Non-preemptive Queueing Model of
Spectrum Handoff Scheme Based on
Prioritized Data Traffic in Cognitive Wireless Networks
Muhammed Enes Bayrakdar and Ali Cß alhan
In this study, a non-preemptive M/G/1 queueing model of a spectrum handoff scheme for cognitive wireless networks is proposed. Because spectrum handoff gives secondary users an opportunity to carry on their transmissions, it is crucially important to determine the actions of primary users. In our queueing model, prioritized data traffic is utilized to meet the requirements of the secondary users. These users’ packets are categorized into three different priority classes: urgent, real-time, and non-real time. Urgent data packets have the highest priority, while non-real time data packets have the lowest priority. Riverbed (OPNET) Modeler simulation software was used to simulate both reactive and proactive decision spectrum handoff schemes. The simulation results were consistent with the analytical results obtained under different load and traffic conditions. This study also revealed that the cumulative number of handoffs can be drastically decreased by exploiting priority classes and utilizing a decent spectrum handoff strategy, such as a reactive or proactive decision-based strategy.
Keywords: Cognitive radio networks, Priority, Queueing model, Spectrum handoff.
I. Introduction
Cognitive radio is a new technology that improves spectrum utilization with the assistance of an exclusive sensing mechanism [1], [2]. Through this mechanism, secondary usersfind and use underutilized spectrum holes without causing any harm to the licensed primary users [2], [3]. When primary users perform actions in licensed channels, secondary users cease their communications in these channels orfind new channels in which to continue their transmissions [4], [5]. Carrying on communications in a new channel in order to prevent harmful interference to primary users is known as a spectrum handoff process [6], [7]. In the spectrum handoff process, secondary users change their communication channels using either reactive or proactive decision-based approaches [6], [8]. In reactive decision-based spectrum handoffs, secondary usersfind new channels in an on-demand way using spectrum sensing mechanisms [9], [10]. In proactive decision-based spectrum handoffs, spectrum handoff channels are decided beforehand [3], [11]. Reactive decision-based spectrum handoffs increase the number of handoffs, while proactive decision-based spectrum handoffs are free of spectrum sensing time [4], [12].
Several research articles about spectrum handoffs and priority classes have recently been published [13], [14]. Shiang and Schaar investigated a dynamic channel-selection strategy for independent wireless users [1]. In their approach, knowledge interchange among users was an essential tool for managing available spectrum resources. However, they did not take priority classes into account. Zahed and others proposed a prioritized, proactive spectrum-handoff decision scheme to reduce total service time [2]. In their study, a preemptive resume-priority
Manuscript received Nov. 22, 2016; revised Jan. 13, 2017; accepted Feb. 9, 2017. Muhammed Enes Bayrakdar (corresponding author, muhammedbayrakdar@ duzce.edu.tr) and Ali Cßalhan (alicalhan@duzce.edu.tr) are with the Department of Computer Engineering, Duzce University, Duzce, Turkey.
This is an Open Access article distributed under the term of Korea Open Government License (KOGL) Type 4: Source Indication + Commercial Use Prohibition+ Change Prohibition (http://www.kogl.or.kr/news/dataView.do?data Idx=97).
M/M/1 queue was utilized, but a number of handoff parameters were not evaluated. Chu and others investigated priority traffic-based dynamic spectrum access for cognitive radio networks using a multi-dimensional Markov chain [3]. In their work, blocking and dropping probabilities were evaluated, but the cumulative number of handoffs was not taken into consideration.
Zhang and others investigated a Markov transition model integrated with a preemptive resume-priority M/M/ 2 queueing network [4]. The cumulative handoff delay was comprehensively evaluated in their study, but the cumulative number of handoffs was not considered. Wang and Wang proposed spectrum-handoff mechanisms for cognitive radio networks [5]. Reactive sensing-based spectrum handoffs and proactive decision-based spectrum handoffs were studied comparatively, while repetitive-spectrum handoff problems were ignored. Lertsinsrubtavee and others proposed a new spectrum-handoff mechanism that aimed to reduce redundant spectrum-handoff processes [6]. In their research, backup channels were utilized to evade repeated spectrum handoffs, while priority was not considered. Arif and others investigated the probabilities of spectrum handoffs for secondary users under different traffic models [7]. In their paper, different distribution models were simulated in to validate theoretical results, while priority classes were disregarded. Yang and others proposed an optimal frequency-selection scheme for secondary users [8]. In their study, the optimal frequency was obtained by considering tradeoffs between the interference and successful transmission probabilities, while successive-spectrum handoff was neglected. Lee and Yeo analyzed both spectrum-handoff delay and channel availability for secondary users [9]. In their work, the effects of spectrum-sensing time, data-transmission time, and the state transitions of primary users were studied comparatively, while priority classes were not taken into account.
In this study, a non-preemptive M/G/1 queueing model of a spectrum-handoff process for cognitive wireless networks is proposed. Because spectrum handoff offers secondary users an opportunity to continue their communications, it is necessary to determine the actions of primary users. In our queueing model, priority-based data traffic is employed to meet the requirements of the secondary users. These users’ packets are grouped into three distinctive priority classes: urgent, real-time, and non-real time. Urgent data packets have the highest priority, while non-real time data packets have the lowest priority.
