Professional Documents
Culture Documents
ABSTRACT Enhancing spectrum utilization efficiency (SUE) to accommodate the multitude of 5G-based
IoT devices within the available scarce spectrum has been a pivotal point of research in the current decade.
Equipped with interweave and underlay modes, cognitive radio networks (CRNs) are envisioned to be the
most promising technology for SUE enhancement. Since the 5G-based IoT is swiftly transforming into
a heterogeneous network, hybrid underlay-interweave mode CRNs appear to be the optimal key for SUE
enhancement. Besides enhancing SUE, ensuring fairness among secondary users (SUs) to equitably utilize
the network’s resources in the wake of primary users’ (PUs) service emergence has been a critical problem in
CRNs. Considering the importance of SUE enhancement, in this paper, we investigate two problems. Firstly,
we precisely analyze, in a novel way, the performance of hybrid underlay-interweave mode of CRNs for SUE
enhancement from the standpoint of SUs. For this purpose, we propose a hybrid underlay-interweave mode
enabled CRN scheme and apply it on a legacy CRN. We model the proposed scheme with continuous time
Markov chain and derive closed form expressions for various SUE-related quality of service parameters. The
analysis is conducted under dynamic reservation of channels for interrupted users, considering the effects of
multi-levels network traffic loads and varying channel failure rates. Secondly, we propose a multi-attribute
based fairness driven algorithm for determination and interruption of SUs’ services to ensure fairness among
services in the network’s resources utilization. We evaluate our proposed scheme under both perfect and
imperfect spectrum sensing scenarios. The obtained results demonstrate that, as compared to the state-of-
the-art, the proposed scheme significantly enhances SUE while the proposed algorithm achieves a noticeable
fairness among SUs’ services.
INDEX TERMS Cognitive radio networks, dynamic spectrum access, dynamic channel reservation, hybrid
underlay-interweave mode, spectrum utilization efficiency.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 63403
A. U. Khan et al.: HBLP: Hybrid Underlay-Interweave Mode CRN for the Future 5G-Based IoT
underutilized. The underutilized part of the licensed spectrum channel failure has been overlooked in performance evalu-
is also termed as white space [8]. To make efficient utilization ation of CRNs.
of the white space, CRNs enable SUs to use the idle parts of Another essential aspect of CRNs investigated by several
the spectrum provided that no interference issues are created researchers is to dynamically reserve a certain number of
for PUs’ services [8]. channels for the support of interrupted or failed services.
Enhancing spectrum utilization efficiency in CRNs is one In [19], a channel reservation scheme is proposed to ana-
of the main objectives of CRNs’ research community in the lyze the performance of CRNs under channel failures. It is
current decade. For instance, in [9], PUs’ traffic allocation found that service retainability can be enhanced through
is optimized for the division and assignment of bandwidth channel reservation. The authors in [20] proposed a static
in such a way as to maximize spectrum utilization effi- channel reservation scheme for PUs. In [21], a channel reser-
ciency using interweave mode (IM). In [10], soft cooperative vation scheme is proposed to analyze the impact of arrival
spectrum sensing is used to enhance spectrum utilization. and service rates of SUs on their blocking, dropping and
For this purpose, the length of SUs’ time slots inside the hand off probability. The authors in [22] proposed a low-
time frame is adjusted. The authors in [11] proposed MIMO complex channel reservation scheme by using one dimen-
based energy and spectrum optimization for 5G using auction sional Markov chain for the enhancement of call level QoS
game. provisioning.
The utility of CRNs lies in dynamic spectrum access Another novel and interesting concept in CRNs is licensed
(DSA). DSA allows SUs to use the idle part of the spectrum shared access (LSA) [23] wherein the licensed and unlicensed
dynamically. Various flavors of DSA have been proposed in users can share the spectrum under certain sharing rules
the literature. In [12], a Q-learning-based dynamic spectrum agreed upon by both the users. When a licensed user grants
access algorithm is proposed. The algorithm addresses the a part of its spectrum to an unlicensed user for a predeter-
problem of imperfect spectrum sensing for idle channel avail- mined amount of time, the licensed user cannot interrupt the
ability detection in distributed CRNs. In [13], a dynamic spec- unlicensed user during that period of time [24]–[28]. In this
trum access algorithm is developed. The algorithm utilizes fashion, channel auction and bidding are avoided in channel
deep multi-user reinforcement leaning. During each time slot, allocation since the licensed user willingly shares a portion
users compare their current states to the spectrum access of its spectrum. Consequently, a predictable scale of QoS
actions obtained from a trained deep-Q network. The objec- satisfaction can be gained by SUs’ services.
tive is to tackle the problem of network utility maximization An important scheme in this domain is proposed in [28].
under dynamic spectrum access. The scheme, which, for brevity, is termed as BLP in this
Under DSA, CRNs operate in one of the three modes: inter- paper, proposes a dynamic channel reservation (DCR) algo-
weave, underlay and overlay [14]. In IM, SUs use only the rithm and evaluates various performance metrics related to
unused portion of the spectrum. In the underlay mode (UM), spectrum utilization efficiency. BLP is based on the concept
PUs and SUs can use the same part of the spectrum, simulta- of LSA and operates in two modes, Mode 1 and Mode 2,
neously, provided the SUs’ transmission power is below the selected by the operator as per his requirements. Mode 1 aims
PUs’ interference threshold level. In overlay mode too, both to enhance service retainability of ongoing services while
PUs and SUs can use the same spectrum, simultaneously. Mode 2 aims to reduce blocking probability of new services.
