Cooperative spectrum sensing based hybrid machine learning technique for prediction of secondary user in cognitive radio networks

Abstract

The spectrum scarcity problem in today’s wireless communication network is addressed through the use of a cognitive radio network (CRN). Detection in the spectrum is made easier by cooperative spectrum sensing (CSS), which is a tool developed by the military. The fusion centre receives the sensing information from each secondary user and uses it to make a global conclusion about the presence of the principal user. Literature has offered several different methods for decision making that lack scalability and robustness. CSS censoring is inspected in the attendance of faded settings in the current study. Rayleigh fading, which affects reporting channels (R), is examined in detail. Multiple antennae and an energy detector (ED) are used by each secondary user (SU). A selection combiner (SC) combines the ED outputs with signals from the primary user (PU), which are established by several antennas on SU, before the joint signal is utilised to make a local result. SUs are expurgated at the fusion centre (FC) using a hybrid Support Vector Machine (SVM) that significantly improves detection performance and reduces the number of false positives. With a minimum false alarm probability of 0.1, error rate of 0.04, spectrum utilization of 99%, throughput of 2.9kbps and accuracy of 99%, proposed model attains better performance than standard SVM and Artificial Neural Network (ANN) models.

Keywords

Energy detector primary user cognitive radio network secondary user cooperative spectrum sensing

1 Introduction

In order to keep up with the changing demands and needs of people, wireless communication technologies are always evolving and improving. It was necessary to address the issue of spectrum scarcity due to the rise in popularity of wireless applications and services. It has been revealed that licenced bands are not utilised at a rate of up to 90% by the US telecommunications body [1], it is the Federal Communications Commission (FCC) (FCC). Findings were made public in the FCC’s report on spectrum efficiency, “FCC Account of the Spectrum Efficiency Working Group.” Spectrum bands that are currently underutilised or perhaps completely unutilized have been studied extensively in recent years. Mitola’s cognitive radio concept is the most well-known in the field [2].

Secondarily, secondary users (SUs) can only access CR networks through opportunistic usage of unused spectrum slots that are not claimed by main users (PUs). The result is that they need to be able to detect the various spectral bands throughout time [3]. It is then imperative that the SUs be evacuated as soon as the PU reenters the bands in order to minimise unwanted interference. For the SUs, another approach requires that their predefined interference temperature be maintained in order to coexist with the PU in licenced bands [4]. In order to ensure the PU’s QoS, SUs always transmit information at a reduced power level [4]. This dramatically improves spectrum efficiency. Spectrum sensing is an significant aspect of CR [5 –7]. Multi-path fading, and the near–far effect all have the potential to compromise spectrum sensing results in a real-world wireless communication environment. Spectrum sensing (CSS) has evolved as an active way to ensure that many SUs work together to detect the various bands of the spectrum. Final decisions are made by the fusion centre (FC) based on the information providing by the SUs [8].

Local information fusion can be separated into two categories. In the first approach, each SU takes a local decision and intelligences the outcomes via 1-bit data, where ’1’ signifies presence and ’0’ denotes absence of the PU [9, 10]. Rules like “OR,” “AND,” and “MAJORITY” are used to come to a conclusion in the final choice. With Soft-decision detection [11, 12], which is more expensive for local transmission bands than 1-bit hard decision detection, each SU sends the sensing data to the FC, It then generates a global test fact by integrating all of the local sensing info together. A compromise between sensing performance and band cost is therefore difficult to find. Hybrid SVMs are employed in this study to evaluate local information fusion (GL-SVM). Four antennas are employed for sensing in the secondary user prediction. For the remainder of the document, the following structure is used: In Section 3, a system model with an explanation of the proposed GL-SVM is mentioned in Section 2, which also includes related research. Section 4 deliberates the testing of the suggested model with an existing technique, and Section 5 concludes the study.

2 Related works

IEDs in a CSS system can only be evaluated if the S-channel is also a Rayleigh, or Rician faded and R is perfect, as in [13] and [14]. While this is true in [13, 14], it isn’t taken into account in [15 –17]. The S- and R-channels are both faded in the current work (Hoyt, Rayleigh, Rician). Using threshold-based filtering, it is possible to halt transmission over severely fading R-channels. For Rayleigh [15] and Hoyt [16] fading, the presentation of CSS with threshold-based editing of CEDs is examined. At FC, IED under the influence of Rayleigh fading is examined in [17], which examines the effects of these logic rules: OR, AND, and majority-logic fusion. When OR-logic rules are used, the overall error probability is higher than the error likelihood with logic fusion for many CR users, according to the findings [18] So we set about creating analytical and simulation frameworks for energy-constrained cognitive radio networks that use censoring with thresholds in Rayleigh, fading channels. Several fading models have been devised contingent on the specific propagation situation and the fundamental communiqué scenario.

Simulating an environment with a lot of reproduced and dispersed waves, Rayleigh distribution is employed. However, the set of reproduced and dispersed waves is conquered by one strong constituent, which is why Nakagami-n is usually used to depict Rayleigh fading channels. [19]. The Nakagami-q distribution, which spans the Gaussian. For strong ionospheric scintillation in satellite links, Hoyt’s distribution is typically utilised [20, 21]. There is an examination of data derived from separate homogeneous Bernoulli likelihood mass functions in [22] for situations where the alternative hypothesis’ success probabilities are unknown. An equation for the pmf for variables is found in [23] and labels the number of successful trials in a series when the individual success probabilities change between trials. Furthermore, its compensations in footings of speed, installation and ease of examination are also covered in this article. A wireless sensor network (WSN) is described in [24 –26].