The main contributions of our work to the existing body of research are as follows: (i) considering priority classes, including urgent, real-time, non-real time classes, (ii)
decreasing the number of handoffs by avoiding repetitive spectrum handoff, (iii) simulating both proactive and reactive decision schemes, (iv) evaluating the blocking probability, forced termination probability, handoff delay, and number of handoffs, (v) adapting a non-preemptive M/G/1 queueing model, and (vi) validating simulation results obtained from Riverbed (OPNET) software with analytical results obtained from MATLAB software.
II. Spectrum Handoff in Cognitive Radio Networks
Cognitive radio is a technology that allows secondary users to utilize spectrum holes in an opportunistic way [13]. Secondary users are commonly considered as visitors to the spectrum because they have no license [2], [9]. Hence, when some parts of the spectrum exploited by secondary users are needed for licensed usage, the communications of secondary users need to either be stopped or carried on in other available parts of the spectrum [15], [16]. Continuing communication in another part of the spectrum is known as a spectrum handoff [17], [18]. Generally, there are three different events for which spectrum handoff happens: when an action of a primary user is discovered on the licensed spectrum, when a secondary user fails the connection because of mobility, or when an available spectrum cannot support the bandwidth and data rate requirements [5], [12].Figure 1 depicts the classification of spectrum handoff techniques. Spectrum handoff processes in cognitive radio networks are mainly comprised of switching handoffs and non-switching handoffs. The switching handoff technique is further subdivided into reactive and proactive decision based schemes in related literature [19], [20].
1. Reactive-Decision Spectrum Handoff
In reactive-decision based spectrum handoffs, secondary users carry out the spectrum-handoff process after
Spectrum handoff process
Switching handoff Non-switching handoff
Reactive decision
Proactive decision
discovering that communication has been lost [3]. This method requires instant spectrum handoff based on spectrum sensing with no planning time and causes considerable performance degradation for secondary networks [14]. Reactive spectrum handoff is commonly used in the immediate appearance of primary users in the channel. The main spectrum-sensing equation is formulated as
X½i ¼ W i½ ; H0; i ¼ 1; 2; 3; . . . ; I S i½ þ W i½; H1; i ¼ 1; 2; 3; . . . ; I
; (1) where H0represents the absence of primary users and H1
represents the existence of primary users. X[i], S[i] and W [i] are a sample of a signal received by the secondary users, a Gaussian-distributed random noise process with a mean of zero, and a signal received by the primary users with a mean of zero, respectively. With the help of (1), an equation for the decision value is derived as
E ¼X I i¼1 X½ij2 H1 R H0 c; (2)
where E represents the decision value andc is a predefined decision threshold.
Pd ¼ P E cH1
: (3)
In (3), Pd represents the probability of detection and
P[ j ] denotes the conditional probability. If the decision value is greater than or equal to the predefined threshold while a primary user is active, then the existence of the primary user has been correctly detected [21].
Pfa¼ P E cH0
: (4)
In (4), Pfarepresents the probability of a false alarm. If
the decision value is greater than or equal to the predefined threshold when a primary user is not active, then the existence of a primary user has been falsely detected [21].
The relationship between Pdand Pfais defined as
Pfa¼ Qð ffiffi I p RSNþ Q1ðPdÞ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1þ 2RSN p Þ; (5)
where RSN represents signal-to-noise ratio and I is the
number of samples. Q() and Q1() denote a general q-function and the inverse of a general q-function.
2. Proactive-Decision Spectrum Handoffs
Proactive decision-based spectrum handoffs are more suitable when user mobility and quality degradation are present on the spectrum [22], [23]. In this scheme, secondary users prepare target channels for spectrum
handoff processes before starting transmissions [11]. After taking the results of long-term observation into account, a set of selected candidate channels is prepared for the forthcoming spectrum handoffs [5]. Proactive-decision spectrum handoff does not spend time on sensing. Therefore, handoff delays and total service times are shorter in proactive schemes than in reactive schemes. However, predetermined target channels may later become unavailable in a proactive decision scheme.
In Table 1, performance metrics of spectrum handoffs are listed. Probability of detection is defined as the probability of correctly detecting primary user activity on the spectrum. Probability of false alarm means the probability of falsely detecting primary user activity on the spectrum when it there is none [21]. Probability of missed detection is the opposite of the probability of detection in that it is the probability of detecting that the spectrum is idle when it is occupied by a primary user. Blocking is defined as obstructing secondary users due to the scarcity of the idle spectrum. Forced termination occurs when there is no available spectrum to carry out spectrum handoff processes [22], [23]. Spectrum handoff delay is formulated as
Dsh ¼ tswþ s; (6)
where Dshand tswdenote the spectrum handoff delay and
the switching time, respectively.s is the spectrum sensing time, which is equal to zero for proactive-decision spectrum handoffs. The average spectrum handoff delay is given as:
E D½ sh ¼ kp tswrsþ E Tp 2 kprsþ E Tp ks tswrskp h i 1 kpE Tp rs ð Þ2 ; (7) where E[ ] denotes the average of the concerned expression, kp and ks represent the arrival rates of the
primary and secondary users, respectively, and Tpandrs
are the mean service time of the primary users and the
Table 1.Performance metrics of spectrum handoffs.
Performance metric Description
Pd Probability of detection
Pfa Probability of false alarm
Pmd Probability of missed detection
Pb Probability of blocking
Pft Probability of forced termination
Dsh Spectrum handoff delay
packet inter-arrival time of the secondary users, respectively.