Certain literature has used hybrid modes [15]–[18]. In [15], To this end, Mode 1 reserves a large number of channels
a hybrid underlay-interweave approach is considered wherein under heavy network traffic load and a small number of
SUs transmit in both the presence and absence of active PUs. channels under low network traffic load. Mode 2, on the other
In [16], a hybrid scheme is proposed for channel access and hand, reserves a large number of channels under low network
channel prediction, with the objective of improving network traffic load, and a small number of channels under heavy
throughput and transmission delay. The scheme is evaluated network traffic load. However, the scheme considers only the
under a high traffic CRN. In [17], the authors compare the IM for channel access which is, though, simple yet ineffi-
performance of the interweave and underlay modes of CRNs cient because of the fact that pure interweave mode leads
through a unified network setup. The authors then propose to protraction in service waiting time for users in addition
a hybrid scheme for channel access in order to improve to enhancing blocking probability and transmission delay.
the network’s throughput and transmission delay. In [18], Moreover, the scheme follows random approach in the selec-
a hybrid interweave-underlay spectrum scheme is proposed. tion of services for interruptions which results in unfairness
In the scheme, users sense the 5GHz license-exempt spectrum among users in the utilization of network’s resources. Besides
and reduce performance degradation induced by imperfect this, the scheme assumes all the channels to be of equal capac-
estimation of the interference link parameters. ity which is unrealistic in CRNs partly because, especially in
Though, most of the literature mentioned above addresses urban areas, more than one type of networks exist where each
most of the problems associated with dynamic spectrum network uses a different band of frequencies. This factor leads
access in CRNs, none of them, however, takes fairness to the realization of diverse frequency channels, with differ-
and equality among SUs into account in SUs’ services ent capacity, used by SUs. Lastly, the scheme is analyzed
interruption happened due to PUs emergence. Moreover, under perfect spectrum sensing only which is unrealistic in
TABLE 1. The list of key notations. terminates the interrupted users. A number of assumptions
are made in modeling the system which are given as under.
• PUs and SUs arrive with Poisson process with rates λP
and λS , respectively [29].
• Channel failure may occur to both the idle and active
channels, in both the RB and the NB. Channel failure
rate per channel is λF .
• Channels’ active time (time before a channel failure
occurs) and service time of PUs and SUs are exponen-
tially distributed [28]. The service rates per channel for
PUs and SUs are µP and µS , respectively.
• Multiple failed channels can be repaired or restored
simultaneously with µR restoration rate per channel. The
restoration time is exponentially distributed.
• The latency caused by spectrum sensing, spectrum hand-
off and service switching to the UM is considered to be
negligible.
• In order for making the CTMC modeling tractable, per-
fect spectrum sensing is assumed [19], [28], [30]–[37].
It is worth noting that certain literature has used Pareto
distribution for modeling incoming traffic [38], [39]. How-
ever, modeling incoming traffic in communication networks
with Poisson process is a more realistic assumption. This is
because of the fact that Poisson process is a natural arrival
process, such that the moment of arrival of a service is
random and the probability of admitting two consecutive
services simultaneously can be practically negligible [29].
Furthermore, Poisson distribution has been widely used in the
literature for modeling the users’ arrival process [40]–[45].
III. THE PROPOSED SCHEME
In this section, we present our proposed HBLP scheme. The
scheme uses hybrid underlay-interweave mode of channel
access in CRNs for enhancing spectrum utilization efficiency.
Furthermore, we also present our proposed MAFD algorithm,
which is used to determine services for interruptions. The
algorithm also ensures availability of the optimal channels
for PUs as and when they require in addition to maintaining
fairness among SUs’ services in the network’s resources
utilization. MAFD algorithm is described in Algorithm 1
and elaborated in Subsection III-A. Thereafter, we revisit
the DCR algorithm, as described in Algorithm 2, in Subsec-
tion III-B. The channels’ allocation procedure is presented in
Subsection III-C and the corresponding CTMC modeling is
presented in Subsection III-D.
A. THE MULTI-ATTRIBUTE BASED FAIRNESS DRIVEN
(MAFD) ALGORITHM
Determination of a service for interruption under the
defined criteria is described in Algorithm 1. The nota-
tion fmaxs represents the function used to scan the active
The PUs’ services operating in the NB and the RB are repre- services for finding the service with highest value of signal
sented by PUNB and PURB , respectively. Similarly, the SUs’ to noise plus interference ratio (SNIR), fmaxc represents
services operating in the NB and the RB are represented the function used to scan the active services for find-
by SUNB and SURB , respectively. The system dynamically ing the service with the highest value of channel capacity
assigns channels to new arriving users and handovers or in bits per second, and fmaxt represents the function
used to scan the active services for finding the service Algorithm 1 Service Determination
with the highest value of active time or service time duration. Input: SUact : Currently active SUs.
Similarly, SUmaxt represents the set of active SUs’ services Output: SUMax : The service to be interrupted.
with the highest values of active time or service time duration, [1] Calculate SUmaxs = fmaxs (SUact )
SUmaxc represents the set of active SUs’ services with the [2] if SUmaxs > 1 then
highest values of channel capacity in bits per second, and [3] Calculate SUmaxc = fmaxc (SUmaxs )
SUmaxs represents the set of active SUs’ services with the [4] if SUmaxc > 1 then
highest values of SNIR. [5] Calculate SUmaxt = fmaxt (SUmaxc )
As can be seen in Algorithm 1, whenever an SU’s service [6] SUMax = SUmaxt
needs to be interrupted, the algorithm analyzes all the ongoing [7] else
SUs’ services and stores in set SUmaxs the SU/SUs’ services [8] SUMax = SUmaxc
with the largest value of SNIR as given below [9] end
[10] else
SUmaxs = max (SUi (γi )), (1)
∀SUi ∈N [11] SUMax = SUmaxs
[12] end
where γi is SNIR of the ith active SU’s service, SUi ,
SUmaxs is the set of active SUs’ services with the high-
est value of SNIR, and N is the total number of
active SUs’ services. We use the free space propagation set SUmaxc and selects the SU’s service with the longest active
model to calculate of SNIR for a service (γ ) as given time or service time duration, as given below.
below [46]
SUmaxt = max (SUi (Tactive )), (7)
∀SUi ∈P
λ 2
pn
γ = 10 × log10 × , (2) SUMax = max(SUmaxt ), (8)
I 4πd
where P denotes the total number of SUs’ services in SUmaxt .
where pn is the transmission power of the intended trans-
Active time or service time duration (Tactive ) for a service can
mitted signal on channel n, I is the received interference
be calculated as
from other users, λ is the wavelength of the frequency
used, and d is the distance between a user availing the Tactive = t2 − t1 , (9)
service and base station. Thus, the service with the high-
est value of γ in SUmaxs , as given blow, is selected for where t1 and t2 indicate the starting time instant and the
interruption. current time instant of the service, respectively.
