You are on page 1of 12

Applied Soft Computing Journal 80 (2019) 285–296

Contents lists available at ScienceDirect

Applied Soft Computing Journal


journal homepage: www.elsevier.com/locate/asoc

New quantum-genetic based OLSR protocol (QG-OLSR) for Mobile


Ad hoc Network

De-gan Zhang, Yu-ya Cui , Ting Zhang
Key Laboratory of Computer Vision and System (Tianjin University of Technology), Ministry of Education, Tianjin 300384, China
Tianjin Key Lab of Intelligent Computing & Novel software Technology, Tianjin University of Technology, Tianjin 300384, China

highlights

• Put forward new improved quantum genetic strategy.


• Present novel Q-Learning strategy has been embedded in our improved quantum genetic strategy.
• New protocol overcomes the shortage of the traditional protocol.
• Prove the property of convergence and global optimization of the algorithm.
• Improve the performance of transmission.
• Support the mobility, self-organization and distributed control of MANET (Mobile Ad hoc Network).

article info a b s t r a c t

Article history: Due to the mobility, self-organization and distributed control of Mobile Ad hoc Network (MANET),
Received 4 November 2017 the routing protocol of MANET must adapt to the rapid changes of the network structure, and
Received in revised form 23 February 2019 ensure the maximum savings of network resources. This paper proposes a kind of new quantum-
Accepted 25 March 2019
genetic based OLSR protocol (QG-OLSR) for MANET. The protocol adopts the MPR (multi-point relay)
Available online 16 April 2019
technology in OLSR (Optimal Link State Routing). By embedding new augmented Q-Learning algorithm
Keywords: and combining the OLSR algorithm to optimize the selection of MPR sets, it can effectively reduce the
Protocol consumption of network topology control, improve the delivery rate of data packets, and reduce the
MANET time delay of the end-to-end packet transmission between nodes. The results of the experimental
Optimal link state routing testings show that the new protocol is reliable and highly efficient, which is suitable for many
Multi-point relay applications of MANET.
Quantum genetic © 2019 Elsevier B.V. All rights reserved.

1. Introduction for notification of link state in the network [14–16]. The infor-
mation with link state is generated by the nodes selected as
Mobile Ad hoc Network (MANET) is composed of a group MPR, which can reduce the flooding of information. And the
of logically equivalent nodes with wireless transceiver device, MPR node receives & forwards the TC message from the other
it does not rely on any basis set [1–3]. Due to the high dy- MPR node. When the MPR node sends its own TC message, the
namic property of the topology, the traditional routing protocol MPR node also needs to add the information of being selected
is not applicable. So the routing problem of MANET is one of as MPR by the neighbor node in the TC packet. In Refs. [17–
hot topics [4–8]. OLSR (Optimized Link State Routing) is a typical 19], the problem of flooding is considered for a specific physical
table-driven routing protocol and is different from the traditional layer. In Refs. [19–21], the flooding problem of heterogeneous
LSR (Link State Routing), its key function is the MPR (Multiple MANET is studied. In Refs. [22,23], MPR is applied to multicast
Point Relay) strategy. The MPR strategy is very important and
protocols. In Refs. [24–26], it is proved that the MPR selection is
useful when the node density is bigger in the network, so OLSR
an NP-complete problem. However, the above references do not
is very suitable for the network with high node density [9–13].
consider redundant information, topology control overhead, and
In OLSR, each node selects a part of its own hop node as
transmission efficiency in a network in a high-density network
its own MPR node set. Only the node selected as MPR will for-
ward the Topology Control (TC) packet. MPR node is responsible environment.
The concept of quantum computation is put forward by Be-
∗ Corresponding author at: Key Laboratory of Computer Vision and System nioff and Feynman [17–20]. Currently, there have been many
(Tianjin University of Technology), Ministry of Education, Tianjin 300384, China.
quantum computing methods [21–25]. Quantum Genetic Algo-
E-mail addresses: gandegande@126.com (D.-g. Zhang), rithm (QGA) [26–29] is based on the vector representation of
2310674826@qq.com (Y.-y. Cui). quantum states, and the probability amplitude of qubits is used

https://doi.org/10.1016/j.asoc.2019.03.053
1568-4946/© 2019 Elsevier B.V. All rights reserved.
286 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296

in gene coding. Therefore, each chromosome can performance a directional information, the Grover’s search algorithm can find
superposition of multiple states at a certain time, and the gene is the specified target item in an unordered √ table with directory
updated by a quantum spine. The quantum solution and the local entries. In the time complexity of O( n) to find the specified
optimal solution of the mutated population are used to optimize target, but before the time complexity of the classic algorithm is
the target solution. The parallel nature of quantum computing can O(n). The problem of solving 0–1 knapsack problem in Ref. [17] is
be used to derive large data factors in a small time complexity, that each gene has only ‘‘0’’ or ‘‘1’’ state. The problem is simple,
which can solve the NP-C problem [29–32]. Therefore, QGA can but in other problem solving, each gene may exist multiple states,
effectively solve multi-objective optimization problems such as so the proposed algorithms are not universal. The solution is
mobile WSN and MANET. The current research focuses on two to design a unitary transformation that can operate on multiple
types of models: Quantum Inspired Genetic Algorithm based on states at the same time. This method has the advantages of sim-
ple encoding, high computational efficiency, but computational
quantum multi-cosmic features, QGA based on Quantum Genetic
complexity is large, and the design of high latitude unitary trans-
and Quantum Genetic Superposition (QGA).
formation is troublesome [28–31]. Another solution is the use of
Q-Learning algorithm [31–34] is an unsupervised active learn-
binary coding in genetic algorithms to quantify the presence of
ing algorithm, which does not require a specific system model,
polymorphic bits, such as two states encoded with a single qubit
and can be adapted to different environments by interacting with and four states encoded with two bits, and the characteristics
the surroundings. At the same time, it is a reinforcement learn- of the method is better versatility, and to achieve easier [32–
ing and heuristic learning algorithm, which belongs to machine 34]. Quantum bit coding makes it possible for a chromosome to
learning method. The learning process of the node is as follows: express multiple states at the same time, which makes QGA have
The source node finds the path to the target node with the largest better diversity characteristics than classical algorithm. Quantum
return value by periodically updating the state activity value (Q- bit coding can also obtain better convergence. With factor or |α|2
Value) in different states. It can adaptively adjust Q-Table with orientation with 0 or 1, the chromosome encoded by the qubit
ensuring the reliability of each hop link to adapt such dynamic will converge to a single state.
network topology of MANET [35–40]. New augmented Q-Learning From the above analysis, we can see that the existed QGA is
strategy has been embedded in our improved QGA. not universal, only to solve specific problems, not suitable for
Our improved QGA adopts the multi-state factor bit coding OLSR routing in the selection of MPR collection and the entire
method to encode the nodes of MANET. The quantum cross- network performance of MANET [35–40]. In this paper, a novel
operation and the quantum not-gate are used to realize the quantum genetic strategy named QG-OLSR protocol based on im-
genetic mutation. The quantum rotation gate and dynamic ad- proved quantum genetic strategy is proposed, and its correctness
justment of the rotation angle strategy are adopted to consider and validity are proved by theory and experiment. The algorithm
the nodal energy information, which avoid premature and local adopts the multi-state factor bit coding method to encode the
convergence. For the NP-completeness of the MPR problem, an nodes in the network. The quantum cross-operation and the
improved global genetic algorithm is used to obtain the global quantum not-gate are used to realize the genetic mutation. The
optimal solution. quantum rotation gate strategy and the dynamic adjustment of
the rotation angle strategy are adopted to consider the nodal
energy information, which avoid premature and local conver-
2. Related works
gence. For the NP-completeness of the MPR problem, an improved
global genetic algorithm is used to obtain the global optimal
When the network of MANET is very sparse, OLSR degraded
solution [41–44]. Because the calculation of MPR is carried out
to the general LSR protocol. In Ref. [15], the image acquisition independently by each node, only the information of link state
method is applied to the MANET, and the node information is within two hops is considered. Therefore, even though the net-
collected by the image instead of only the OLSR, an OLSR pro- work is very large and the nodes are very dense, the topology
tocol based on image enhancement is proposed. The protocol in information of two hops does not consume a large amount of
Ref. [20] can find a better solution of the MPR set problem in a energy and can achieve a better calculation efficiency. Since the
short time defined as O(n log(n)), but because of the limitations of selection of MPR sets is optimized and the information such as
the greedy protocol, the solution is not necessarily the optimal so- node energy is taken into account in the selection process, the
lution. In Ref. [21], they proposed a new Neighborhood State Self- redundant information such as topology control information in
Adaptive Update (NSSAU) protocol based on the loop control the- the network will be reduced, resulting in a decrease of network
ory to establish the mapping relationship between network topol- topology control overhead and an increase of effective packet de-
ogy changes and neighbor-like updates. The OLSR-NSSAU routing livery rate [45–48]. The average end-to-end delay of the effective
protocol is proposed in conjunction with the OLSR protocol. packet becomes shorter.
Shor proposed quantum decomposition strategy and Grover It is well known that Q-Learning strategy used as a kind of self-
proposed database search strategy [17–21]. Shor proposed quan- learning mechanism can find a shortest path from source node
tum decomposition strategy and the use of quantum computing to destination node through constant interaction with outside of
parallelism can quickly decompose large numbers of prime fac- MANET. Based on this idea, our improved QGA is adopted new
tors . However, the optimal time complexity of traditional classi- augmented Q-Learning strategy in this paper.
cal algorithm is . This will make it easy for the quantum computer
3. New QG-OLSR protocol
to crack the widely used RSA public key encryption system, it will
be a serious threat to the banking, network and other information The new optimized link state routing protocol (we name QG-
security and national security. Therefore, Shor’s quantum decom- OLSR protocol) is based on our improved QGA, which can reduce
position algorithm has attracted worldwide attention. Grover’s the redundant information in the MANET. The global convergence
database search algorithm can provide substantial secondary of the MPR set selection is guaranteed by quantum crossing
acceleration for many (but not all) heuristic algorithms. The and quantum not-gate variation, and the quantum rotation gate
Grover quantum search algorithm neglects the nature of the is used to update the data to improve the data transmission
search elements in the search, and focuses on those elements efficiency in the topology of MANET.
of the index. At the same time, it can also effectively decipher The QG-OLSR protocol is based on the hierarchical routing
the DES encryption system, with the potential to accelerate the approach with the idea of Edsger Dijkstra’s shortest path, a router
search password system key purposes [22–27]. When searching in a MANET does not transfer the ‘‘route entry’’ to the neighbor,
D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296 287