Wanlu Lei et al., [27], proposes an in-depth enhancement study model to realize real-time spectrum assignment using the mathematical terms of mixed integers and nonlinear programming. Rodney Martinez Alonso et al. [28] have proposed a dynamic management algorithm that aims to improve the spectrum utilization of units. Prabhat Thakur et al. [29], advises on potential questions related to random channel perception using the revised and improved electoral approach. Varun et al., [30], call for implementation of an intelligent machine learning system to improve prediction accuracy for channel assortment.

Hurmat Ali Shah et al., [31] an efficient spectrum-sharing approach based on predicted machine learning. This predictive technology learns the current state of the primary user from a spectral context. Sensitivity details are kept in the appropriate categories, and then the current sensitivity level is divided into such categories. According to the classification results, the primary user is recognized as present or absent. The options identified by the FC were combined to implement the final solution using a new decision-making technique that takes into account the reliability of CRN users.

Koushik A.M et al., [32] develop a learning-based transmission spectrum to explore whether a channel with error rate in the datagram is inactive. When learning checks are moved from the Q table to the expert node, the boost takes more time to define the learning channel [20].

Goli Sai Sumith et al [33] projected a decision-making to optimize the optimal channel for delivery, which would be comparable to and superior to the existing multi-attribute decision-making (MADM) methods. This algorithm gives a high effect when the PU power is high when using the plot against static distortion. In both cases, the algorithm selects the same channel.

Dr. Ruby et al., [34] Research primarily focuses on spectrum distribution and migration selection. The problem of interference often occurs in PU and SU. It includes an intelligent method for spectrum allocation and relay selection to increase efficiency and signal-to-noise ratio (SNR). A forward based random paging selection scheme is used to select the best solution, determine the free space, create an efficient SU connection, and combine it with PSO technology. Zahra Beheshti [35] discusses the S-shaped transport function of BPSO, which is used to solve the problem of local. This algorithm increases performance efficiency and speed of collection.

SP Rajamohan and K. Umamageswari [36] influences the hybrid approach of the IPBSO concept and promotes its products based on high demand rating. This process was created to provide a credible advantage subcommittee. Duplicate reviews were identified and classification accuracy was found to be better.

Sha-sha Guo et al. Al., [37] obtainable the binary Search Location 00 and 10. There are some issues with local and global optimism. So the Z-shaped transfer function of BPSO was implemented and tested with several definitions to avoid this problem. The effect of this transmission process is to improve its accuracy and speed.

Shanidul Hoque et al. [38] Comes with M/G/1 sequencing to reduce Total Service Time (TST) for the seasoned user and to provide for Total Delivery Delay (CHD) during a SU transaction. This has been proposed using fixed, variable and random spectrum transmission. Explore PU performance and cache size based on delays.

Anastassia Gharib et al. [39] discussed about multi-band cooperative spectrum sensing to support secondary users to access the spectrum. The presented approach divides the sensing tasks based on the secondary user’s residual energy, capability, and channel conditions. For each channel, a leader has been selected by formulating the scenario into an optimization problem. Also, the presented approach includes new secondary users in the sensing process and assigns multiple channels with minimized energy consumption.

Ahmed et al. [40] presented a cooperative spectrum sensing model for cognitive radio networks based on Maximum A Posteriori estimation. The presented approach considers the secondary users as sensing nodes and gets individual decisions about the primary user occupancy status. The decisions are received by the fusion centre and posteriori estimations makes final decision which defines the primary user occupancy. Minimized interference and improved detection were achieved by the presented model compared to existing approaches.

Shunchao et al. [41] presented a clustering algorithm for cooperative spectrum sensing in cognitive radio networks. The presented approach includes k-medoids and mean shift algorithm to avoid fraudulent sensing data. The sensing status is transferred as a local energy vector to the fusion center where all the energy vectors are fused for further process. The presented clustering-based global decision model enhances the robustness of the sensing process compared to traditional approaches.

Mehran et al. [42] presented a soft decision algorithm for cooperative spectrum sensing in cognitive radio networks. The presented approach initially takes soft decision based on sensor measurements. Further a novel fast soft decision is presented based on machine learning algorithms. The decision boundaries are used to classify the power spectral density measurement vectors. The final decision function includes power spectral density of sensors to obtain faster solutions. Improved detection probability, less training time are attained by the presented model which is better than linear support vector machine based approach.

3 System model

The proposed model is portrayed in Fig. 1. A secondary FC and a PU round out the setup. PU and FC are expected to have a single transmit/receive antenna, but each SU has M receive antennas, an ED, and a solitary transmit antenna, according to our assumptions. The R-channel is used by all SUs to communicate their sensor data to the FC. Furthermore, each SU gets M antennae’s signal and processes it with M-eds for each of its antennas. SC is used to integrate the energy values at the production of the EDs right before making important decisions.

Fig. 1

CSS with censoring.

Assuming that all SUs have the same value of (the local detection threshold), the highest is initially designated and associated with a local detection threshold, indicated as, before a hard local judgement on the PU status is made. There are two possibilities for R-channel sound: perfect (no censorship). A noisy-Rayleigh faded censored S- and R-channels are both present in the presence of censoring.