3. Non-switching Spectrum Handoffs
In non-switching spectrum handoffs, secondary users await the availability of their present spectrums instead of changing spectrums [3]. Non-switching handoffs result in long handoff delays because of potential waiting times [3]. Moreover, long waiting times cause substantial performance degradation for secondary users [9]. Because the channel-switching time is zero in non-switching handoff processes, the delay expression is formulated as:
E D½ sh ¼ kp E Tp 2 kprsþ E Tp ks ð Þ h i 1 kpE Tp rs ð Þ2 : (8)
Two different priority methods are commonly used for spectrum handoffs: preemptive and non-preemptive priority approaches [19]. In a preemptive priority approach, the communications of low-priority secondary users are interrupted [19]. This approach is commonly utilized in real-time applications [10]. In a non-preemptive priority approach, spectrum handoff occurs after the ongoing transmissions of low-priority secondary users have been completed [3]. A non-preemptive priority approach is commonly utilized in applications, which do not require immediate actions [6].
III. Proposed Queueing Model
Our proposed network model is a cognitive wireless network environment that consists of primary users, secondary users, and base stations, as shown in Fig. 2. Since primary users are licensed users, they have the right to access their time slots at any time. Secondary users communicate through the base station only when primary users have empty time slots. As a queueing model for secondary users, a non-preemptive M/G/1 prioritized-data traffic model composed of different priority classes, including urgent, real-time, and non-real time classes.
Primary base station
Secondary user (Urgent)
Primary user
Secondary base station
Primary user
Primary user
Primary user Primary user
Primary user Primary user Primary user Primary user Primary user Secondary user (Real time) Secondary user (Non-real time) Secondary user
(Urgent) Secondary user
(Real time) Secondary user (Non-real time)
Fig. 2.Network model of the proposed system.
Packet of secondary user Yes No Yes No Yes No Yes No Yes Yes No No Yes No Yes No Is there an urgent packet? Is there a non real time packet?
All the packets are transmitted and the queue is empty.
Find an idle time slot and transmit the packets.
Find an idle time slot and transmit the packets.
Find an idle time slot and transmit the packets. Is spectrum handoff required? Is spectrum handoff required? Is spectrum handoff required? Is there a new packet in the queue? Is there a new packet in the queue? Get the priority
class of the packet.
Push the packet into the queue regarding priority.
Is there a real time packet?
All of the included performance metrics are derived analytically. In our simulation scenarios, a reactive decision-based scheme and a proactive decision-based scheme are applied and compared.
Figure 2 shows the primary base station, the secondary base station, ten primary users, and six secondary users with different priority classes. For our simulation scenarios, packet inter-arrival time is the longest for urgent data packets and the shortest for non-real time data packets. Urgent data packets consist of industrial, science, and medical data as our network model is based on the ISM (Industrial, Science, Medical) band. Real-time data packets consist of instant messaging, online video-conferencing, web meetings, and other related types of communication, while non-real time data packets are any other forms of data communication.
Figure 3 depicts a block diagram of the queueing model. The queueing strategy pushes packets into the queue according to their priorities. If the priority of a packet is urgent, then the transmission of the packet is started. If a time slot is required by a primary user while it is transmitting, a spectrum handoff process is carried out, and transmission continues in any other time slot. Once all of the urgent packets in the queue have been transmitted, real-time and non-real time data packets are transmitted using idle time slots. In contrast to the transmission of urgent packets, newly-arrived real-time and non-real time data packets are always checked before initiating a spectrum handoff.
1. An Analytical Model of the Primary Network
In our work, primary users communicate via their base stations using TDMA as a medium-access control protocol. Because TDMA is a non-collision technique, the TDMA frame is divided into time slots of equal lengths, as illustrated in Fig. 4. Every primary user in the network has a unique time slot for communication. When the frame is divided into ten parts, the slot time for primary users becomes 0.1 s.
In our network model, the packet transmission interval for primary users is equal to the time slot length. Each primary user in the network environment is an independent
source, resulting in a Poisson distribution. Therefore, the load offered to primary users is represented by
Gp ¼ kpNp; (9)
where Npand Gpare the number of primary users and the
primary user load, respectively. The packet generation probability of primary users expressed according to the Poisson distribution is
PMðmÞ ¼
GpmeGp
m! : (10)
As can be seen from (10), the packet generation probability is derived using the load offered to primary users, where m denotes an event occurring in a definite interval.
2. An Analytical Model of the Secondary Network In our network model, secondary users employ a slotted Aloha random access scheme to utilize the time slots left idle by primary users. Secondary users’ packets are PU0 PU1 PU2 PU3 PU4 PU5 PU6 PU7 PU8 PU9
Time Slot time
0.1 s
Frame time, 1 s
Fig. 4.Frame structure of the primary network.
Total frame time
PU EMPTY SU PU EMPTY PU PU PU PU EMPTY
PU PU EMPTY PU SU PU PU PU PU EMPTY
PU PU PU PU SU PU EMPTYEMPTY PU PU
PU PU PU PU EMPTY PU SU PU PU PU
Sensing
delay Data transmission Handoff
delay Data transmission
Proactive decision spectrum handoff Reactive decision spectrum handoff Handoff delay N ex t fra me
PU: primary user (active) EMPTY: time slot (idle) SU: secondary user (active) : spectrum handoff process
Fig. 5.Frame structure of the secondary network.