Making selection of services for interruption following
SUMax = max(SUmaxs ). (3) MAFD algorithm, as described above, leads to achieve-
ment of a noticeable level of fairness among SUs’ ser-
The notation SUMax represents the service selected to be vices in the utilization of network’s resources. On the other
interrupted. If multiple SUs’ services have the same value hand, random selection of services for interruption leads to
of γ , the algorithm further analyzes the set SUmaxs and unfairness and irregularity in the utilization of network’s
selects the SU/SUs’ services with the highest value of channel resources.
capacity.
SUmaxc = max (SUi (Ci )), (4) 1) TIME COMPLEXITY ANALYSIS FOR MAFD ALGORITHM
∀SUi ∈M We analyzed MAFD algorithm for time complexity and found
where M is the total number of SUs’ services in SUmaxs and that the time complexity of the algorithm is O(1), as shown
Ci is the capacity of the channel being used by the ith SU’s in Table 2.
service within SUmaxs . The channel capacity is calculated
using Shannon’s formula [47] B. THE DCR ALGORITHM
The DCR algorithm is described in Algorithm 2. The algo-
C = B log2 1 + γ
(5) rithm makes channel reservation based on the current traffic
load, φ. The notation φ is a ratio between the number of
where B is the channel bandwidth and C is the channel active PUs and SUs’ services, and the number of non-failed
capacity of a service in bits per seconds. So, the service channels in the network. The current traffic load is granulated
in SUmaxc with the highest value of channel capacity, as given into k + 1 levels, where k ∈ Z + . For instance, with k = 3,
below, is selected for interruption. the current traffic load is represented by four levels as very
SUMax = max(SUmaxc ). (6) high, high, medium, and low. Initially, φ is calculated on
Line 1 and then a resultant traffic load level is determined.
If multiple SUs’ services exist in SUmaxs with the same Mode 1 of the algorithm enhances retainability of current
value of channel capacity, the algorithm further analyzes the services by reserving a large number of channels, subject
Algorithm 2 The DCR Algorithm TABLE 3. Time complexity analysis for the DCR algorithm.
1) NEW PU ARRIVAL
TABLE 2. Time complexity analysis for MAFD algorithm.
A new PU arrival triggers Algorithm 2 to determine a suitable
number of channels for the RB and NB, and to allocate
the channels to each band accordingly. Thereafter, the new
arriving PU is assigned a channel as described in Algorithm 3.
The arriving PU is assigned an idle channel in the NB if
there is any. Otherwise, an SUNB , determined by Algorithm 1,
is interrupted by the PU. The interrupted SU hands over to the
RB if an idle channel exists there. Otherwise, the interrupted
SU adopts the UM. If the SU is unable to communicate in
the UM, it terminates. Similarly, non-availability of any idle
channel or SUNB results in blockage for the new arriving PU.
2) NEW SU ARRIVAL
A new SU arrival triggers Algorithm 2 to determine a suitable
to Rmax , under a high value of φ. Mode 2 enhances channel number of channels for the RB and NB, and to allocate
availability for new services by reserving a small number the channels to each band accordingly. Thereafter, the new
of channels under a high value of φ. To prevent spectrum arriving SU is assigned a channel as described in Algorithm 3.
handover of existing users from the RB to the NB, and to An idle channel in the NB is assigned to the SU. Non-
maintain a positive value for R0 , the DCR algorithm selects availability of any idle channel in the NB leads to the initi-
the higher value between R0 and NR , as shown in Line 16. The ation of the UM or blockage of the SU.
notation NR represents the number of existing users in the RB.
In Line 17, the availability of idle channels in the entire 3) CHANNEL FAILURE
network is examined in order to effectively set the number Idle channels, as well as active channels, may fail during the
of channels in the RB (RCH ) according to the value of R0 CRN’s operation. Failure of an idle channel induces no oper-
obtained in Line 16. ation in the network. However, failure of an active channel
brings about several events in the network depending on the
1) TIME COMPLEXITY ANALYSIS FOR DCR ALGORITHM type of the service that fails. In the following, description
Time complexity analysis of Algorithm 2 is carried out and of the failures and the resultant events associated with the
found to be O(n), as shown in Table 3. various types of services is presented.
and and
X
π(x),
X
suum(CHA) = 1 − x ∈ S. (15) suum(RFTP ) = λP π(x), x ∈ S. (22)
A(x)=T or A(x)=T ; sum >0;
An (x)=T −RCH (x) suum =0
Similarly, the rate of forced termination of SUs’ services
The blocking probability is contrary to channel availabil-
operating in the UM, owing to SUs’ services operating in the
ity. The blocking probability for SUs operating in the IM,
IM, is represented by suum(RFTS ) and is expressed as
SUBL , and the blocking probability for SUs operating in the
UM, suumBL , are respectively given by
X
suum(RFTS ) = λS π(x), x ∈ S. (23)
A(x)=T ; sum >0;
X
SUBL = π(x), x ∈ S, (16)
A(x)=T or
An (x)=T −RCH (x) 3) SERVICE RETAINABILITY FORMULATION
and The effective channel assignment rate for a new arriving SU is
X represented by 3S , and is calculated as 3S = λS SUCHA [48].
suum(BL) = π(x), x ∈ S. (17) Hence, the forced termination probability of SUs’ services,
A(x)=T or SUFT , becomes
An (x)=T −RCH (x) suum =0
0
(SURFTC + SURFT )
C. SERVICE RETAINABILITY SUFT = C
. (24)
3S
Retainability of a service can be expressed in terms of the
where SURFTC = SURFT + suum(RFTP ) + suum(RFTS ) and
forced termination probability (FTP) of that service as given 0 0
SURFT = SURFT + su0umRFT . Thus, the service retainability
by C
of SUs’ services can be expressed as
Service retainability = 1 − FTP. (18)
SURET = 1 − SUFT . (25)
The FTP depends on the rate of forced termination of a
service owing to channel failures and high priority service D. NETWORK UNSERVICEABLE PROBABILITY
arrivals, both of which are defined in the following. Network unserviceable probability, NUP, integrates the
impacts of both blocking and forced termination probability.