The - i-th antenna (i = 1, … . M) at respectively SU can be uttered as: $y_{i} (n) = {\begin{matrix} n_{i} (n) H_{0} \\ h_{i} s (n) + n_{i} (n) H_{1} \end{matrix}$ (1)

The PU is assumed to be the source of a s (n) unidentified signal with energy E s n_i in this scenario. a complex with a circular symmetry With zero mean symmetric variance, Gaussian random variables can be characterised as complex random variables. Fading considered to be independent and identically distributed [9]. It’s a given that that I and {h_i} are distinct. The i - th antenna at each SU has a decision variable, W_i, which is used to decide if the PU is current or not: [43 –46]: $W_{i} = {| y_{i} |}^{p}$ (2) the ED parameter is p > 0. For p = 2, W_i can be reduced to the CED’s decision variable, W_i False alarm and miss detection likelihoods can be expressed as shadows: $P_{f} = \int_{λ}^{\infty} f_{Z} | H_{0} (z) dz = 1 - F_{Z |} H_{0} (λ); Z ⩾ 0,$ (3) $P_{m} = \int_{0}^{λ} f_{Z} | H_{1} (z) dz = F_{Z |} H_{1} (λ); Z ⩾ 0,$ (4) Decision variable Z is represented by $f_{Z} | H_{0} (z)$ and $f_{Z} | H_{1} (z)$ respectively, as conditional probability density functions (PDFs). Each antenna’s ED CDF can be written as [43 –46].: $F_{W_{i} | H_{j}} (x) = \Pr [{| y_{i} |}^{p} | H_{j} ⩽ x]_{j = 0, 1}$ (5) where Pr [.] denotes the likelihood of some event occurring. SC diversity is used to produce the supreme value of M assessed for dissimilar variety twigs (i.e., ${W_{i}}_{i = 1}^{M}$ ) evaluated by each SU, i.e., i = 1. Hypothesis H 0 yields the conditional CDF with SC as follows [27 –30]. $F_{z | H_{0}} (z) = {[1 - \exp (- \frac{z^{2 / p}}{σ_{n}^{2}})]}^{M}$ (6) Following the SC output, a binary hard detector is used to determine whether or not there is a attendance or nonappearance of a PU [39 –42]: $Z \begin{matrix} \begin{matrix} 1 \\ > \end{matrix} \\ \begin{matrix} < \\ 0 \end{matrix} \end{matrix} λ$ (7)

For example, the detection threshold $λ = λ_{n} σ_{n}^{p}$ , and noise standard deviation σ_n can both be written in terms of the normalised detection threshold and the noise standard deviation, respectively. An IED with the same discovery threshold is expected to exist in all SUs. In order to determine λ we must first normalise using a factor involving p [39]. The false alarm probability P_f at an SU can be stated as a function of (3), (6), and (7). $P_{f} = 1 - {[1 - \exp (- \frac{λ^{2 / p}}{σ_{n}^{2}})]}^{M}$ (8)

In a fading situation, P_f is the same. As a result, it will not be revisited.

3.1 Rayleigh faded environment

In a fading situation, P f is the same because there is no PU signal. As a result, it will not be revisited. $σ_{h}^{2}$ , i.e., $h_{i} \sim C N (0; σ_{h}^{2})$ [30]. $P_{m}^{Ra} = {[1 - \exp (- \frac{λ^{2 / p}}{σ_{n}^{2} (1 + {\bar{γ}}_{s})})]}^{M}$ (9) where ${\bar{γ}}_{s} = E_{s} σ_{h}^{2} / σ_{n}^{2}$ is the SNR of the link. For example, if we use (9), the optimal discovery threshold (abbreviated as _opt) corresponds to a total error rate of $(P_{m}^{Ra} + Pf)$ of a solitary SU is minimalized, can be gotten.

To find λ_opt, a first order partial derivative of $P_{m}^{Ra} + Pf$ with respect to λ is achieved for fixed values of p and ${\bar{γ}}_{s}$ and the result is set to zero, i.e., $\frac{\partial (P_{m}^{Ra} + Pf)}{\partial λ} = 0$ . For a generic situation with M antennas, it is difficult to track _opt analytically. Close-form languages of _opt for M = 1 and 3 can be found, however,: $[σ_{n}^{2} (1 + {\bar{γ}}_{s})$ In $(1 + {\bar{γ}}_{s}) / {\bar{γ}}_{s}]^{\frac{p}{2}}$ and ${[σ_{n}^{2} (1 + {\bar{γ}}_{s}) In (1 + \frac{{\bar{γ}}_{s}}{2 {\bar{γ}}_{s}})]}^{p / 2}$ .

3.2 Local Sensing (ED) Model

No prior knowledge of the source signal is required to utilise ED as the most commonly used algorithm. So, these are the major arguments for utilising the single-user technique that we described earlier.

