Spectrum mobility as an essential issue has not been fully investigated in mobile cognitive radio networks (CRNs). In this paper, a novel support vector machine based spectrum mobility prediction (SVM-SMP) scheme is presented considering time-varying and space-varying characteristics simultaneously in mobile CRNs. The mobility of cognitive users (CUs) and the working activities of primary users (PUs) are analyzed in theory. And a joint feature vector extraction (JFVE) method is proposed based on the theoretical analysis. Then spectrum mobility prediction is executed through the classification of SVM with a fast convergence speed. Numerical results validate that SVM-SMP gains better short-time prediction accuracy rate and miss prediction rate performance than the two algorithms just depending on the location and speed information. Additionally, a rational parameter design can remedy the prediction performance degradation caused by high speed SUs with strong randomness movements.
1. Introduction
Cognitive radio (CR) as a solution for the next generation wireless networks brings new hope to address the wireless spectrum inefficiency problem which has attracted a great deal of attention in recent years [1–4]. In general, CR paradigms are classified in three types: interweave, underlay, and overlay. In interweave or opportunistic spectrum access (OSA) model [5, 6], CUs can use the licensed spectrums opportunistically when the spectrums are detected idle by spectrum sensing. It is very sensitive to PU traffic pattern and it relies on the detection error for the models [6, 7]. Thus, it is essential to investigate the spectrum mobility which is the foundation of resource allocation and network construction.
In a CRN, the spectrum mobility for CUs includes two aspects: spectrum mobility in the time domain and spectrum mobility in the space domain [8]. The time-varying and space-varying characteristics of the spectrum mobility lead to the problem that it is hard to access the licensed spectrums for CUs in a real network. Time-varying characteristic is because of the random variations of PUs’ arrivals and departures. Thus, some related literatures have focused on the impact of PUs’ activity on CRNs [9–11]. In [12], a selective opportunistic spectrum access scheme is proposed with the aid of PUs’ traffic prediction techniques. The scheme can estimate the probability of a channel being idle and choose the best order of spectrum sensing to maximize spectrum efficiency. The definition of channel availability vector is introduced to characterize the state information of licensed channels [13]. And a prediction-based sensing approach is presented to maximize system throughput which reduces the sensing time. In [14], a forecast scheme of call arrival rate and call holding time for PUs is proposed. CUs can reduce the frequency hopping rate through the traffic pattern prediction of PUs.
In the space domain, the movement of CUs directly results in the changing of the spectrum availability. Nevertheless, the movement of CUs, as one of the most important factors in wireless communication systems, is not adequately discussed for CRNs in existing works. A mobility model describing airborne nodes is proposed in [15]. And a stability-capacity-adaptive routing scheme is proposed to achieve high throughput and small transmission time based on the model. In [16], an optimal power control algorithm in mobile CR ad hoc networks is proposed. Without causing harmful interference to PUs, the network achieves maximized throughput based on the algorithm in the legacy network. In [17], a cluster-based routing protocol which can increase throughput and reduce data delivery latency is presented to mend the route in mobile CRNs. In [18], a general scheduling framework with the mobility information is conducted to solve maximum throughput channel scheduling problem for mobile CRNs. And two polynomial time optimal algorithms are proposed and evaluated by using the mobility trace obtained from a real public transportation system.
However, few of existing works investigate the following two issues: (1) considering time domain and space domain characteristics of spectrum mobility together and (2) considering the prediction of spectrum mobility. In practice, a CRN should be forward looking rather than reactive [19]. And a prediction-based CRN can not only improve system performance but also minimize interference to PUs [20–22], because spectrum detecting may take a long time or delay. In [23], a neural network based channel status predictor using multilayer perceptron is proposed. The system spectrum utilization is improved and the sensing energy is saved greatly by predicting the idle channels. In [24], a channel handoff scheme based on SVM is presented to reduce the handoff time. The channel handoff caused by the random movement of PUs and CUs is considered in the prediction design. In [25], a binary time series approach is used to predict the future occupancy of neighboring channels. This approach performs very well for deterministic occupancy even without updating data.