1) THE RATE OF FORCED TERMINATION OF SUs’ SERVICES Let SUNUP denotes NUP for SUs’ services. For mathematical
OWING TO CHANNEL FAILURES expression of SUNUP , we assume to be the probability of
The rate of forced termination of SUs’ services, owing to successfully finishing an SU service. Thus, we have
0
channel failure, operating in the IM (SURFT ) is given by
X SUNUP = 1 −
0
SURFT = λF (T − f ) π(x), x ∈ S. (19) = 1 − (1 − SUBL )(1 − suum(BL) )(1 − SUFT )
A(x)=T
(bn >0 ) or (an =bn =0; br >0) = 1 − SUBL (1 − SUFT − suum(BL)
Similarly, the rate of forced termination of SUs’ services, + SUFT suum(BL) ) − SUFT (1 − suum(BL) )
owing to channel failure, operating in the UM (su0um(RFT ) ) can − suum(BL) . (26)
be expressed as
E. HANDOFF PROBABILITY
X
su0um(RFT ) = λF π(x), x ∈ S. (20)
A(x)=T
We define handoff probability for SUs’ services to be the
suum >0 likelihood that, in the event of SUs’ services interruptions
happened due to channel failures or PUs’ arrivals, at least one
2) THE RATE OF FORCED TERMINATION OF SUs’ SERVICE vacant channel or low priority service is available to be used
OWING TO HIGH PRIORITY SERVICES ARRIVALS for channel handoff. Mathematically,
High priority service arrivals also cause terminations of low X X
priority services. Note that suum services may also face ter- Pr (HO) = λP π(x) + λF π(x)
mination because of the interference limit violation. The An (x)=T −RCH (x);bn >0 An (x)=T −RCH (x);bn >0
Ar (x)<RCH (x) Ar (x)<RCH (x)
interference limit violation arises when an existing or new X
arriving high priority service receives interference of the + λF π(x), x ∈ S. (27)
above-threshold level from a user operating in the UM. The Ar (x)=RCH (x);
An (x)<T −RCH (x) or bn >0 or an >0
rates of forced termination of SUs’ services, operating in
the IM, owing to PUs arrivals, SURFT , and the rates of forced
termination of SUs’ services, operating in the UM, owing to V. RESULTS AND DISCUSSION
PUs arrivals, suum(RFTP ) , are respectively given by In this section, we present the numerical results of our
X proposed scheme, HBLP. For performance comparison,
SURFT = λP π(x), x ∈ S, (21) BLP scheme, as proposed in [28], is considered. The com-
A(x)=T ; bn >0 parison is carried out under various network traffic loads
VOLUME 8, 2020 63411
A. U. Khan et al.: HBLP: Hybrid Underlay-Interweave Mode CRN for the Future 5G-Based IoT
TABLE 5. The list of symbols and their values used in the numerical
results.
A. CHANNEL AVAILABILITY
Fig. 2a shows channel availability plotted as a function of PU
arrival rate (λP ). As can be seen in the figure, the respective
modes of HBLP outperform their BLP counterparts at all
values of λP . For instance, both Mode 1 and Mode 2 of HBLP FIGURE 2. Channel availability for SUs as a function of λP and λF .
respectively achieve 22% and 17% improvement over their
BLP counterparts at λP = 15. In Fig. 2b, channel availability
is plotted as a function of channel failure rate (λF ) wherein
SURET plotted as a function of λF . Both the modes of HBLP
HBLP outperforms BLP at all values of λF . For instance, both
outperform their BLP counterparts. For instance, at λF =
Mode 1 and Mode 2 of HBLP respectively achieve 13% and
0.25, Mode 1 and Mode 2 of HBLP respectively achieve 6%
14% improvement over their BLP counterparts at λF = 0.6.
and 10% improvement over their BLP counterparts.
B. SERVICE CAPACITY
Fig. 3a shows SUs’ services capacity (SUCAP ) plotted as a D. NETWORK UNSERVICEABLE PROBABILITY
function of λP where both the modes of HBLP outperform Fig. 5a and 5b show SUs’ services NUP (SUNUP ) plotted
their BLP counterparts. For instance, Mode 1 and Mode 2 of as a function of λP and λF , respectively. As can be seen in
HBLP achieve 114% and 107% enhancement over their BLP the figures, both the modes of HBLP outperform their BLP
counterparts at λP = 15. Similarly, Fig. 3b shows SUCAP plot- counterparts. For instance, in Fig. 5a, 38% improvement is
ted as a function of λF wherein HBLP significantly improves achieved by HBLP over BLP in both the modes at λP = 15.
SUCAP . For instance, at λF = 0.6, Mode 1 and Mode 2 of Similarly, in Fig. 5b, Mode 1 and Mode 2 of HBLP achieve
HBLP respectively achieve 49% and 44% enhancement over 30% and 32% improvement, respectively, over their BLP
their BLP counterparts. counterparts, at λF = 0.15.
FIGURE 4. Retainability for SUs as a function of λP and λF . FIGURE 5. Network unserviceable probability as a function of λP and λF .
the network’s services, in Mode 1, because any interruption both BLP and HBLP follow the same procedure for channel
to SUNB leads the interrupted SUNB to make handoff to the reservation and the same hierarchy of service priorities in
RB, where SURB , because of its priority level, can access any channel assignment and interruption or termination.
idle channel available in the network if handoff is required.
