Spectrum Sensing Based on Hybrid Spectrum Handoff in Cognitive Radio Networks

The rapid advancement of wireless communication combined with insufficient spectrum exploitation opens the door for the expansion of novel wireless services. Cognitive radio network (CRN) technology makes it possible to periodically access the open spectrum bands, which in turn improves the effectiveness of CRNs. Spectrum sensing (SS), which allows unauthorized users to locate open spectrum bands, plays a fundamental part in CRNs. A precise approximation of the power spectrum is essential to accomplish this. On the assumption that each SU’s parameter vector contains some globally and partially shared parameters, spectrum sensing is viewed as a parameter estimation issue. Distributed and cooperative spectrum sensing (CSS) is a key component of this concept. This work introduces a new component-specific cooperative spectrum sensing model (CSCSSM) in CRNs considering the amplitude and phase components of the input signal including Component Specific Adaptive Estimation (CSAE) for mean squared deviation (MSD) formulation. The proposed concept ensures minimum information loss compared to the traditional methods that consider error calculation among the direct signal vectors. The experimental results and performance analysis prove the robustness and efficiency of the proposed work over the traditional methods.


Introduction
The phrase "Spectrum Handoff" or "Spectrum Handover" refers to the procedure used in the cognitive radio (CR) network for users to change spectrum bands.A transceiver can intelligently determine which communication channels are in use and which ones are not in CR, a form of wireless communication [1].The transceiver then immediately switches to open channels, avoiding busy ones [2].Moreover, it increases spectrum efficiency and the consumer's quality of service (QoS) through avoiding occupied channels.With the explosive expansion of wireless communication industries [3], a significant demand exists for establishment of novel wireless networks in licensed and unlicensed frequency spectra.Recent research demonstrates that the current fixed spectral assignment approach leads to subpar spectrum utilization [4][5][6].Cognitive radio networks (CRNs) have emerged as a viable technique to solve this issue by allowing access to the sporadic intervals of vacant frequency bands, often known as white space or spectrum gaps, and therefore improving spectrum efficiency (SE) [7][8][9].In the most basic sense, every CR user in a CRN must first determine if licensed users, also known as primary users (PUs), are present and if not, Entropy 2023, 25, 1285 2 of 18 whether the spectrum is accessible.Spectrum sensing (SS) is a kind of radio frequency (RF) environment sensing that is typically used to accomplish this [10][11][12].
SS has two goals: first, CR users must get out of interfering negatively with PUs by moving to an open band to a reasonable level [13][14][15].Second, to attain the essential throughput and QoS, CR users should effectively locate and utilize the spectrum gaps [16][17][18].Therefore, the effectiveness of primary and cognitive radio networks depends on the detection accuracy in SS [19,20].
The performance of detection could be determined primarily depending upon two metrics: false alarm (FA) probability indicates the probability of a CR user stating that a PU is available while the spectra are free, and detection probability indicates the probability of CR user portraying that a PU is available while the spectra are indeed engaged by a PU [21].As a detection miss leads to intervention with PUs and a FA would lessen the SE, it is typically necessary for optimum detection performance where the probability of detection is increasingly subjected to an FA probability [22].The performance of detection in SS may be considerably hampered by a variety of issues, including receiver uncertainty, shadowing, and multipath fading [23].
The main contributions of this study is as follows.
This study proposed a component-specific cooperative spectrum sensing model (CSC-SSM) which considers the amplitude and phase components of the input signal to decrease the information loss in CRNs.
The component-specific adaptive estimation (CSAE) is proposed for calculating the mean squared deviation (MSD).
This paper is structured as follows: Section 2 describes the existing component-specific cooperative spectrum sensing (CSS) models.Section 3 explains the proposed CSAE.The component-specific adaptive estimation (CSAE) for MSD formulation is described in Section 4, whereas Section 5 presents the results.Finally, Section 6 provides the conclusion of this paper.