It is possible to represent each SU’s signal as a binary hypothesis problem: $Y (n) = W (n) : H_{0}$ (10) $Y (n) = hX (n) + W (n) : H_{1}$ (11) where Y (n) represents the predicted signal samples at the detector, W (n) represents the Rayleigh Faded Environment, h represents the channel gain, and X represents the PU signal samples (n). According to the test statistics, the detector must select one of the hypotheses. Test statistics are used to assess the energy of the detected signal: $T (Y) = \frac{1}{N} \sum_{n = 1}^{N} {(Y [n])}^{2}$ (12) where N = f_s is Sample size, and sampling frequency are all expressed in terms of f s. In order to reach a conclusion, the test statistics are likened to a preset threshold k over an observational period and are based on the received signal. The following are possible values for the cutoff and decision statistics: $λ = \frac{[\frac{Q^{- 1} (Pf)}{\sqrt{N}} + 1]}{SNR}$ (13) where Q (.) depicts standard Gaussian’s balancing distribution function $Z = {\begin{matrix} T (Y) ⩽ λ : H_{0} \\ T (Y) > λ : H_{1} \end{matrix}$ (14)

The existence or absenteeism of a PU signal can be detected by each SU. If Z is greater than the threshold, PU exists; else, it does not exist. Chance of Pm, and probability of false alarm (Pfa) are three major performance evaluation factors that can be used to measure system performance (Pf). They can be expressed mathematically as follows: $P_{d} = P {T (Y) > λ | H_{1}$ (15) $P_{m} = P {T (Y) 〈 λ | H_{1}$ (16) $Pf = P {T (Y) > λ | H_{0}$ (17) $Pe = Pf + Pm$ (18) where Pe constitutes likelihood of error which is merger of Pf and Pm.

3.3 Cooperative spectrum sensing model

A cooperative network of M units, where K users engage in the sensing process, is studied. The signal received at FC by kth SU should be referred to as: $y_{k} = h_{k} m_{k} + n_{k}; k \in {1, 2, \dots M}$ (19) It’s the noise and channel coefficients that are used in this equation. The PSK message signal (m k) is determined by the kth SU’s judgement. Specifically, m k = 1 signifies the presence of PU whereas 0 indicates its absence. If y_k > 0, FC takes a decision (u_k) favors $H_{1}$ in decision making as subsequent: $u_{k} = {\begin{matrix} 1; H_{1} \\ 0; H_{0} \end{matrix}$ (20)

When a fusion rule is utilised, the separate local one-bit judgments are merged via FC. To aid in global decision-making, there are many MSUs (k) that have different variations of the fusion rule (). Classification and expression of the most often used fusion rules:

OR Rule: (k = 1) $H_{1}$ is preferred if there is at least one SUB reporting presence of PU; otherwise, $H_{1}$ is preferred if there is no SUB reporting presence of the P.C. Mathematically speaking, the OR can be stated as: $\land = {\begin{matrix} H_{1}; if \sum_{k = 1}^{M} u_{k} ⩾ 1 \\ H_{0}; if \sum_{k = 1}^{M} u_{k} < 1 \end{matrix}$ (21)

AND Rule: (k = N) A decision in favour of $H_{1}$ is made by FC when all SUs report the existence of a PU, but if the number of decisions is M, a decision in favour of $H_{0}$ is made. Mathematically, the AND rule is represented as follows: $\land = {\begin{matrix} H_{1}; if \sum_{k = 1}^{M} u_{k} = 1 \\ H_{0}; if \sum_{k = 1}^{M} u_{k} < 1 \end{matrix}$ (22)

Majority Rule: $(k = \frac{M}{2})$ FC takes a judgement in favour of $H_{1}$ i.e. PU if the sum of choices favouring $H_{1}$ is greater than or equivalent to the sum of decisions favouring H 0 or the other way around. There are two possible ways to numerically describe the MAJORITY rule: $\land = {\begin{matrix} H_{1}; if \sum_{k = 1}^{M} u_{k} ⩾ M / 2 \\ H_{0}; if \sum_{k = 1}^{M} u_{k} < M / 2 \end{matrix}$ (23)

In a cooperative context, the performance evaluation parameters can be expressed as follows: $Global false alarm probablity, Qf = P (u = 1 | H_{0})$ (24) $Global detection probablity, Qd = P (u = 1 | H_{1})$ (25) $\begin{matrix} Global miss - detection probablity, \\ Qm = P (u = 0 | H_{1}) = 1 - Qd \end{matrix}$ (26) $Qe = Qf + Qm$ (27) where Qe constitute global probability of error.

ANN Model: Predictor layers (input), intermediate and forecasting layers make up a neural network (output layer). In reaction to fresh information, the number of neurons in the buried layer can be changed. Using neural pattern recognition, inputs are categorised into target categories, the network is trained, and its performance is evaluated using confusion matrices. Only one neuron is required in the output layer since it must make a final binary judgement of 0 or 1. At random, each sample or dataset is split into three separate groups for the next three procedures.

Training: samples are provided to the network during training, and the network is changed based on errors.

Validation: While generalisation progress is halted, samples are collected to assess the network’s generalisation or learning.

Testing: These samples have no bearing on training, hence they serve as a stand-alone indicator of network performance.

Proposed GL-SVM Model:

In this research work, hybrid SVM that consists of Gaussian SVM and Lagrangian SVM (GL-SVM) is used and Fig. 2 shows the flow of projected GL-SVM.

Fig. 2

Workflow of proposed GL-SVM.

The features are extracted from the sensing channel by using conventional energy detector model and fed into the proposed GL-SVM model. When compared with cyclo and matched filter, the reason for choosing the energy detector is simplicity in construction and provides accurate classification, while using GL-SVM. In this way, the proposed model differs from the traditional SVM which enhances the detection accuracy. Further to improve the detection accuracy the proposed model uses double phase heterogeneous kernel that yield weights from the investigation of the train and test data.