Furthermore, SURB can interrupt active PUNB or SUNB when G. PERFORMANCE EVALUATION UNDER IMPERFECT
handoff is required. All these factors lead to the rapid increase SPECTRUM SENSING
in Pr (HO) with increase in λF in Mode 1. In the preceding scenario of the proposed scheme, perfect
The reason for the initial quick rise of Pr (HO) in Mode 2 is spectrum sensing was assumed. However, results achieved
that, in Mode 2, a large number of channels are reserved in under perfect spectrum sensing do not always coincide
the RB, under low traffic load. However, as the arrival rate of with the results achieved under imperfect spectrum sensing
PUs’ services increases, Mode 2 initiates reserving a fewer because of the anomalies in the network functionalities intro-
number of channels. Since SUNB are prone to interruptions duced by the latter type of sensing. In this section, we intro-
under new PUs’ arrivals, they are more likely to be terminated duce the imperfect spectrum sensing in our model for making
instead of handoff in the face of PUs’ arrivals and channel the model more realistic.
failures, given that fewer channels are assigned to the RB. The In imperfect spectrum sensing, an SU is unable to correctly
only significant contribution for Pr (HO) is made by channel detect an ongoing or new arriving PU on the same channel
failures happened to SURB . However, as fewer SURB can be the particular SU is using. This leads to loss of transmission
contained in the network when operating in Mode 2, their power and data in addition to interference generation for
contribution to Pr (HO) is of low level. the PU. To formulate this mathematically, we assume that
It is worth noting that the Pr (HO) analysis made for PUs’ transmission intensity can be characterized by binary
Mode 1 and Mode 2, as depicted in Fig. 9, is equally appli- stochastic hypotheses, such that the active and idle channels
cable to both BLP and HBLP. It is because of the fact that are represented by 1 and 0, respectively [49]. Thus, the idle
FIGURE 6. Fairness among SUs in the utilization of service time. FIGURE 7. Fairness among SUs in service’s SNIR.
FIGURE 8. Fairness among SUs in the utilization of service’s FIGURE 9. Handoff probability as a function of λP and λF .
capacity (bps).
the fact that in the case of miss detections, SUs assume that
the channel under consideration is idle and start using that
channel. Though, after occupation of such channel, and hav-
ing used the channel for a short and specified time duration,
the occupant SU is terminated. Yet this factor adds to the
overall enhancement of channel availability.
Fig. 11 indicates the service capacity of SUs plotted as
a function of PU arrival rates assuming imperfect spectrum
sensing. As shown in the figure, the service capacity is not
significantly changed according to the increase in Pm . The
reason for this is that service capacity indicates the average
number of services completed per unit time. Though high
values of Pm ensure enhancement in channel availability, FIGURE 10. Channel availability vs PU arrival rate under imperfect
the number of services completed per unit time still do not get sensing.
enhanced. It is worth noting that an SU after commencement
of transmission on an already occupied channel is shortly
terminated. Thus, the number of services completed per Pm decreases retainability of services. The reason is
unit time still remains unchanged even under high values that retainability signifies interruptibility, which decreases
of Pm . because of miss detection. Note that once an SU occupies a
Fig. 12 indicates service retainability of SUs plotted as channel under miss detection, such that the channel is being
a function of PU arrival rate, assuming imperfect spec- used by another PU, the SU is shortly interrupted. This factor
trum sensing. It is observed in the figure that increase in leads to decrement in service retainability level.
H. CRITICAL DISCUSSION
In this paper, we have proposed a hybrid underlay-interweave
mode enabled scheme to enhance SUE in CRNs. The scheme
utilizes dynamic reservation of channels in order to accom-
modate interrupted users, and allows SUs to switch to the UM
when idle channels are unavailable. The scheme is modeled
using CTMC, and evaluated under various network traffic
loads and channel failure rates. For performance evalua-
tion, essential SUE-related performance metrics including
FIGURE 11. Service capacity vs PU arrival rate under imperfect sensing. channel availability, service capacity, service retainability,
handoff and network unserviceable probability are consid-
ered from the perspective of SUs. Moreover, the scheme
is evaluated under the impacts of perfect and imperfect
spectrum sensing. In addition, we have proposed a multi-
attribute based fairness-driven algorithm which facilitates
fairness among SUs’ services in the utilization of network’s
resources in terms of services’ SNIR, services’ capacity
and services’ utilization time. For performance comparison,
BLP scheme [28] is considered.
The numerical results confirm the effectiveness of our
proposed scheme. For instance, a maximum of 22% enhance-
ment is achieved by our proposed scheme in channel avail-
ability. Similarly, service capacity is doubly increased by
our proposed scheme. A reasonable enhancement of 10% is
observed in service retainability. In network unserviceable
FIGURE 12. Service retainability vs PU arrival rate under imperfect probability, as high as 38% improvement is noted. Further-
sensing.
more, the scheme ably achieves fairness among SUs in the
utilization of network’s resources.
The scheme is ideal for deployment in the 5G-based IoT
where resource constrained devices can take advantage of
the UM of CRN for energy efficiency enhancement. Fur-
thermore, it enables device-to-device communication in a
congested environment where devices need to transmit their
data regularly.
Nonetheless, our proposed scheme bears certain limita-
tions. For instance, the UM is considered for SUs only and,
thus, performance evaluation is made from the standpoint of
SUs. Moreover, only three parameters, such that services’
SNIR, services’ capacity, and service time duration, are used
in the evaluation and selection of services for interruption.
Lastly, the PUs’ services performance enhancement brought
about by Algorithm 1 has not been considered. These limi-
FIGURE 13. NUP vs PU arrival rate under imperfect sensing. tations are easily addressable and are the focus of our future
work.
this objective, the scheme dynamically reserves a number [14] S. Kusaladharma and C. Tellambura, An Overview of Cognitive Radio
of channels for accommodating the interrupted services, and Networks. Atlanta, GA, USA: American Cancer Society, 2017, pp. 1–17.
