOFCA) a particular grid square are equivalent with respect to forwarding
protocol [25], residual energy levels, the distance to the base station, packets, and only one equivalent node is active at a time. The node
and the density of parameters are considered fuzzy input parameters shifts among a sleeping state, a discovery state and an active state to
in the calculation of the cluster-head competition radius. Fuzzy logic prolong the lifetime of the network. The dynamic directional routing
schemes with different parameters and rules are employed by the (DDR) protocol [39] uses geographic information to limit the direction
LEACH protocol using fuzzy logic (LEACH-FL) [26], the energy-aware and distance of transmission when a node chooses its parent node. This
fuzzy unequal clustering (EAUCF) protocol [27], the energy conserved method reduces the energy consumption of unnecessary processes to
unequal clustering with fuzzy logic (ECUCF) protocol [28] and other prolong the lifetime of the network. The advantage of this approach is
approaches. that it is easy and requires little energy to discover neighbor nodes and
The optimization of cluster-head selection becomes difficult as the determine the parent nodes, but the disadvantage is that the solution
scale of the network increases. Metaheuristic algorithms have been in- may be the local optimal solution; therefore, the hot-spot problem
troduced to help solve this problem due to their excellent performance is difficult to avoid. This problem can be fatal for a node with no
in solving large-scale problems. Theant colony optimization (ACO) continuous energy source.
algorithm is used in the multipath routing (MPR) protocol [29] for the Another method employed by tree-based routing protocols is load
selection of cluster-heads, a genetic algorithm (GA) is applied in the balancing with a graph-based approach. In a tree structure, it is in-
GP-LEACH protocol [30], the harmony search (HS) algorithm is applied evitable that the nodes closer to the root bear higher loads than nodes
in the HS-LEACH protocol [30], the application-specific low power further from the root and die earlier; therefore, load balancing is the
routing (ASLPR) protocol [31] combines a GA and a simulated anneal- focus of most tree-based routing protocols. The RSTR protocol [40]
ing (SA) algorithm to optimize certain parameters used to determine
introduced the concept of virtual potential energy by considering resid-
the threshold for application-based cluster heads, and the butterfly
ual energy and the local densities of nodes and used virtual potential
optimization algorithm (BOA) and ACO are combined in [32] to re-
energy in the establishment of a minimum spanning tree to balance the
duce overall energy consumption. Some similar approaches, such as
energy consumption of the whole network and prolong the lifetime of
the heuristic algorithm for clustering hierarchy (HACH) protocol [33]
the network. Mao et al. [41] proposed a directed acyclic graph (DAG)
and the GA-based threshold sensitive energy-efficient routing proto-
based method. In this method, a weak node is defined, and a criterion
col (GATERP) [34], use metaheuristic algorithms to obtain optimal
function is designed to evaluate node data transmission to find and
solutions. The advantages of routing protocols based on metaheuristic
protect weak nodes in the network, resulting in improved balance in
algorithms are that they can not only handle large-scale problems but
energy consumption among nodes; consequently, the network lifetime
also provide global optimal solutions; however, the disadvantages are
can be prolonged. The methods used in tree-based routing protocols
high computational complexity and long computing times.
to prolong the lifetime of a network mainly focus on load balancing.
Approaches based on fuzzy logic and metaheuristic algorithms have
However, other mechanisms are rarely employed due to the topology
been combined by some researchers to take advantage of the benefits
of the tree structure, especially because nodes closer to the base station
of both schemes. Some representative methods include the fuzzy and
must play the role of repeaters; therefore, these nodes consume more
ant colony optimization-based combined MAC, routing, and unequal
energy than the other nodes.
clustering cross-layer (FAMACROW) protocol [35], the systems inter-
For the application scenario proposed in this paper, the cluster-
operability framework (SIF) protocol [36], and the LEACH based on
based method and the tree-based method have valuable merits. The
the Sugeno fuzzy inference system (LEACH-SF) protocol [37], but they
tree-based approach can effectively solve the problem that some nodes
are characterized by high implementation complexity.
cannot communicate directly with a base station, which is caused by
Cluster-based approaches have been improved by researchers, and a
number of effective protocols suitable for various application scenarios long transmission distances or obstacles. The cluster-based approach
have been developed. However, these methods are not fit for the pro- provides processes to prolong the lifetime of a network. However, as
posed application scenario in this paper. First, cluster-based approaches shown in Table 1, the existing protocols cannot be directly used in the
are usually used in one-hop networks in which the nodes can commu- application scenario proposed in this paper.
nicate directly with the base station, but in the application scenario Therefore, a hybrid approach that combines a cluster-based method
proposed in this paper, some nodes cannot directly communicate with and a tree-based method is proposed for the new application scenario.
the base station. Second, one of the most important ways to prolong
the lifetime of a network is to reduce the amount of data transmitted 3. Network model
by performing data aggregation; however, data aggregation cannot
be applied in the application scenario proposed in this paper since 3.1. Necessary assumptions
the aggregation process will inevitably lead to information loss, and
without data aggregation, traditional cluster-based routing protocols For the scenario researched in this paper, the network model is
will lose the ability to prolong the network lifetime. based on the following assumptions.

2.2. Tree-based routing protocols (a) The function of the network is to monitor a specific area and
periodically collect raw data from a target area with nodes.
Tree-based routing protocols are mainly used in dynamic networks, (b) The whole network is static. None of the nodes nor the base
and in these schemes, not all sensor nodes can directly contact the base station move after deployment.

68
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Table 1
Comparison with existing protocols for the application scenario proposed in this paper.
Classification Subclassification Protocol Operating with Obstacles allowed Oriented to prolong Effective in raw
redundant nodes in the target area the network lifetime data collection
ClusterL-based Classical approaches LEACH yes no yes no
routing protocols LEACH-C
TL-LEACH
T-LEACH
DS-LEACH
LEACH-EP
LEACH-DT
Fuzzy-based MOFCA yes no yes no
approaches LEACH-FL
EAUCF
ECUCF
Metaheuristic-based MPR yes no yes no
approaches GP-LEACH
HS-LEACH
ASLPR
HACH
GATERP
Hybrid fuzzy-based FAMACROW yes no yes no
and SIF
metaheuristic-based LEACH-SF
approaches
Tree-based – GAF yes no yes yes
routing protocols – DDR – no no yes
– RSTR yes yes no yes
Proposed – HTC-RDC no yes yes yes

(c) All the nodes are homogeneous. Each node has sensing, trans- • FNL (first node loss): The time (or round in an experiment) when
mission and forwarding functionalities that are enabled when the base station cannot receive messages from all of the nodes.
needed according to the routing protocol. For users, the nodes FNL must be caused by the death of the first node, but there may
are identical, and users do not need to determine whether the be more than one node that loses contact with the base station at
nodes are relay nodes. FNL.
(d) The locations where the nodes are deployed are random. Al- • HNL (loss of half of the nodes): The time (or round in an experi-
though the nodes are deployed intentionally, the result of the ment) when the base station can no longer receive messages from
deployment can be considered random. more than half of the nodes.
(e) The nodes are deployed with a relatively high spatial density to
According to the above assumptions, every single node in a WSN
ensure that each node has at least one path to reach the base
in the proposed scenario in this paper is crucial. The death of the
station.
first node is the shortest board of the barrel effect, so FNL is the key
(f) The nodes can only select the same node as the data-receiving
performance indicator of the WSN, and this parameter indicates the
node at a time and cannot distribute the data in their cache to
time when the WSN stops working. HNL is retained as a reference for
different receiving nodes.
evaluating network performance.
(g) Node batteries have limited energy supplies and are not
rechargeable, but the base station has a permanent energy sup-
3.3. The mathematical model of the energy consumption of the network
ply. The energy value of the batteries is defined as the residual
energy that can be accurately measured by the nodes. The energy consumption model of the radio module of a node is as
(h) The data that the nodes collect are raw data that can be com- follows. When a transmitter transmits 𝑙-bit data to a receiver located
pressed before transmission but cannot be aggregated. at distance 𝑑 from the transmitter, the energy consumption of the
(i) The radio wave propagation channel is symmetric, and the dis- transmitter (𝑐𝑡 ) is:
tances between the nodes and the base station can be computed
through the received signal strength indicator (RSSI) value. 𝑐𝑡 (𝑙, 𝑑) = 𝑙 × 𝐸𝑒𝑙𝑒𝑐 + 𝑙 × 𝐸𝑓 𝑠 × 𝑑 2 (1)
(j) The transmission power of each node is continuously adjustable where 𝐸𝑒𝑙𝑒𝑐 represents the energy expenditure required to run the
according to the transmission distance. transmitter or the receiver circuitry per bit and 𝐸𝑓 𝑠 represents the
(k) Each node is not redundant. The death of any node will cause amplifier energy factor for the free-space model.
the whole network to lose its expected functionalities. The energy consumption of the receiver (𝑐𝑟 ) when receiving 𝑙-bit
data is:
3.2. Evaluation criteria
𝑐𝑟 (𝑙) = 𝑙 × 𝐸𝑒𝑙𝑒𝑐 (2)
For static networks, the lifetime of the network can be described by Based on these assumptions and the energy consumption model
the first node dies (FND), half of the nodes die (HND), and last node of the radio module of the above node, if only the radio module is
dies (LND) metrics [26,41–43]. In the scenario proposed in this paper, considered to consume the energy, the mathematical model of the
a node that cannot communicate directly with the base station may lose energy consumption of the network can be established as follows.
contact with the base station, even if it is still alive, due to the death We assume that the network consists of 𝑁 nodes and that there are
of all of its relay nodes. The alive/dead states can no longer accurately 𝑀 feasible schedules by which the data at each node can be transmitted
describe the network lifetime in this application scenario, so two new to the base station directly or through a relay. In the feasible schedule
metrics are introduced. 𝑆𝑖 (𝑖 ∈ 1, 2, … , 𝑀), the energy consumption of node 𝑗(𝑗 ∈ 1, 2, … , 𝑁) in

