1. introduction
With the progressive development of wireless devices and gadgets, the traditional spectrum resource allocation scheme intensifies spectrum scarcity, which in turn reduces the utilization of the spectrum significantly (Feng & Weilian, 2018; M. G & V, 2021). The fixed channel allocation policy regulates the wireless networks, in other word, the governmental agencies regulate the spectrum for the authorize holders or services on a long-term basis for a large geographical region. Furthermore, referring to the Federal Communication Commission (FCC) report, temporal and geographical variation of the licensed spectrum exploitation varies from 15% to 85% (Liu, Wang, Chen, & Guo, 2018a, 2018b). Even though the traditional fixed spectrum allocation worked well in the past, however, the effectiveness of the traditional spectrum assignment policies in recent years reduced dramatically due to the significant access to limited spectrum resources in wireless mobile services (L. Zhang, Xie, & Chen, 2020).
Licensed users experience different needs for traffic and bandwidth for various amounts of time (Tarek, Benslimane, Darwish, & Kotb, 2020). Besides the new technologies, the bandwidth requirements for different types of applications keep on changing (Mishra, Sagnika, Singh, & Mishra, 2019). The inefficient spectrum utilization with the restricted spectrum availability necessitated a novel communication model for opportunistically exploiting the current wireless spectrum resources. Therefore, for solving these spectrum inefficiency problems, Cognitive Radio (CR) that is contingent on Dynamic Spectrum Access (DSA) was proposed (Singh & Dutta, 2020). The so-called NeXt Generation (xG) Networks, also called dynamic spectrum access networks (DSANs) are capable of providing high bandwidth for their users in an opportunistic manner (Salehi & Solouk, 2022).
CR is the key enabling technology for the xG networks (Motta, Banerjee, & Sharma, 2023; Rajesh Babu, Garg, & Chakraborty, 2022). This technology was first introduced by Dr. Joseph Mittola in his PhD dissertation in 2000 (Salih & Shakir, 2022). CR is a Software Defined Radio (SDR) that can intelligently detect the spectrum bands (holes) for the secondary users (SUs) (also known as unlicensed users or cognitive users) and opportunistically utilizing the spectrum bands without interfering with the licensed users or so-called primary users (PUs) (Latif et al., 2021). A CR device can regulate its operating parameters for instance (power transmission, communication technology, operating frequency, modulation scheme, … etc.). So it can adapt itself to the environment and achieve the desired objectives (Tarek et al., 2020). Spectrum allocation (also known as spectrum allotment or spectrum assignment) which is a part of the process of spectrum management is frequently addressed as a vital research challenge in the CRNs (Latif et al., 2021).
An efficient spectrum allocation system can determine the efficiency of the whole cognitive radio network along with maximizing the spectrum utilization (Feng & Weilian, 2018). Efficient allocation scheme that aims to maximize the spectrum utilization and minimize the interference is considered to be an optimization problem (Latif et al., 2021). Various traditional models have been utilized for modeling the spectrum resource allocation problem for instance graph model, game model, interference temperature model, auction model, and among other models. Graph-based models are the most commonly used methods for modeling the spectrum allocation problem (Feng & Weilian, 2018). Although the graph-based allocation optimization problem is a non-deterministic polynomial (NP)-hard combinatorial optimization problem (Feng & Weilian, 2018; Liu et al., 2018b; Sehli & Babes, 2020). Therefore, inexact methods (also called approximate methods) are used to solve such problems, like heuristic and meta-heuristic optimization algorithms (Feng & Weilian, 2018).
In this paper, different optimization algorithms have been studied and has been applied for allocating the spectrum in CRNs. The algorithms were examined to show their capabilities for achieving the near optimal solution and to maximize the network utilization with interference-free allocation. Indeed, after studying each algorithm individually it was observed that there is a need to enhance the existing optimization algorithms as they tend to fall early into local optimum. Hence, an improved hybrid optimization algorithm is proposed. Therefore, the key contributions of this work can be listed as follows:
1. Modeling the problem of the spectrum allocation using graph coloring model.
2. Proposing an improved PSO algorithm that combines the Particle Swarm Optimization algorithm with Random Neighborhood Search. This algorithm improves the local search capabilities of the PSO, thus preventing the solution from falling early into local optimum.
3. Comparing the achievement validation of the anticipated hybrid (PSO-RNS) algorithm with other optimization algorithms such as: Greedy algorithm, RNS algorithm, CSGC algorithm, and classical PSO algorithm.