[15] H. Mokhtarzadeh, A. Taherpour, A. Taherpour, and S. Gazor, ‘‘Throughput
allocates channels to users in a dynamic, efficient and pri- maximization in energy limited full-duplex cognitive radio networks,’’
oritized fashion. Furthermore, we proposed a multi-attribute IEEE Trans. Commun., vol. 67, no. 8, pp. 5287–5296, Aug. 2019.
based fairness driven algorithm for the selection and inter- [16] P. Thakur, A. Kumar, S. Pandit, G. Singh, and S. N. Satashia, ‘‘Performance
analysis of high-traffic cognitive radio communication system using hybrid
ruption of SUs’ services which maintains fairness among spectrum access, prediction and monitoring techniques,’’ Wireless Netw.,
SUs’ services in the utilization of network’s resources. The vol. 24, no. 6, pp. 2005–2015, Aug. 2018.
scheme is modeled using CTMC and evaluated for various [17] F. Mehmeti and T. Spyropoulos, ‘‘Performance analysis, comparison, and
optimization of interweave and underlay spectrum access in cognitive
spectrum utilization efficiency related QoS parameters under radio networks,’’ IEEE Trans. Veh. Technol., vol. 67, no. 8, pp. 7143–7157,
the effects of various network traffic loads and channel failure Aug. 2018.
rates. Additionally, the scheme is evaluated under the impacts [18] P. Zuo, T. Peng, W. Linghu, and W. Wang, ‘‘Optimal resource allocation
of perfect and imperfect spectrum sensing. The numerical for hybrid interweave-underlay cognitive SatCom uplink,’’ in Proc. IEEE
Wireless Commun. Netw. Conf. (WCNC), Apr. 2018, pp. 1–6.
results demonstrate that the proposed scheme significantly [19] I. A. M. Balapuwaduge, F. Y. Li, and V. Pla, ‘‘Significance of channel
improves all the considered parameters. failures on network performance in CRNs with reserved spectrum,’’ in
In future, the scheme can be further analyzed to character- Proc. IEEE Int. Conf. Commun. (ICC), May 2016, pp. 1–6.
[20] T. Chakraborty and I. S. Misra, ‘‘Design and analysis of channel reserva-
ize the performance enhancement of PUs’ services resulting tion scheme in cognitive radio networks,’’ Comput. Electr. Eng., vol. 42,
from the multi-attribute based service selection and inter- pp. 148–167, Feb. 2015.
ruption approach of our proposed algorithm. Besides this, [21] N. M. El Azaly, E. F. Badran, M. R. M. Rizk, and M. A. Mokhtar,
‘‘Performance enhancement of steady-state Markov analysis for cognitive
the UM can also be considered for PUs’ services, and mul- radio networks via channel reservation,’’ Alexandria Eng. J., vol. 56, no. 4,
tiple parameters can be included in the service evaluation for pp. 469–475, Dec. 2017.
selection and interruption. Lastly, probability of false alarm, [22] J. Wang and A. Huang, ‘‘Low-complexity channel reservation design
scheme in cognitive radio networks,’’ in Proc. Int. Conf. Wireless Commun.
in addition to miss detection probability, can also be incorpo- Signal Process. (WCSP), Oct. 2012, pp. 1–5.
rated in spectrum sensing analysis to extend the scheme for [23] E. Markova, I. Gudkova, A. Ometov, I. Dzantiev, S. Andreev,
gaining more insightful observations. Y. Koucheryavy, and K. Samouylov, ‘‘Flexible spectrum management in a
smart city within licensed shared access framework,’’ IEEE Access, vol. 5,
pp. 22252–22261, 2017.
REFERENCES [24] ECC Report 205. (Feb. 2014). Licensed Shared Access (LSA).
[1] Y. B. Zikria, S. W. Kim, O. Hahm, M. K. Afzal, and M. Y. Aalsalem, [Online]. Available: https://www.ecodocdb.dk/download/baa4087d-
‘‘Internet of Things (IoT) operating systems management: Opportunities, e404/ECCREP205.PDF
challenges, and solution,’’ Sensors, vol. 19, no. 8, p. 1793, Apr. 2019. [25] M. Mustonen, M. Matinmikko, M. Palola, S. Yrjölä, and K. Horneman,
[2] Y. Xiao, M. Krunz, and T. Shu, ‘‘Multi-operator network sharing for ‘‘An evolution toward cognitive cellular systems: Licensed shared access
massive IoT,’’ IEEE Commun. Mag., vol. 57, no. 4, pp. 96–101, Apr. 2019. for network optimization,’’ IEEE Commun. Mag., vol. 53, no. 5, pp. 68–74,
[3] Y. Wang, Y. Wang, F. Zhou, Y. Wu, and H. Zhou, ‘‘Resource allocation in May 2015.
wireless powered cognitive radio networks based on a practical non-linear [26] A. Ponomarenko-Timofeev, A. Pyattaev, S. Andreev, Y. Koucheryavy,
energy harvesting model,’’ IEEE Access, vol. 5, pp. 17618–17626, 2017. M. Mueck, and I. Karls, ‘‘Highly dynamic spectrum management within
[4] F. Hu, B. Chen, and K. Zhu, ‘‘Full spectrum sharing in cognitive radio licensed shared access regulatory framework,’’ IEEE Commun. Mag.,
networks toward 5G: A survey,’’ IEEE Access, vol. 6, pp. 15754–15776, vol. 54, no. 3, pp. 100–109, Mar. 2016.
2018. [27] A. Chouayakh, A. Bechler, I. Amigo, L. Nuaymi, and P. Maillé, ‘‘Auc-
[5] Z. Na, Z. Pan, M. Xiong, X. Liu, W. Lu, Y. Wang, and L. Fan, ‘‘Turbo tion mechanisms for licensed shared access: Reserve prices and revenue-
receiver channel estimation for GFDM-based cognitive radio networks,’’ fairness trade offs,’’ ACM SIGMETRICS Perform. Eval. Rev., vol. 46, no. 3,
IEEE Access, vol. 6, pp. 9926–9935, 2018. pp. 43–48, Jan. 2019.
[6] D. Wang, B. Song, D. Chen, and X. Du, ‘‘Intelligent cognitive radio in [28] I. A. M. Balapuwaduge, F. Y. Li, and V. Pla, ‘‘Dynamic spectrum reser-
5G: AI-based hierarchical cognitive cellular networks,’’ IEEE Wireless vation for CR networks in the presence of channel failures: Channel allo-
Commun., vol. 26, no. 3, pp. 54–61, Jun. 2019. cation and reliability analysis,’’ IEEE Trans. Wireless Commun., vol. 17,
[7] X. Zhang, X. Zhang, L. Han, and R. Xing, ‘‘Utilization-oriented spectrum no. 2, pp. 882–898, Feb. 2018.
allocation in an underlay cognitive radio network,’’ IEEE Access, vol. 6, [29] F. Gebali, Analysis of Computer Networks, 2nd ed. Cham, Switzerland:
pp. 12905–12912, 2018. Springer, 2015.