69
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Assume that the residual energy of each node is 𝐸𝐴 = 𝐸𝐵 = 1000𝐸2 ,

and if the network only performs based on 𝑆1 , the lifetime of this
network is

min{𝐸𝐴 ∕𝑐𝐴 , 𝐸𝐵 ∕𝑐𝐵 } = 111 (11)

Fig. 3. An example of the mathematical model.

The energy consumption of the other 3 feasible schedules can be
calculated in the same way. The results are shown in Table 2.
According to Eq. (3), the longest lifetime of the network can be
one round is 𝑆𝑖 . Let 𝑟𝑖 be the total number of rounds in which schedule
obtained as the optimal solution of the following optimization problem:
𝑆𝑖 is applied by the network. The mathematical model of the network
is:
∑ ∑4
max 𝑧 = 𝑟𝑖 max 𝑧= 𝑖=1 𝑟𝑖 = 𝑟1 + 𝑟2 + 𝑟3 + 𝑟4
⎧ ∑ ⎧
⎪ 𝑐𝑖𝑗 × 𝑟𝑖 ≤ 𝐸𝑗 , 𝑗 = 1, … , 𝑁 (3) ⎪ 9𝐸2 ⋅ 𝑟1 + 18𝐸2 ⋅ 𝑟2 + 𝐸2 ⋅ 𝑟3 + 10𝐸2 ⋅ 𝑟4 ≤ 1000𝐸2
𝑠.𝑡. ⎨ ⎪ (12)
⎪ 𝑟𝑖 ∈ N, 𝑖 = 1, … , 𝑀 𝑠.𝑡. ⎨ 4𝐸2 ⋅ 𝑟1 + 𝐸2 ⋅ 𝑟2 + 8𝐸2 ⋅ 𝑟3 + 15𝐸2 ⋅ 𝑟4 ≤ 1000𝐸2
⎩ ⎪
⎪ 𝑟1 , 𝑟2 , 𝑟3 , 𝑟4 ∈ N
where 𝑧 represents the objective value, which reflects the lifetime of ⎩
the network; 𝐸𝑗 represents the residual energy of node 𝑗; and the first
constraint indicates that the energy consumption of each node can be The optimal solution to this linear integer programming problem is
no greater than the residual energy of the node.
To solve this optimization problem, all feasible schedules should be ⎧ 𝑟∗1 = 103,
listed. The value of 𝑀 is ⎪ 𝑟∗2 = 0,
⎪
∏
𝑁 ⎨ 𝑟∗3 = 73, (13)
𝑀= 𝑠𝑗 (4) ⎪ 𝑟∗4 = 0
𝑗=1 ⎪
⎩ 𝑧 ∗= 𝑟∗1 + 𝑟∗2 + 𝑟∗3 + 𝑟∗4 = 176
where 𝑠𝑗 represents the number of different choices for node 𝑗 to select
a path to the base station. Thus, if the routing protocol instructs the network to choose the
If the optimization problem can be accurately solved and the op- combination of 𝑆1 and 𝑆3 for 103 rounds and 73 rounds, respectively,
timal solution is 𝑟∗𝑖 , the objective value under the optimal solution the network will reach the maximum lifetime.
conditions is 𝑧∗ , and the routing protocol should include a function However, the value of 𝑠𝑗 in extreme cases is
that enables the network to produce a schedule 𝑆𝑖 corresponding to
𝑟∗𝑖 , where 𝑟∗𝑖 is not zero and directs the network to employ 𝑆𝑖 for no ∑
𝑁−1
𝑠𝑗 = 𝐴𝑘𝑘 (14)
more than 𝑟∗𝑖 rounds. We call these schedules candidate schedules.
𝑘=0
Since there is no analytical solution to the linear programming
problem, we use a simple example to illustrate the energy consumption where 𝐴𝑘𝑘 represents the full arrangement of 𝑘 nodes. In this extreme
of the network, as shown in Fig. 3. There are two nodes, Node A and case, any two nodes can communicate with each other.
Node B, that transmit 𝑙-bit data to the base station (or a relay node) C According to Eqs. (4) and (14), if 𝑁 = 5, then 𝑀 = 1545 ≈ 8.7×1010 .
in each round. The distance between Node A and Node B is 𝑑, and the It is obvious that this is an NP-hard problem, so it is difficult to
distance between Node B and Node C is 2𝑑. obtain the exact solution when the node number is large. Metaheuristic
When 𝑁 = 2, there are two choices for each node, direct transmis- algorithms may be helpful for obtaining a satisfactory solution, but a
sion or indirect transmission; therefore, 𝑠1 = 2, and 𝑠2 = 2. According large computing scale is unavoidable.
to Eq. (4), 𝑀 = 4.
Notably, the residual battery power of each node is consumed by
Let 𝑆1 be a schedule in which Node A and Node B both choose direct
functions other than radio wave emission and reception. The other
transmission, which means that A=>C and B=>C. According to Eq. (1),
the energy consumption of Node A is modules, such as sensor and processor modules, also consume energy,
and the energy consumption of the sensor and processor modules is
𝑐𝐴 (𝑙, 𝑑) = 𝑙 × 𝐸𝑒𝑙𝑒𝑐 + 𝑙 × 𝐸𝑓 𝑠 × (3𝑑)2 (5) usually correlated with environmental factors based on where the nodes
Let 𝑙 × 𝐸𝑒𝑙𝑒𝑐 = 𝐸1 and 𝑙 × 𝐸𝑓 𝑠 × 𝑑2 = 𝐸2 , then are deployed; these factors can be difficult to quantify. Therefore,
even if the optimal solution of this model is obtained, it is of no
𝑐𝐴 (𝑙, 𝑑) = 𝐸1 + 9𝐸2 (6) practical significance. Although the mathematical model of the energy
consumption of the network is not perfect, it can be used to draw some
The energy consumption of Node B is
useful conclusions.
𝑐𝐵 (𝑙, 𝑑) = 𝑙 × 𝐸𝑒𝑙𝑒𝑐 + 𝑙 × 𝐸𝑓 𝑠 × (2𝑑)2 Based on the above analysis, the routing protocol is characterized
(7)
= 𝐸1 + 4𝐸2 by the following properties:
In the scenario proposed in this paper, it is reasonable that 𝐸2 ≫ 𝐸1
(𝑑 > 100 m); therefore, the approximate values can be obtained as • A schedule 𝑆 can be generated by this protocol. Schedule 𝑆 is
equal to or approximate to a schedule corresponding to 𝑟𝑚𝑎𝑥 ,
𝑐𝐴 (𝑙, 𝑑) ≈ 9𝐸2 (8) where 𝑟𝑚𝑎𝑥 is the greatest or close to the greatest 𝑟∗𝑖 . Since 𝑟∗𝑖
is difficult to obtain when the scale of a network is large, a
𝑐𝐵 (𝑙, 𝑑) ≈ 4𝐸2 (9) satisfactory solution will suffice.
• This approach is characterized by low computational complexity.
The total energy consumption of 𝑆1 is
• This approach is adaptive; even if the energy consumption of the
𝑐𝑇 𝑜𝑡𝑎𝑙 (𝑙, 𝑑) = 𝑐𝐴 (𝑙, 𝑑) + 𝑐𝐵 (𝑙, 𝑑) ≈ 13𝐸2 (10) other two modules is considered, the approach is still effective.