but advertises some link status to the neighbor. Instead of simply Where |α|2i + |β|2i = 1, i = 1, 2, . . . , m, this representation has
learning routes from neighboring routers, the router running the the advantage that it can represent any state of the superposition,
routing protocol divides the routers into areas, collects link state that is, if a system has m quantum bits, then the system at a time
information for all routers in the area, and generates the topology that can be 2m states. For example, the amplitude of a three-bit
of MANET based on the state information. representation is as follows:
The quantum heredity means that quantum state vector ex- 1.0
[ ]
√1 1
2 2

pression is used to genetic coding, the use of quantum revolving (3)
door to achieve the evolution of chromosomes, to achieve better
√1 1.0 2
3
2
results than that of conventional genetic algorithm. However, Then the system state can be expressed as:
since all individuals evolve toward a target, there is a high prob- √ √
ability that they will fall into the local optimum if there is no 1 3 1 3
√ |000⟩ + √ | 001⟩ + √ | 100⟩ + √ | 101⟩ (4)
crossover operation. 2 2 2 2 2 2 2 2
The quantum crossover means that according to the rapid The above results indicate the probability of the system is in
evolution of the population characteristics of the use of quantum state 000, state 001, state 100, and state 101 is 1/8, 3/8, 1/8, and
bit population crossing operation, it takes the first locus of an 3/8, respectively. So the gene chain with three bit vectors can
individual as the first locus of the new individual, and takes the represent four states at the same time. Due to the state superpo-
second locus of the adjacent individual as the second locus of the sition of the bits, the evolutionary computation represented by
new individual, repeating until the child have the same number the bits is more diverse than the conventional method. In Eq. (3),
of individuals with parents. A descendant is generated for each only one gene chain can represent the four states, but using the
individual until the number of offspring and parent are equal. traditional method requires at least four gene chains to represent,
The quantum not-gate variation means that a qubit has two namely (000), (001), (100) and (101). According to the properties
independent states, ‘‘0’’ and ‘‘1’’, α and β are the probability for of Eq. (1) and |α|2i + |β|2i = 1, i = 1, 2, . . . , m, it can be seen that
states amplitude of the two states, respectively which indicates as the value of |α|2 or |β|2 tends to 0 or 1, the bit gene chain is
the probability amplitude of the two states, respectively. The converged to an independent state, and the diversity disappear
quantum non-gate is an interchange of the probability amplitude slowly.
of collapse to ‘‘0’’ and the probability amplitude of collapse to ‘‘1’’.
3.2. New improved QGA
3.1. Principle of QGA
QGA utilizes quantum bits to preserve the chromosomal genes
of individuals in a population.{In the t-generation, the population
QGA is based on the combination of quantum computation and can be expressed as Q (t) = qt1 , qt2 , . . . , qtn , wherenis the total
}
genetic algorithm. QGA is an algorithm based on the superposi- number of individuals in the population at the tth generation, qti
tion of quantum bit and quantum state, and uses polymorphic is the ith individual, and each individual represents the MPR set of
gene bit coding method and quantum rotation gate to find the the node, and the chromosome gene chain of the individual m as:
optimal solution, which has the advantages of fast convergence
α1t ⏐⏐ α2t ⏐⏐ ...⏐⏐ αmt
[ ⏐ ⏐ ⏐ ]
and global optimal solution. qti = (5)
The basic genetic algorithm adopts the proportional selection β1t ⏐ β2t ⏐ ...⏐ βmt
strategy. The probability of crossover and the mutation respec- Where i = 1, 2, . . . , m, m is the number of quantum bits, that is,
tively are Pc and Pm , whose values are the smallest, definition the length of chain of quantum chromosomes.
length of the pattern H is δ (H), and its order is o(H). The pop- A kind of new augmented Q-Learning strategy has been em-
ulation of the generation contains m(H , t + 1), and the following bedded in our improved quantum genetic strategy. Different from
inequality holds. traditional way of Q-learning, each node has a Q-Table (as ML(i, j))
in our new augmented Q-Learning strategy. Nodes complete the
f (H) δ (H)
m(H , t + 1) ≥ m(H , t) × × (1 − Pc × − Pm × o(H)) task of learning by exchanging beacon information and updating
f¯ l−1 Q-Table. The beacon packet sent by nodes contains not only its
In the basic genetic algorithm, data grows exponentially in own speed, location and other information, but also the maxi-
which order of low, defined length of short and the fitness value mum Q-Value of neighbor nodes to the destination node, i.e., the
maximum value in a column. Q-Table is a two-dimensional table,
are greater than the average fitness value. In contrast, data de-
whose size is determined by the number of neighbor nodes and
creases exponentially which the fitness value is less than the
the number of destination nodes. It is obvious that it has a
average fitness value.
good scalability. The value in Q-Table is updated by periodically
QGA minimum information storage unit is a quantum com- exchanging beacon packets between nodes. The task of learning
puter with only two states, called qubits. A qubit has only ‘‘0’’ is distributed to each node, which makes the approach quickly
and ‘‘1’’ two independent states, it can also be a superposition of converge to the optimal path, and the changes of the topology of
two states, can be expressed as: MANET can be timely adjusted. In our improved QG algorithm, we
have added the convergence condition of the approach to keep it
|ψ⟩ = α|0⟩ + β|1⟩ (1)
suitable to the mobile or distributed computing environment of
Where α and β are complex numbers representing the probability MANET.
amplitudes of the two corresponding states. |α|2 represents the Our studied Q-Learning environment is that the whole
probability when the qubit state is ‘‘0’’, and |β|2 represents the MANETs environment as the learning environment of agent; The
probability when the qubit state is ‘‘1’’. In [the]QGA, a bit is defined learning agent is each node of MANET; State space S is composed
α of other nodes except for itself; Activity spaceA is that the beacon
as a set of complex(α, β), expressed as , α and β according packet transmitted from one node to another node is defined as
β
to Eq. (1), then a chain of chromosomal genes with m bits can an activity; Immediate reward R is that the agent carries out an