The train data is derived from the power signal, while test data sequence is used from the received energy signal statistics. OR fusion rule is used in the weight combination process so as to assist in an efficient cooperative sensing strategy. The essence behind the utilization of two kernels is to maximize the hyperplane margin that provides a fine separation between two decisions related to presence or absence of SU in the sensed signal. The maximization function is obtained as: $O (F) = \underset{w, b}{max_{︸}} H_{M}$ (28) In the above Equation, the hyper plane margin is reflected as H_M, while the parameters of hyper-plane for weight and offset is represented by w, b and it is given as: $w . x + b = 0$ (29)

Where x represents the set of data points on the 2D space. Based on the above Equation, the spectrum sensing hypothesis get transformed into $\begin{matrix} w . x + b = + 1 if H_{M} ⩾ T_{h} and \\ w . x + b = - 1 if H_{M} ⩽ 1 \end{matrix}$ (30)

The Equation (31) represents the optimal hyperplane that needs an ideal classification. Hence, the value taken by hyperplane depends on the preciseness of H_M which for an ideal condition is formulated as $H_{M} \in Ideal = \frac{1}{2} {| w |}^{2} = \min$ (31) In other words, the optimization constraints for a standard SVM can be obtained as $O_{F_{M}} = \min \frac{1}{2} {| w |}^{2} + β \sum_{k = 0}^{p} E_{err}$ (32)

The objective function for SVM is denoted as O_{F
_M}, the learning rate is presented as β and error incurred in classification is given as E_err. If classification is not accurate or precise, E_err ≠ 0. GL-SVM is used to minimize these classification error, where the application of LSVM and GSVM to the proposed spectrum sensing problem, the solution is formulated as:

$\begin{matrix} O_{F_{M}} (L_{k}) = \min \frac{1}{2} {| w |}^{2} + β \sum_{k = 0}^{p} E_{err} - \sum_{k = 0}^{p} α_{L} \\ [r_{x} (x (t), y (t) (w . x^{T} + b) - 1 + E_{err}] - \sum_{k = 0}^{p} μ_{L} E_{err} \end{matrix}$ (33) $\begin{matrix} O_{F_{M}} (G_{k}) = \min \frac{1}{2} {| w |}^{2} + β \sum_{k = 0}^{p} E_{err} - \sum_{k = 0}^{p} α_{G} \\ [r_{x} (x (t), y (t) (w . x^{T} + b) - 1 + E_{err}] - \sum_{k = 0}^{p} μ_{G} E_{err} \end{matrix}$ (34)

The support vectors are x^T. The weights derived from both kernels after processing on that, feature vector are fused using OR rules to identify whether the SU is present or absent as: $F_{decision} (t) = sign (w . x^{T} + b)$ (35) Where sign reflects the signum function with a range of (+1,-1) and the weight computation is detected as follows: $w = \sum_{k = 0}^{p} α_{K} r_{X} (x (t), y (t))$ (36)

Finally, the proposed GL-SVM identify the SU with efficient manner.

4 Results and conversation

In this unit, the projected GL-SVM is tested with standard ANN, SVM and other techniques. The project is mainly focused on secondary users, hence it is higher than primary users. Four antennas are used and remaining simulation parameters are explained in Table 1.

Table 1
Simulation strictures and its description

Parameters Values

Area for deployment 250×250 m²

Initial node energy 0.5 Joules

No of Secondary Users 1000

No of Primary Users 3

Rule Used OR, AND, Majority, ANN, SVM, Proposed SVM

Number of Antennas (M) 4

Fading Channel Rayleigh

SNR –12dB

probability of false alarm 0.01

Fading Parameter (K) 1

Lambda (Normalized 30

detection threshold)

Parameters	Values
Area for deployment	250×250 m²
Initial node energy	0.5 Joules
No of Secondary Users	1000
No of Primary Users	3
Rule Used	OR, AND, Majority, ANN, SVM, Proposed SVM
Number of Antennas (M)	4
Fading Channel	Rayleigh
SNR	–12dB
probability of false alarm	0.01
Fading Parameter (K)	1
Lambda (Normalized	30
detection threshold)

4.1 System evaluation

Initially, the likelihood of false alarm is considered to test the efficiency of proposed model and graphical representation is provided in Fig. 3.

Fig. 3

False Alarm Probability (FAP).

From the above figure, it is clearly proved that the proposed SVM has less FAP than various existing rules such as OR, AND, Majority, ANN and standard SVM. When the secondary users are 350, the FAP of proposed model is 0.1 and OR has 0.3FAP. When the number of users upsurges, the FAP of proposed model gets decreased. The graphical representation of proposed model with other rules in terms of error rate is shown in Fig. 4.

Fig. 4

Error Rate (ER).

When the sum of secondary users are 1000, the ER for proposed model is 0.04 and the same model has 0.13 of error rate, while the secondary user is only 100. The existing ANN and SVM has less error rate than majority, AND, OR. When compared with all techniques, OR has high ER, i.e. it has 0.24 of ER when the users are 1000. Figure 5 shows the utilization of spectrum for all techniques.

Fig. 5

Spectrum utilization.