70
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Table 2
The energy consumption of the schedules in one round.
Schedule Node A Node B Total energy Lifetime
consumption
Mode Energy consumption Mode Energy consumption
approximate value approximate value
(Exact value) (Exact value)
a
𝑆1 Direct(A=>C) 9𝐸2 (𝐸1 + 9𝐸2 ) Direct(B=>C) 4𝐸2 (𝐸1 + 4𝐸2 ) 13𝐸2 111
a
𝑆2 Direct(A=>C) 18𝐸2 (3𝐸1 + 18𝐸2 ) Indirect(B=>A=>C) 𝐸2 (𝐸1 + 𝐸2 ) 19𝐸2 55
a
𝑆3 Indirect(A=>B=>C) 𝐸2 (𝐸1 + 𝐸2 ) Direct(B=>C) 8𝐸2 (3𝐸1 + 8𝐸2 ) 9𝐸2 125
a
𝑆4 Indirect(A=>B=>C) 10𝐸2 (3𝐸1 + 10𝐸2 ) Indirect(B=>A=>C) 5𝐸2 (3𝐸1 + 5𝐸2 ) 15𝐸2 100
a The term directly related to the lifetime of the network.

4.1. Load balancing

In the literature, nodes usually have the same residual energy when
a network is initiated. Therefore, maintaining a load balance will
maintain equality or approximate equality in the residual energy among
nodes. As introduced in Section 2, in cluster-based protocols, fuzzy
logic is used to maintain a load balance. In this paper, fuzzy logic is
not directly used. However, based on the concept of fuzzy logic, some
definitions, such as equivalent nodes, are introduced to achieve load
balancing. In tree-based protocols, some nodes bear heavier loads than
others due to their position in the topology, so it is inevitable for them
to die earlier. A load balance can be maintained by regulating the
selection of the receiving nodes when the farther nodes send data to
the relay nodes. In this paper, the concept of in-degree based on graph
theory is used to help achieve a load balance.

4.2. Schedule updating

Fig. 4. The optimal solutions under different initial conditions. For a specific schedule, after certain rounds of operation, the cu-
mulative effect of unbalanced energy consumption will deteriorate the
balance of the residual energy. In the example in Fig. 3, no single
4. Analysis of the mechanisms used to prolong the network life- schedule can maintain the residual energy balance under the condition
time that the initial energy of the two nodes is equal.
Therefore, after a certain number of rounds, a new schedule should
Since it is difficult to obtain candidate schedules by solving op- be generated based on the current status of the network and exe-
timization problems at large scales, candidate schedules will be con- cuted instead of the previous schedule. This process is called schedule
structed by the routing protocol instead. The construction of the can- updating. The essence of schedule updating is to implement a combi-
didate schedule set is based on the nature of the candidate schedules. nation of different schedules that can complement each other in energy
The nature of the candidate schedules is discussed first. consumption.
From Table 2, even if only the most energy-efficient schedule (𝑆3 , In addition, there are two important effects of schedule updating
A=>B=>C, B=>C, total energy consumption 9𝐸2 ) is used, the lifetime in practical applications. The first is coping with topological changes
of the network is 125. However, through a combination of schedules, caused by node death and environmental changes. For example, a
the lifetime of the network can be prolonged up to 176, a 40.8% in- relay node may die due to component failure, energy depletion, or
crease. It can be inferred that energy savings are not the most important loss of communication in some direction due to a new obstacle. At
factor that affects the lifetime of the whole network. this time, it is necessary to update the schedule to allow the nodes in
For further analysis, another example is designed. In the example the network to contact the base station again. The second is dealing
in Fig. 3, it is assumed that the residual energy of Node A is fixed at with the uncertainty of energy consumption. The energy consumption
𝐸𝐴 = 1000𝐸2 , the residual energy of Node B varies from 0 to 10000𝐸2 , of the sensor module and the processor module of a sensor node can
and other premises remain unchanged. The optimal solutions of these be considered uncertain, so the energy consumption of all sensor nodes
10000 models are obtained by a CPLEX solver and demonstrated in can be formulated as a Markov process; in this case, the current state of
Fig. 4.
the network is the only factor that needs to be considered in schedule
In Fig. 4, it is worth noting that although 𝑆2 is the least energy-
generation. This approach is one way to address uncertain energy
efficient case, when 𝐸𝐵 is small, 𝑆2 should still be applied because Node
consumption.
B consumes the least energy in 𝑆2 . As the amount of energy used by
Node B increases, the proportion of 𝑆3 that consumes more energy at
Node B than at Node A increases. It can be concluded that the residual 4.3. Relay node insertion for energy savings
energy rather than energy efficiency is one of the most important
factors that influences the combinations of schedules. Additionally, When the network scale is large, that is, the number of nodes
the schedule with the greatest 𝑟∗𝑖 tends to minimize the difference in is large, under the premise of load balancing, there may still be a
residual energy between the two nodes. This result can be interpreted as variety of optional schedules. At this time, choosing an energy-efficient
the global optimum when the residual energy of the nodes is balanced. schedule can provide more choices for subsequent operations.
There have been many mechanisms used in previous protocols that A model is established to analyze the principle of energy savings,
let nodes with higher remaining energy levels bear greater loads than as shown in Fig. 5. It is assumed that Node 0 at coordinates (0, 0) will
other nodes. However, different mechanisms can be improved and transmit 𝑙-bit data to Node 𝑛 at coordinates (0, 𝑑) via (𝑛 − 1) relay nodes
applied to the scenarios proposed in this paper. and that the coordinates of Node 𝑖 are (𝑥𝑖 , 𝑦𝑖 ).

71
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

𝑐𝑡𝑜𝑡𝑎𝑙 reaches the minimum value.

Eqs. (19) and (20) indicate that even under the optimal conditions
in which relay nodes are uniformly distributed, the energy savings
achieved by increasing the number of relay nodes is limited.
However, in practical applications, the deployment of nodes is
influenced by the locations of the targets to be sensed, so a uniform
distribution of nodes is difficult to achieve. Furthermore, even under
the optimal conditions, when a relay node is added, the reduction in
energy consumption is

𝛥𝑐𝑡𝑜𝑡𝑎𝑙 = 𝑐𝑡𝑜𝑡𝑎𝑙 (𝑛) − 𝑐𝑡𝑜𝑡𝑎𝑙 (𝑛 + 1)