activity and get, its range is [0, 1]. Because the destination node
be expressed as:
can directly reach the destination node, the reward value is 1.
α1 ⏐⏐ α2 ⏐⏐ . . .⏐⏐ αm
[ ⏐ ⏐ ⏐ ]
Eq. (6) defines the initial value R of the entire focus network case
(2)
β1 ⏐ β2 ⏐ . . .⏐ βm of MANET as follows:
288 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296
{
R=
1 s ∈ Nd
(6) node, R is the reward value, maxy∈Nx Qx (d, y) is the maximum Q-
0 else Value betweenx and its neighbor node to the destination node
D. Discount factor γ is an important parameter and it affects
where Nd represents one hop neighbor node set of destination
the reward value which a node obtains from carrying out an
node D. The reward value of an activity for all the neighbor nodes
activity according to the equation above. Considering that the link
of the destination node is 1. In learning process, the reward value
stability is an important parameter, we put the link reliability r (l)
which can be obtained from a state transition to another state is
between nodes, as a discount factor, that is γ = r (l). In MANETs,
indicated by Q-Value Q (s, a) (s ∈ S , a ∈ A) and its range is [0, 1]. the available bandwidth, as an important parameter, determines
Each learning agent maintains a two-dimensional table, which the rate of packet transmission. As defined in [31] the bandwidth
is used to record the Q-Value of the reachable destination node BW is calculated as follows:
and its one hop neighbor node. This two-dimensional table is
n × SB × 8
called Q-Table (shown in Table 1). In Q-Table, the first line repre- BW (bps) = (8)
sents for all possible destination nodes’ ID, which is expressed by T
Di ; the first column shows the one hop neighbor nodes’ ID, which where n represents the number of packets which the node send
is expressed by Ni .Q (D1 , N1 ) represents the Q-Value between the and receive, SB is the size of the packet and represented in byte,
node and the neighbor node N1 when it reaches the destination T is the time interval. Assuming that the maximum available
node D1. The size of Q-Table as shown in Table 1 is determined bandwidth of the node is a fixed value, which is defined as
by the number of neighbor nodes of MANET and the number of maxBW , the bandwidth factor can be calculated as follows:
destination nodes. It is obvious that Q-table has a good scalability. maxBW −BW
BF = (9)
The value in Q-Table is updated by periodically exchanging bea- maxBW
con packets between nodes. The task of learning is distributed to
As the factor that affects learning speed, the bandwidth factor
each node, which makes the algorithm quickly converge to the
is changed with the change of effective bandwidth, and deter-
optimal path, and the changes of the topology of MANET can be mines the learning progress of each node of MANET. We modify
timely adjusted. Eq. (7) to obtain a new heuristic function as follows:
Learning tasks are assigned to each node agent of MANET, { }
and the learning process is mainly updating the Q-Table in the Qs (d, x) ← (1 − BF ) · Qs (d, x) + BF · R + γ · max Qx (d, y) (10)
agent, meanwhile, updating the pair of state activity of Q-Value y∈Nx

Q (s, a) (s ∈ S , a ∈ A). The standard Q-Learning function is shown According to Eq. (10), considering that the bigger the hops
in Eq. (7) as follows: number is, the smaller the reward value is. So the final reward
value is based on the number of hops, link reliability and band-
{ }
Qs (d, x) ← Qs (d, x) + R + γ · max Qx (d, y) (7) width. By adding bandwidth and link state, the optimal path from
y∈Nx
source node to destination node can be obtained in the dynamic
where Qs (d, x) is the Q-Value to be updated, S is the node, x is network.
its neighbor node, Nx is x’s neighbor node. D is the destination Our improved QG algorithm may be described as follows:
D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296 289

Table 1 denoted by N12 , N22 , . . . , Nn2


2
, where n2 is the number of two-hop
Q-Table. neighbor nodes. Normally, n2 is a multiple of{ m.
Because the selection of set P(t), P(t) = qt1 , qt2 , . . . , qtn is a
}
D1 D2 ...
N1 Q(D1,N1) Q(D2,N1) ... better set of Q (t) and qti is the MPR set of node i. For a node i,
N2 Q(D1,N2) Q(D2,N2) ...
according to each node in the two-hop neighbor table NB_table2
... ... ... ...
of the node i, the node i selects the nodes covering them in the
one-hop neighbor table NB_table1 of the node i, and selects these
one-hop neighbors into the MPR set of the node i.
3.3. Initialization & encoding If the jth hop neighbor node of the node i is selected into the
MPR set of i, the jth hop neighbor corresponding to the gene chain
During evolutionary computation, there are many different of the node i is obtained in the chromosome gene chain of the
encoding ways to encode genes. The classical encoding methods node i after initializing 0–1 coding. The probability amplitude of
include sequence encoding, real encoding, adaptive coding, out- state 1 is set to 1, βjt = 1, the probability amplitude of state 0 is
of-order encoding, symbol encoding. For the gene chain of QGA set to 0, αjt = 0, and the rest continues to follow the initialization
for OLSR put forward in this paper, we use the 0–1 method principle, αit and βit (i = 1, 2, . . . , m, i ̸ = j) hold the initialization
to encode, considering the global and genetic cross mutation value √1 .
2
problem, setting chromosome length equal to the total number This method of setting the value is in accordance with the
of nodes in the topology of MANET. property |α|2i + |β|2i = 1, i = 1, 2, . . . , m, and has a heuristic
By the using initialize Q (t) to do the initialization work, the effect. Due to the fact that nodes want to select the one-hop
probabilities of all the one-hop neighbor nodes of i are equal to neighbor nodes that cover more than two-hop neighbors as their
that of the MPR set of i for node i, so the probability amplitude of own MPR set, then one-hop neighbor node j is likely to be in
the chromosomal locus of all individuals qti in population is Q (t) the smallest MPR set of node i. In fact, according to the two-hop
and αit are both initialized to the value √1 . That is to say, for the neighbor table, the MPR set is selected in the node i hop node. The
2
node i, its initial chromosome gene chain is expressed as follows: more the two-hop neighbors are covered by one-hop neighbor
⎡( )0 ⏐⏐ ( )0 ⏐⏐
. . .⏐ √1 0
⏐( ) ⎤ nodes, the greater the probability of the two-hop neighbors being
⏐ √1 ⏐
√1 selected. So the virtuous loop owns a heuristic effect.
. . .⏐ 2 m ⎥