The key contributions of this paper are as follows: (1) we first take the two issues discussed above into account at the same time. And an effective joint feature vector extraction scheme is originally designed through the theoretical analysis on joint information of CUs’ mobility and PUs’ working activities. (2) Based on the extracted joint feature vector, a novel SVM-based spectrum mobility prediction scheme considering the time-space domain of spectrum mobility together is proposed for mobile CRNs in order to ameliorate the traditional prediction methods only utilizing the location and speed information directly. (3) Finally, simulations are conducted to confirm the effectiveness of the proposed prediction mechanism. The new prediction mechanism achieves higher short-time prediction performance than the conventional algorithms with little training nodes, which is vital in CRNs.
The rest of the paper is organized as follows. The system model is described in Section 2 and the spectrum availability of SUs is discussed in Section 3. In Section 4, a spectrum mobility prediction scheme is proposed based on SVM. The simulation results are shown along with a discussion in Section 5. At last, Section 6 concludes the paper.
2. System Model2.1. Mobile CRN System Model
In this paper, we consider a mobile CRN scenario where Nc CUs coexist with Np PUs illustrated in Figure 1. Assume that each PUp(p=1:Np) has a licensed access to a spectrum cp with a coverage radius Rp. Thus, the number of PUs is equal to the number of spectrums in the network. Each CUc(c=1:Nc), with an interference radius r, can exploit locally unused licensed spectrum opportunistically without causing any interference to the corresponding PUp. Suppose that the CRN assigns spectrums periodically with an allocation interval time Tc which is the interval time between two times of spectrum allocation. We also assume spectrum sensing is ideal in this paper.
System model of a mobile CRN.
Figure 1 gives out an instantaneous snapshot of a mobile CRN deployment with 20 mobile CUs. Two PUs are located in the area. The activity of each PUp(p=2 in Figure 1) is characterized as an on/off (busy/idle) model. The busy time and idle time of PUpcan be modeled by the exponential distribution with means αp and βp, respectively [26–28]. The probability density function (PDF) can be written, respectively, as
(1)fON(t,αp)=1αpe-t/αp,t≥0,fOFF(t,βp)=1βpe-t/βp,t≥0.
In this paper, a random mobility model which characterizes the movement of CUs in a two-dimensional space is considered [29]. The movement of each CUc consists of a sequence of random length intervals called mobility epochs during which CUc moves at a constant speed in a constant direction. And the mobility epoch lengths Te are independently exponentially distributed with mean 1/λe. The probability distribution function can be expressed as
(2)Me(x)=P(Te≤x)=1-e-λex.
During each epoch, the mobile direction of CUc is uniformly distributed over [0,2π) and the speed of CUc is uniformly distributed over [0,vmax]. We assume mobility is uncorrelated among all the CUs in a network. And it is reasonable to assume that epoch length, speed, and direction are uncorrelated in the model. Figure 2 shows a mobility trajectory of one given CUc as an example.
A mobility trajectory example for one given CUc.
Definition 1.
Given a licensed spectrum cp and an instantaneous time t, the instantaneous spectrum availability ISApc(t) for one CUc can be defined as
(3)ISApc(t)={-1Dp,c(t)<Rp+rc∩(αp(t)=-1)1Dp,c(t)<Rp+rc∩(αp(t)=1)1Dp,c(t)≥Rp+rc.
ISApc(t)=1 means that licensed spectrum cp is instantaneously available at t for CUc and ISApc(t)=-1 means that licensed spectrum cp is not instantaneous available at t for CUc, where Dp,c(t) represents the Euclidean distance between PUp and CUc at t. αp(t) represents the time-varying working activity of PUp, denoted as
(4)αp(t)={1PUpisidleatt-1PUpisbusyatt.
Definition 2.
Given a licensed spectrum cp and a continuous period T, the continuous spectrum availability CSApc(T) for one CUc can be defined as
(5)CSApc(T)={(c,p)∣ISApc(t)=1fromt0tot0+T},
where t0 is a given reference time. CSApc(T)=1 means that spectrum cp is available to CUc not only at t0 but also at any time between t0 and t0+T. CSApc(T)=1 means that spectrum cp is not available to CUc at some time between t0 and t0+T. In practice, T could be a slot or some slots, during which CUc can achieve activation and access to the network. Apparently, we focus more on CSApc(T) because a continuous time period (or a short-time duration) rather than an instantaneous time is more meaningful for OSA or resource allocation in CRN.