The utilization of spectrum is high in the proposed model than existing rules and techniques. Next to proposed model, ANN and standard SVM uses average spectrum than OR, AND, majority rules. The OR has less spectrum utilization and provides poor performance, which is proved in the above figure. Figure 6 shows the throughput analysis of proposed model.

Fig. 6

Throughput.

When the user upsurges, the throughput is also augmented for all techniques. The proposed SVM reaches 2.9kbps of throughput and Majority rule obtained 2.5kbps of throughput, when the user are 1000. This analysis shows proposed SVM achieved better performance than existing techniques. The probability for detection is graphically represented in Fig. 7.

Fig. 7

Detection probability.

The probability must be high for better performance, here, ANN and proposed SVM shows better detection probability than other techniques. When the user reaches 650, the detection probability is stable for both proposed model and ANN. However, the majority rules reaches the stable probability only, when the number of users are 800. However, the OR with AND rules didn’t the reach the probability. Figure 8 depicts the misclassification rate.

Fig. 8

Misclassification rate.

The rate of misclassification must be less for the better performance model. When the users are less, the misclassification rate is very less for proposed model, i.e. 0.26 when the user is 50. But the same proposed model gained the misclassification rate as 0.54, when the users are 1000. Not only proposed model, the OR rules attained high probability rate i.e. 0.71, when the users are 1000. Figure 9 represents the sensing time for all techniques.

Fig. 9

Sensing time.

The good model should take less time for sensing, but the OR takes 4 ms of sensing time, when the users are 600. For the same number of users (i.e.600), the AND takes 3.4 ms, majority rules consumed 3 ms. But, the ANN and standard SVM attained nearly 1 ms to 1.3 ms of sensing time, where the proposed model takes only 0.8 ms for sensing for 600 users. This shows the better performance of proposed model over existing techniques. The graphical representation for energy consumption using different techniques is provided in Fig. 10.

Fig. 10

Energy consumption.

The best algorithm should consume less energy even though when the number of secondary users increases. When compared with OR, AND, majority and standard SVM, the ANN consumes less energy, but higher than proposed model. This analysis shows the proposed model achieved good performance. Figures 11 12 provides the graphical representation of probability of detection with false alarms and Signal to Noise Ratio (SNR).

Fig. 11

Chance of detection vs false alarm.

Fig. 12

Chance of detection Vs SNR.

The ANN and proposed model provides stable detection probability, when the false alarm reaches 0.7. Comparing with all techniques, OR with AND rules provides less performance. Accuracy comparison for different techniques is given in Fig. 13.

Fig. 13

Accuracy.

When the number of users are 900, the accuracy of proposed SVM is 99% and it achieved 83% of accuracy when the user is only 100. The OR achieved only 64% of accuracy when the user is 100 and its maximum accuracy reaches upto 85%, when the user is high (i.e.1000).

Further to validate the performance of proposed work with recent research works a comparative analysis is presented in Table 2. From the comparative analysis it can be observed that the proposed model attains maximum accuracy compared to existing approaches. From this simulation scenario, it is proves that the proposed model achieved better presentation than other existing techniques.

Table 2

Performance comparative analysis

S.No	Method	Accuracy
1	HMM [47]	67.43%
2	MLP [47]	80.57%
3	LSTM [47]	84.08%
4	H2BMM [48]	68.2%
5	Advanced H2BMM [48]	86.48%
6	Proposed Model	99%

5 Conclusion

With SUs censored by R-channel quality, this study examined the presentation of energy detector-based CSS situations. The average miss detection probability is strongly influenced by the SU selection filtering threshold. It has been determined that the performance of both channel approximation has been associated. An optimal censoring threshold can be established based on the values of pertinent network characteristics, such as the restriction of energy detector, the normalised detection threshold, the sum of antennas, the regular S and R-channel SNRs. Increasing the sum of obtainable SUs, the sum of antennas, and the average S and R-channel SNRs. CSS systems for energy cognitive radio, for example, could benefit from these findings. Further this research work can be enhanced by adopting the feature benefits of deep learning techniques to reduce the misclassification rate and error rate in the spectrum sensing process.

6 Declaration

Funding –The author did not receive support from any organization for the submitted work.

Conflicts of Interest - The author has no relevant financial or non-financial interests to disclose.

Ethics Approval –The paper is an original contribution of research and is not published elsewhere in any form or language.

Consent Statement –All authors mentioned have contributed towards the research work, drafting of the paper as well as have given consent for publishing of this article.

Availability of Data & Material –The author hereby declare that no specific data sets are utilized in the proposed work. The have also agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Consent to publication –all authors listed above have consented to get their data and image published.

Author’s contribution –Mr.Sivagurunathan P T–Research proposal –construction of the work flow and model –Final Drafting; Dr. N.Sathish Kumar–Survey of Existing works –Improvisation of the proposed model–Formulation of pseudocode.

Code Availability –Since, future works are based on the custom codes developed in this work, the code may not be available from the author.

The authors have no relevant financial or non-financial interests to disclose.

No Humans or Animals were involved in the experimentation.

References

FCC, Federal Communications Commission Spectrum Policy Task Force, Report of the Spectrum Efciency Working Group. Technical report. USA (2002).

Mitola

and Maguire

G.A.

, Cognitive radio: making software radios more personal, IEEE Pers. Commun. Mag. 6(4) (1999), 13–18. https://doi.org/10.1109/98.788210

Harrington

, Federal communications commission, Bell Labs Tech. J. 9 (2015), 1.