[8] A. Ali, L. Abbas, M. Shafiq, A. K. Bashir, M. K. Afzal, H. B. Liaqat, [30] K. Zheng, L. Zhao, J. Mei, B. Shao, W. Xiang, and L. Hanzo, ‘‘Survey of
M. H. Siddiqi, and K. S. Kwak, ‘‘Hybrid fuzzy logic scheme for efficient large-scale MIMO systems,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 3,
channel utilization in cognitive radio networks,’’ IEEE Access, vol. 7, pp. 1738–1760, 3rd Quart., 2015.
pp. 24463–24476, 2019. [31] Y. Xu, A. Anpalagan, Q. Wu, L. Shen, Z. Gao, and J. Wang, ‘‘Decision-
[9] S. Huang, D. Yuan, and A. Ephremides, ‘‘Bandwidth partition and allo- theoretic distributed channel selection for opportunistic spectrum access:
cation for efficient spectrum utilization in cognitive communications,’’ Strategies, challenges and solutions,’’ IEEE Commun. Surveys Tuts.,
J. Commun. Netw., vol. 21, no. 4, pp. 353–364, Aug. 2019. vol. 15, no. 4, pp. 1689–1713, 4th Quart., 2013.
[10] M. Ali and H. Nam, ‘‘Optimization of spectrum utilization in cooperative [32] C. Altay, H. B. Yilmaz, and T. Tugcu, ‘‘Cooperative sensing analysis under
spectrum sensing,’’ Sensors, vol. 19, no. 8, p. 1922, 2019. imperfect reporting channel,’’ in Proc. IEEE Symp. Comput. Commun.
[11] S. Ghosh, D. De, and P. Deb, ‘‘Energy and spectrum optimization for 5G (ISCC), Jul. 2012, pp. 770–775.
massive MIMO cognitive Femtocell based mobile network using auction [33] T. Yucek and H. Arslan, ‘‘A survey of spectrum sensing algorithms for
game theory,’’ Wireless Pers. Commun., vol. 106, no. 2, pp. 555–576, cognitive radio applications,’’ IEEE Commun. Surveys Tuts., vol. 11, no. 1,
May 2019. pp. 116–130, 1st Quart., 2009.
[12] F. Li, K.-Y. Lam, Z. Sheng, X. Zhang, K. Zhao, and L. Wang, ‘‘Q-learning- [34] X. Li, Q. Zhao, X. Guan, and L. Tong, ‘‘Optimal cognitive access of
based dynamic spectrum access in cognitive industrial Internet of Things,’’ Markovian channels under tight collision constraints,’’ IEEE J. Sel. Areas
Mobile Netw. Appl., vol. 23, no. 6, pp. 1636–1644, 2018. Commun., vol. 29, no. 4, pp. 746–756, Apr. 2011.
[13] O. Naparstek and K. Cohen, ‘‘Deep multi-user reinforcement learning for [35] D. Niyato, E. Hossain, and P. Wang, ‘‘Optimal channel access management
distributed dynamic spectrum access,’’ IEEE Trans. Wireless Commun., with QoS support for cognitive vehicular networks,’’ IEEE Trans. Mobile
vol. 18, no. 1, pp. 310–323, Jan. 2019. Comput., vol. 10, no. 4, pp. 573–591, Apr. 2011.
[36] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, ‘‘Green cellular networks: GHULAM ABBAS (Senior Member, IEEE)
A survey, some research issues and challenges,’’ IEEE Commun. Surveys received the B.S. degree in computer science from
Tuts., vol. 13, no. 4, pp. 524–540, 4th Quart., 2011. the University of Peshawar, Pakistan, in 2003,
[37] Q. Zhao, S. Geirhofer, L. Tong, and B. M. Sadler, ‘‘Opportunistic spectrum and the M.S. degree in distributed systems and
access via periodic channel sensing,’’ IEEE Trans. Signal Process., vol. 56, the Ph.D. degree in computer networks from the
no. 2, pp. 785–796, Feb. 2008. University of Liverpool, U.K., in 2005 and 2010,
[38] R. N. Yadav, R. Misra, and S. Bhagat, ‘‘Spectrum access in cognitive smart- respectively. He was a Research Associate with
grid communication system with prioritized traffic,’’ Ad Hoc Netw., vol. 65,
Liverpool Hope University, U.K. Since 2011, he
pp. 38–54, Oct. 2017.
has been with the Faculty of Computer Sciences
[39] W. Ahmed, J. Gao, and M. Faulkner, ‘‘Performance evaluation of a cogni-
tive radio network with exponential and truncated usage models,’’ in Proc. and Engineering, GIK Institute of Engineering
4th Int. Symp. Wireless Pervas. Comput., Feb. 2009, pp. 1–5. Sciences and Technology, Pakistan. He is currently working as an Associate
[40] S. L. Castellanos-Lopez, F. A. Cruz-Pérez, and G. Hernandez-Valdez, Professor and the Director of the Huawei ICT Academy, GIK Institute of
‘‘Performance of cognitive radio networks under ON/OFF and Poisson Engineering Sciences and Technology. He is a Co-Founding Member of the
primary arrival models,’’ in Proc. IEEE 22nd Int. Symp. Pers., Indoor Telecommunications and Networking (TeleCoN) Research Laboratory, GIK
Mobile Radio Commun., Sep. 2011, pp. 609–613. Institute of Engineering Sciences and Technology. His research interests
[41] T. M. C. Chu, H. Phan, and H. Zepernick, ‘‘On the performance of include computer networks, and wireless and mobile communications. He
underlay cognitive radio networks using M/G/1/K queueing model,’’ IEEE is a Fellow of the Institute of Science and Technology, U.K. and the British
Commun. Lett., vol. 17, no. 5, pp. 876–879, May 2013. Computer Society.