𝐸𝑓 𝑠
= (2𝐸𝑒𝑙𝑒𝑐 ⋅ 𝑛 ⋅ 𝑙 + ⋅ 𝑑 2 ⋅ 𝑙)
Fig. 5. The model for energy-saving analysis.
𝑛
𝐸𝑓 𝑠 (21)
− [2𝐸𝑒𝑙𝑒𝑐 ⋅ (𝑛 + 1) ⋅ 𝑙 + ⋅ 𝑑 2 ⋅ 𝑙]
𝑛+1
1
=[ 𝐸 ⋅ 𝑑 2 − 2𝐸𝑒𝑙𝑒𝑐 ] ⋅ 𝑙
𝑛(𝑛 + 1) 𝑓 𝑠
Eq. (21) indicates that as 𝑛 increases, the marginal utility rapidly
decreases rapidly. When one relay node is established, the marginal
utility reaches a maximum level close to 50%. In addition, if the
number of optional nodes is large, the computational complexity for
schedule design will increase exponentially as the number of relay
nodes increases.
Based on the above analysis, if some nodes can be selected as
relay nodes and reduce energy use in data transmission, selecting one
node only is satisfactory for balancing effectiveness and computational
complexity.
Fig. 6. The optimal solution of the energy-saving analysis model.
In addition to the above mechanisms, Fig. 4 illustrates another
important rule: direct transmission (A=>C or B=>C) always exists.
Additionally, when the residual energy levels of two nodes are similar,
According to Eqs. (1) and (2), the total energy consumption during
the proportion of S1 (both of the nodes choose the direct transmission
data transmission is
mode) will be relatively high. In other words, when the difference in the
∑
𝑛−1
residual energy levels of the nodes is not large, the direct transmission
2
𝑐𝑡𝑜𝑡𝑎𝑙 = 2𝑛 ⋅ 𝐸𝑒𝑙𝑒𝑐 ⋅ 𝑙 + 𝑛 ⋅ 𝑙 ⋅ 𝐸𝑓 𝑠 ⋅ 𝑑𝑖(𝑖+1) (15)
𝑖=0
mode should be chosen. A reasonable explanation for this phenomenon
is that if data aggregation cannot be performed, that is, there is no data
where 𝑑𝑖(𝑖+1) represents the distance between Node 𝑖 and Node (𝑖 + 1).
reduction in the wireless channel, any node acting as a relay node will
Additionally,
experience at least twice as much energy consumption as other nodes,
∑
𝑛−1 ∑
𝑛−1
which will exacerbate the imbalance in energy consumption.
2
𝑑𝑖(𝑖+1) = [(𝑥𝑖+1 − 𝑥𝑖 )2 + (𝑦𝑖+1 − 𝑦𝑖 )2 ] (16)
Another way of interpreting this rule is that a node should choose
𝑖=0 𝑖=0
the direct transmission mode if there is no better option to balance and
There is a minimum value in this function, and the following
save energy. This is one of the key concepts considered in designing
equations are satisfied:
the routing protocol.
⎧ 𝜕 ∑ 2
𝑛−1
⎪ ( 𝑑𝑖(𝑖+1) ) = 0 𝑖 = 1, 2, … ...𝑛 − 1
⎪ 𝜕𝑥𝑖 𝑖=0 5. Hybrid tree-based and cluster-based routing protocol for raw
⎨ (17)
⎪ 𝜕 ∑
𝑛−1
data collection
⎪ ( 𝑑2 ) = 0 𝑖 = 1, 2, … ...𝑛 − 1
⎩ 𝜕𝑦𝑖 𝑖=0 𝑖(𝑖+1)
The operation of the HTC-RDC protocol consists of five phases. The
The solution of the equation set is
{ relationships among the phases and subphases are illustrated in Fig. 7.
𝑥𝑖 = 𝑑𝑛 ⋅ 𝑖
𝑖 = 1, 2, … ...𝑛 − 1 (18)
𝑦𝑖 = 0

Thus, when the relay nodes are uniformly distributed along the 5.1. The initialization of the network
connection between Node 0 and Node 𝑛, as shown in Fig. 6, the total
energy consumption is minimized. Since some nodes cannot communicate directly with the base sta-
The lowest total energy consumption is tion, they should find a path to the base station first. Although some
𝑑 dynamic routing protocols, such as the optimized link state routing
𝑐𝑡𝑜𝑡𝑎𝑙−𝑚𝑖𝑛 = 2𝑛 ⋅ 𝐸𝑒𝑙𝑒𝑐 ⋅ 𝑙 + 𝑛 ⋅ 𝑙 ⋅ 𝐸𝑓 𝑠 ⋅ ( )2
𝑛 (OLSR) [42] and ad hoc on-demand distance vector (AODV) [43]
(19)
𝐸𝑓 𝑠 ⋅ 𝑑 2 ⋅ 𝑙 protocols, can realize path discovery from a node to the base station,
= 2𝐸𝑒𝑙𝑒𝑐 ⋅ 𝑙 ⋅ 𝑛 +
𝑛 the overhead of energy consumption is large; therefore, these methods
When are only applicable to nodes with a continuous energy supply. A method
√ √ of advertising from the base station to all nodes is employed in the
⎢ 𝐸 𝑑2 ⎥ ⎡ 𝐸𝑓 𝑠 𝑑 2 ⎤
𝑛=⎢ ⎥ 𝑛=⎢ ⎥
𝑓𝑠 proposed protocol for network initialization so that all nodes can find
or (20)
⎢ 2𝐸𝑒𝑙𝑒𝑐 ⎥ ⎢ 2𝐸𝑒𝑙𝑒𝑐 ⎥ a path to the base station. This phase is divided into two subphases.
⎣ ⎦ ⎢ ⎥

72
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Fig. 7. The phases and subphases of the operation of the HTC-RDC protocol.

5.1.1. Advertising
The advertising process is illustrated in Fig. 8. Every node should
have an attribute named Hierarchy. The base station level is Hierar-
chy 0.
Step 1: The base station starts the initialization process. The base
station generates an integer in the current round of advertising and then
advertises a message with the power that is the same as the maximum
power of the radio module of the node. The message includes the round
number, the ID of the advertiser, the hierarchy level and the residual
energy information of the advertiser (the residual energy of the base
station can be set as infinity).
Step 2: The nodes maintain the initialization process, as shown in
Fig. 9. The nodes that receive a message should memorize the message,
compute the distance between the node and the source node based
on the RSSI, record the distance information, and then perform the Fig. 8. The advertising process.
following operations.

• Case 1: If it is the first advertisement message in the most recent

round, clear the hierarchy information, and set the hierarchy message. Thus, the problem that some nodes cannot communicate
level to the hierarchy value contained in the message plus 1. In directly with the base station can be effectively solved, regardless
addition, memorize the source node according to the information of whether this lack of communication is caused by a distance
in the advertisement message; this node is treated as a parent that exceeds the transmission radius of the node or the existence
node. of obstacles.
Then, the node should generate a new message with the same
In the advertising process, each node will advertise once and only
structure, but the message should be revised to contain the rel-
once. The result of the advertising phase is that each node possesses
evant node information. After a certain time interval, the nodes
advertise the new message with a carrier sense multiple-access local topology information. Each node belongs to a specific hierarchy
with collision avoidance (CSMA/CA) protocol. and has at least one parent node, with or without sibling and child
• Case 2: If there are already messages stored from previous adver- nodes. The topology of the whole network is shown in Fig. 10.
tising rounds, then a node should perform the following tasks. The hierarchy of a node indicates the number of hops required for
If the hierarchy level in the message is less than the hierarchy a message from the node to reach the base station, and the lower the
level of the node, record the source node as a parent node. hierarchy level is, the closer to the base station the message is; that is,
If the hierarchy level in the message equals the hierarchy level of as long as a node transmits data to a node with a lower hierarchy level,
the node, record the source node as a sibling node. it can transmit data to the base station.
If the hierarchy level in the message is greater than the hierarchy
level of the node, record the source node as an offspring node. 5.1.2. Node information collection
As long as there is at least one path for each node to reach the If each node retains only one edge with one parent node only in the
base station, each node will receive at least one advertisement graph of the topology, the graph will form a tree structure. In the order

73
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

A better option involves all nodes providing their own information

via a feedback message to the base station, and an optimized central-
ized routing protocol can be established and distributed to all the nodes
by the base station. Another advantage of the centralized method is
that the computational overhead is borne by the base station, which
has an unlimited energy supply. However, at this time, each node only
has access to local topology information. To ensure that the topology
information can be collected by the base station, each node retains
an edge with its parent node, which has the highest residual energy
when sending a feedback message to the base station. The feedback
message sent from each node should contain all the necessary node
information for schedule creation, including the residual energy level
and topological information, such as hierarchy, parent nodes, sibling
nodes and offspring nodes.
The initialization of the network ends when all the feedback mes-
sages are collected by the base station. Since there are no expected data
transmitted to the base station, the energy consumption of the process
is a form of functional consumption by the network.

5.2. Schedule creation

After the initialization of the network, data from every node can
be collected in the same way as described for the feedback process.
Since this approach is not effective in prolonging the lifetime of the
network, an optimized schedule should be created by the base station
based on all the network information. One of the most important tasks
in schedule creation is to set the next-hop node for each node.