Head of the queue Tail of the queue Urgent packet arrival Real-time packet arrival Non real-time packet arrival
Urgent Urgent … Real-time Real-time … real-timeNon real-timeNon real-timeNon real-timeNon …
divided into three different priority classes: urgent, real-time, and non-real time. The possible actions of secondary users in the cognitive wireless network environment are as follows: if the time slot being utilized will remain idle for the next slot, the secondary user continues to use it; if the next slot is not empty, the secondary user tries tofind an available time slot for the transmission. In the case that the secondary user finds an available channel, a spectrum handoff process is carried out to continue the transmission. If the secondary user cannot discover an available time slot, it is essential to stop the transmission and wait for the availability of a later time slot.
In Fig. 5, the spectrum handoff process for secondary users is illustrated. In our cognitive radio wireless network model, it is assumed that each secondary user produces packets according to a Poisson distribution process. However, the average packet lengths of secondary users are shorter than those of primary users due to the spectrum-sensing and spectrum-handoff delays. In a proactive decision-based scheme, the sensing delay is zero, because predetermined channels are utilized when making a spectrum handoff.
In Fig. 6, the queue structure of the secondary network is illustrated. In the queue, arrangement of the packets according to priority classes is performed as in (6). For equal priority classes, a newly arrived packet is pushed to the back of all of the other packets of the same priority in the queue using the FCFS (First Come First Served) algorithm. The blocking probability of the secondary network, Pb, is calculated as follows:
Pb ¼ Gpþ ð1 GpÞ ks
Gbs
ksGbsþ Ts;
(11)
where Ts is the mean service time of secondary users and
Gbs is the equivalent workload of each base station. The
forced termination probability of the secondary network is derived from the blocking probability as
Pft¼ ð1 PbÞ
Pb1=Gbs1
TsDshþ Pb1=Gbs1
; (12)
where Ts is the mean service time of secondary users and
Gbsis the equivalent workload of each base station.
3. A Simulation Model of the Proposed Network The Riverbed (OPNET) Modeler simulation software includes several functions such as designing, simulation, and data aggregation [24], [25]. Configuration of the wireless network is carried out in three stages: the network
stage, the node stage, and the process stage. In the network stage, the topology of the network is formed. The node stage defines the behavior of the node and controls the data flow in different modules of the node. In the process stage, there are state machines that consist of states and the transitions between them. The source code of Riverbed (OPNET) Modeler is written with the C programming language [25].
Our simulation model was tested for three different channel models tofind the probabilities of detection and false alarm. In Fig. 7, a comparison of channel models is shown in so that the effects of environmental conditions on the proposed network are visible. We have utilized an AWGN channel model for our graphical results to detect the differences between priority classes. Because detection and sensing are very important parameters for cognitive radio networks, channel models are investigated utilizing the probabilities of detection and false alarm.
In Fig. 8, the end to end delay performance is illustrated to help determine the user capacity. It is clear that for more than 25 users, jitter effects occur in the network environment, so the number of users in our simulation scenarios was chosen accordingly.
For our network model, we used the Riverbed (OPNET) Modeler simulation software to design, model, and simulate the proposed network environment. All of the digital modulation schemes were evaluated. The QPSK linear modulation technique was chosen because it provides same data rate as BPSK does for half of the required bandwidth. The FSK scheme was not chosen because of bandwidth inefficiency. M-ary modulation schemes were not preferred because of their requirement of higher transmission power. Considering the importance
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Probability of false alarm (Pfa)
Probabi lity of detect ion ( Pd ) Rician Rayleigh AWGN
of both bandwidth and power efficiency for our network, the QPSK technique was the best match.
In our simulation scenario, there were ten primary users and 6 secondary users with randomly distributed priority classes in the wireless network area. A free-space loss-channel propagation model was employed to determine the power of the received packets for the reactive-decision based scheme. The rest of the parameters utilized in the simulation are specified in Table 2.
The process model of the secondary base station used in the simulation scenarios is illustrated in Fig. 9. Because the Riverbed (OPNET) simulation software is event driven, after the state variables are defined, their first values are assigned for the initial state [24]. Then, the process goes into an idle state and waits for an interrupt
representing the occurrence of a new event. In the priority state, after the priority of a packet is obtained, the packet is pushed into the queue accordingly. In the receive state, when a packet is received by the secondary base station, the source and destination addresses of the packet are checked. If the packet belongs to the base station, the data statistics are updated. Otherwise, the packet is discarded. The process enters the send state at the beginning of each time slot to transmit ready packets from the queue.
The process model utilized in the simulation scenarios for secondary users is demonstrated in Fig. 10. Because the Riverbed (OPNET) simulation software is event-driven, after all of the variables are defined, their first values are allocated in the initial state. Then, the process
Initial
state
Idle
state
Send
state
Handoff
state
Receive
state
Queueing
state
Fig. 10.Process model of secondary users in the Riverbed (OPNET) simulation software.
500 1,000 1,500 2,000 2,500 3,000 0.08 0.10 0.12 0.14 0.16 0.18 0.20 0.22 0.24 Time (s) E n d to end delay (s ) 25 secondary users 50 secondary users 100 secondary users
Fig. 8. Network user capacity.
Table 2.Simulation parameters.