Our objective is to give out the spectrum mobility prediction for all the CUs based on the joint theoretical analysis of CUs’ mobility and PUs’ working activities under the mobile model. Obviously, it is a complicated nonlinear problem which cannot be solved by normal algorithm. In this paper, a new prediction scheme is presented to solve this problem by SVM.
2.2. Support Vector Machine
As a highly competitive learning method, SVM is gaining popularity in many fields based on the statistical learning theory [30–32]. SVM adopts structural risk minimization principle which has been shown superior to empirical risk minimization principle used by traditional neural networks [24]. Moreover, the generalization ability of SVM is strong [33]. SVM is initially used to solve the classification problem. Assume there is l training sample data denoted as
(6)D={(xi,yi)∣xi∈Rd}i=1l,
where xi is an input vector containing multiple features. yi∈{-1,+1} is a class indicator. d is the dimension of sample data. Optimal hyper plane is constructed as
(7)w·xi+b=0,
where w are weights and b is offset argument. The samples on H1 and H2 are support vectors. The according equations are w·xi+b=+1 and w·xi+b=-1, respectively. So, the classification margin is 2/∥ω∥. For our nonlinear problem, the representation of the sample data has to be changed from the original input space to a higher dimensional space which is referred to as the feature space. This quadratic programming (QP) problem can be expressed as
(8)minΦ(w,ξ)=12∥w∥2+C∑i=1lξi,s.t.myi[(w·xi)+b]≥1-ξi,i=1:l.ξi is relaxation factor and C is cost parameter which is a given value. A Lagrange function is constructed to solve the above constrained optimization problem (8) as follows:
(9)Q(w,b,α,β,ξ)=12∥w∥2+C∑i=1lξi-∑i=1lαi[ci(xi·w+b)-1+ξi]-∑i=1lβiξi,
where αi and βi are Lagrange multipliers. In order to get the solution of the original problem, we calculate the partial derivative for different variables as
(10)∂Q(w,b,α,β,ξ)∂w=w-∑i=1lαicixi=0,∂Q(w,b,α,β,ξ)∂b=∑i=1lαici=0,∂Q(w,b,α,β,ξ)∂ξ=C-αi-βi=0.
Based on (10), the original optimization problem shown in (9) is transformed to a dual optimization problem as
(11)maxJ(α)=∑i=1lαi-12∑i=1,j=1lαiαjcicjK(xi,xj),s.t.M0≤αi≤C,s.t.M∑i=1lαici′=0,i=1,2,…,l.K(xi,xj) is a chosen kernel function which will be discussed later in the prediction scheme section. Thus, we can obtain the optimization w*(12)w*=∑i=1lαi*ci′xi,
where αi* can be solved by (11). And b which does not appear in the dual problem can be calculated through the original constraint. Consider
(13)b*=12[maxci=-1(〈w*·xi〉)+minci=+1(〈w*·xi〉)].
Therefore, the final prediction output expression can be written as
(14)D(l′)=sign(∑i=1lαiciK(xi,xj)+b).
3. Joint Feature Vector Extraction
The common idea is utilizing SVM via domain information such as location and speed directly. However, it does not make good use of the CUs’ mobility characteristic and PUs’ working activities information. The traditional methods, thus, result in low prediction accuracy performance, which will be discussed in the simulation part.
In this section, CSApc(T) as joint feature vector is extracted for SVM through theoretical deduction. Obviously, CSApc(T) is related to a period T. In this paper, we focus mainly on spectrum allocation interval time Tc which is meaningful for a real CRN. Two situations need to be investigated in order to derive CSApc(Tc): (1) CUc is in the coverage area of PUp at t0 and (2) CUc is not in the coverage area of PUp at t0, respectively.
For the first situation, let CUc predict a period Tp during which CUc can use cp continuously. And our idea is to derive CSApc(Tc) through computing CSApc(Tp). In fact, CSApc(Tp) includes two main situations: (1) C1(Tp) representing the situation that CUc does not move into PUp′ coverage scope between t0 and t0+Tp and (2) C2(Tp) representing the situation that CUc moves into PUp′ coverage scope at t0+Ts(0≤Ts≤Tp) while the activity of PUp is inactive between t0+Ts and t0+Tp. We believe C1(Tp) and C2(Tp) dominate the main situations although other complicated situations as small probability events also exist.