The rest of this paper is organized as follows: Section 2 presents the related studies that have been conducted in the research field of the spectrum allocation in CRNs. Section 3, presents the allocation system model and the utility function. Section 4, presents the proposed optimization algorithms for allocating the spectrum in CRNs. Section 5, presents the simulation results and analysis for the given optimization algorithms. Lastly, the conclusions are drawn in section 6.
2. Literature review
Recently, the efficient spectrum allocation scheme has gained a huge attention in the research area of CRNs. A large number of works have been conducted by researchers for finding a near-optimal solution and proposing a new fast algorithm to solve this intractable problem in the research field. In this section various optimization algorithms that have been applied for spectrum allocation in the CRNs with the latest research conducted in this field has been reviewed.
Peng, Zheng, & Zhao (2006) proposed a Color Sensitive Graph Coloring (CSGC) technique to address the problem of the spectrum allocation. The system model was based on a graph coloring model. Three objective functions were presented and three policy labeling rules with both collaborative and non-collaborative rules each corresponds to a different objective function (Maximum Sum Reward (MSR), Maximum-Minimum Reward (MMR), and Maximum Proportional Fairness MPF). This algorithm has shown its capability not only in improving the spectrum performance, but also improved the utilization and fairness of the network.
Teng, Xie, Chen, & Zhang (2020) proposed an improved spectrum allocation scheme of a dynamic-time varying that based on chaotic binary particle swarm optimization. The chaotic map idea was introduced in the improved algorithm for optimizing the initial population and the optimal position for all particles in each generation. Moreover, the global ergodicity of the chaotic map was utilized to overcome the shortcomings of the PSO algorithm. The experimental results showed that a fast and efficient spectrum allocation can be realized for satisfying the demands of cognitive radio communication.While Yesaswini & Annapurna, (2021) in their study presented a comparison between the two optimization algorithms that is particle swarm optimization and genetic algorithm for assigning the spectrum in CRNs. The results showed that PSO algorithm provided more spectrum utilization as compared to the GA based approach with less number of iterations to obtain the near optimal result.
Moreover, Satria, Mustika, & Widyawan (2018) presented a solution for spectrum allocation in CRN using Modified-ACO algorithm. The proposed solution depends on the intensity of the pheromone in the path used by the ants for making decision about the channel selection. The simulation results showed that the ants have improved the spectrum allocation and achieved fairness in the CR environment, as well as it increased the throughput of the system.
Mishra et al. (2019) proposed a PSO algorithm for allocating the spectrum for cognitive system. The author focused on maximizing the total transfer rate along with the subjected constraints such as individual transfer rate, total power usage, and bit error rate. The presented algorithm has shown its capability to increase the real-time total transfer rate and required less iterations to obtain the near optimum result as compared to GA based approach. Additionally, X. Zhang, Zhang, & Wu (2020) proposed an adaptive PSO algorithm based on sawtooth propagation technique (SA) for allocating the spectrum in the (OFDMA) cellular network system. They assessed the network utility performance with two forms, one with no fairness for the linked users and the other with considering the fairness with the linked users. The simulation results showed validation of the proposed algorithm in terms of utilization and fairness among the linked users in the cellular network system.
M. G & V (2021) studied and analysed the performance of the PSO algorithm taking into consideration several parameters such as individual power rate, entire power usage, and bit error rate (BER). These parameters were considered to confirm the fairness among the users along with the efficient energy with minimum BER. The simulation results showed the ability of the proposed algorithm to offer larger allocation rate instantaneously with less number of iterations.
Moreover, Tian, Deng, & Xu (2020) proposed an Immune Parallel Artificial Bee Colony Optimization (IPABCO) algorithm. The authors adopted the immune theory and the parallel theory for solving the problem of the spectrum allotment in the cognitive radio sensor networks. The proposed algorithm was based on the classical graph coloring model that aims at optimize the network efficiency. The anticipated algorithm showed its capabilities in terms of maximizing the total network benefits. In addition, it showed superiority in terms of having fast convergence speed and high optimization ability as compared to PSO and ACO algorithms.
However, an enhanced ABCO algorithm was presented in Agarwal, Vijay, & Bagwari (2021), for spectrum allocation in cognitive radio network. The proposed method first predicted the channel of the PUs, then it performed the spectrum sensing, thus the probability has been reduced, and the overall throughput of the network has increased. The simulation results showed that the proposed method had better efficiency by 11.48% in terms of spectrum usage compared to the binary artificial bee algorithm.