Parameter Value
Slot length for primary users 100 ms
Number of slots in a frame 10
Frequency for primary and secondary users 2.4 GHz
Bandwidth for primary and secondary users 2 MHz
Data rate for primary and secondary users 1 MBps
Modulation scheme QPSK
Transmit power for secondary users 100 mW
Probability of detection 0.91–0.99
Probability of false alarm 0.01–0.09
Packet size of primary users 12 Kb
Packet size of secondary users 10 Kb
Spectrum handoff delay 4 ms
Initial state
Idle
state Receive state
Send state Priority
state
Fig. 9.Process model of the secondary base station in the Riverbed (OPNET) simulation software.
goes into the idle state and waits for an interrupt signifying the occurrence of a new event [24]. In the handoff state, after an idle time slot is identified, the highest-priority packet carries out the spectrum handoff process. In the queueing state, after the source and destination information of the nodes is added to the packets coming from the upper stages, they are pushed into the queue according to priority. In the receive state, when a packet is received by the secondary user, the source and destination addresses of the packet are controlled. If the packet belongs to the secondary user, the data statistics are updated. Otherwise, the packet is discarded. The process enters the send state at the beginning of each time slot to transmit packets waiting in the queue.
IV. Analysis of Numerical Results
A performance evaluation of the spectrum handoff process with priority classes was carried out for both reactive and proactive decision schemes by using simulation scenarios. Owing to their importance, the cumulative handoff delay, blocking probability, forced termination probability, and cumulative number of handoffs were taken into account when evaluating the spectrum handoff scheme. The analytical and simulated results of our proposed network were obtained with MATLAB software and Riverbed (OPNET) simulation software, respectively.
Figure 11 shows the cumulative handoff delays for proactive-decision spectrum handoff when the arrival rate of primary users increases from 0.02 to 0.2. The cumulative handoff delay for secondary users is higher
when the mean service time of primary users is 0.9. The cumulative handoff delay of secondary users is lower when the mean service time of primary users is 0.4. The mean service time of primary users was chosen to realize its effects on cumulative handoff delay.
Figure 12 shows the cumulative handoff delays of proactive-decision and reactive-decision spectrum handoffs for increasing arrival rates of primary users. The cumulative handoff delay for the reactive-decision based scheme is higher than that of the proactive-decision based scheme because of the difference in spectrum-sensing time. The simulated results for cumulative handoff delay are almost the same as the analytical results obtained for different mean service times of primary users.
Figure 13, shows the cumulative handoff delay for the reactive-decision based scheme considering different priority classes for a mean primary users’ service time of
E[Tp]=0.9 (Ana.-Proact.) E[Tp]=0.4 (Ana.-Proact.) E[Tp]=0.9 (Sim.-Proact.) E[Tp]=0.4 (Sim.-Proact.) E[Tp]=0.9 (Ana.-React.) E[Tp]=0.4 (Ana.-React.) E[Tp]=0.9 (Sim.-React.) E[Tp]=0.4 (Sim.-React.) 0.25 0.20 0.15 0.10 0.05 C u mulative hando ff delay 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 Arrival rate of primary users
E[Tp] = 0.9 (ana.-proact.) E[Tp] = 0.4 (ana.-proact.) E[Tp] = 0.9 (sim.-proact.) E[Tp] = 0.4 (sim.-proact.) E[Tp] = 0.9 (ana.-react.) E[Tp] = 0.4 (ana.-react.) E[Tp] = 0.9 (sim.-react.) E[Tp] = 0.4 (sim.-react.)
Fig. 12.Comparison of cumulative handoff delay results.
Urgent (Ana.) Real Time (Ana.) Non-Real Time (Ana.) Urgent (Sim.) Real Time (Sim.) Non-Real Time (Sim.) 0.25 0.20 0.15 0.10 0.05 C u mulative hando ff delay 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 Arrival rate of primary users
Urgent (ana.) Real time (ana.) Non-real time (ana.) Urgent (sim.) Real time (sim.) Non-real time (sim.)
Fig. 13.Cumulative handoff delay results for priority classes.
E[Tp]=0.9 (Analytical) E[Tp]=0.4 (Analytical) E[Tp]=0.9 (Simulation) E[Tp]=0.4 (Simulation) 0.25 0.20 0.15 0.10 0.05 C u mulative hando ff delay 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
Arrival rate of primary users E[Tp] = 0.9 (ana.)
E[Tp] = 0.4 (ana.)
E[Tp] = 0.9 (sim.)
E[Tp] = 0.4 (sim.)
Fig. 11.Cumulative handoff delay results for proactive decision spectrum handoff.
0.9 s. The cumulative handoff delay of urgent traffic is the lowest because of its high priority. Real-time traffic waits in the queue longer than urgent traffic does, so the cumulative handoff delay of real-time traffic is higher than that of urgent traffic. Non-real time packets wait much longer in the queue than any other kind of packet because of their low priority.
Probability of blocking is defined as the probability of obstructing secondary users due to the scarcity of idle spectrum. In this case, secondary users are not able tofind available time slots because of the intense load offered to the primary users. Figure 14 shows the blocking probability for different arrival rates of secondary users when the offer loads for primary users are 0.2, 0.3, and 0.4. An increase in the load offered to primary users raises the blocking probability for secondary users. When the arrival rate of secondary users is 0.2, the blocking probabilities for secondary users are 0.4, 0.47, and 0.55 when offered loads for primary users are 0.2, 0.3, and 0.4, respectively. Simulation results closely match the analytical results because of the sensitivity of the graphic.