5.2.1. Establishment of an optimized tree structure

Based on the analysis in Section 4, the direct transmission modes
should be considered first. For a node, the direct transmission path is
established by setting the next hop as a parent node because a node
with a lower hierarchy is inevitable in the path to the base station.
If there is only one parent node for a node, such as for nodes with
Hierarchy 1, the node sets the only parent node as its next-hop node.
Fig. 9. Flowchart of senor node processing for an advertisement message. If there is more than one parent node, the next-hop node should be
the parent node with the highest residual energy level or an equivalent
node with the lowest in-degree. An equivalent node is defined as a
node that meets the relevant energy equivalence and cost equivalence
criteria.
(1) Energy equivalence criterion Denote the parent node that has
the most residual energy as the candidate node and a parent node that
may be an energy-equivalent node as an alternative node. The energy
equivalence criterion is as follows:
𝐸can
𝛼= ≤ 𝛼0 (22)
𝐸alt
where 𝐸can represents the residual energy of the candidate node, 𝐸alt
represents the residual energy of an alternative node, and 𝛼0 is a
parameter defined a priori according to experimental results.
(2) Cost equivalence criterion The cost equivalence criterion is as
follows:
𝑐
𝛽 = can ≤ 𝛽0 (23)
𝑐alt
Fig. 10. The topology of the network after the advertising process.
where 𝑐alt represents the energy cost correlated with the candidate
node, 𝑐can represents the energy cost correlated with an alternative
node, and 𝛽0 is a parameter defined a priori according to experimental
of hierarchy from high to low, each node forwards the data it receives results.
from its offspring node and sends its own data to the parent node; 𝑐alt and 𝑐can refer to different specific expressions in two different
this is the simplest tree-based routing protocol. However, the simplest situations.
(i) If the residual energy of the candidate node is lower than that of
routing protocol may not be the most efficient in terms of prolonging
the node that selects its next-hop node, which tends to be a reasonable
the lifetime of the network because no mechanism that can prolong the situation because a node closer to the root in the tree structure tends to
lifetime of the network is employed. bear a greater load than nodes farther from the root, the overall energy

74
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

direct transmission paths need to be determined, and the possible direct

transmission paths DA, EA, EB and EC are indicated by dotted arrows.
For Node D, since it has only one parent node, its direct transmission
path can be determined as DA. However for Node E, since there are 3
possible direct transmission paths, the direct transmission path needs to
be determined by further analysis. In this example, it is assumed that
the residual energy levels of Nodes A, B, and C are 𝐸𝐴 , 𝐸𝐵 , and 𝐸𝐶 ,
respectively, and that 𝐸𝐴 > 𝐸𝐵 > 𝐸𝐶 ; therefore,
𝐸𝐵 𝐸𝐶
𝛼𝐵𝐴 = ≥ 𝛼0 , 𝛼𝐶𝐴 = ≥ 𝛼0 (27)
𝐸𝐴 𝐸𝐴
That is, Node A is the candidate node, and Nodes B and C are
alternative nodes that meet the energy equivalence criterion.
Under the condition that 𝐸𝐴 < 𝐸𝐸 , 𝛽 should be calculated according
to Eq. (25) as follows:
Fig. 11. Example of the establishment of an optimized tree structure.
𝑐𝐵 𝑑 2 + 𝑑𝐵𝑂
2
𝛽𝐵𝐴 = ≈ 𝐸𝐵 (28)
𝑐𝐴 2 2
𝑑𝐸𝐴 + 𝑑𝐴𝑂
cost should be considered. Therefore, Eq. (22) can be specifically
𝑐𝐶 𝑑 2 + 𝑑𝐶𝑂
2
expressed as: 𝛽𝐶𝐴 = ≈ 𝐸𝐶 (29)
𝑐𝐴 2 2
𝑑𝐸𝐴 + 𝑑𝐴𝑂
𝑐 𝑐alt_node + 𝑐alt_next−hop
𝛽 = can = (24) It is assumed that 𝛽𝐵𝐴 ≤ 𝛽0 and 𝛽𝐶𝐴 ≤ 𝛽0 ; that is, Node B and Node C
𝑐alt 𝑐can_node + 𝑐can_next−hop
are both equivalent nodes. Because the in-degree of Node A is 1 and the
where 𝑐alt_node represents the energy cost of the node when it transmits in-degrees of Node B and Node C are both 0, the next-hop node of Node
data to the alternative node, 𝑐alt_next−hop represents the energy cost of E should be selected between Node B and Node C. Then, comparing 𝛽𝐵𝐴
the alternative node when it transmits data to the next-hop node of and 𝛽𝐶𝐴 , 𝑑𝐸𝐵2 +𝑑 2 > 𝑑 2 +𝑑 2 ; therefore, Node C is the next-hop node
𝐵𝑂 𝐸𝐶 𝐶𝑂
the alternative node, 𝑐can_node represents the energy cost of the node of Node E.
when it transmits data to the candidate node, and 𝑐can_next−hop represents
Under the condition that 𝐸𝐴 > 𝐸𝐸 , should be calculated as follows:
the energy cost of the candidate node when it transmits data to the
next-hop node of the candidate node.
In the scenario proposed in this paper, according to Eq. (1), Eq. (24) 𝑐𝐵 𝑑2
𝛽𝐵𝐴 = ≈ 𝐸𝐵 (30)
𝑐𝐴 2
𝑑𝐸𝐴
can be approximately formulated as:

𝑐can
2
𝑑alt_node 2
+ 𝑑alt_next−hop 𝑐𝐶 𝑑2
𝛽= ≈ (25) 𝛽𝐶𝐴 = ≈ 𝐸𝐶 (31)
2 2 𝑐𝐴 2
𝑑𝐸𝐴
𝑐alt 𝑑can_node + 𝑑can_next−hop
It is assumed that 𝛽𝐵𝐴 ≤ 𝛽0 and 𝛽𝐶𝐴 ≤ 𝛽0 ; that is, Node B and Node C
where 𝑑alt_node represents the distance between the node and the alter-
are both equivalent nodes. Because the in-degree of Node A is 1 and the
native node, 𝑑alt_next−hop represents the distance between the alternative
in-degrees of Node B and Node C are both 0, the next-hop node of Node
node and the next-hop node of the alternative node, 𝑑can_node represents
E should be selected between Node B and Node C. Then, comparing 𝛽𝐵𝐴
the distance between the node and the candidate node, and 𝑑can_next−hop 2 < 𝑑 2 ; therefore, Node B is the next-hop node of Node E.
represents the distance between the candidate node and the next-hop and 𝛽𝐶𝐴 , 𝑑𝐸𝐵 𝐶𝑂
node of the candidate node.
(ii) If the residual energy of the candidate node is higher than 5.2.2. Optimization by incorporating a cluster structure
that of the node that selects the next-hop node, energy consumption Based on the analysis in Section 4, if the cluster structure can be
tends to be unbalanced, and the node should be protected in energy integrated into the established tree structure, more mechanisms may
consumption. In this case, Eq. (22) can be specifically expressed as: be used to further prolong the lifetime of the network. The expected
cluster structure should balance the residual energy of nodes, promote
2
𝑑alt_node
𝑐can energy savings or both.
𝛽= ≈ (26)
𝑐alt 2
𝑑can_node It can be inferred that such a cluster structure is formed because
some nodes become cluster heads, while some other nodes choose
The alternative nodes that meet the above two criteria are equiv- indirect paths via the cluster heads; these indirect paths have the
alent nodes. If there are equivalent nodes for the candidate node of a following properties:
given node, the node with the lowest in-degree among the candidate
node and the equivalent nodes should be set as the next-hop node. • These paths will increase the balance of residual energy or pro-
Furthermore, if there is more than one node with the same minimum vide greater energy efficiency as a direct transmission path.
in-degree, the node that minimizes 𝛽 should be selected. • The relay node in an indirect path, that is, the cluster head, should
The establishment process must be performed for every node from a be a sibling node; otherwise, the path will be a direct transmission
lower hierarchy to a higher hierarchy since the direct transmission path path or a path with the wrong direction to an offspring node.
of a parent node may be used as a reference in the process. Compared • If a node is selected as a relay node by another node, it can no
with the tree structure with which a feedback message is collected, this longer choose an indirect transmission path; otherwise, for the
tree structure is optimized by load balancing. node that selects it as a relay node, there will be more than two
To illustrate the establishment process, a simple example is shown relay nodes, which will hinder energy savings.
in Fig. 11.
Nodes A, B, and C are classified as Hierarchy 1, so they all have The approach used to search for the best path involves determining
only one parent node: the base station (denoted as Node O), and their if the indirect path with any sibling node as the relay node is better
direct transmission paths AO, BO and CO are deterministic and are than the relevant direct transmission path. The criteria for selecting the
indicated by normal arrows in Fig. 11. Nodes D and E are classified best path are shown in Fig. 12. Node A and Node B are classified as
as Hierarchy 2 since they may have more than one parent node, so the Hierarchy 𝑖, Node C and Node D are classified as Hierarchy 𝑖 + 1, and

75
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Fig. 12. Illustration of the process of searching for a better path.

the direct transmission paths of Node C and Node D are determined as

C=>A and D=>B. Node C is searching for a better transmission path,
and as the sibling node of Node C, C=>D=>A is an alternative indirect
path.
A better indirect path can be identified based on Alg. 1. Fig. 13. Hybrid structure of trees and clusters.