Mitola

, Cognitive radio for flexible mobile multimedia communications, Mobile Netw. Appl. 6 (2001), 435–441.

Digham

F.F.

, Alouini

M.S.

and Simon

M.K.

, On the energy detection of unknown signals over fading channels, IEEE Trans. Commun. 55 (2007), 21–24.

Zeng

and Liang

Y.C.

, Eigenvalue-based spectrum sensing algorithms for cognitive radio, IEEE Trans. Commun. 57 (2009), 1784–1793.

and Lu

, Performance Analysis of Decomposed Cramer-Von Mises Detector for Blind Spectrum Sensing under Noise Uncertainty. In Proceedings of the 10th International Conference on Wireless Communications and Signal Processing (WCSP 2018), Hangzhou, China, 26–28 September 2018; pp. 1–6.

Abdelmohsen

and Walaa

, Advances on spectrum sensing for cognitive radio networks: Theory and applications, IEEE Commun. Surv. Tut. 19 (2017), 1277–1304.

Chaudhari

, Lunden

, Koivunen

and Poor

H.V.

, Cooperative sensing with imperfect reporting channels: hard decisions or soft decisions? IEEE Trans. Signal. Process 60 (2012), 18–28.

10.

Axwll

, Leus

, Larsson

E.G.

and Poor

H.V.

, Spectrum sensing for cognitive radio: State-of-the-art and recent advances, IEEE Signal Process. Mag. 29 (2012), 101–116.

11.

, Zhao

and Li

Y.G.

, Soft combination and detection for cooperative spectrum sensing in cognitive radio networks, IEEE Trans. Wirel. Commun. 7 (2008), 4502–4507.

12.

Paula

A.D.

and Panazio

, Cooperative spectrum sensing under unreliable reporting channels, Wirel. Netw. 20 (2014), 1399–1407.

13.

Nallagonda

, Chandra

, Roy

S.D.

and Kundu

, Detection performance of cooperative spectrum sensing over Hoyt and Rican faded sensing channels, IEICE Communications Express (ComEx) 2(7) (2013), 319–324.

14.

Nallagonda

, Chandra

, Roy

S.D.

and Kundu

, On performance of cooperative spectrum sensing based on improved energy detector with multiple antennas in Hoyt fading channel. In Proc. of Annual IEEE India Conference (INDICON), IIT Bombay, India (2013), 1–6.

15.

Nallagonda

, Roy

S.D.

, Kundu

, Ferrari

and Raheli

, Cooperative spectrum sensing with censoring of cognitive radios in fading channel under majority logic fusion. In Cognitive Communication and Cooperative HetNet Coexistence, Chapter 7, M.G. Di Benedetto and F. Bader, Eds. Springer, Switzerland (2014), 133–162.

16.

Nallagonda

, Roy

S.D.

, Kundu

, Ferrari

and Raheli

, Cooperative spectrum sensing with censoring of cognitive radios with majority logic fusion in Hoyt fading. In Proc. of IEEE International Conference on Advanced Networks and Telecommunications Systems (ANTS), Chennai, India (2013), 1–6.

17.

Nallagonda

, Roy

S.D.

and Kundu

, Cooperative spectrum sensing with censoring of improved energy detector based cognitive radios in Rayleigh faded channel. International Journal of Wireless Information Networks (IJWIN), Springer, 21(1) (2014), 74–88.

18.

Sun

, Zhang

and Leataief

K.B.

, Cooperative spectrum sensing for cognitive radios under bandwidth constraints. In Proc. of IEEE Wireless Communications and Networking Conference (WCNC), Kowloon (2007), 1–5.

19.

Simon

M.K.

and Alouini

M.S.

, Digital Communication over Fading Channels. 2nd edition, John Wiley and Sons, NJ, USA (2004).

20.

Chandra

, Bose

and Bose

M.K.

, Performance of non-coherent MFSK with selection and switched diversity over Hoyt fading channel, Wireless Personal Communications 68(2) (2013), 379–399.

21.

Chandra

, Bose

and Bose

M.K.

, Symbol error probability of non-coherent M-ary frequency shift keying with postdetection selection and switched combining over Hoyt fading channel, IET Communications 6(12) (2012), 1692–1701.

22.

Ciuonzo

, De Maio

and Salvo Rossi

, A systematic framework for composite hypothesis testing of independent Bernoulli trials, IEEE Signal Processing Letters 22(9) (2015), 1249–1253.

23.

Fernandez

and Williams

, Closed-form expression for the poisson-binomial probability density function, IEEE Transactions on Aerospace and Electronic Systems 46(2) (2010), 803–817.

24.

Chen

, Jiang

, Kasetkasem

and Varshney

P.K.

, Channel aware decision fusion in wireless sensor networks, IEEE Transactions on Signal Processing 52(12) (2004), 3454–3458.

25.

Lin

, Chen

and Varshney

P.K.

, Decision fusion rules in multi-hop wireless sensor networks, IEEE Transactions on Aerospace and Electronic Systems 41(2) (2005), 475–488.

26.

Niu

, Chen

and Varshney

P.K.

, Fusion of decisions transmitted over Rayleigh fading channels in wireless sensor networks, IEEE Transactions on Signal Processing 54(3) (2006), 1018–1027.

27.