⎢ 2 ⏐ 2
q0i = ⎣ ( )10 ⏐ ( )20 ⏐

1 1
. . .⏐⏐ ( 1 )0 ⎦
3.4.2. On fitness function
⏐ ⏐

2
⏐ √2 ⏐
1 2
. . . ⏐ √2 m
The smallest MPR set in the OLSR is typically considered
Where i = 1, 2, . . . , m, m is the total number of nodes in to reduce flooding messages in the MANET, thereby reducing
the topology of MANET. A qubit of the chromosomal gene chain overall energy consumption. However, only minimizing the MPR
qti |t = 0 represents a linear superposition of all possible states setting standard may result in excessive energy consumption of
with the same probability, expressed by the equation as follows: some nodes in the MPR set, resulting in partial network failure,
2m affecting the overall data transmission efficiency and lifetime
∑ 1 of the network. Therefore, it is necessary to comprehensively
|ψq0j ⟩ √ |Sk ⟩ (11)
2m minimize the MPR set and make a comprehensive individual
k=1
adaptive assessment of the node energy information. According to
Where Sk is the state, represented by the binary string ⟨x1, x2, . . . , the aforementioned heuristic selection rule, the fitness equation
xn⟩, xi is 0 or 1, and m is the chromosome gene chain length, of the node j in the set is as follows:
i = 1, 2, . . . , m.
Fitness_NB(i, j) = NB_table1.size − q[j].size (12)
3.4. Selection of initial population Where Fitness_talble1.size denotes the number of nodes in the
hop list of the node j, and q[j].size denotes the size of the MPR
According to the OLSR protocol, the MPR set of the source node set selected by the individual j in the population Q (t).
should satisfy the following conditions: (1) each terminal in the In Eq. (12), the fitness of node j is figured out because the
MPR set is a single-hop neighbor at the source end; (2) the source MPR set of node i is to be selected. Node j is a one-hop neighbor
end propagates the message to all two-hop neighbor endpoints node of node i. According to the above heuristic rule, if nodes
through the MPR set. The heuristic rule is used to select MPR set are selected as nodes in MPR set, which cover more than two-
of node i, which is mainly based on the one-hop and two-hop hop neighbor node, node j needs to be evaluated. Similarly to the
neighbor nodes table of node. In addition, considering the nodal following Eq. (13), the node j in the MPR set of i is selected, and
energy, the node fitness function is put forward, and the node the energy fitness function of node j needs to be evaluated.
i is selected as the node in the MPR set to evaluate the fitness. Selecting the MPR set based on the above heuristic rules may
According to the above two rules, the MPR set of the node i is result in excessive local energy consumption in the network. In
selected to ensure the optimal selection of the MPR set. And the this paper, according to the node energy information, the node
redundant information is transmitted less, the topology lifetime energy fitness function can be calculated as follows:
of MANET is longer.
N N1
∑ ∑
Fitness_ENE(i, j) = Ccst − Wij × Costij × (µj + c + ∆Ccst ) (13)
3.4.1. On heuristic rules
i j
The nodes in the topology of MANET are divided into the
following three categories: (1) Node i; (2) One-hop neighbor In Eq. (13), N is the total number of nodes in the network
node of node i and stored in one-hop neighbor table NB_table1, topology and N1 is the number of hop neighbors in node j.
denoted by N11 , N21 , . . . , Nn1
1
, where m is the one-hop neighbor Costij × (µj + c + ∆Ccst ) represents the cost of the entire link,
number of node i; (3) The two-hop neighbor node of node i, that Costij represents the cost of the edge, i.e. the energy consumed to
is to say, the nodes that need to reach by two hops, and do not send and receive data between the node i and the node j, µ is
include the nodes in one-hop neighbor node table, and store these the compression ratio, and c is an arbitrary constant, ∆Ccst is the
nodes in the two-hop neighbor node table NB_table2 of node i, noise volume. The cost of the edge between the nodes Costij can
290 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296

be expressed by radio terminal energy consumption ETX (k, dij ) 1


2
.
How can these surely be optimal values? The value of w1 and
and ERX (k, dij ) with the energy consumption of the noise ∆E as w2 were changed based on the different applications by machine
follows: learning, which introduced in the references [9–12]. Based on the
learning process, the optimal value can be obtained. If equation
Costij = ETx (k, dij ) + ERx (k, dij ) + ∆E (14)
(18) is treated as a multiple objective approach, it should be
Based on the radio energy consumption model as Eq. (14), the divided or extended the dimension of the parameters, so that
energy consumption per node for k bits is as follows: non-dominated solutions can be formed to be another dimension
(k, d) = ETx −elec(k) + ETx−amp (k, d) + ∆ETx (k, d)
ETx{ of thought for the design. In the simulation experiment or actual
application, according to different application scenarios, differ-
kEelec + kEfs × d2 + ∆Enoise , d < d0 (15)
= ent network performance requirements of MANET, different and
kEelec + kEmp × d4 + ∆Enoise , d ≥ d0
√ parameters can be set.
In Eq. (15), d0 = fs E
, k is the number of transmitted Based on the result of the node fitness function, the node j is
Emp
selected as the node in MPR set of node i, and the equation is as
packets, d is the transmission distance, ∆ETx (k, d) and ∆Enoise are
follows:
the energy consumption of the noise. When the distance between
the source node and the destination node of MANET is less than Fitness(i, j)
MPR (i, j) = ∑gen _size (19)
or equal to the threshold value d0 , the node adopts the free
j=1 Fitness(i, j)
space mode; when the distance between the source node and
the destination node of MANET is greater than the threshold Where gen _size is the total number of individuals in the popula-
value d0 , in contrast, the node adopts the multi-path attenuation tion, that is to say, gen _size is equal to the total number of nodes
mode; Eelec (nj/bit) is the radio frequency energy consumption in the network of MANET. So the selection probability that the
coefficient; Efs and Emp are the energy consumption coefficients node j is selected as the MPR of node i is MPR(i, j).
of the circuit amplifier in free space mode and multi-path fading In summary , the MPR set of node i is selected by the heuristic
mode respectively. The energy consumption value of the node rule, and the MPR set selected by the heuristic rule is filtered and
receiving a k-bit the message is calculated as follows: optimized by the fitness function of node j. Since the heuristic
rules only depend on the coverage of the nodes to select the
ERx (k) = ERx−elec (k) = kEelec (16)
MPR set, the energy information of the nodes of MANET is not
The weighting factor Wij in Eq. (13) is used to protect nodes taken into account. However, MPR sends TC control messages
with low energy of MANET, so that they do not have to send too periodically, which needs to consume energy. If only the MPR
much data. Since node j is optimized, node j is selected as the set judged by the distance information is selected, the probability
MPR set of node i, and node j with less energy can be protected. of some nodes being selected is large, and the energy consump-
The calculation of Wij equation is as follows: tion of these nodes is large. It may result in excessive energy
⎧ E Eav g consumption in the regional nodes. The topology of MANET is
j
× × (−2) Eav g /Ej ≤ 0.45
1 ∑k partially paralyzed and network data transmission efficiency is
⎪ 1 ∑k
i=1 Ei i=1 Ei

k k


affected. Therefore, the heuristic rule is adopted to select the MPR
⎨ E Eav g
j
wij = 1 ∑k E × 1 ∑k ×2 Eav g /Ej ≥ 1.9 (17)
⎪ k i=1 i k i=1 Ei nodes, according to the one-hop and two-hop neighbor table of

⎪ Ej Eav g nodes. And the distance fitness of the nodes of MANET is obtained
× others