C1(Tp) also contains two parts: (1) Pout1 representing the situation that the velocity of CUc does not change from t0 to t0+Tp and (2) Pout2 representing the other situations. Consider
(15)C1(Tp)=Pout1+Pout2.
From (2), Pout1 can be easily obtained as
(16)Pout1=1-Me(Tp)=e-λeTp.
It is difficult to get the accurate value of Pout2 because we cannot know the velocity change information (the change in time, speed, and direction) at any time for CUc. However, the approximate value E(Pout2) by estimating Pout2 can be derived [34]. When Tp<Tc, CUc has to change its movement speed and direction (or any of them) before t0+Tp, which makes CUc away from PUp. Therefore, CSApc(Tc) can be obtained as
(17)CSApc(Tc)≈E(Pout2)=1λeTp+εa+e-λeTp(12pawayλeTp-1λeTp-εa-1),
where paway denotes the probability that CUc moves away from PUp after the first velocity (speed and direction) change. In practice, εa≥0 represents other situations (small probability events except the situations discussed above). For example, CUc changes its speed three times while CUc still does not move into PUp’s coverage scope from t0 to t0+Tc. εa is used to balance the equation and we will discuss it later.
When Tp≥Tc, CSApc(Tc) for CUc can be obtained as
(18)CSApc(Tc)≈C1(Tc)+E(Pout2)=1λeTc+εa+e-λeTc(12pawayλeTc-1λeTc-εa).
For the second situation, the movement of CUc and the working state of PUp should be investigated simultaneously. Different from the first situation, we believe that the prediction of the idle state for PUp is more essential due to the original position of CUc. Similar to the first situation, let CUc predict a continuous period Tpout that CUc will not move out of PUp’s coverage boundary from t0 to t0+Tpout. It is noted that Tpout is different from the Tp mentioned above.
When Tpout<Tc, CSApc(Tc) for CUc can be obtained as
(19)CSApc(Tc)=pin·∫t0t0+TpoutfOFF(t,βp)dt+εin.
Similar to εa above, εin is used to denote all the other small probability events. ∫t0t0+TpoutfOFF(t)dt represents the idle probability of cp between t0 and t0+Tpout. pin represents the probability that CUc moves out of PUp′ coverage boundary before t0+Tp. pin consists of two parts. Pin1 denotes the situation that the velocity of CUc remains unchanged from t0 to t0+Tp. And Pin2 denotes the other situations. Pin can be easily obtained as
(20)Pin=1λeTpout+εb+e-λeTp(12pbλeTpout-1λeTpout-εb),
where εb≥0 tries to represent all the other situations. pb represents the probability that CUc moves away from PUp after the first change in velocity. According to (1), (19), and (20), CSApc(Tc) for CUc can be obtained as
(21)CSApc(Tc)=(1λeTpout+εb+e-λeTpout(12pbλeTpout-1λeTpout-εb))×(e-μpt0-e-1/βp·(t0+Tpout))+εin.
When Tpout≥Tc, CSApc(Tc) is mainly determined by PUp’s working activity. Thus, CSApc(Tc) for CUc can be obtained as
(22)CSApc(Tc)=∫t0t0+TcfOFF(t,βp)dt+εc=e-1/βp·t0-e-1/βp·(t0+Tc)+εc.εc denotes all the other spectrum availability situations. Thus, we obtain joint feature vector sets Sic={CSApc(Tc)i}i=1l according to different situations based on (17), (18), (21), and (22). Moreover, we can get the ultimate prediction expression according to (14)
(23)D(l′)=sign(∑i=1mαici′K(Sic,xj)+b),s.t.Sic={CSApc(Tc)i}i=1l.
4. Spectrum Mobility Prediction Scheme
In this section, a new SVM-based spectrum mobility prediction scheme is proposed based on the analysis and deduction above. The main steps of the proposed prediction scheme are as follows.
Step 1 (CRN initialization).
Initialize the original locations of PUs and CUs randomly in the two-dimensional deployment area. The coordinates of PUs are not changed once generated. And PUs’ initial working states are stochastic. Initialize the original speed, direction, and epoch lengths for each CUc. Assume the maximum velocity for CUs is vmax. The beginning time of the system is set to t0. The parameters C, ξi of SVM are initialized. Simulations are based on many times to make sure of the accuracy of the result. In addition, set Tc, αp, βp, εaway, εa, εb, εc, paway, pbλe, Rp, and rc.