While, Feng & Weilian (2018) pioneered the Fireworks Algorithm (FA) to address the spectrum allocation problem in the cognitive radio networks. The system was based on a graph coloring model with considering multiple objective functions. Two binary coding layers were considered for individual fireworks. The given algorithm has shown its capability in avoiding the local optimum and enhancing the system fairness, utilization, and convergence speed.
Moreover, Latif et al. (2021) proposed an improved evolutionary algorithm which is a differential evolution based particle swarm optimization algorithm with a repair process (DE-PSO-RP). Where the repair process is considered for excluding the interference among the secondary users, hence increasing the spectrum resources in the CRNs. The authors compared the results obtained by the anticipated algorithm with that obtained by the other algorithms such as PSO with repair process (PSO-RP) and differential evolution with repair process (DE-RP). The simulation results showed that the proposed algorithm improved the spectrum utilization in the CRNs.
3. System Model
In the conditions of the dynamic spectrum access like cognitive radio networks, the channel allocation problem is described by enabling unlicensed users (SUs) to access the spectrum with legacy users (PUs). Hence, dynamic spectrum systems increase the available spectrum's capacity and commercial value. Based on restrictions agreements upheld by PUs, secondary users are able to access the available spectrum in a non-interfering way. This section presents the system model aided by a sample model. The suggested model assumes that the environmental factors like user location and spectrum availability will not change during the period of spectrum allocation. This is because users quickly adapt to environmental changes despite the sluggish changes in the spectrum environment.
3.1 System Sample Model
In this section a sample model describing
the proposed system model is presented. Figure 1, illustrates the sample model
deployment of a single cell with a base station to provide a wireless
connection to a residential community. In this sample model there are four
randomly distributed primary users (PU-Ⅰ,
PU-Ⅱ, PU-Ⅲ, PU-Ⅳ) and four randomly
distributed secondary users (SU-Ⅰ, SU-Ⅱ, SU-Ⅲ, SU -Ⅳ).
Randomly every primary user 
occupies a channel 
with a protection area
(radius) denoted as 
. It is assumed that the four secondary users opportunistically access the channels
available in the network without affecting the transmission of the primary
users only in two cases; either when PU is absent, or SU lies outside the
protection area of the PU. Otherwise, there will be harmful interference.
Accordingly, as demonstrated in Figure 1, SU-Ⅰ
cannot access the channel occupied by PU-Ⅰ since
the coverage area for both users are overlapped. However, SU-Ⅰ can utilize the channel occupied by PU-Ⅰ when it is absent (i.e., PU-Ⅰ not transmitting data). On the
other hand, SU-Ⅱ can opportunistically access all the channels available
in the network without affecting the PUs transmission as it allocates far away
from all primary user’s protection areas. As a result, a list of available
channels will be attained for every secondary user, and each channel will
provide a different reward for the secondary users which is the protection
range that a user can get after utilizing that channel. Additionally, for
preventing the interference among the secondary users, each channel will list a
pair of secondary users that cannot utilize the same channel simultaneously.
Therefore, (SU-Ⅲ) and (SU-Ⅳ) cannot utilize channel m at the same
time as their protection ranges overlapped. Hence, the system model output
parameters are Availability matrix (L), Reward matrix (B), and interference
matrix(C).
Figure 1: System Sample Model.