⎩ 1 ∑k ∑k
1
k i=1 Ei k i=1 Ei according to the number of neighbors. And the energy fitness
Where Ej is the energy of current node j of MANET; Eav g is the of the node of MANET is obtained by using the node energy
average energy of all the hop nodes of node j; Ei is the energy fitness function containing the energy information. In the end, the
of the superior node i of the node j, the superior node is the optimal MPR set of each node is selected.
Selector MPR node of the node j; setting the parameters k, k is
the total number of nodes MPR Selector j. For making the node 3.5. Crosstalk and mutation
energy consumption more balanced, when the energy of node j
is more than about twice average energy of its one-hop neighbor
3.5.1. Crossover operation
node, Eav g /Ej ≤ 0.45, the weight will be multiplied by (-2) to
The traditional cross-gene operation methods include single-
reduce the cost of the whole link. And the node j is encouraged
cross, multiple-point cross, uniform cross, shuffle to cross, narrow
to become the MPR node of other nodes i of MANET. That is to
proxy cross [33,34]. However, the traditional crossover strategy
say, Eav g /Ej ≥ 1.9, the weight will be multiplied by about 1.9 to
increase the link cost, reduce the fitness of node j, and restrain it does not apply in respect of this approach, for example, in the
from becoming other node i MPR node. single-cross, there are continuous genetic locus of different in-
The Eq. (17) shows that Wij protects the node j with less dividuals’ gene chains, which can better reflect the individual
residual energy and suppresses it as the MPR node of other nodes traits and individual fitness. In the process of biological evolution,
i by using the weighted penalty method, then the node j does not the crossover rate of the offspring generated by a single point of
need to send more TC messages. Thus, the nodes and topology of crossing is small, which is only applicable to the evolutionary pro-
MANET will have longer lifetimes. cess of the environment with slow change, and is not applicable
Based on the node fitness function calculated by the above to the MANET with fast topology change. For the characteristics
heuristic rule and the nodal fitness function calculated from the of high dynamic topology of MANET, the quantum cross-policy
node energy information, the comprehensive fitness function of is selected. The steps include the following ones: numbering
the node of MANET can be calculated as follows: and sorting all nodes in the network topology; the first gene of
the first individual is taken as the first gene locus of the new
Fitness(i, j) = w1 × Fitness_NB(i, j) + w2 × Fitness_ENE(i, j) (18)
individual, and the second gene of the adjacent individual is taken
Where w 1 = 1/2, w 2 = 1/2 are the equilibrium heuristic as the second gene locus of the new individual, until the new
selection rule and the fitness of node energy of MANET. According individual has the same number of genes; do it until the new
to the statistical result, the default value of w1 and w2 were set to population has the same genetic size as the parent.
D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296 291

3.5.2. Mutation operation ∆θ i, S(α i, β i) represents the rotation direction, and ∆θ i repre-
After the probability of individual mutation for individual mu- sents the rotation angle. The rotation angle adjustment strategy
tation operation is calculated, and the probability of mutation is to compare the fitness f (xi) of the current measured value of
of the gene locus of the individual chain is calculated, then the the individual qti with the fitness value f (b i) of the current target
mutation operation of quantum not-gate is carried out. Due to value of the individual.
the low probability of mutation, some individuals may not mutate
at all, in order to reduce the waste of computing resources, the 3.7. Detailed steps of new QG-OLSR protocol
probability of mutation at the individual level is judged first,
and then the probability of mutation of the gene locus of in- The improved quantum genetic algorithm with augmented
dividual gene chain is judged. Similar descriptions in genetic Q-Learning strategy applied to the OLSR protocol can optimize
biotechnology refer to the probability of an individual infecting an the selection of the MPR set of the source node in the protocol.
infectious disease, the probability of chromosome mutation and It can effectively reduce the redundant information in the net-
the probability of inheritance to offspring after the disease has work topology, improve the packet delivery rate between nodes
been determined. and reduce the end-to-end delay. According to the aforemen-
Suppose that L is the individual length of the chromosome,N tioned equations and theories, the detailed steps of new QG-OLSR
is the number of individuals in the population, and pinitial is the protocol are as follows:
initial mutation probability. Then the probability of mutation of (1) Based on the improved QGA with new augmented Q-
individual i in the population can be calculated as follows: Learning strategy, each node in the topology of MANET is rep-
resented by every quantum bit of quantum chain. The initial
p(i) = 1 − (1 − pinitial )L (20)
population is denoted by Q (t) = qt1 , qt2 , . . . , qtn , where qti repre-
Where i = 1, 2, . . . , N. Set a random variable φ ∈ [0, 1], sents the gene chain of the ith individual in the population, Eq. (5)
when p(i) ≥ φ individual variation, p(i) < φ when not mutated is the representation of the individual gene chain.
sometimes. Due to the variation, the mutation probability of the (2) Initialization and encoding. When the population Q (t) is
gene on the individual selected for mutation also needs to be initialized, the gene chain is encoded by 0 or 1, and the proba-
modified accordingly to ensure that the expected number of gene bility amplitudes α and β denoting ‘‘0’’ and ‘‘1’’ are initialized to
mutations across the population is equal. In traditional mutation the value √1 , then the Q (t) of the i individual qti in the population
2
mode, the expectation number of mutation is n × L × pinitial, can be expressed as Eq. (11).
and the probability of new gene mutation is p′ , then the expected (3) Selecting the initial population. According to the charac-
number of mutation under the new mutation mode is (n × p(i)) × teristics of MPR strategy in OLSR protocol, the heuristic rule and
(L × p′ ). Mutation expectation values can be calculated as follows: the node fitness function are used to select each node MPR set.
⃝1 Heuristic rules: the node in the network of MANET is di-
n × L × pinitial = n × p(i) × L × p′ (21) vided into three categories: node i, one-hop node of i which
p = p/p(i) = pinitial /(1 − pinitial )
′ L
(22) is stored in NB_table1; two-hop node of i which is stored in
NB_table2 Based on new augmented Q-Learning strategy, for
By comparing the traditional variant of n × L and the new node i, the nodes that cover all NB_table2 are selected in
variant of n × p(i) × L, we can know that the difference is NB_table1 according to each node in NB_table2, and these nodes
n × L × (1 − p(i)). Therefore, the longer the chromosome gene in
chain, the lower the computational complexity. According to the NB_table1 are selected into the MPR set of node i. The probability
quantum not-gate and the calculated gene mutation probability amplitudes of the state ‘‘1’’ at the corresponding positions of the
p′ , the values of the amplitudes α and β of the chromosomal gene chain of the node i are modified to 1, and the probability
loci are exchanged with probability p′ . The quantum mutation amplitude of the state ‘‘0’’ is changed to 0, and the rest remains
operation actually changes the state of the superposition of the unchanged.
quantum bit states. So the mutation operation is effective for all ⃝2 Fitness degree rules: according to the above heuristic rule
the superimposed states of the chromosome. and by using Eq. (12) to calculate the distance fitness of the node,
the energy fitness of the node is calculated by using Eq. (13).
3.6. Updating of the gene chain Eqs. (18) and (19) are used to calculate the overall fitness of nodes
of MANET to determine whether node j in NB_table1 is suitable
According to the while loop of the above QGA, we can obtain for MPR node of node i based on the new augmented Q-Learning
a set P(t) of solutions and evaluate for each solution using the strategy.
above fitness function by observing the Q (t − 1) state and the (4) Doing the crossing and mutation operation of Gene strands.
previous process description. The UpdateQ (t) operation is done In order to avoid falling into the local optimal solution, we use
by using the appropriate quantum gate U(t), which is the optimal quantum crossing to perform gene chain crossing operation to
solution to update P(t). The quantum gate U(t) can be chosen keep the convergence by augmented Q-Learning strategy, and
by the specific application cases. Currently, there are many kinds use quantum not-gate to realize gene chain mutation operation,
of quantum gates of the quantum bit gate, which are commonly and judge the probability of mutation of individual gene chain in
used in quantum rotation gate, quantum not-gate, quantum con- consideration of the probability of individual level mutation. The
trol gate, Hadamard gate and so on. Based on the characteristics probability of gene-site mutation is calculated by Eqs. (20)–(22).
of QGA, it is more suitable to select quantum rotation gate. The (5) Doing the update operation of Gene chains. The chromoso-
procedure for adjusting the quantum revolving door is as follows: mal gene by quantum rotation gate in parent offspring produced
⏐α ′ ⏐ ⏐cos(θi ) − sin(θi )⏐ ⏐αi ⏐ by the probability amplitudes of α and β are updated by Eq. (23),
⏐ ⏐ ⏐ ⏐ ⏐ ⏐
⏐ i′ ⏐ = ⏐ (23)
⏐βi ⏐ ⏐sin(θi ) cos(θi ) ⏐ × ⏐βi ⏐ and the update strategy with augmented Q-Learning updates
⏐ ⏐ ⏐
rotation angle θ .
⏐ αi ⏐
⏐ ⏐
(6) Checking the convergence of the whole process based on
Where ⏐ ⏐ is the probability amplitude of the ith bit in the
βi ⏐