Algorithm 1 Algorithm for Alternative Path Searching

Input: every sibling node of Node C;
5.3. Schedule distribution
Output: the indirect path (if it exists) or null (the direct transmission
path should be retained);
1: for every sibling node denoted as D do The schedule should be distributed from the base station to all the
2: if D=>B is a direct transmission path then nodes so that every node knows when and to which node it should send
3: if 𝐸𝐵 ≥ 𝐸𝐴 then its data. The distribution process involves a multicast method. The base
4: if 𝑑𝐶𝐷 <𝑑𝐶𝐴 then station broadcasts the schedule, and then any node that receives the
5: if 𝐸𝐷 ∕𝐸𝐶 > 𝛼0 then schedule multicasts the schedule to its offspring nodes if it has offspring
6: record CD as an alternative path nodes. The power of multicasting only needs to be high enough to allow
7: end if
the schedule to reach the farthest offspring node. This power usually
8: end if
does not need to be the maximum transmission power of nodes, and a
9: end if
node with no offspring nodes does not need to multicast the schedule;
10: end if
11: end for
therefore, the overhead of the distribution process will be much less
12: denote the number of alternative paths as 𝑛 than that associated with the initialization of the network.
13: if 𝑛 == 0 then
14: return null
5.4. Data collection
15: else if 𝑛 == 1 then
16: return alternative path
17: else Each node sends its data to its next-hop node in a given time
18: filter out the alternative paths with the smallest in-degree value slot according to the schedule. This process ensures the steady-state
of Node D operation of the network and will last for a certain period until the
19: update 𝑛 as the number of filtered alternative paths schedule updates.
20: if 𝑛 == 1 then
21: return alternative path
22: else 5.5. Schedule updating
23: 2 + 𝑑2 )
return the alternative path that has the smallest (𝑑𝐶𝐷 𝐷𝐴
value
24: end if There are two cases in which the schedule should update in different
25: end if ways. In the first case, the network has collected data for a certain fixed
number of rounds, which is determined a priori. In this case, the new
This algorithm still works in the case where Node B and Node A
coincide (they are the same node, which is a common parent node of schedule is created based on the known topology and residual energy
Node C and Node D). In this cluster structure, a node selects a sibling information for nodes, which is extracted from the data received from
node as a relay node when the data received at the same hierarchy all nodes. The advantage of this method is a reduction in energy con-
level are collected. If there are cluster structures, the topology of the sumption. The base station, which has a permanent power supply, is the
network is now a hybrid structure of trees and clusters, as illustrated only energy consumer in schedule creation, and nodes only consume
in Fig. 13. energy in schedule distribution; therefore, this process consumes much
less energy than network initialization. The second case involves the
5.2.3. Formulating the schedule topological alteration of the network. In this case, not all data from
Based on the hybrid tree and cluster structure, which includes all nodes can be received. Therefore, the topological information of the
the nodes and the edges connected to the next-hop nodes, the schedule
network is incomplete and cannot support schedule creation. Therefore,
is formed after allocating a time slot for each node. The function of
the network should be reinitialized. The advertisement and feedback
time slot allocation is to enable the nodes to send all of the received
processes must be repeated, a new schedule must be created with the
data and their own data to their next-hop nodes after receiving data
from all of their offspring nodes and sibling nodes who select them new network information, and the schedule must then be distributed to
as the relay node; thus, the data flows from the highest hierarchy to all nodes. The advantages of this method are simplicity and robustness.
the lowest hierarchy (the base station). The purpose of the time slot Regardless of what factors make the topology change, the method will
allocation is to reduce the standby time and the energy waste caused work. The disadvantage of this method is that it consumes more energy
by data retransmission after channel conflicts. than the alternative method.

76
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

6.2. Comparative methods

Since pure cluster-based protocols cannot support WSNs in the

scenario proposed in this paper, two tree-based protocols are imple-
mented for comparison. One method is the minimum hop transmission
(MHT) protocol. In the MHT protocol, each node chooses the path
corresponding to the lowest energy consumption (ignoring the energy
consumption of circuitry), which consists of only one node in each
hierarchy. This method is intuitive and simple. The other method is the
robust self-organizing tree-based routing (RSTR) protocol, which has
become one of the most popular approaches in recent years. However,
the sleep-wake mechanism in RSTR is not applied since this mechanism
is not suitable for the scenario proposed in this paper.

6.3. Results

6.3.1. The schedules created by different protocols

For the same experimental environment, the schedules created by
the MHT, RSTR, and HTC-RDC protocols in 𝑟𝑜𝑢𝑛𝑑 0 and 𝑟𝑜𝑢𝑛𝑑 400
are shown in Fig. 15. Fig. 15(a)–(c) show that at the beginning of the
experiment, all nodes choose direct transmission paths, and the effects
of the three protocols are similar. However, at 𝑟𝑜𝑢𝑛𝑑 400, the cyan
arrow in Fig. 15(f) indicates that in the HTC-RDC protocol, some nodes
select nodes of the same hierarchy as relays; additionally, as shown in
Fig. 14. Application scenario for the experiment.
Fig. 15(e), the RSTR protocol will not utilize this approach. Fig. 15(d)
shows that because an updating mechanism is not adopted, some nodes
Table 3
in the network with the MHT protocol have already died.
Experimental parameters.
Parameters Value
6.3.2. The optimization of parameters
Target area 2000 m × 2000 m There are two system parameters 𝛼0 and 𝛽0 that need to be manually
Node number 100
determined a priori. The optimal setting of these two parameters is
Node transmission radius 2000 m
Coordinates of the base station (1000, −1000) related to the topology of the network, but a reasonable choice would
Obstacle 1 location (end points) (0, 1500), (500, 1500) be values greater than 1 because 𝛼0 and 𝛽0 are used to avoid the
Obstacle 2 location (end points) (1000, 500), (1500, 500) selection of a next-hop node that is a hot spot; moreover, to ensure
Initial energy of node battery 2000J a balance in residual energy, these values should not be much greater
𝐸𝑒𝑙𝑒𝑐 50 nJ/bit
than 1, thus limiting the next-hop node choices to those with relatively
𝐸𝑓 𝑠 10 pJ/bit/m2
Header length of the data frame 240 bit (30 Byte)
high levels of residual energy. An experiment is performed to optimize
Length of data collected by a 6 × 105 bit the setting of the two parameters in advance. The result is shown in
single node per time interval Fig. 16.
The result indicates that the lifetime of the network is not strictly
linearly related to the setting of these two parameters, and when the
values 𝛼0 and 𝛽0 are approximately 1.5, the effect is best. In our
6. Experiment results
experiments, both 𝛼0 and 𝛽0 are set to 1.5.

6.1. Experiment setup 6.3.3. The lifetime of the network

For the same node distribution scenario, the lifetimes of the WSN
using the MHT, RSTR, and HTC-RDC protocols are shown in Fig. 17.
To verify and analyze the behavior of the proposed HTC-RDC pro- The blue line represents the number of alive nodes, and the red line
tocol, an application scenario is designed, as shown in Fig. 14. One represents the number of nodes that can successfully transmit data to
hundred sensor nodes are randomly deployed in the target area. Two the base station. There are some points on the red line that are below
line segments are set in the target area to simulate the occlusion the standard curve because at those times, some nodes are alive, but
their data cannot be received by the base station due to the death of
effect of mountains, buildings, etc. The effect of the obstacles in the
their parent nodes. In the next round, the point returns to the standard
experiment hinders the transmission of the radio wave among the nodes
curve because the schedule is updated.
(or the base station) when the corresponding node edges intersect the At the initial stage of the experiment, the number of nodes from
obstacles. Some nodes cannot directly communicate with the base sta- which the base station can collect data equals the total number of
tion because of obstacles, and some nodes are beyond the transmission nodes. This indicates that the HTC-RDC protocol can achieve its ex-
radius of the base station, so they have to communicate with the base pected function that enables the network to work in a large area beyond
station through a relay. the influence of the obstacles.
The experiment is carried out in an environment using Python Taking the FNL proposed in this paper as the evaluation criterion,
programming language, and the parameters of the experiment are the network lifetime corresponds to the first drop of the red line and
is indicated by the orange vertical line in Fig. 17. The FNLs of HTC-
shown in Table 3. The input of the experiment is the coordinates of all
RDC and RSTR are significantly greater than that of MHT, and the FNL
nodes that are generated by the random function, and the output is the
of HTC-RDC (488 rounds) is greater than that of RSTR (422 rounds).
number of sensor nodes from which the base station can receive data For reference, the HNL value is indicated by a green vertical line. The
in each round. The output of the experiment will indicate 1) whether FNL value of the network using the HTC-RDC protocol is closer to HNL,
the network achieves the expected function, i.e., whether the network which indicates that the HTC-RDC protocol makes the network load
can collect data from all nodes, and 2) the lifetime of the network. more balanced.