3.2 Allocation Model and System Utility Function
Suppose a network of secondary users
indexed from n=1 to N and primary users indexed from k=1 to K that are
competing for a number of channels indexed from m=1 to M. Furthermore, based on
the location and usage of the channel by adjacent primary users, the channel
availability and rewards of each secondary user can be determined. To adjust
the interference range, each secondary user modifies the transmit power on
channel m which is denoted as 
to avoid the interference constraints imposed by PUs. Thereby, the secondary user
will regulate the interference range to the maximum permissible level using a
fixed power control scheme. As a result, the transmission power of the SU will
be bounded by two boundaries: the minimum transmit power 
which corresponds to
minimum interference rage 
and the maximum transmit
power 
which corresponds to
maximum interference range 
. Hence, it is assumed
that channel 
will be available for the
secondary user 
only if (Ghasemi & Ghasemi, 2020):
|
|
|
Where is
the protection range of the secondary user on channel ,
is the protection range
of the primary user on channel , and 
is the distance between
secondary user and primary user . Furthermore, it is assumed that 
uses the out of band
mechanism to determine the primary users' location and power in order to
correctly adjust their 
to avoid interference
with the 
. If such power
adjustment is utilized, the secondary user’s interference ranges will be equal
to the minimum value between the maximum permissible transmit power and the
minimum distance from all the primary users which occupies that channel,
according to Peng et al. (2006):
|
|
|
[pWhere 
represents the distance
of secondary user from channel m occupied
by number of PUs, is the maximum
permissible transmit power of SU, is the Euclidian
distance between the 
secondary user and 
primary user, and 
is the primary user’s
coverage area (interference range). Additionally, after determining the
availability of the channels for the secondary users based on interference
ranges of the with the 
for a given topology,
the interference among the secondary users will be determined. This can
be implemented by measuring the distance from the centers of the protection
range of secondary user and 
on channel which is denoted as 
. This distance indicates
the position of the secondary users with respect to each other. If this
distance was less or equal than the sum of radii of the protection ranges of 
and 
, that is:
|
|
|
Then secondary users and cannot utilize channel simultaneously because
there will be interference due to overlapped regions of the protection area for
both users. Where 
is the 
protection area and 
is the protection area. The key
component of the proposed model is presented as follows (Ghasemi & Ghasemi, 2020):
Channel availability matrix: 
is an N by M binary
matrix that indicates weather channel is available for
secondary users or not. Depending on the
distance if 
, it means channel is not available to ( 
), otherwise the channel
is available ( 
Channel reward matrix: 
is an N by M matrix that
gives the reward matrix for each secondary user. In addition, this matrix signifies
the bandwidth that a user can attain after
occupying channel with the assumption that
there is no interference from neighbors. Relying on the sample in Figure 1, the reward for each secondary user
can be one of the following:
1.
Coverage
area of SU using channel 
:
|
|
2.
The
Capacity using channel (with
the assumption that the SNR is a function of 
):
|
|
|
Where 
represents the reward of
secondary user n after occupying channel m. It is worth noting that if 
, then 
. For the given problem, (Equation
)) which represents the
coverage area of the secondary users is considered as a reward for the
secondary users).
Interference constraint matrix: 
, is an N by N by M
matrix representing the interference constraints among the secondary users. In
this matrix each channel will have a N-by-N matrix that represents the
interference among users. Where 
represents the
interference of and occupying channel m
simultaneously. If 
, it means that 
and 
cannot utilize channel simultaneously. This
matrix is a channel availability dependent matrix that is:
and 
if 
. In this paper the
geometry model has been used in which two users will conflict if there was no
sufficient distance between them that is: 
. Moreover, this
constraint is a channel specific in which two users might have conflict on one
channel but not in another.
Conflict-free channel allocation
matrix: 
, is a N by M matrix
representing the final channel allocation matrix. Where 
represents the
occupation of secondary users n to channel m. That is: if 
, it means that channel is allocated to user , otherwise channel cannot be allocated to
user 
This matrix necessitates
to satisfy all the constraints imposed by the PUs.
Radio interface limit 
: Represents the quantity
of channels that the secondary user can allocate.
User reward
representing the reward
vector that each secondary user can get after assigning the channels.
Network utilization U(
): This represents the objective, as the key objective of the
spectrum assignment problem is to increase the utilization of the network.
3.3 Utility Function
After defining system model for the given network, the following utility function (i.e., objective function or fitness function) can define the spectrum allocation problem (Teng et al., 2020):
|
|
(6) |
The utility functions for a specific application type can be acquired based on extensive surveys. Alternatively, based on traffic pattern and fairness inside the network the utility functions can be designed. The single-hop-flows is considered for addressing the utility, as they are the simplest format of wireless transmission. In addition, in this work the utility function has been expressed as Maximum-Sum-Reward (MSR), the aim of this objective function is maximizing the network utilization without taking into consideration the fairness. The expression of this objective function is represented as follows (Peng et al., 2006):
|
|
(7) |
To facilitate the simulation, the Mean Reward fitness function has been considered instead of the Maximum-Sum-Reward as follows:
|
|
(8 |
The conflict-free assignment matrix requires to satisfy the subjected
constraints bellow:
|
|
(9 |
|
||
|
|
(10) |
|
||
|
|
(11) |
|||
The first constraint (Equation (
) states that the channel
has to be available in order to be allocated for the cognitive user. Otherwise,
the user-channel pair will not be considered in the final solution. The second
constraint (Equation (10)) shows that the highest quantity of
channels that can be allocated to SU for the final conflict-free allocation
matrix have to be less or equal than 
. Meanwhile, the third
constraint (Equation (11)) represents the incompatibility
constraint, in other word, each channel might contain pairs of incompatible
users that cannot utilize the same channel simultaneously to avoid
interference. Therefore, for achieving a conflict-free assignment matrix, the proposed
algorithms necessitate first to satisfy all the aforementioned constraints
presented in this section.