Step 2 (operate the mobile CRN model).
PUs’ working states obey an exponential on/off model. The PDF satisfies (1). And the mobility of CUs follows the mobile model mentioned in Section 2.1. The mobility epoch lengths are independently exponentially distributed with mean 1/λe. It is noted that wrap-around technique is adopted during simulation in order to make the total number of SUs unchanged in the simulation area.
Step 3 (calculate the joint feature vectors Sic for SVM).
Here, there are three situations to be investigated. Firstly, CUc does not move into the coverage of PUp at t0 granted that CUc moves with vmax towards PUp. Secondly, CUc is out of the coverage of PUp at t0, but CUc may move into the coverage of PUp between t0 and t0+Tc. Finally, CUc is in the coverage of PUp at t0. The detail calculation steps for Sic are described in Algorithm 1.
Algorithm 1: Joint feature vectors extraction for CUc.
Calculation steps of Sic
if Dp,c(t0)≥Rp+rc+vmax·Tc
Spectrum cp is available for CUc.
else if Rp+rc≤Dp,c(t0)<Rp+rc+vmax·Tc
if Tp<Tc
Calculate CSApc(Tc) by (17).
else
Calculate CSApc(Tc) by (18).
end
else if 0<Dp,c(t0)<Rp+rc
if Tpout<Tc
Calculate CSApc(Tc) by (21).
else
Calculate CSApc(Tc) by (22).
end
end
The vectors Sic are constructed for SVM.
Step 4 (execute prediction by SVM).
Firstly, a SVM prediction model is generated according to the history input vectors Sic from Step 3. Secondly, put the data to be predicted into the generated SVM model. Then, compute the prediction results and record the results. Here, we adopt the RBF kernel as mapping function for SVM in simulation. Because the RBF kernel function tends to obtain more robust results than other kernels and can reduce numerical difficulties, the RBF kernel function can be defined as
(24)K(x,y)=exp(-γ∥x-y∥2),γ>0.
Step 5 (the system resets).
Execute Steps 1 to 4 until simulation numbers are satisfied for testing. Then, the operation stops. Compute the prediction performance: the prediction accuracy rate paccuracy and the miss detection probability pmiss. paccuracy is defined as
(25)paccuracy=|{i∣yi′·f(xi′)>0}|l′×100%,
where {xi′}1l′ are testing data which are to be predicted. And yi′∈{-1,+1} are true labels for testing data. l′ is the total number of testing data. f(xi′)∈{-1,+1} are the predicted decision values. Here, {-1,+1} represents the busy/idle working activity for a given PU. |·| represents the element numbers for a given set. Actually, paccuracy reflects the accuracy degree of spectrum prediction mechanism. The higher the paccuracy is, the better the prediction effect is.
In addition, the miss prediction rate pmiss (the rate that the spectrum is predicted to be idle while it is actually busy) is investigated for the proposed prediction mechanism. Because pmiss can reflect the actual interference to the PUs to some extent. The smaller the pmiss is, the better the prediction mechanism is. pmiss can be defined as
(26)pmiss=|{i∣yi′·f(xi′)<0,f(xi′)=1}|l′×100%.
Note that paccuracy+pmiss≤100%.
5. Simulation Results and Analysis
In this section, experimental results of the prediction performances for our proposed scheme are investigated. Simulation parameters are shown in Table 1. We compare the proposed prediction mechanism with the traditional prediction schemes (SVM with initial location coordinates of CUs and SVM with initial location coordinates and speed of CUs) to evaluate the prediction performances under different parameters.
Simulation parameters.
Parameter
Value
Total simulation number for testing
1000
Simulation area
5000 m × 5000 m
Rp
1000 m
rc
500 m
Kernel function
RBF kernel
C
10
ξi
0.01
Tc
1 s~15 s
αp
1/3 s
βp
1/3 s
pa=pb=paway
0.5
εa=εb=εc=εin=εaway
0~0.2
1/λe
3 s
CUs’ maximum velocity vmax
0 m/s~50 m/s
Total number of PUs
2
Total number of training CUs
20~240
Total number of testing CUs
1000
Note that we assume εa=εb=εc=εin=εaway during simulation for simplicity, because they are very small positive values that are set to balance the according equations. In addition, the total testing number is set to 1000 in order to avoid randomness during simulation.