Figure 15 shows the blocking probabilities under different arrival rates of secondary users when the load offered to primary users is high. An increase in the load offered to primary users enhances the blocking probability for secondary users. When the arrival rate of secondary users is 0.3, the blocking probabilities for secondary users under different primary loads approach one another, with values between 0.75 and 0.85. Taking Figs. 14 and 15 into consideration, the blocking probability of secondary users reaches a steady state as the load offered to primary users increases.
Forced termination occurs when there is no spectrum available to carry out a spectrum handoff process. In cases of forced termination, secondary users’ packets are dropped and must be transmitted again. Figure 16 shows the forced termination probabilities for different arrival rates of secondary users when the loads offered to primary users are low, such as 0.2, 0.3, and 0.4. The load offered to primary users greatly affects the forced termination probability. As this load increases, the forced termination probability of secondary users increases at almost the same rate. When the arrival rate of secondary users is 0.2, the forced termination probabilities for secondary users are about 0.0001, 0.005, and 0.016, with increasing values of offered primary loads.
Figure 17 shows the forced termination probability results under different arrival rates of secondary users
Gp=0.35 (Analytical) Gp=0.45 (Analytical) Gp=0.55 (Analytical) Gp=0.35 (Simulation) Gp=0.45 (Simulation) Gp=0.55 (Simulation) 0.9 0.8 0.7 0.6 0.5 B locking probab ility 0.3 0.05
Arrival rate of primary users 0.4 0.10 0.15 0.20 0.25 0.30 Gp= 0.35 (ana.) Gp= 0.45 (ana.) Gp= 0.55 (ana.) Gp= 0.35 (sim.) Gp= 0.45 (sim.) Gp= 0.55 (sim.)
Fig. 15.Blocking probability results for high primary loads.
Gp=0.2 (Analytical) Gp=0.3 (Analytical) Gp=0.4 (Analytical) Gp=0.2 (Simulation) Gp=0.3 (Simulation) Gp=0.4 (Simulation) 0.8 0.7 0.6 0.5 0.4 B locking proba b ility 0.2 0.05
Arrival rate of primary users
Gp= 0.2 (ana.) Gp= 0.3 (ana.) Gp= 0.4 (ana.) Gp= 0.2 (sim.) Gp= 0.3 (sim.) Gp= 0.4 (sim.) 0.3 0.10 0.15 0.20 0.25 0.30
Fig. 14.Blocking probability results for low primary loads.
Gp=0.2 (Analytical) Gp=0.3 (Analytical) Gp=0.4 (Analytical) Gp=0.2 (Simulation) Gp=0.3 (Simulation) Gp=0.4 (Simulation) 0.035 0.030 0.025 0.020 0.015 Forc ed te rm in at io n p roba b ility 0 0.05
Arrival rate of secondary users 0.010 0.10 0.15 0.20 0.25 0.30 0.005 Gp= 0.2 (ana.) Gp= 0.3 (ana.) Gp= 0.4 (ana.) Gp= 0.2 (sim.) Gp= 0.3 (sim.) Gp= 0.4 (sim.)
Fig. 16.Forced termination probability results for low primary loads.
when the load offered to primary users is high, such as 0.35, 0.45, and 0.55. An increase in the load offered to primary users raises the forced termination probability for secondary users. For high loads offered to primary users, the forced termination probability approaches 0.1. When the arrival rate of secondary users is 0.3, the forced termination probabilities for secondary users are about 0.02, 0.05, and 0.11, with increasing values of offered primary loads.
Figure 18 shows the cumulative number of spectrum handoff results when the mean service time of primary users is 0.9 s. To illustrate the differences between the reactive-decision and proactive-decision based schemes, the cumulative numbers of handoffs are shown comparatively. When the proactive-decision based scheme is employed, the maximum number of handoffs is about
320, and the maximum number of handoffs is almost 550 when the reactive-decision based scheme is utilized. The cumulative number of handoffs for the reactive-decision based scheme is higher than for the proactive-decision based scheme. This is because the proactive scheme uses predetermined slots for spectrum handoff, while reactive scheme suffers from high false alarm and missed detection probabilities.
Figure 19 demonstrates the throughput results for different priority classes and the base station. For different priority classes, the throughputs are close to one another. Because of the congestion and packet loss in the base station, the throughput there is nearly 7 packets per second. In Fig. 20, the packet-loss ratios are shown for different priority classes and the base station. The base station suffers more packet loss than users with any priority class do. Priority classes do not affect the packet
Base station (Sim.) Urgent (Sim.) Real Time (Sim.) Non-Real Time (Sim.) Base station (Ana.) Urgent (Ana.) Real Time (Ana.) Non-Real Time (Ana.) 0.40 0 500 1,000 1,500 2,000 2,500 3,500 0.10 0 3,000 0.15 0.35 0.25 0.45
Base station (sim.) Urgent (sim.) Real time (sim.) Non-real time (sim.) Base station (ana.) Urgent (ana.) Real time (ana.) Non-real time (ana.) 0.30 0.20 0.05 4,000 Simulation time (s) Pac k et los s ra tio
Fig. 20.Packet loss ratio.
Proactive (Ana.) Reactive (Ana.) Proactive (Sim.) Reactive (Sim.) 500 450 250 200 0 500 Simulation time (s) 100 1,000 1,500 2,000 2,500 3,500 Proactive (ana.) Reactive (ana.) Proactive (sim.) Reactive (sim.) 50 0 3,000 150 400 350 300 550 Cumulative nu mber of hando ff
Fig. 18.Cumulative number of handoff results.