[42] Y. Gao and Y. Jiang, ‘‘Performance analysis of a cognitive radio network
with imperfect spectrum sensing,’’ in Proc. IEEE Conf. Comput. Commun. ZIAUL HAQ ABBAS received the M.Phil. degree
Workshops (INFOCOM), Mar. 2010, pp. 1–6. in electronics from Quaid-e-Azam University,
[43] E. W. M. Wong and C. Foh, ‘‘Analysis of cognitive radio spectrum Pakistan, in 2001, and the Ph.D. degree from the
access with finite user population,’’ IEEE Commun. Lett., vol. 13, no. 5, Agder Mobility Laboratory, Department of Infor-
pp. 294–296, May 2009. mation and Communication Technology, Univer-
[44] T. Jiang, H. Wang, M. Daneshmand, and D. Wu, ‘‘Cognitive radio-based sity of Agder, Norway, in 2012. He was a Visiting
smart grid traffic scheduling with binary exponential backoff,’’ IEEE Inter- Researcher with the Department of Electrical and
net Things J., vol. 4, no. 6, pp. 2038–2046, Dec. 2017. Computer Engineering, University of Minnesota,
[45] K. A. Mehr, J. M. Niya, and N. Akar, ‘‘Queue management for two- USA. He is currently an Associate Professor with
user cognitive radio with delay-constrained primary user,’’ Comput. Netw., the Faculty of Electrical Engineering and a Co-
vol. 142, pp. 1–12, Sep. 2018. Founding Member of the Telecommunications and Networking (TeleCoN)
[46] F.-H. Tseng, L.-D. Chou, H.-C. Chao, and J. Wang, ‘‘Ultra-dense small Research Laboratory, GIK Institute of Engineering Sciences and Technology.
cell planning using cognitive radio network toward 5G,’’ IEEE Wireless
His research interests include energy efficiency in hybrid mobile and wireless
Commun., vol. 22, no. 6, pp. 76–83, Dec. 2015.
communication, 5G mobile systems, ad hoc networks, traffic engineering in
[47] S. Jing, D. N. C. Tse, J. B. Soriaga, J. Hou, J. E. Smee, and R. Padovani,
‘‘Multicell downlink capacity with coordinated processing,’’ EURASIP J. wireless networks, performance evaluation of communication protocols and
Wireless Commun. Netw., vol. 2008, no. 1, Dec. 2008, Art. no. 586878. networks by analysis and simulation, and cognitive radio.
[48] L. Jiao, I. A. M. Balapuwaduge, F. Y. Li, and V. Pla, ‘‘On the performance MUHAMMAD TANVEER received the M.S.
of channel assembling and fragmentation in cognitive radio networks,’’ degree in computer science, in 2017. He is
IEEE Trans. Wireless Commun., vol. 13, no. 10, pp. 5661–5675, Oct. 2014. currently pursuing the Ph.D. degree in computer
[49] M. T. Masonta, M. Mzyece, and N. Ntlatlapa, ‘‘Spectrum decision in
science from the GIK Institute of Engineering Sci-
cognitive radio networks: A survey,’’ IEEE Commun. Surveys Tuts., vol. 15,
ences and Technology, Pakistan. He is a mem-
no. 3, pp. 1088–1107, 3rd Quart., 2013.
[50] Y.-C. Liang, Y. Zeng, E. C. Y. Peh, and A. T. Hoang, ‘‘Sensing-throughput
ber of the Telecommunications and Networking
tradeoff for cognitive radio networks,’’ IEEE Trans. Wireless Commun., (TeleCoN) Research Laboratory, GIK Institute
vol. 7, no. 4, pp. 1326–1337, Apr. 2008. of Engineering Sciences and Technology. His
[51] S. M. Kay, Fundamentals of Statistical Signal Processing. research interests include communication net-
Upper Saddle River, NJ, USA: Prentice-Hall, 1993. works and network security.
[52] S. Atapattu, C. Tellambura, and H. Jiang, ‘‘Spectrum sensing via energy SAMI ULLAH received the M.S. degree in
detector in low SNR,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2011,
computer science from the IBMS, The Univer-
pp. 1–5.
sity of Agriculture, Peshawar, Pakistan. He is
[53] A. Kumar, P. Thakur, S. Pandit, and G. Singh, ‘‘Analysis of optimal thresh-
old selection for spectrum sensing in a cognitive radio network: An energy
currently pursuing the Ph.D. degree in com-
detection approach,’’ Wireless Netw., vol. 25, no. 7, pp. 3917–3931, puter Science with the Ghulam Ishaq Khan
Oct. 2019. Institute of Engineering Sciences and Technol-
[54] P. Thakur, A. Kumar, S. Pandit, G. Singh, and S. Satashia, ‘‘Analysis of ogy (GIKI), Topi, Pakistan. He is also a Lec-
high-traffic cognitive radio network with imperfect spectrum monitoring turer with Shaheed Benazir Bhutto University,
technique,’’ Comput. Netw., vol. 147, pp. 27–37, Dec. 2018. Sheringal, Pakistan. He is a member of the
Telecommunications and Networking (TeleCoN)
Research Laboratory, GIK Institute of Engineering Sciences and Technology.
ABD ULLAH KHAN received the B.S. degree ALAMGIR NAUSHAD received the B.S. degree
(Hons.) in telecommunication from the University in computer systems engineering from the Uni-
of Science & Technology Bannu, in 2013, and the versity of Engineering and Technology, Peshawar,
M.S. degree in electrical engineering from COM- Pakistan, in 2011, and the M.S. and Ph.D. degrees
SATS University Islamabad, Pakistan, in 2016. He in computer system engineering from the GIK
is currently pursuing the Ph.D. degree from the Institute of Engineering Sciences and Technology,
GIK Institute of Engineering Sciences and Tech- Pakistan, in 2014 and 2019, respectively. He is
nology, Pakistan. Part of his Ph.D. degree is from currently an Assistant Professor with the National
the National University of Science of Sciences and University of Sciences and Technology, Quetta,
Technology, Islamabad. He is currently an Active Pakistan, and a member of the Telecommunica-
Member with the Telecommunications and Networking (TeleCoN) Research tions and Networking (TeleCoN) Research Laboratory, GIK Institute of
Laboratory, GIK Institute of Engineering Sciences and Technology. His Engineering Sciences and Technology. His research interests include mobile
research interests include wireless sensor networks, body area networks, ad hoc and cellular networks.
cognitive radio networks, and vehicular ad hoc networks.
63420 VOLUME 8, 2020