the convergence condition by augmented Q-Learning strategy. If
gene chain, θ i is the rotation angle, and the selection strategy is so, the processing is over, otherwise, the process will go to step
similar to the reference [15]. The rotation angleθ i = S(α i, β i) × (3).
292 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296

Table 2
Parameters for experimental testing.
Node distribution range 500 m × 500 m
The total number of nodes 200
Node initial energy 0.9 J
Node communication radius 105 m
Node movement speed [0 Vmax], Vmax=30 m/s
Packet transmission rate [0 Smax], Smax=15 p/s
Packet length 1500 bit
Node communication distance 85
threshold
Energy consumption factor of 4.5×10−8 J/bit
circuit
Energy consumption coefficient Efs: 0.9×10−11 J/(bit*m−2 )
of channel propagation model
Compression ratio Emp: 1.2×10−15 J/(bit*m−4 )
Initial mutation probability 0.45

Fig. 2. Topological control overhead of motion velocity.

needs to send their own message information packet. According


to the number of data packets sent by the source node and the
number of data packets received by the destination node, the data
packet delivery rate is calculated.
(3) The average delay of end-to-end of MANET: The average
delay time after the data flow is generated from the node or after
the data is received, until data is successfully received by the next
hop node.
Figs. 2–4 show the influence of various routing protocol on
the network performance under the different speed of nodes
of MANET. Fig. 2 is the network topology control overhead of
MANET, Fig. 3 is the packet delivery rate, Fig. 4 is the average end
to end delay. The nodes are randomly distributed in the range of
500×500 in MANET, and the speed of nodes of MANET is evenly
distributed on [0, Vmax]. The nodes of MANET generate and send
Fig. 1. Network topology diagram of experimental testing.
packets at a rate of 15 packets per second and each packet size is
1500 bits (see Fig. 1).
As shown in Fig. 2, in both the traditional routing protocol
(7) Checking whether the processing meets the requirements and the improved routing protocol, the network topology control
and the error threshold is less than 5% or not, If so, the protocol overhead of MANET increases with the increase of node motion
will exit or do EXIT operation, otherwise, the processing will go rate. This is because the faster the node moves, the bigger the
to step (2). network topology changes. For table-driven protocol routing, the
proportion of the total information of the communication that
4. Experimental testing
needs to be maintained by the structure is larger, resulting in
an increase in network topology control overhead. It can be seen
In our experiments of MANET with the parameters as Table 2,
from the figure that the control cost of QG-OLSR proposed in this
the QG-OLSR protocol is simulated and analyzed by MATLAB
paper is greatly improved compared with that of OLSR. Compared
platform. We compare this protocol with the classical OLSR proto-
with OLSR-NSSAU, the control cost of QG-OLSR decreases about
col, the proposed OLSR-NSSAU (Neighborhood State Self-Adaptive
Update) protocol and the CE-OLSR (Cartography Enhanced-OLSR) 9.3%, and the control overhead of CE-OLSR decreases about 2.8%.
protocol [13–18]. The packet-to-packet delivery rate and the av- Compared with the OLSR-NSSAU, the control overhead of the
erage end-to-end delay between the nodes and the network QG-OLSR is about 8.8% lower than that of the OLSR-NSSAU, and
lifetime are analyzed under the banner of different node speed, the control cost of the QG-OLSR is 3.2% lower than that of the
different node packet sending rate, different network node den- CE-OLSR when the node moves at a high speed of V = 30 m/s.
sity, routing topology control overhead, packet-to-packet delivery As shown in Fig. 3, in both the traditional routing protocol and
rate. the improved routing protocol, the rate of successful packet deliv-
The main experimental performance testing objects of MANET ery between nodes decreases with the increase of moving speed
are as follows: of the node in MANET. In the static state of the node, since the
(1) The overhead of topology control: because OLSR is the topology of the network does not change, the number of packets
table-driven routing protocol, it needs to periodically send rout- requiring topology maintenance is small, so the delivery rate of
ing group information and maintain the routing table to other the effective packet in the total packet is high. With the increase
nodes of MANET. Therefore, the analysis of the packet overhead of the node movement speed, when the node moves at a high
of maintaining the routing topology information is carried out. speed of V=30m/s, the QG-OLSR is about 0.16% higher than the
(2) The rate of data packet delivery: each node not only peri- OLSR; the QG-OLSR is about 0.10% higher than the OLSR-NSSAU;
odically sends routing maintenance information packets, but also the QG-OLSR is about 0.05% higher than the CE-OLSR.
D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296 293

Fig. 3. The data packet delivery rate of different nodes. Fig. 5. Topology control overhead for transmission rates.

Fig. 6. Packet delivery rate of data packet delivery rate.


Fig. 4. Average end-to-end delay of moving speed of nodes.

of the node is constant, and the control cost for maintaining the
As shown in Fig. 4, the OLSR, OLSR-NSSAU, CE-OLSR, and
topology structure is constant. However, with the increase of data
QG-OLSR protocols do not differ significantly between node rest
rate, the proportion of routing information needed to maintain
and node low-speed motion. For table-driven routing, each node
the topology in the total data information of the network be-
knows the path to each other node. When the network topology
comes smaller and smaller. Therefore, as the packet sending rate
changes little and the packet generation rate is the same, the
increases, the network control overhead is slightly lower. OLSR-
network congestion is basically the same, and the delay between
NSSAU and CE-OLSR are more efficient than OLSR in network
the nodes is basically the same. When the nodes are moving at
topology control cost of QG-OLSR.
high speed of V = 30m/s, the average end-to-end delay is larger As shown in Fig. 6, when the node speed is constant and the
than that nodes keep static, because the routing table needs to direction of motion is random, the packet delivery rate is slightly
be calculated for the table-driven route. It can be seen from the decreasing for a variety of routing protocols, and the QG-OLSR
figure that the average end-to-end delay is improved by about is decreased by about 0.22% with the increase of the node data
2.6 m/s compared with OLSR and OLSR-NSSAU when the node packet transmission rate. This is because when the node sends
moves at high speed, and the delay is about 1 ms compared with a large amount of data packets, the network will cause a larger
CE-OLSR. amount of data, resulting in network congestion slightly. And the
Figs. 5 and 6 show the influence of various routing algorithms QG-OLSR has higher packet delivery rate than OLSR, OLSR-NSSAU
on network performance under different packet sending rates. and CE-OLSR.
Fig. 5 shows the network topology control overhead, Fig. 6 shows Figs. 7–9 show the simulation results of the network topo-
the packet delivery rate. The nodes are randomly distributed in logical range of 200 nodes to 500 m × 500 m. Node velocity is
the range of 500 m × 500 m. Each node moves at a uniform speed 15m/s, and packet generation rate is 11 per second. The network
of 20m/s and the direction of movement is random. The rate at topology control overhead, packet delivery rate, and end-to-end
which packets are generated by the node is in the range of 0 to delay between nodes are analyzed under different node densities.
15 packets per second, and the size of each packet is 1500 bits. As shown in Fig. 7, for many different routing protocols, rout-
As shown in Fig. 5, when the node speed is constant, the ing control overhead of nodes in smaller network is small, be-
network topology control cost tends to decrease slightly as the cause in larger network, in order to enable all nodes to receive
packet sending rate increases. This is because the oscillation the messages from others. According to the characteristics of
degree of the topology structure is the same when the velocity MANETs and OLSR routing, you need to send more data packets
294 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296