Related Works
In 2018, Muthukkumar and Manimegalai [24] examined the collaboration between secondary users (SUs) and main users using the Priority-Based Two-Stage Detection Model (PBTSDM).SUs in distributed CSS continually sensed among themselves and used an entropy-based energy detection approach to jointly determine whether or not PUs were present.The outcomes displayed that applying the suggested technique considerably improved the accuracy of energy efficiency (EE) and sensing time.However, noise uncertainty was a concern.
In 2017, Atmaca et al. [25] used cooperative spectrum sensing to maximize the throughput of Carrier Sense Multiple Access (CSMA) in Random Access CRNs (RACRNs).A CRN was simulated using the CSMA media access control (MAC) system in this study, with a particular emphasis on examining its throughput performance.In the identical network-level condition, throughput performances of CRNs were achieved and compared.Nevertheless, the network load needed to be concentrated more.
In 2019, Sharifi [26] offered an effective protection strategy using the Attack Aware CSS (ACSS).The concept was based on the assessment of attack strength, where attack population and assault strength were correlated.The chance that a particular sensor was malicious is equal to the ratio of malevolent sensors to all sensors, which was known as the attack strength.The suggested method predicted attack strength and used the Bayesian hypothesis test to enhance collaborative sensing performance, supposing malicious sensor activity or an attack plan.However, strong interference might affect PUs.
In 2021, Ye and Jiang [27] proposed a study on cluster-based CRNs that included an ideal linear-scaled CSS.Different weight values for cooperative nodes were assigned in this system depending on the signal-to-noise ratio (SNR) of CR users and the historic sensing accuracy.Additionally, the CR users could be grouped, and the cluster heads chosen to Entropy 2023, 25, 1285 3 of 18 collect the local sensing data were the users with superior channel characteristics.The suggested approach provided superior sensing performance while also increasing detection probability and lowering error probability, according to the simulation findings.More experimental platforms need to be considered to confirm the feasibility of this approach.
In 2021, Devi and Umamaheswari [28] included the use of the M/G/1 queuing model and the Spectrum Binary Particle Swarm Optimization (Spec BPSO) algorithm for the prediction of an efficient spectrum handoff method.Cluster-based CSS (CBCSS) was employed to increase SU effectiveness and decrease channel congestion.This research project also provided a framework for observing how main user behavior affected spectrum handoff performance delays with potential CRN interruptions.Nevertheless, metaheuristic schemes were not focused on.
In 2020, Rajaganapathi and Nathan [29] developed the accurate CSS and optimal relay selection (ORS) system, which enhanced the SUs using a hybrid CRN throughput.The precision of choosing the underlay/overlay technique to convey information was increased by an accurate CSS approach.When an underlying transmission strategy is chosen, SUs employ relays to reduce interference.An optimal relay selection approach was applied in this case to optimize relay choice.The throughput was improved by the suggested system, according to the numerical data.In the future, optimization concepts can be included to ensure more enhanced results.
To effectively use the report time slot by increasing the detecting time of SUs, in 2021, Hossain et al. [30] suggested the idea of Multiple Reporting Channels (MRCs) for clustered CRNs.In this method, each cluster was given a reporting channel for reporting purposes.The designated single reporting channel was used by all the SUs in every cluster to progressively transmit their sensing findings to the associated CH, extending the SUs' sensing time length.This method considerably improved all SUs' sensing times compared to non-sequential reporting and also reduced all cluster heads' (CHs') reporting time delays compared to sequential single-channel reporting.Multiple PUs as well as ML concepts were not taken into account.
In 2018, Jaglan et al. [31] deployed Artificial Neural Networks (ANNs) at fusion centers, which resulted in a notable improvement in detection accuracy and a decrease in the FA rate when compared to traditional methods.It was determined that the suggested ANN technique can handle CRN scalability while maintaining performance.Additionally, the SNR of each SU was taken into account while making decisions at the fusion center.Furthermore, the suggested method was evaluated for resilience against security attacks (malicious users) and unintentional mistakes happening at SUs.A minimal amount of FA issues occurred.
In 2022, Arshid et al. [32] deployed a user transmission system that senses available channels through cooperative spectrum sensing.Energy economy was achieved by optimizing the energy consumption of the sensing process.For spectrum managing, a threshold method based on main user traffic patterns was presented.A CSS was also explained and executed to find the best channel with the highest throughput and least amount of energy use.The suggested method improved throughput and energy efficiency while maintaining the handoff delay, and preventing false alarms and missed detection.
In 2022, Bani and Kulkarni [33] deployed a hybrid detector (HD) to identify spectrum holes using the available resources.An energy detector (ED) and matched detector (MD) served as the foundation for the HD architecture.The HD was able to sense the signal more accurately than a single detector like an ED.Whether or not the primary user information was accessible in this case, HD functioned under both circumstances.Under heterogeneous conditions, HD was analyzed both with and without spectrum sensing.The IEEE Wireless Regional Area Network (WRAN) 802.22 standard served as the foundation for the HD's design specifications.OR rules produced the best outcomes for the HD model.