Wanlu Lei , Yu Ye and Ming Xiao , Deep reinforcement learning based spectrum allocation in integrated access and backhaul networks, IEEE Transactions on Cognitive Communications and Networking 6(3) (2020).

28.

Rodney Martinez Alonso , David Plets , Margot Deruyck , Luc Martens , Glauco Guillen Nieto and Wout Joseph , Dynamic interference optimization in cognitive Radio Networks for rural and suburban areas, Wireless Communications and Mobile Computing, Wiley, (2020).

29.

Prabhat Thakur , Alok Kumar , Pandit

, Singh

and Satashia

S.N.

, Performance analysis of cognitive radio networks using channel-prediction-probabilities and improved frame structure, Digital Communications and Networks, Elsevier 4(4) (2018), 287–295.

30.

Varun

and Annadurai

, PALM-CSS: a high accuracy and intelligent machine learning based cooperative spectrum sensing methodology in cognitive health care networks, Journal of Ambient Intelligence and Humanized Computing, Springer (2020).

31.

Shah

H.A.

and Koo

, Reliable machine learning based spectrum sensing in cognitive radio networks, Wireless Sensor Networks for Smart Communications 2018 (2018), 1–17.

32.

Koushik

A.M.

, Bentley

, Fei , Hu and Kumar

, A hardware test bed for learning-based spectrum handoff in cognitive radio Networks, Journal of Network and Computer Applications 106 (2018), 68–77.

33.

Sumith

G.S.

, Mysore

and Nandakumar

, Enhanced model for spectrum handoff in cognitive radio networks. Proceedings of the 2nd international conference on electronics, communication and aerospace technology (ICECA) (2018).

34.

Ruby

, Vijayalakshmi

and Kannan

, Intelligent relay selection and spectrum sharing techniques for cognitive radio networks, Cluster Computing, Springer 22 (2017), 10537–10548.

35.

Beheshti

, A time–varying mirrored S-shaped transfer function for binary particle swarm optimization, Information Sciences 512 (2020), 1503–1542.

36.

Rajamohana

S.P.

and Umamaheswari

, Hybrid approach of improved binary particle swarm optimization and shuffled frog leaping for feature selection, Journal of Computers and Electrical Engineering, Elsevier 67 (2018), 497–508.

37.

Guo

S.-S.

, Wang

J.-S.

and Guo

M.-W.

, Z-shaped transfer function for binary particle swarm optimization algorithm, Computational Intelligence and Neuroscience Hindawi 2020 (2020), 1–21.

38.

Hoque

, Shekhar

, Sen

and Arif

, Analysis of handoff delay for proactive spectrum handoff scheme with PRP M/G/1/K queuing system in cognitive radio networks, Institute of Engineering and Technology Communications 13(6) (2019), 706–711.

39.

Anastassia Gharib , Waleed Ejaz and Mohamed Ibnkahla , Enhanced multiband multiuser cooperative spectrum sensing for distributed CRNs, IEEE Transactions on Cognitive Communications and Networking 6(1) (2020), 256–270.

40.

Ahmed Tohamy , Usama Sayed Mohamed , Mohammad Abdellatif

, Taha Khalaf

and Mohamed Abdelraheem , Cooperative spectrum sensing using maximum a posteriori as a detection technique for dynamic access networks, IEEE Access 8 (2020), 156408–156421.

41.

Shunchao Zhang , Yonghua Wang , Pin Wan , Jiawei Zhuang , Yongwei Zhang and Yi Li , Clustering algorithm-based data fusion scheme for robust cooperative spectrum sensing, IEEE Access 8 (2020), 5777–5786.

42.

Mehran Golvaei and Mohammad Fakharzadeh , A fast soft decision algorithm for cooperative spectrum sensing, IEEE Transactions on Circuits and Systems II: Express Briefs 68(1) (2021), 241–245.

43.

Singh

, Bhatnagar

M.R.

and Mallik

R.K.

, Cooperative spectrum sensing in multiple antenna based cognitiveradio network using an improved energy detector, IEEE Communications Letters 16(1) (2012), 64–67.

44.

Singh

, Bhatnagar

M.R.

and Mallik

R.K.

, Cooperative spectrum sensing with an improved energy detector in cognitive radio network. In Proc. of IEEE National Conference on Communication (NCC), Bangalore, India (2011), 1–5.

45.

Singh

, Bhatnagar

M.R.

and Mallik

R.K.

, Optimization of cooperative spectrum sensing with an improved energy detector over imperfect reporting channels. In Proc. of IEEE Vehicular Technology Conference (VTC), San Francisco, CA, USA (2011), 1–5.

46.

Nallagonda

, Chandra

, Roy

S.D.

and Kundu

, Detection performance of cooperative spectrum sensing over Hoyt and Rican faded sensing channels, IEICE Communications Express (ComEx) 2(7) (2013), 319–324.

47.

Pei Liang Zuo , Xing Wang , Wangdan Linghu , Rong Sun , Tao Peng and Wenbo Wang , Prediction-Based Spectrum Access Optimization in Cognitive Radio Networks. 018 IEEE 29th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) (2018), 1–7.

48.

Yanxiao Zhao , Zhiming Hong , Yu Luo , Guodong Wang and Lina Pu , Prediction-based spectrum management in cognitive radio networks, IEEE Systems Journal 12(4) (2018), 3303–3314.