4. Optimization algorithms for spectrum allocation in CRNs
4.1 Spectrum Allocation using RNS Algorithm
Local search is based on the concept of the neighborhoods. This algorithm can be used for finding solutions for highly complex problems where their analytical model contains a huge number of variables and constraints or little theoretical knowledge is available. The neighborhoods require to be well defined for efficiently searching them. Thus, the local search can be implemented to find a good solution quickly for large scale instances of such problems. The local search is implemented with a number of iterations on the same problem instance using different (e.g. random generated) initial solutions. The size of the neighborhood can determine how the local search can effectively improve the solution in each iteration (Crama, Kolen, & Pesch, 1995).
The Random Neighborhood Search (RNS)
algorithm is used for generating a population of candidate solutions for the
given optimization problem (Saleh, Ahmed, & Nashat, 2020). Algorithm I illustrates the
steps of the RNS method for allocating the spectrum in CRNs. This algorithm
starts by initializing the number of iterations (Step2) in which it represents
the population of candidate solutions. Step2 to Step7 defines the main steps
for implementing this method which starts by randomly selecting a channel m and
allocating it to a secondary user. After that checking the interference with
other secondary users on channel m. If the selected user
has interference with user on channel 
, then allocate the
channel to user 
and delete it from
its channel list 
and from user channel list
. Otherwise, if
there was no interference with user 
, then just
allocate channel m to user 
, and delete it
from its channel list (Step 4). Repeat until the availability matrix is empty,
then evaluate the fitness function using Equation (8
for evaluating the
Mean Reward. After that update the iteration and repeat the previous steps
(from 3 to 5). If maximum iteration is met, then select the best solution which
represents the final conflict-free assignment matrix. Otherwise, repeat the
procedure.
|
Algorithm I: RNS Algorithm |
|
Inputs: Availability matrix(L), Reward matrix(B), and Interference matrix (C) |
|
Output:
Conflict-free allocation matrix (A) |
|
Let M be the number of available channels |
|
Step1: Initialize parameters set and simulation model Step2: Set the value of maximum iteration. |
|
Step3: For |
|
Step4: For the selected
channel m, check if |
|
Step5: Repeat Step2 for
all |
|
Step6: Evaluate the
Fitness Function according to equation (8 |
|
Step7: If a predefined maximum iteration is met, then select the best solution among all iterations, otherwise return to Step3. Step8: End
|
4.2 Spectrum Allocation using PSO Algorithm
Particle Swarm Optimization is a swarm
intelligent optimization algorithm that was basically introduced for the first
time in 1995 by Kennedy and Eberhart. This algorithm is utilized for solving
the optimization problems as it imitates the attitude of the flock of birds and
school of fishes as they continuously change their location to search for food
relying on their current location. Recently, the researchers and scholars have
adopted PSO algorithm for solving different problems in CRNs, although it has revealed
a massive flexibility for solving the spectrum allocation problem and searching
the near optimal solution (Mishra et al., 2019; Salih & Shakir, 2022). Since the spectrum allocation is a discrete space
optimization problem, the binary PSO has been utilized with this problem. Therefore, for simulating the PSO behavior to solve the
spectrum allocation problem and find near optimal solution, the birds will be
represented by a population of particles. This algorithm starts by defining the
number of particles or the population size denoted by P, and the length of
particles is set to D = 
, which counts the number of ones in
the availability matrix. Binary coding mechanism is used to randomly
generate particle
position at
iteration 
as 
, where 
is the 
bit in the particle position vector, and 
. This mechanism is
analogous to that used in Genetic Algorithm (GA), which is regarded as a
possible solution in the optimization problem. Similarly, the velocity of
particle
is implied by 
, where 
, and 
is a system parameter that prevent 
from the continuous value . Additionally, this parameter has
not been set to a small value to prevent high mutation rate. Then for each
particle map the 
element in 
to in the pre-conflict assignment matrix, according to the example
in Figure 2 (Teng et al., 2020).