77
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Fig. 15. The schedules created by the MHT, RSTR and HTC-RDC protocols in 𝑟𝑜𝑢𝑛𝑑 0 and 𝑟𝑜𝑢𝑛𝑑 400.

Table 4
The statistics of the Monte Carlo experimental results.
Protocol Average FNL (round) Average HNL (round) HNL-FNL (round)
MHT 122.69 687.36 564.67
RSTR 490.394 600.332 109.938
HTC-RDC 546.544 579.218 32.674

data from all nodes. The statistics of the 500 results are shown in
Table 4. The statistics indicate that the FNL of the network with
the HTC-RDC protocol is enhanced by approximately 11.4% over the
existing protocols.

6.4. Discussion

6.4.1. The effect of the tree structure and cluster structure

From the experimental results, it can be concluded that in the
scenario proposed in this paper, the tree structure can enable the
network to achieve its expected function, and the incorporation of
the cluster structure into the tree structure is an effective method
for prolonging the lifetime of the network. However, the statistics in
Table 4 show that although the FNL of the HTC-RDC protocol is the
greatest, the HNL is the least in the three protocols. This phenomenon
Fig. 16. The influence of different combinations of 𝛼0 and 𝛽0 on the lifetime of the
network.
shows that although HTC-RDC prolongs the lifetime of the network, it
does not make the network more energy efficient. Under the condition
of raw data collection without data aggregation, the incorporation of
the cluster structure mainly balances the load of the network, so it is
6.3.4. Monte Carlo Experiments
reasonable that the FNL and HNL values are closer and that the FNL
To verify the performance of the proposed protocol for different value is larger but the HNL value is smaller.
node distributions, 500 Monte Carlo experiments were performed. The
experiments used the same target area and the same obstacles, and 500 6.4.2. The effectiveness of HTC-RDC protocol
different node distributions were generated by the random function. Although the HTC-RDC protocol prolongs the network lifetime by
The results show that in each experiment, the base station can collect only 11.4%, Table 4 shows that the FNL and HNL values are already

78
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

Fig. 17. The lifetime of the WSN with the MHT, RSTR, and HTC-RDC protocols under the same experimental conditions.

very close, which means there is little room for further optimization. References
Another fact is that in multihop networks, some nodes with a lower
hierarchy level must be relay nodes and bear a greater load, so re- [1] D. Miorandi, S. Sicari, F. De Pellegrini, I. Chlamtac, Internet of things: Vision,
gardless of the applied routing protocol, their energy will dissipate applications and research challenges, Ad Hoc Netw. 10 (7) (2012) 1497–1516.
earlier. Therefore, achieving an 11.4% improvement is an effective [2] R. Manchanda, K. Sharma, Energy efficient compression sensing-based clustering
framework for IoT-based heterogeneous WSN, Telecommun. Syst. 74 (3) (2020)
improvement.
311–330.
In addition, if applied in practical applications, the HTC-RDC pro- [3] C. Zygowski, A. Jaekel, Optimal path planning strategies for monitoring coverage
tocol will have three other merits: (1) The application of the schedule holes in wireless sensor networks, Ad Hoc Netw. 96 (2020) 101990.
will result in considerable energy savings compared with traditional [4] J. Amutha, S. Sharma, J. Nagar, WSN strategies based on sensors, deployment,
protocols. Each node can turn its radio module off when it does not sensing models, coverage and energy efficiency: Review, approaches and open
need to send or receive a signal according to the schedule. For example, issues, Wirel. Pers. Commun. 111 (2) (2020) 1089–1115.
without a schedule, the nodes must remain awake to listen to the [5] M. Younis, K. Akkaya, Strategies and techniques for node placement in wireless
sensor networks: A survey, Ad Hoc Netw. 6 (4) (2008) 621–655.
environment, gather information from other nodes and potentially
[6] M.Y. Aalsalem, W.Z. Khan, W. Gharibi, M.K. Khan, Q. Arshad, Wireless sensor
receive data. (2) The HTC-RDC protocol is a centralized method, so the networks in oil and gas industry: Recent advances, taxonomy, requirements, and
base station can control every node in the network. This will make it open challenges, J. Netw. Comput. Appl. 113 (2018) 87–97.
possible to develop more functions for the network. (3) The network [7] Y. Mekonnen, S. Namuduri, L. Burton, A. Sarwat, S. Bhansali, Review—Machine
using the HTC-RDC protocol does not require redundant nodes, which learning techniques in wireless sensor network based precision agriculture, J.
will reduce the cost of the network and make it easier to use. Electrochem. Soc. 167 (3) (2020) 037522.
[8] A. Shahanaghi, Y. Yang, R.M. Buehrer, Stochastic link modeling of static wireless
sensor networks over the ocean surface, IEEE Trans. Wireless Commun. 19 (6)
7. Conclusion (2020) 4154–4169.
[9] A. Alaiad, L. Zhou, Patients’ adoption of WSN-based smart home healthcare
In this paper, a new application scenario of WSNs is proposed based systems: An integrated model of facilitators and barriers, IEEE Trans. Prof.
on the requirements of practical applications. In this application sce- Commun. 60 (1) (2017) 4–23.
nario, (1) each sensor node is crucial, (2) the data to be collected from [10] S.-K. Yang, Y.-M. Shiue, Z.-Y. Su, C.-G. Liu, A novel authentication scheme
the sensor nodes are raw data instead of aggregated data, and (3) some against node captured attack in WSN for healthcare scene, in: 2019 IEEE Eurasia
Conference on Biomedical Engineering, Healthcare and Sustainability, ECBIOS,
of the sensor nodes cannot directly communicate with the base station. IEEE Press, 2019, pp. 39–42.
For this application scenario, a mathematical model based on optimiza- [11] G.M.E. Rahman, K.A. Wahid, LDAP: Lightweight dynamic auto-reconfigurable
tion theory is formulated. Since the model cannot be accurately solved protocol in an IoT-enabled WSN for wide-area remote monitoring, Remote Sens.
when the scale of the network is large, a new centralized generated 12 (19) (2020) 3131.
protocol with a hybrid tree and cluster structure for raw data collection, [12] S. Al-Sodairi, R. Ouni, Reliable and energy-efficient multi-hop LEACH-based
termed HTC-RDC, is proposed. The experimental results indicate that clustering protocol for wireless sensor networks, Sustain. Comput. Inform. Syst.
20 (2018) 1–13.
the WSN with the proposed protocol can collect raw data from all the
[13] M. Kaur, A. Munjal, Data aggregation algorithms for wireless sensor network: A
nodes and that the lifetime of the WSN can be prolonged by an average review, Ad Hoc Netw. 100 (2020) 102083.
of 11.4% compared with that achieved by the existing approaches. In [14] V. Seedha Devi, T. Ravi, S.B. Priya, Cluster based data aggregation scheme for
typical cases, the protocol can achieve satisfactory performance. latency and packet loss reduction in WSN, Comput. Commun. 149 (2020) 36–43.
[15] C. Thomson, I. Wadhaj, Z. Tan, A. Al-Dubai, Towards an energy balancing
CRediT authorship contribution statement solution for wireless sensor network with mobile sink node, Comput. Commun.
170 (2021) 50–64.
[16] R.R. Rout, S.K. Ghosh, Adaptive data aggregation and energy efficiency using
Yangbin Zhang: Conceptualization, Methodology, Software. Lihua
network coding in a clustered wireless sensor network: An analytical approach,
Liu: Conceptualization, Funding acquisition, Resources, Supervision, Comput. Commun. 40 (2014) 65–75.
Writing – review & editing. Mao Wang: Validation, Writing – review & [17] W.R. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficient communi-
editing. Jibing Wu: Validation, Writing – review & editing. Hongbin cation protocol for wireless microsensor networks, in: Proceedings of the 33rd
Huang: Supervision, Writing – review & editing. Hawaii International Conference on System Sciences, IEEE Press, 2000, pp.
223–233.
Declaration of competing interest [18] W.B. Heinzelman, A.P. Chandrakasan, H. Balakrishnan, An application-specific
protocol architecture for wireless microsensor networks, IEEE Trans. Wireless
Commun. 1 (4) (2002) 660–670.
The authors declare that they have no known competing finan- [19] D. Zhixiang, Q. Bensheng, Three-layered routing protocol for WSN based on
cial interests or personal relationships that could have appeared to LEACH algorithm, in: Wireless, Mobile and Sensor Networks, IET Conference on.
influence the work reported in this paper. IET, Stevenage, 2007, pp. 72–75.