Research Gaps
Users of CR pooled their sensory data through cooperation in order to make judgements that were more accurate when combined than when taken separately.Due to multipath fading and shadowing, the SNR of the received primary signal was very low, making the identification difficult.Since receiver sensitivity is the ability to sense weak signals, the receiver was subjected to strict sensitivity criteria, which greatly increased the implementation complexity and hardware cost.
More crucially, while the SNR of the PU signal was below what is known as an SNR wall, the detecting performance could not be increased by raising the sensitivity.Fortunately, CSS significantly decreased the sensitivity required and the hardware restriction difficulties.CSS was used to alleviate multipath fading-and shadowing-related performance loss without raising the cost of CR device installation.The cooperative advantage, however, extended beyond enhanced detection performance and loosened sensitivity requirements [34].
As was previously said, cooperative sensing led to cooperative gain, but there were a variety of conditions that restricted this benefit.For instance, their observations were coupled when CR users were stopped by the same obstruction and were under spatially correlated shadowing.Cooperation amongst more spatially connected CR users functioned as well for detection.This brought up the question of user selection in cooperative sensing [35].
The influence of nearby SUs' behavior on an SU was not taken into consideration in the conventional spectrum handoff method; additionally, the spectrum handoff condition in a single field was only carried out in CRNs [36] and the hybrid spectrum access setup merging interweave mode by underlay/m-mode which was not discussed here.Thus, a paradigm is suggested to address the inadequacies of the aforementioned existing spectrum handoff methodologies.

Component-Specific CSS Model
Spectrum handoff is regarded as the primary problem in spectrum mobility when a PU appears and SUs use this specific PU as a licensed channel.Spectrum handoff is an essential part of CRNs that enables resilient service for secondary consumers and is designed to assist secondary users in locating suitable target channels to carry out communication.The proposed CSCSSM model manages transmission power and chooses the channels with the longest holding time to avoid the spectrum handoff.
Assume P to be PUs and S to be SUs.The power spectrum discharged by every PU is captured as a linear grouping of certain basic operations.Now, Gaussian is used as a base operation.Every SU, through SS, effectively identifies the entire spectrum from every PU region.The power spectrum from PU p is modeled in Equation (1).and constraints ω m , σ m refer to the central frequency and standard deviation; g ω = g 1 e jω , g 2 e jω , g 3 e jω . . .g A e jω refers to a vector with base operations; scalars a pm refer to coefficients of the base extension for user p; and p = a p 1 , a p 2 . . ., a p A refers to a vector with aspects involved in the linear grouping of the base operations.Equation (1) can estimate the necessary part of the power spectra if A is adequate.
The power spectra from SU s is identified through PU p which is attenuated owing to transmission path loss implied by q ps .The path loss coefficient is identified and described earlier in a training phase among PUs by every SU.Training is typically repetitive at certain periods since the coefficients vary (gradually) in time, owing to the movement of the node.If the broadcasted spectrum moves from PU to SU, the previous power spectra are evaluated by the receiver of the SU s, denoted as q ps K p e jω .Therefore, the entire power spectra from every PU at SU s are modeled as in Equation ( 2).
In Equation (2), o s = T 1 , T 2 . . .T P T (P.A × 1) and v s,ω = q sP ⊗ g ω (1 × P.A) and σ 2 s is the receiver noise.Observe that T p implies that a pm k is included in the power spectra composition of PU p; therefore, o s concatenates the a pm k of every PU p.At every time period i, s notices the received power spectra in a discrete frequency {ω r } in a period [0, π] under the size and noise u s,r by mean zero and covariance matrix C u s of size O × O as shown in Equations ( 3 In Equation ( 6), u s refers to model noise and/or measurement with mean zero and C u s of size O × O.At O diverse frequencies, the measurements are taken and therefore, the matrix has O rows.Consequently, in Equation ( 6), a linear model is attained for computing constraints significance in o s .The steps for processing are described below.1.
The power spectrum of PU, denoted by p, is subjected to path loss attenuation [37].