Fig. 7. Different node density, routing control overhead. Fig. 9. Different node density, average end-to-end delay.

less and the efficiency of data transmission decreases, leading to


an increase in end-to-end packet delay between nodes. From the
figure, it can be seen that the OLSR delay is about 54%, the OLSR-
NSSAU is about 47%, the CE-OLSR is about 44%, and the QG-OLSR
is about 40% when the node density in the network becomes
small increasingly. Compared with other protocols, the QG-OLSR
protocol has a better efficiency.
Aforementioned NSGA-II [29] is one of the fast and elitist
successful multi-objective genetic algorithms in optimization of
current generation mobile WSN and MANET. Based on OLSR with
NSGA and NSGA-II, we do many relative experiments to discuss
the implication of the same in our proposed work and compare
with QGA’s implication by taking few conflicting issues like mo-
bility, throughput, delay, etc. Here we give Fig. 10 to show their
respective performances.
As can be seen from Fig. 10, the QG-OLSR protocol has better
efficiency than the other five protocols. From our experimental
Fig. 8. Different node density, packet delivery rate.
test and analysis (the process is similar to aforementioned de-
scription), we can see that because the calculation of MPR of
QG-OLSR protocol is carried out independently by each node, only
the LS information within two hops is considered, therefore, even
to maintain a highly dynamic network topology. OLSR increases though the network is very large and the nodes are very dense,
about 28%, OLSR-NSSAU increases about 25%, CE-OLSR increases the topology information of two hops does not consume a large
about 23%, and QG-OLSR increases about 21% for different routing amount of energy and can achieve a better calculation efficiency.
protocols. By comparison, the QG-OLSR protocol has a smaller Since the selection of MPR sets is optimized and the information
ratio of the increase of the topology maintenance and control such as node energy is considered in the selection process, the
overhead caused by the increase of the topology of MANET. redundant information such as topology control information in
Fig. 8 shows the impact of different routing protocols on the the network will be reduced, resulting in a decrease of network
packet delivery rate for different node densities. It can be seen topology control overhead and an increase of effective packet
from Fig. 8 that the packet delivery rates of different routing delivery rate.
protocols have a slight decrease with the node density becoming
smaller, but the difference is not significant. The main reason 5. Conclusions
is that the topology range becomes larger, the node density be-
comes smaller, and the effective routing between nodes becomes This paper proposed a kind of new quantum-genetic based
less. From the figure, it can be seen that the OLSR protocol reduces OLSR protocol (QG-OLSR) for MANET. Because the traditional
the packet delivery rate by about 2.6%, the OLSR-NSSAU reduces QGA strategy needs to be improved, New augmented Q-Learning
packet delivery rate by about 1.7%, and the CE-OLSR reduces strategy has been embedded in our improved quantum genetic
packet delivery rate by about 0.9% , QG-OLSR reduces packet strategy. The nodes in the network are encoded and initialized
delivery rate by about 0.7%. Compared with other routing pro- by the quantum gene bit. The MPR set of each node is selected
tocols, QG-OLSR has the least impact on the packet delivery rate by using heuristic rules and node fitness rules (nodal energy in-
when the density of the nodes is small, in the same simulation formation). The probability of mutation in the gene is considered
environment. in advance when the gene mutation probability is calculated.
Fig. 9 shows the effect of routing protocols on the average The quantum rotation gate is used to update the gene chain,
end-to-end delay between nodes with different node densities. and the MPR set of the repair strategy is added. The results of
The Fig. 9 shows the delay of different routing protocols. As the the experimental testings show that the new protocol is reliable
network topology increases, the distribution of nodes is frag- and highly efficient, which is suitable for many applications of
mented, and the data transmission path between nodes becomes MANET.
D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296 295