Figure 3 shows paccuracy among different algorithms versus training node number. The proposed SVM-SMP converges at about 40 training data faster than SVM-location algorithm (SVM-LA) with a convergent result at about 100 training data. paccuracy of SVM-location-speed algorithm (SVM-LSA) is much worse than the other comparison algorithms which shows that the initial speed parameters have a bad effect on the prediction performance. It is caused by the time-varying characteristic of CUs’ velocity (speed or direction).
Prediction accuracy rate paccuracy for CUs versus training node number.
As shown in Figure 4, paccuracy decreases with the increasing of Tc for the three different algorithms. paccuracy of the proposed SVM-SMP is better than SVM-LA when Tc is relatively small (1≤Tc≤4.5). However, paccuracy of SVM-SMP is worse than SVM-LA when Tc is big enough (Tc>4.5). It is because the proposed prediction scheme is based on the short-time prediction idea according to the feature vector extraction analysis in (17), (18), (21), and (22). Moreover, wrap-around technique makes CUs stay at their original positions with big probability at the simulation boundary. Thus, SVM-LA decreases not that fast with the increasing of Tc. And the SVM-SMP works well when the prediction time is not long and vice versa. Note that the short-time prediction performance is mainly focused on in this paper because CR itself should achieve communication in a very short time.
Prediction accuracy rate paccuracy for CUs versus Tc.
As illustrated in Figure 5, pmiss is studied versus Tc for different algorithms. pmiss of SVM-SMP is nearly equal to 0 when Tc is small (1≤Tc≤5), which is better than SVM-LA. However, pmiss of SVM-SMP increases very fast when Tc is big enough. Figure 5 shows the good short-time miss prediction rate characteristics of the new algorithm which is very essential to CR.
Miss prediction rate pmiss for CUs versus Tc.
In Figures 6 and 7, paccuracy and pmiss versus vmax are investigated between SVM-SMP and SVM-LA. From Figure 6, paccuracy of SVM-SMP is better than that of SVM-LA when 1≤Tc≤4s. However, paccuracy of SVM-SMP is worse than that of SVM-LA when Tc=5s and vmax>42m/s. It shows that the proposed SVM-SMP lose the advantages when the prediction time and speed are too big simultaneously. As shown in Figure 7, pmiss of SVM-SMP is less than 0.1%. Comparatively, pmiss of SVM-LA is approximately 1%. Thus, SVM-SMP shows good pmiss performance with vmax changing.
Prediction accuracy rate paccuracy for CUs versus vmax.
Miss prediction rate pmiss for CUs versus vmax.
In Figures 8 and 9, we investigate paccuracy and pmiss versus λ between SVM-SMP and SVM-LA. From Figure 8, paccuracy of SVM-SMP decreases obviously with the increasing of λ when vmax is big (such as 50m/s). However, paccuracy of SVM-SMP does not change very obviously with the increasing of λ when vmax is small (such as 10m/s). It is due to the fact that the bigger the λ is, the stronger the irregular movements of CUs are. Thus, it is difficult for the prediction when CUs are moving with high speed and strong irregular movements. As shown in Figure 9, pmiss of SVM-SMP is much better than that of SVM-LA when vmax is small (such as 10m/s, 30m/s). However, pmiss of SVM-SMP is worse than that of SVM-LA when vmax=50m/s and λ>13, which validates the performance degradation of the prediction performance again when vmax is relatively big with strong irregular movements. It is because SVM-SMP is based on the assumption of weak irregular movements for CUs.
Prediction accuracy rate paccuracy for CUs versus λ.
Miss prediction rate pmiss for CUs versus λ.
In Figures 10 and 11, paccuracy and pmiss versus βp are investigated for SVM-SMP and SVM-LA. In Figure 10, paccuracy of SVM-SMP is obviously better than paccuracy of SVM-LA when the prediction time is short such as 1 s and 3 s. However, paccuracy of SVM-SMP is worse than paccuracy of SVM-LA when the prediction duration time is Tc=5s and the mean idle time is βp<4. In Figure 11, pmiss of SVM-SMP is nearly equal to 0, which is much better than SVM-LA. In addition, the prediction performance (paccuracy and pmiss) does not improve significantly for the two algorithms with the increasing of βp when βp is big enough according to Figures 10 and 11.