2.
The path loss attenuation is subjected to the total power spectrum and thus, the power spectrum model is obtained.

3.
The measurement model per SU s is computed based on the power spectrum of PU, path loss attenuation, and total power spectrum [38], and the model as shown in Equations ( 7)- (10).
Entropy 2023, 25, 1285 The factor for path loss is modeled as in Equation (10).
In Equation (10), b ps,i refers to the Euclidean distance from s to p at i; b o refers to a reference distance that is b o = 1; and n designs [39] the attenuation surroundings in CRN [40].Therefore, the values for path loss among SU s and P PUs are modeled as in Equation (11).
q s,i = [q 1s,i , q 2s,i ....q Ps,i ] In the assessment of q s,i , a relevant Gaussian noise of mean zero and SD σ q is considered; accordingly, qs,i = q s,i + n s .If SU s changes, q s varies its distance from PUs which also varies accordingly.
For estimating the spectrum, it is adequate to approximate the constraint vector, which factorizes the base operations.Depending upon the network data {b s,i , V s,i }, the issues are treated as an assessment of numerous benefits, and assistance is presumed among the nodes for processing information in a dispersed manner as per the Adapt Then Combine (ATC) policy.The aforesaid policy estimates the centralized outcomes if every node desires to approximate a similar vector of constraints.
Every vector { o s } S s=1 includes constraints which are important for the entire model's constraints of mutual importance to node subset together with other nodes s, and constraints of local importance for node s.In particular, subsets of constraints in o s account or:

•
A global constraint vector associated with the frequency band in power spectra of every PU that impacts every node present in the CRN.

•
In a case where J diverse subsets of general constraints is considered, the observation model offered in Equation ( 6) is rewritten as Equation (12).
Conventionally, every node tries to resolve by using the subsequent optimization issue [41] as shown in Equation (13).
arg min As per the concept, the amplitude and phase components are considered separately and the optimization issue is defined as shown in Equations ( 14)-( 16) based upon f and ζ 1 , ζ 2 , . .., ζ J in which, I s refers to a well-organized set of index j related with vector ζ j , which is of interest to node s; V s f and V s c j refer to matrices of sizes O × M f and O × M c j , respectively, and includes columns of V s,i related with f and ζ s,j .arg min