Figure 2: Code Mapping Example
Then the interference is examined in each channel to achieve a conflict-free
assignment matrix. After that the fitness function will be evaluated according
to Equation (8. Then define 
which is the local best solution that particle has attained at
iteration , and 
, is the global best
solution among all particles. At 
, all
particle’s velocities and positions will be updated using below expressions
(Teng et al., 2020):
|
|
(12) |
|
|
(13) |
Where 
is the d -dimensional
velocity of the particle in the following
generation, it is determined by the previous velocity value of particle , the cognition part: 
and the social part: 
. Where 
is the inertial weight, 
and 
are the acceleration
coefficients, 
and 
are uniform random
numbers generated in the interval of [0,1]. The value of is limited to a value in
the interval of [0,1] according to the following sigmoid function (Teng et al., 2020):
|
S ( |
(14) |
Equation (13) defines that
the probability of bit 
, if 
S (), Otherwise, 
, where 
is a uniform random
number distributed in the interval of [0,1]. Additionally, to avoid S () from approaching 0 or 1, the value of is constrained in the
limit of 
, that is (Zhao, Xu, Zheng, & Shang, 2009):
|
|
(15 |
Equation (15) defines that if the value of 
is greater than then set the value of 
, or if the value of 
is less than the 
, then set the value of 
.
4.3 Proposed Hybrid PSO-RNS Algorithm for Spectrum Allocation in CRNs
In this section a new algorithm is proposed
which is hybrid PSO with the Random Neighborhood Search (PSO-RNS) algorithm.
The performance of this algorithm has been compared with other optimization
algorithms (such as: Greedy algorithm, RNS algorithm, CSGC algorithm, PSO
algorithm) that have been examined individually to evaluate its capability to
fit the given objective function and to validate the performance of the new
proposed algorithm. As aforementioned, RNS is used to enhance the initial
starting solution by providing a population of candidates to choose the best
solution among all candidates with a predefined maximum iteration. While in
PSO, particles can search globally and find the initial near optimal solution. As a result, in later algorithm, PSO tends to fall into
the local optimum during the search process. Therefore, RNS is performed for
each generation of particle velocity, allowing particles to achieve the near
global optimum via evolution during the search iteration. Algorithm II illustrates the main steps for
the proposed hybrid system. It starts by initializing all the parameters with
defining the fitness function and the population size 
(Step 1 to 3). At iteration 
, randomly generate
particle 
position vector 
using a
binary coding mechanism. Additionally,
generate particle velocity vector 
, where the bit values of this vector are randomly generated in the
interval of
(Step4). Then map the elements in the position vector to the pre-conflict
assignment matrix then perform RNS algorithm utilizing Algorithm I (Step5). Where RNS algorithm
provides additional search for finding the best position vector which is
generated initially by the PSO algorithm. Hence, RNS with a predefined maximum
iteration randomly selects the channels for the secondary users. The interference
is checked among all the secondary users, so if two users interfere if they
utilize the same channel, randomly allocate the channel to one of them and
delete it from the other user channel’s list. Thus, the free-assignment matrix
will be generated for each particle. After a predefined maximum iteration is
met for the RNS algorithm it will select the best solution for each particle
among all iterations which in turn enhances the solution for all particles.
Then evaluate the fitness function for all particles. Then it will search for the local best solution 
and global best solution 
(Step7). Update the
iteration t and all particles’ velocities and positions according to equation (12) and (13) respectively, then reassess the fitness
function (Step8). If the current solution vector is superior to the previous
best solution, then the particle position and global best position is updated
(Step9). Finally (Step10), terminate the algorithm if a predefined maximum
iteration is met and choose the best solution among all iterations, otherwise
repeat Step7. Figure 3 demonestrate the
flow chart of hybrid PSA with RNS algorithm for spectrum allocation in CRNs.
|
Algorithm II: Hybrid PSO with RNS Algorithm |
|
Inputs: Availability matrix(L), Reward matrix(R), and Interference matrix (C) |
|
Output:
Conflict-free allocation matrix (A) |
|
Let M be the number of available channels |
|
Step1: Initialize system parameters set, simulation model, and PSO parameter |
|
Step2:
Identify the Fitness Function |
|
Sep3: Set
population size P and the length of particle |
|
Step4: Set
iteration |
|
Step5:
Map the |
|
Step6:
Find the best position for each particle Step7: Set
iteration Step8: Update the particle position if the current particle’s position is better than the previous best position. Then update the global best position. Step9: Find the best solution if a predefined maximum iteration is met, otherwise return to Step7. |