[13] Di. Niu, An asynchronous fixed-point algorithm for resource sharing with
coupled objectives, IEEE/ACM Trans. Netw. 24 (5) (2016) 2593–2606.
[14] Y. Zhao, L.M. Xu, M.L. Shi, On-demand multicast routing protocol with
multi-point relay (ODMRP-MPR) in mobile ad-hoc network, Porc. of the
ICCT 1 (1) (2003) 80–89.
[15] Mohamed. Belhassen, Performance evaluation of a cartography enhanced
OLSR for mobile multi-hop ad hoc networks, Proc. of the Wirel. Adv. Netw.
1 (1) (2011) 90–99.
[16] Bin. Sun, Wenzhi. Zhang, Modify OLSR for MANET/INTERNET inter-
connecting with mobiel gateways, Porc. of IC-NIDC 1 (1) (2009)
60–70.
[17] Kuk-Hyun. Han, Genetic quantum algorithm and its application to combi-
national optimization problem, IEEE Proc. of the 2000 Congr. Evol. Comput.
1 (1) (2000) 1354–1360.
[18] Muhammad. Safdar, Comparative study of routing protocols in mobile ad
hoc networks, Int. J. Comput. Sci. Trends Technol. 4 (2) (2016) 263–275.
[19] Yanan. Zhu, A new constructing approach for a weighted topology of
wireless sensor networks based on local-world theory for the internet of
things(IOT), Comput. Math. Appl. 64 (5) (2012) 1044–1055.
[20] Ke. Zheng, A novle multicast routing method with minimum transmis-
sion for WSN of cloud computing service, Soft Comput. 19 (7) (2015)
Fig. 10. Data packet delivery rate of different nodes.
1817–1827.
[21] Si. Liu, Novel unequal clustering routing protocol considering energy
balancing based on network partition & distance for mobile education, J.
Acknowledgments Netw. Comput. Appl. 88 (15) (2017) 1–9, http://dx.doi.org/10.1016/j.jnca.
2017.03.025.
[22] A. Hafslund, Secure extension to the OLSR protocol, OLSR Interop &
This research work is supported by National Natural Science Workshop San Diego: CA USA 1 (1) (2004) 25–33.
Foundation of China (Grant No. 61571328), Tianjin Key Natural [23] H.N. Saha, Modified fidelity based on-demand secure (MFBOD) routing
Science Foundation, China (No.18JCZDJC96800), Major projects of protocol in mobile ad-hoc network, Found. Comput. Decis. Sci. 40 (4)
(2015) 267–298.
science and technology in Tianjin, China (No.15ZXDSGX00050),
[24] Danyang. Qin, Lin. Ma, Topology property based on network tomography
Training plan of Tianjin University Innovation Team, China for wireless mobile multihop communication network, Int. J. Distrib.
(No.TD12-5016, TD13-5025), Major projects of science and tech- Sensor Netow. 2014 (6) (2014) 33–38.
nology for their services in Tianjin, China (No.16ZXFWGX00010, [25] A. Malossini, E. Blanzieri, T. Calarco, Quantum genetic optimization, IEEE
No.17YFZCGX00360), the Key Subject Foundation of Tianjin, China Trans. Evol. Comput. 12 (2) (2007) 231–241.
[26] R. Nowotniak, J. Kucharski, GPU-based tuning of quantum-inspired genetic
(No.15JCYBJC46500), Training plan of Tianjin 131 Innovation Tal- algorithm for a combinatorial optimization problem, Bull. Polish Academy
ent Team, China (No.TD2015-23). Sci. Technol. 60 (2) (2012) 323–330.
[27] C. Patvardhan, S. Bansal, Quantum-inspired evolutionary algorithm for
Conflict of interest difficult knapsack problems, Memetic Comput. 7 (2) (2015) 1–21.
[28] T.C. Lu, An adaptive population multi-objective quantum-inspired evolu-
tionary algorithm for multi-objective 0/1 knapsack problems, Inform. Sci.
No author associated with this paper has disclosed any po- 243 (18) (2013) 39–56.
tential or pertinent conflicts which may be perceived to have [29] H.Q. Zhang, Multiobjective optimization problems with complicated pareto
impending conflict with this work. For full disclosure statements sets, MOEA/D and NSGA-II, IEEE Trans. Evol. Comput. 13 (2) (2009)
refer to https://doi.org/10.1016/j.asoc.2019.03.053. 284–302.
[30] F.A. Adwan, M.A. Shraideh, M. Rasol, A genetic algorithm approach for
breaking of simplified data encryption standard, Int. J. Secur. Appl. 9 (9)
References (2015) 295–304.
[31] Z. Jin, Y.A. Ma, Q-learning-based delay-aware routing algorithm to extend
[1] Xiazi. Yi, . AsmaaAdnane, Multipath optimized link state routing for mobile the lifetime of underwater sensor networks, Sensors (Basel) 17 (7) (2017)
ad hoc networks, Ad Hoc Netw. 9 (1) (2011) 28–47. 1660.
[2] Degan Zhang, Guang Li, Ke Zheng, An energy balanced routing method [32] K.A. Linn, E.B. Laber, Interactive q-learning for quantiles, J. Am. Stat. Assoc
based on forward-aware factor for wireless sensor network, IEEE Trans. 112 (518) (2017) 638–649.
Ind. Inf. 10 (1) (2014) 766–773. [33] M. Zheng, Inferring gene regulatory networks based on a hybrid parallel
[3] B. Mans, N. Shrestha, Performance evaluation of approximation algorithms genetic algorithm and the threshold restriction method, Interdiscip. Sci. 10
for multipoint relay selection, in: Proc. Of the 3rd Annual Mediterraneam (1) (2018) 221–232.
Ad Hoc Networking Workshop, 2004, 1(1), pp. 50–59. [34] Songran. Liu, Zhe. Li, A modified genetic algorithm for community
[4] Ali Moussaoui, A link-state QoS routing protocol based on link stability for detection in complex networks, in: 2017 International Conference on Algo-
mobile ad hoc networks, J. Netw. Comput. Appl. 39 (3) (2014) 117–125. rithms, Methodology, Models and Applications in Emerging Technologies
[5] Xiaodong Song, Xiang Wang, Extended AODV routing method based (ICAMMAET), 2017, 1 (1), pp. 1–8..
on distributed minimum transmission (DMT) for WSN, Int. J. Electron. [35] Ting. Zhang, A kind of effective data aggregating method based on
Commun. 69 (1) (2015) 371–381. compressive sensing for wireless sensor network, EURASIP J. Wireless
[6] P. Sondi, Design guidelines for quality of service support in optimized link Commun. Networking 2018 (159) (2018) 1–15, http://dx.doi.org/10.1186/
state routing-based mobile ad hoc networks, Ad Hoc Netw. 11 (3) (2013) s13638-018-1176-4.
298–323. [36] S. Zhou, A low duty cycle efficient MAC protocol based on self-adaption
[7] Hiroshi. Mineno, Integrated protocol for optimized link state routing and and predictive strategy, Mobile Netw. Appl. 23 (4) (2018) 828–839.
localization: OLSR-L, Simul. Model. Pract. Theory 19 (8) (2011) 1711–1722. [37] Hongli. Niu, Novel PEECR-based clustering routing approach, Soft Comput.
[8] S.H. Zou, H.T. Wu, S.D. Cheng, A simple and efficient broadcasting scheme 21 (24) (2017) 7313–7323.
for mobile ad hoc networks, J. Softw. 16 (8) (2005) 1104–1111. [38] Degan. Zhang, A new approach and system for attentive mobile learning
[9] Xiang. Wang, Xiaodong. Song, A kind of novel VPF-based energy-balanced based on seamless migration, Appl. Intell. 36 (1) (2012) 75–89.
routing strategy for wireless mesh network, Int. J. Commun. Syst. 30 (6) [39] Si. Liu, Novel dynamic source routing protocol (DSR) based on genetic
(2017) 1–15, http://dx.doi.org/10.1002/dac.2889. algorithm-bacterial foraging optimization (GA-BFO), Int. J. Commun. Syst.
[10] Stefano. Rosati, Dynamic routing for flying ad hoc networks, IEEE Trans. 31 (18) (2018) 1–20, http://dx.doi.org/10.1002/dac.3824.
Veh. Technol. 65 (3) (2016) 1690–1700. [40] Yameng. Tang, Novel reliable routing method for engineering of internet
[11] G. Cervera, Multipath routing strategy to prevent flooding disruption of vehicles based on graph theory, Eng. Comput. 9 (2018) http://dx.doi.
attacks in link state routing protocol for MANET, J. Netw. Comput. Appl. org/10.1108/EC-07-2018-0299.
36 (2) (2013) 744–755. [41] Ting. Zhang, Novel self-adaptive routing service algorithm for application
[12] Zhen. Ma, New AODV routing method for mobile wireless mesh network of VANET, Appl. Intell. 11 (2018) http://dx.doi.org/10.1007/s10489-018-
(MWMN), Int. Autom. Soft Comput. 22 (3) (2016) 431–438. 1368-y.
296 D.-g. Zhang, Y.-y. Cui and T. Zhang / Applied Soft Computing Journal 80 (2019) 285–296

[42] Degan. Zhang, Hui. Ge, New multi-hop clustering algorithm for vehicular [46] Chen. Chen, Yuya. Cui, New method of energy efficient subcarrier allocation
ad hoc networks, IEEE Trans. Intell. Transp. Syst. 7 (2018) http://dx.doi. based on evolutionary game theory, Mobile Netw. Appl. 9 (2018) http:
org/10.1109/TITS.2018.2853165. //dx.doi.org/10.1007/s11036-018-1123-y.
[43] Xiaodan. Zhang, Design and implementation of embedded un-interruptible [47] Jinxin. Gao, Novel approach of distributed & adaptive trust metrics
power supply system (EUPSS) for web-based mobile application, Enterp. for MANET, Wirel. Netw. 1 (2019) http://dx.doi.org/10.1007/s11276-019-
Inform. Syst. 6 (4) (2012) 473–489. 01955-2.
[44] Xiaodong. Song, New agent-based proactive migration method and system [48] Si. Liu, Dynamic analysis for the average shortest path length of mobile
for big data environment (BDE), Eng. Comput. 32 (8) (2015) 2443–2466. ad hoc networks under random failure scenarios, IEEE Acces. 1 (2019)
[45] Degan. Zhang, Xiang. Wang, A novel approach to mapped correlation of ID http://dx.doi.org/10.1109/ACCESS.2019.2896699.
for RFID anti-collision, IEEE Trans. Serv. Comput. 7 (4) (2014) 741–748.

You might also like