Prediction accuracy rate paccuracy for CUs versus μ.
Miss prediction rate pmiss for CUs versus μ.
Figure 12 shows the impact of ε on the prediction accuracy rate paccuracy for SVM-SMP. When λ=1, the maximum of paccuracy occurs at about ε=0 which is relatively small. With the increasing of λ, the maximum value position of paccuracy moves to the right. When λ=10, the maximum of paccuracy occurs at about ε=0.1. It is because the small probability events happen more times when SUs’ randomness movements are strong (λ is big). The bigger the λ is, the bigger the maximum value position of paccuracy occurs for ε. Therefore, we can obtain better prediction performance by adjusting ε for randomness movements of different strength.
Prediction accuracy rate paccuracy for CUs versus ε.
As shown in Figure 13, miss prediction rate pmiss is studied versus ε for SVM-SMP. When λ=5 and λ=10, the minimum of pmiss occurs at about ε=0.07 and ε=0.1, respectively. It shows that different optimal ε corresponds to SUs’ randomness movements of different strength for pmiss. The simulation results further confirm the impact of ε on the prediction accuracy rate in Figure 12.
Miss prediction rate pmiss for CUs versus ε.
6. Conclusions
In this paper, a new spectrum mobility prediction algorithm is proposed in mobile CRNs. SVM theory is adopted to improve the spectrum mobility prediction performance, which takes into account time- and space-varying characteristics together. Moreover, new extracted feature vectors based on the theoretical analysis are input into SVM. Simulation results confirm that the convergence speed of our SVM-SMP is faster than SVM-LA and SVM-LSA. Meanwhile, SVM-SMP shows better short-time prediction performance than SVM-LA and SVM-LSA, which is essential to real mobile CRNs. In addition, the prediction performance degradation caused by SUs’ high speed and strong randomness movements can be made up by choosing the proper parameters.
As known to us, how to choose the best parameters (C and ξi) quickly is still an open problem in SVM. We will further analyze the impact of ε on prediction performance. It leaves us to investigate in the future.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This research is supported by National Natural Science Foundation and Civil Aviation Administration of China (Grant nos. 61101122 and 61071104), National High Technology Research and Development Program of China (863 Program) (Grant no. 2012AA120802), and Postdoctoral Science-Research Development Foundation of Heilongjiang Province (Grant no. LBH-Q12080).
KaderF.AsaduzzamanHoqueM.Hybrid spectrum sharing with cooperative secondary user selection in cognitive radio networks201323220812100HaykinS.Cognitive radio: brain-empowered wireless communications20052322012202-s2.0-1384429640810.1109/JSAC.2004.839380SunS.JuY.YamaoY.Overlay cognitive radio OFDM system for 4G cellular networks20132026873WangC. X.HaiderF.GaoX.Cellular architecture and key technologies for 5G wireless communication networks2014522122130ZhaoQ.SwamiA.A decision-theoretic framework for opportunistic spectrum access200714414202-s2.0-3454858097210.1109/MWC.2007.4300978ZouY.YaoY.ZhengB.A cooperative sensing based cognitive relay transmission scheme without a dedicated sensing relay channel in cognitive radio networks20115928548582-s2.0-7865135154810.1109/TSP.2010.2090876LetaiefK. B.ZhangW.Cooperative communications for cognitive radio networks20099758788932-s2.0-6734918810810.1109/JPROC.2009.2015716AkyildizI. F.LeeW.VuranM. C.MohantyS.A survey on spectrum management in cognitive radio networks200846440482-s2.0-4264909260810.1109/MCOM.2008.4481339WangB.JiZ.LiuK. J. R.ClancyT. C.Primary-prioritized Markov approach for dynamic spectrum allocation200984185418652-s2.0-6734918212310.1109/T-WC.2008.080031AsghariV.AissaS.Adaptive rate and power transmission in spectrum-sharing systems2010910327232802-s2.0-7795790161810.1109/TWC.2010.090210.100291JungE.LiuX.Opportunistic spectrum access in multiple-primary-user environments under