Component-Specific Adaptive Estimation (CSAE) for MSD Formulation
Here, the diffusion technique ATC which includes an adaptation and a combination phase is exploited.The key phases of the ATC method are as follows: 1.
Consider φ ; fulfill The adaptation stage and combination stage at i th iteration is shown in Equations ( 17) and ( 18), respectively.
For every j ∈ I s , ζ . When the algorithm ends, ϕ s, f and ϕ s,ζ j k approximate the required o f and ζ o j k.Presuming a clique topology, i.e., λ s ∩ Γ j = Γ j for every s ∈ Γ j , the even combination rule forms combination weights as in Equations ( 19) and (20).
In conventional work, the adaptive weighting method is deployed as in Equations ( 21) and (22).
As per our concept, the amplitude and phase components are considered separately and the adaptive weighting mechanism is defined as shown in Equations ( 23) and (28).
e j( Ŵ(g:r,s,1)−W(g:r,s,1)) (29) e j( Ŵ(g:e,s,i)−W(g:e,s,i)) ( In Equation (23), u refers to a smaller positive value between [0, 1] and γ s,l and δ s,l refers to variance in the evaluation of common and global interest constraints.Subsequently, the weights related to both common and global parameter evaluation process is performed as shown in Equations ( 37) and (38).
Algorithms 1 and 2 show the pseudocode for CSAE and MSD estimation.Output: MSD: S × (J + 1) × iter, Ŵ : M × S × iter Input: S, O, M, J, iter, µ, B, m t , b, W, V aug , Γ Step 1: Initialization Ŵ = L M×S×iter , Ŵ(:, :, 1) = randn(M, S, 1)e = L O×S×iter , MSD = L S×J+1×iter for s = 1 : S do g = 0, r = 0 for j = 1 : Z α(d) and B β(d) are computed as shown in Equations ( 29) and ( 30) end end Step2: Iterative Part for i = 2 : iter do Adaptation Step for each node for s = 1 : S do e(:, s, i) = b(:, s, i) − V aug (:, :, s, i) Ŵ(:, s, i − 1) do Ŵ(:, s, i) = Ŵ(:, s, i − 1) + µV H aug (:, :, s, i)e(:, s, i) end for s = 1 : S do Global: Adaptive Weight Estimation for l = 1 : S do Z α(d) and B β(d) are computed as shown in Equation (31) and Equation (32 Elect only global constraint vectors from every user: ϕ f = Ŵ(1 : m t (1), :, i) Concatenate a global set of constraints from every user ϕ f = ϕ f (:) Combining step for Global ϕ f = R ϕ f , Ŵ(1 : m t (1), :, i) = reshape ϕ f , m t (1), S General: Adaptive Weights Estimation z = 0, y = m t (1) for j = 1 : Z α(d) and B β(d) are computed as shown in Equations ( 33) and (34 Elect userconcerned for j th subset of constraints: d = f ind(R(:, j + 1) = 0) Elect j th subset of M h j general constraints from user in d = ϕ ζ j = Ŵ(z : y, h, i) Concatenate j th subset of general constraints from every user ϕ ζ j = ϕ ζ j (:)  The MSD error evaluation of the CSCSSM was compared to that of the PBTSDM, ORS-ACSS, Spec BPSO-QM, and ATC methods by adjusting the σ to 0.1.Also, a network with Q = 3 PUs or 5 PUs, and K = 7 SUs, 11 SUs, or 15 SUs was simulated, and the findings are displayed in Figure 2. On examining Figure 2c, it is evident that the CSCSSM maintained the MSD error value for at the time 9 as approximately −27.48 dB, which is better than PBTSDM with −16.62 dB, ORS-ACSS with −19.78 dB, Spec BPSO-QM with −21.67 dB, and ATC with −23.54 dB.Simultaneously, at time 8, the CSCSSM generated an MSD of −28.42 dB as seen in Figure 2e; meanwhile, the standard methodologies scored the lowest MSD, notably, PBTSDM = −16.84dB, ORS-ACSS = −18.93dB, ATC adaptive weights = −19.48dB, and ATC = −26.74dB.As a result, the CSCSSM had reduced and minimized MSD errors when compared with the current methodologies.The MSD error evaluation of the CSCSSM was compared to that of the PBTSDM, ORS-ACSS, Spec BPSO-QM, and ATC methods by adjusting the σ to 0.1.Also, a network with Q = 3 PUs or 5 PUs, and K = 7 SUs, 11 SUs, or 15 SUs was simulated, and the findings are displayed in Figure 2. On examining Figure 2c, it is evident that the CSCSSM maintained the MSD error value for at the time 9 as approximately −27.48 dB, which is better than PBTSDM with −16.62 dB, ORS-ACSS with −19.78 dB, Spec BPSO-QM with −21.67 dB, and ATC with −23.54 dB.Simultaneously, at time 8, the CSCSSM generated an MSD of −28.42 dB as seen in Figure 2e; meanwhile, the standard methodologies scored the lowest MSD, notably, PBTSDM = −16.84dB, ORS-ACSS = −18.93dB, ATC adaptive weights = −19.48dB, and ATC = −26.74dB.As a result, the CSCSSM had reduced and minimized MSD errors when compared with the current methodologies.The effectiveness of the CSCSSM was assessed compared to the PBTSDM, ORS-ACSS, SpecBPSO-QM, and ATC methods by varying the α from 0.1 to 1 in terms of the MSD error measure and is presented in Table 1.Here, it a network of Q = 3 PUs or 5 PUs and K = 7 SUs, 11 SUs, or 15 SUs was simulated.In particular, while the Q was fixed to 3 PUs and K was fixed as 11 SUs, the CSCSSM recorded an MSD error of −27 dB (α = 0.9), whereas the value for PBTSDM was −13 dB, ORS-ACSS was −21 dB, SpecBPSO-QM was −24 dB, and ATC was −25 dB.For Q = 5 PUs and K = 7 SUs, the CSCSSM had the lowest MSD error rate of −23 dB (α = 0.8).Meanwhile, the conventional methodologies had the highest MSD error values: PBTSDM (0.8 dB), ORS-ACSS (−13 dB), SpecBPSO-QM (−17 dB), and ATC (−20 dB).The CSCSSM performed well in the MSD error measurements compared to the conventional algorithms, indicating that the MSD errors of the established algorithms is extremely high.The comparison of CSCSSM to PBTSDM, ORS-ACSS, SpecBPSO-QM, and ATC for both datasets is represented in Figure 3.The MSD error evaluation was carried out while fixing the σ to 0.2 and a network was designed to simulate Q = 3 Pus or 5 PUs and K = 7 The effectiveness of the CSCSSM was assessed compared to the PBTSDM ACSS, SpecBPSO-QM, and ATC methods by varying the α from 0.1 to 1 in terms