Gp=0.35 (Analytical) Gp=0.45 (Analytical) Gp=0.55 (Analytical) Gp=0.35 (Simulation) Gp=0.45 (Simulation) Gp=0.55 (Simulation) 0.12 0.10 0.08 0.06 Forced term inatio n probabi lity 0 0.05
Arrival rate of secondary users 0.04 0.10 0.15 0.20 0.25 0.30 0.02 0 Gp= 0.35 (ana.) Gp= 0.45 (ana.) Gp= 0.55 (ana.) Gp= 0.35 (sim.) Gp= 0.45 (sim.) Gp= 0.55 (sim.)
Fig. 17.Forced termination probability results for high primary loads.
Base station (Sim.) Urgent (Sim.) Real Time (Sim.) Non-Real Time (Sim.) Base station (Ana.) Urgent (Ana.) Real Time (Ana.) Non-Real Time (Ana.) 9.0 8.5 7.0 5.0 500 Simulation time (s) 6.0 1,000 1,500 2,000 2,500 3,500 5.5 0 3,000 6.5 8.0 7.5 9.5 T h ro ughput ( p acket/s ) 4,000 Base station (sim.) Urgent (sim.) Real time (sim.) Non-real time (sim.) Base station (ana.) Urgent (ana.) Real time (ana.) Non-real time (ana.)
loss, and the loss ratio of the network is acceptable. In Fig. 21, the bit error rates for the different channel models are shown. As expected, AWGN gives the lowest bit error rate among all the channel models.
V. Conclusion
In this paper, a non-preemptive M/G/1 queueing model for the spectrum handoff process of cognitive radio networks is proposed. Because spectrum handoff offers secondary users an opportunity to continue their transmissions, it is necessary to carefully detect the actions of primary users. In the queueing model, prioritized data traffic is exploited to meet the necessities of the secondary users. Secondary users’ packets are grouped into three distinctive priority classes: urgent, real-time, and non-real time. Urgent data packets have the highest priority, while non-real time data packets have the lowest priority. A cognitive radio network that consisted of primary users, secondary users, and base stations was designed. Riverbed (OPNET) Modeler simulation software was utilized to simulate both reactive-decision and proactive-decision spectrum handoff processes. Simulation results were shown to be consistent with the analytical results obtained under various offered loads and traffic conditions. This study also found that the cumulative number of handoffs could be significantly decreased by utilizing priority classes and suitable spectrum handoff process, such as reactive-or proactive-decision based. In future studies, a new queueing model may be designed to overcome starvation problem for low-priority secondary users. Moreover, varying packet size may be considered for the data traffic of secondary users.
Acknowledgments
This work was supported by the Scientific Research Projects Office of Duzce University, Turkey (2016.07.02.514, Channel Modeling with Body Fading Effect for Cognitive Radio based Body Area Networks).
The authors would like to thank the editor and the reviewers for their valuable comments and recommendations.
References
[1] H.P. Shiang and M.V.D. Schaar,“Queuing Based Dynamic
Channel Selection for Heterogeneous Multimedia
Applications over Cognitive Radio Networks,” IEEE Trans. Multimedia, vol. 10, no. 5, Aug. 2008, pp. 896–909. [2] S. Zahed, I. Awan, and G. Min, “Prioritized Proactive
Scheme for Spectrum Handoff Decision in Cognitive Radio Networks,” Int. Conf. Broadband, Wireless Comput., Commun. Applicat., Victoria, Canada, Nov. 12–14, 2012, pp. 335–341.
[3] T.M.C. Chu, H. Phan, and H.J. Zepernick, “Dynamic
Spectrum Access for Cognitive Radio Networks with Prioritized Traffics,” IEEE Commun. Lett., vol. 18, no. 7, July 2014, pp. 1218–1221.
[4] L. Zhang et al.,“Modeling for Spectrum Handoff Based on Secondary Users with Different Priorities in Cognitive Radio Networks,” IEEE Int. Conf. Wireless Commun. Signal Process., Huangshan, China, Oct. 25–27, 2012, pp. 1–6.
[5] L.C. Wang and C.W. Wang, “Spectrum Handoff for
Cognitive Radio Networks: Reactive Sensing or Proactive Sensing,” IEEE Int. Performance, Comput. Commun. Conf., Austin, TX, USA, Dec. 7–9, 2008, pp. 343–348.
[6] A. Lertsinsrubtavee, N. Malouch, and S. Fdida,
“Controlling Spectrum Handoff with a Delay Requirement in Cognitive Radio Networks,” Int. Conf. Comput. Commun. Netw., Munich, Germany, July 30–Aug. 2, 2012, pp. 1–8.
[7] W. Arif et al., “A Comprehensive Analysis of Spectrum Handoff Under Different Distribution Models for Cognitive Radio Networks,” Wireless Personal Commun., vol. 85, no. 4, Dec. 2015, pp. 2519–2548.
[8] Z. Yang, Y. Song, and D. Wang, “An Optimal Operating
Frequency Selection Scheme in Spectrum Handoff for Cognitive Radio Networks,” Int. Conf. Comput., Netw. Commun., Anaheim, CA, USA, Feb. 16–19, 2015, pp. 1066–1070.