the packet collision constraint20122025015142-s2.0-8485998262710.1109/TNET.2011.2164933YuanG.GrammenosR. C.YangY.WangW.Performance analysis of selective opportunistic spectrum access with traffic prediction2010594194919592-s2.0-7795227824310.1109/TVT.2009.2039155YinS.ChenD.ZhangQ.LiS.Prediction-based throughput optimization for dynamic spectrum access2011603128412892-s2.0-7995284292910.1109/TVT.2010.2101090LiX.ZekavatS. A.Traffic pattern prediction and performance investigation for cognitive radio systemsProceedings of the IEEE Wireless Communications and Networking Conference (WCNC '08)March 20088948992-s2.0-51649096850HuangX.WangG.HuF.KumarS.Stability-capacity-adaptive routing for high-mobility multihop cognitive radio networks2011606271427292-s2.0-7996037106410.1109/TVT.2011.2153885SongY.XieJ.Optimal power control for concurrent transmissions of location-aware mobile cognitive radio ad hoc networksProceedings of the IEEE Global Telecommunications Conference (GLOBECOM '09)December 2009162-s2.0-7795154629210.1109/GLOCOM.2009.5425877TalayA. C.AltilarD. T.United nodes: cluster-based routing protocol for mobile cognitive radio networks2011515209721052-s2.0-8015518767910.1049/iet-com.2010.0285ZhangL.ZengK.MohapatraP.Opportunistic spectrum scheduling for mobile cognitive radio networks in white spaceProceedings of the IEEE Wireless Communications and Networking Conference (WCNC '11)March 20118448492-s2.0-7995930037810.1109/WCNC.2011.5779242ThomasR. W.DaSilvaL. A.MacKenzieA. B.Cognitive networksProceedings of the 1st IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks (DySPAN '05)November 20053523602-s2.0-3374777190910.1109/DYSPAN.2005.1542652JiangH.LaiL.FanR.PoorH. V.Optimal selection of channel sensing order in cognitive radio2009812973072-s2.0-6134915119110.1109/T-WC.2009.071363LiZ.YuF. R.HuangM.A distributed consensus-based cooperative spectrum-sensing scheme in cognitive radios20105913833932-s2.0-7684908935110.1109/TVT.2009.2031181YücekT.ArslanH.A survey of spectrum sensing algorithms for cognitive radio applications20091111161302-s2.0-7034918561810.1109/SURV.2009.090109TumuluruV. K.WangP.NiyatoD.A neural network based spectrum prediction scheme for cognitive radioProceedings of the IEEE International Conference on Communications (ICC '10)May 2010152-s2.0-7795534215610.1109/ICC.2010.5502348GuoJ.JiH.LiY.LiX.A novel spectrum handoff management scheme based on SVM in cognitive radio networksProceedings of the 6th International ICST Conference on Communications and Networking in China (CHINACOM '11)August 20116456492-s2.0-8485824824610.1109/ChinaCom.2011.6158234YarkanS.ArslanH.Binary time series approach to spectrum prediction for cognitive radioProceedings of the IEEE 66th Vehicular Technology Conference (VTC '07)October 2007156315672-s2.0-4764909501610.1109/VETECF.2007.332ZhaoQ.TongL.SwamiA.ChenY.Decentralized cognitive MAC for opportunistic spectrum access in ad hoc networks: a POMDP framework20072535896002-s2.0-3424722918510.1109/JSAC.2007.070409LeeW.AkyildizI. F.Optimal spectrum sensing framework for cognitive radio networks2008710384538572-s2.0-5514909808310.1109/T-WC.2008.070391NingG.NintanavongsaP.Time prediction based spectrum usage detection in centralized cognitive radio networksProceedings of the IEEE Wireless Communications and Networking Conference (WCNC '12)2012300305McDonaldA. B.ZnatiT. F.Mobility-based framework for adaptive clustering in wireless ad hoc networks1999178146614872-s2.0-003265100110.1109/49.780353VapnikV. N.1995New York, NY, USASpringerCristianiniN.TaylorJ. S.2000Cambridge, UKCambridge University PressLiangY.ReyesM. L.LeeJ. D.Real-time detection of driver cognitive distraction using support vector machines2007823403502-s2.0-3424986534410.1109/TITS.2007.895298WuC.YuQ.YiK.Least-squares support vector machine-based learning and decision making in cognitive radios201261728552863JiangS.HeD.RaoJ.A prediction-based link availability estimation for routing metrics in MANETs2005136130213112-s2.0-2984445224610.1109/TNET.2005.860094