=
g ω p , p = 1, 2.....P (1) In Equation (1), A refers to the number of CRs present in the network; K p refers to the summation of signals received at each CH; g m e jω = e − (ω−ωm )

Figure 1 .
Figure 1.Assessment of network MSD (dB) of the CSCSSM versus traditional schemes for w (global) and Common ς1, ς2 using a network with (a) Q = 3 PUs, K = 7 Sus; (b) Q = 3 PUs, K = 11 Sus; (c) Q = 3 PUs, K = 15 SUs while fixing the σ to 0.05.5.2.Analysis of Network MSD for the CSCSSM and the Conventional Methods with a Network of Q = 3 PUs or 5 PUs and K = 7 SUs, 11 SUs, or 15 SUs Simulated while Fixing the σ to 0.1 Network MSD for the CSCSSM and the Conventional Methods with a Network of Q = 3 PUs or 5 PUs and K = 7 SUs, 11 SUs, or 15 SUs Simulated While Fixing the σ to 0.2The comparison of CSCSSM to PBTSDM, ORS-ACSS, SpecBPSO-QM, and ATC for both datasets is represented in Figure3.The MSD error evaluation was carried out while fixing the σ to 0.2 and a network was designed to simulate Q = 3 Pus or 5 PUs and K = 7 SUs, 11 SUs, or 15 SUs.According to Figure3a, the CSCSSM generated an MSD error rate at time 10 of −32.84 dB, while for the PBTSDM, it was −19.56 dB; ORS-ACSS, it was −25.01 dB; SpecBPSO-QM, it was −28.65 dB; and ATC, it was −29.89 dB.Considering Figure3eat time 7, the models PBTSDM, ORS-ACSS, SpecBPSO-QM, and ATC achieved an MSD error value of −11.24 dB, −19.82 dB, −22.56 dB, and −23.74 dB, although the CSCSSM reported an MSD error of −26.18 dB.This implies the MSD error value is diminished in the CSCSSM in contrast to the previous schemes.

5. 4 .
MSD Error Analysis of CSCSSM and Conventional Methods with a Network of Q = 3 PUs or 5 PUs and K = 7 SUs, 11 SUs, or 15 SUs Simulated by Varying the α