79
Y. Zhang, L. Liu, M. Wang et al. Computer Communications 188 (2022) 66–80

[20] J. Hong, J. Kook, S. Lee, D. Kwon, S. Yi, T-LEACH: The method of threshold- Yangbin Zhang received his B.S. degree in electronic engi-
based cluster head replacement for wireless sensor networks, Inform. Syst. Front. neering from the National University of Defense Technology
11 (5) (2009) 513–521. (NUDT), Changsha, China, in 2009. He is currently pursuing
[21] J. Bagherzadeh, M. Samadzamini, A clustering algorithm for wireless sensor the M.S. degree in management science and engineering
networks based on density of sensors, in: International Conference on Advances with NUDT. His research interests include Internet of things
in Mobile Computing & Multimedia, ACM, 2009, pp. 594–598. and data engineering.
[22] J. Jia, Z. He, J. Kuang, Y. Mu, An energy consumption balanced clustering
algorithm for wireless sensor network, in: 2010 6th International Conference
on Wireless Communications Networking and Mobile Computing, WiCOM, IEEE,
2010, pp. 1–4.
[23] S.H. Kang, T. Nguyen, Distance based thresholds for cluster head selection in
wireless sensor networks, IEEE Commun. Lett. 16 (9) (2012) 1396–1399.
[24] F. Fanian, M. Kuchaki Rafsanjani, Cluster-based routing protocols in wireless
sensor networks: A survey based on methodology, J. Netw. Comput. Appl. 142 Lihua Liu received her B.S. and Ph.D. degrees in infor-
(2019) 111–142. mation and communication engineering from the National
[25] S.A. Sert, H. Bagci, A. Yazici, MOFCA: Multi-objective fuzzy clustering algorithm University of Defense Technology (NUDT), Changsha, China,
for wireless sensor networks, Appl. Soft Comput. 30 (2015) 151–165. in 2004 and 2012, respectively. She is currently an Associate
[26] R. Ge, H. Zhang, S. Gong, Improving on LEACH protocol of wireless sensor Professor in the Science and Technology on Information Sys-
networks using fuzzy logic, J. Inform. Comput. Sci. 7 (3) (2010). tems Engineering Laboratory, NUDT. Her research interests
[27] H. Bagci, A. Yazici, An energy aware fuzzy approach to unequal clustering in include target recognition, data fusion, data engineering,
wireless sensor networks, Appl. Soft Comput. 13 (4) (2013) 1741–1749. and intelligent decision.
[28] K. Sundaran, V. Ganapathy, P. Sudhakara, Fuzzy logic based unequal clustering
in wireless sensor network for minimizing energy consumption, in: 2017 2nd In-
ternational Conference on Computing and Communications Technologies, ICCCT,
IEEE, 2017, pp. 304–309.
[29] J. Yang, M. Xu, W. Zhao, B. Xu, A multipath routing protocol based on clustering
and ant colony optimization for wireless sensor networks, Sensors 10 (5) (2010)
4521–4540. Mao Wang received his B.S. degree in computer science
[30] M. Karimi, H.R. Naji, S. Golestani, Optimizing cluster-head selection in wireless and technology from Beijing Institute of Technology (BIT),
sensor networks using genetic algorithm and harmony search algorithm, in: 20th Beijing, China, in 2011. Then he received his M.S. and
Iranian Conference on Electrical Engineering, IEEE, 2012, pp. 706–710. Ph.D. degrees in computer science and technology from
[31] M. Shokouhifar, A. Jalali, A new evolutionary based application specific routing the National University of Defense Technology (NUDT),
protocol for clustered wireless sensor networks, AEU - Int. J. Electron. Commun. Changsha, China, in 2013 and 2018, respectively. He is
69 (1) (2015) 432–441. currently a Lecturer in the Science and Technology on
[32] P. Maheshwari, A.K. Sharma, K. Verma, Energy efficient cluster based rout- Information Systems Engineering Laboratory, NUDT. His
ing protocol for WSN using butterfly optimization algorithm and ant colony research interests include image retrieval, and evolutionary
optimization, Ad Hoc Netw. 110 (2021) 102317. computation.
[33] M.O. Oladimeji, M. Turkey, S. Dudley, HACH: Heuristic algorithm for clustering
hierarchy protocol in wireless sensor networks, Appl. Soft Comput. 55 (2017)
452–461.
[34] N. Mittal, U. Singh, B.S. Sohi, An energy-aware cluster-based stable protocol for
wireless sensor networks, Neural Comput. Appl. 31 (11) (2019) 7269–7286. Jibing Wu received his B.S. and M.S. degrees in information
[35] S. Gajjar, M. Sarkar, K. Dasgupta, FAMACROW: Fuzzy and ant colony optimiza- system engineering from the National University of Defense
tion based combined mac, routing, and unequal clustering cross-layer protocol Technology (NUDT), Changsha, China, in 2010 and 2012,
for wireless sensor networks, Appl. Soft Comput. 43 (2016) 235–247. respectively, and the Ph.D. degree in management science
[36] Z.M. Zahedi, R. Akbari, M. Shokouhifar, F. Safaei, A. Jalali, Swarm intelligence and engineering from NUDT, in 2017. He is currently
based fuzzy routing protocol for clustered wireless sensor networks, Expert Syst. an Associate Professor in the Science and Technology on
Appl. 55 (2016) 313–328. Information Systems Engineering Laboratory, NUDT. His re-
[37] M. Shokouhifar, A. Jalali, Optimized sugeno fuzzy clustering algorithm for search interests include heterogeneous information network,
wireless sensor networks, Eng. Appl. Artif. Intell. 60 (2017) 16–25. information management, and computational intelligence.
[38] Y. Xu, J. Heidemann, D. Estrin, Geography-informed energy conservation for ad
hoc routing, in: International Conference on Mobile Computing & Networking,
ACM, 2001, pp. 70–84.
[39] R. Almesaeed, A. Jedidi, Dynamic directional routing for mobile wireless sensor
networks, Ad Hoc Netw. 110 (2021) 102301.
[40] Q. He, J. Mou, B. Lin, A robust self-organizing tree-based routing protocol for Hongbin Huang received his B.S. and Ph.D. degrees in
wireless sensor networks, Math. Probl. Eng. 2021 (2021) 1–13. information system engineering from the National Univer-
[41] M. Yuxing, Z. Huiyuan, Y. Dongmei, Weak node protection to maximize the sity of Defense Technology (NUDT), ChangSha, China, in
lifetime of wireless sensor networks, J. Syst. Eng. Electron. 29 (4) (2018) 1996 and 2007, respectively. He is currently a Professor
693–706. in the Science and Technology on Information Systems En-
[42] P. Jacquet, P. Muhlethaler, T. Clausen, A. Laouiti, A. Qayyum, L. Viennot, gineering Laboratory, NUDT. His research interests include
Optimized link state routing protocol for ad hoc networks, in: Proceedings. IEEE cyber–physical system, intelligent decision, computational
International Multi Topic Conference, 2001. IEEE INMIC 2001. Technology for intelligence, and information management.
the 21st Century, IEEE, NEW YORK, 2001, pp. 62–68.
[43] C.E. Parkins, Ad hoc on-demand distance vector routing, in: Proc IEEE Wmcsa99
Feb, Vol. 6, no. 3, 1999, pp. 94–95.

80