[9] D.J. Lee and W.Y. Yeo, “Channel Availability Analysis
of Spectrum Handoff in Cognitive Radio Networks,”
AWGN (Sim.) Rician (Sim.) Rayleigh (Sim.) AWGN (Ana.) Rician (Ana.) Rayleigh (Ana.) 0.010 0 500 Simulation time (s) 1,000 1,500 2,000 2,500 3,500 0.002 0 3,000 0.004 0.008 0.006 0.012 B it e rror ra te AWGN (sim.) Rician (sim.) Rayleigh (sim.) AWGN (ana.) Rician (ana.) Rayleigh (ana.)
IEEE Commun. Lett., vol. 19, no. 3, Mar. 2015, pp. 435– 438.
[10] L.C. Wang, C.W. Wang, and C.J. Chang,“Modeling and
Analysis for Spectrum Handoffs in Cognitive Radio Networks,” IEEE Trans. Mobile Comput., vol. 11, no. 9, Sept. 2012, pp. 1499–1513.
[11] C.W. Wang and L.C. Wang,“Modeling and Analysis for
Proactive Decision Spectrum Handoff in Cognitive Radio Networks,” IEEE Int. Conf. Commun., Dresden, Germany, June 14–18, 2009, pp. 1–6.
[12] C.W. Wang, L.C. Wang, and F. Adachi, “Modeling and
Analysis for Reactive Decision Spectrum Handoff in Cognitive Radio Networks,” IEEE Global Telecommun. Conf., Miami, OH, USA, Dec. 6–10, 2010, pp. 1–6. [13] L.C. Wang, C.W. Wang, and C.J. Chang,“Optimal Target
Channel Sequence Design for Multiple Spectrum Handoffs in Cognitive Radio Networks,” IEEE Trans. Commun., vol. 60, no. 9, Sept. 2012, pp. 2444–2455.
[14] Y. Liu, B. Pei, and H. Wang, “An Intelligent Spectrum Handoff Scheme based on Overflow Queuing Theory in
Cognitive Radio Networks,” URSI General Assembly
Scientific Symp., Beijing, China, Aug. 16–23, 2014, pp. 1– 4.
[15] M. Barooah et al., “An Architectural Framework for
Seamless Handoff between IEEE 802.11 and UMTS Networks,” Wireless Netw., vol. 19, no. 4, May 2013, pp. 411–429.
[16] F. Liu et al.,“Immune Optimization Algorithm for Solving Vertical Handoff Decision Problem in Heterogeneous Wireless Network,” Wireless Netw., vol. 19, no. 4, May 2013, pp. 507–516.
[17] Y. Konishi et al., “Performance Analysis of Dynamic
Spectrum Handoff Scheme with Variable Bandwidth
Demand of Secondary Users for Cognitive Radio
Networks,” Wireless Netw., vol. 19, no. 5, July 2013, pp. 607–617.
[18] J.H. Chu, R.T. Ma, and K.T. Feng, “Stochastic Spectrum Handoff Protocols for Partially Observable Cognitive Radio Networks,” Wireless Netw., vol. 20, no. 5, July 2014, pp. 1003–1022.
[19] M. Li, T. Jiang, and L. Tong, “Spectrum Handoff
Scheme for Prioritized Multimedia Services in Cognitive Radio Network with Finite Buffer,” IEEE Int. Conf. Dependable, Autonomic Secure Comput., Chengdu, China, Dec. 21–22, 2013, pp. 410–415.
[20] M. Mehrnoush, R. Fathi, and V.T. Vakili, “Proactive
Spectrum Handoff Protocol for Cognitive Radio Ad Hoc Network and Analytical Evaluation,” IET Commun., vol. 9, no. 15, Oct. 2015, pp. 1877–1884.
[21] S. Atmaca, “Improving TDMA Channel Utilization in
Random Access Cognitive Radio Networks by Exploiting Slotted CSMA,” Wireless Personal Commun., vol. 71, no. 4, Aug. 2013, pp. 2417–2430.
[22] M. Fahimi and A. Ghasemi,“Analysis of the PRP M/G/1
Queuing System for Cognitive Radio Networks with Handoff Management,” Iranian Conf. Electr. Eng., Tehran, Iran, May 20–22, 2014, pp. 1047–1051.
[23] Y. Wu et al., “Spectrum Handoffs with Mixed-Priority
Queueing Model over Cognitive Radio Networks,” IEEE
Global Conf. Signal Inform. Process., Austin, TX, USA, Dec. 3–5, 2013, pp. 1194–1197.
[24] A. Karahan et al.,“Effects of Transmit-Based and Receive-Based Slot Allocation Strategies on Energy Efficiency in WSN MACs,” Ad Hoc Netw., vol. 13, Feb. 2014, pp. 404– 413.
[25] Riverbed (OPNET) Modeler Software, Accessed 2016. http://www.riverbed.com/products/steelcentral/steelcentral-riverbed-modeler.html
Muhammed Enes Bayrakdar was born in 1988. He received his MS degree from Kocaeli University, Kocaeli, Turkey. He is now a research assistant at the Computer
Engineering Department of Duzce
University, Turkey. His research interests are cognitive radio networks, spectrum handoff, priority classes, probability theory, queueing theory, and Riverbed (OPNET) simulation software.
Ali Cß alhan received his MSc and PhD
degrees from the University of Kocaeli, Turkey in 2006 and 2011, respectively. Since 2011, he has been a member of the
Computer Engineering Department of
Duzce University. His research interests are wireless communications, cognitive radio networks, body area networks, and software-defined networks.
