1736 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO.

3, MARCH 2021

Coexistence Mechanism Between eMBB and

uRLLC in 5G Wireless Networks
Anupam Kumar Bairagi , Member, IEEE, Md. Shirajum Munir , Graduate Student Member, IEEE,
Madyan Alsenwi , Nguyen H. Tran , Senior Member, IEEE, Sultan S. Alshamrani ,
Mehedi Masud , Senior Member, IEEE, Zhu Han , Fellow, IEEE,
and Choong Seon Hong , Senior Member, IEEE

Abstract— Ultra-reliable low-latency communication (uRLLC) first sub-problem with lower complexity. Finally, the significance
and enhanced mobile broadband (eMBB) are two influential of the proposed approach over other baseline approaches is
services of the emerging 5G cellular network. Latency and relia- established through numerical analysis in terms of the MEAR
bility are major concerns for uRLLC applications, whereas eMBB and fairness scores of the eMBB UEs.
services claim for the maximum data rates. Owing to the trade-
off among latency, reliability and spectral efficiency, sharing of Index Terms— Ultra-reliable low latency communications
radio resources between eMBB and uRLLC services, heads to (uRLLC), enhanced mobile broadband (eMBB), coexis-
a challenging scheduling dilemma. In this paper, we study the tence, penalty successive upper bound minimization (PSUM),
co-scheduling problem of eMBB and uRLLC traffic based upon transportation model (TM), resource scheduling.
the puncturing technique. Precisely, we formulate an optimization
problem aiming to maximize the minimum expected achieved rate I. I NTRODUCTION
(MEAR) of eMBB user equipment (UE) while fulfilling the provi-
sions of the uRLLC traffic. We decompose the original problem
into two sub-problems, namely scheduling problem of eMBB UEs
and uRLLC UEs while prevailing objective unchanged. Radio
T HE wireless industries are going through different kinds
of emerging applications and services along with the
explosive trends of mobile traffic [1]. High-resolution video
resources are scheduled among the eMBB UEs on a time slot
basis, whereas it is handled for uRLLC UEs on a mini-slot streaming, virtual reality (VR), augmented reality (AR),
basis. Moreover, for resolving the scheduling issue of eMBB UEs, autonomous cars, smart cities and factories, artificial intelli-
we use penalty successive upper bound minimization (PSUM) gence (AI) based services are some of these categories. It is
based algorithm, whereas the optimal transportation model foreseen that the mobile application market will flourish in
(TM) is adopted for solving the same problem of uRLLC UEs. a cumulative average growth rate (CAGR) of 29.1% dur-
Furthermore, a heuristic algorithm is also provided to solve the
ing 2015 − 2020 [2]. Energy efficiency, latency, reliability,
Manuscript received February 4, 2020; revised June 18, 2020 and data rate, etc are distinct for separate applications and ser-
November 5, 2020; accepted November 14, 2020. Date of publication vices. To handle these diversified requirements, International
November 24, 2020; date of current version March 17, 2021. This work was
partially supported by the National Research Foundation of Korea (NRF) Telecommunication Union (ITU) has already classified 5G ser-
grant funded by the Korea government (MSIT) (No. 2020R1A4A1018607) vices into Ultra-reliable low-latency communication (uRLLC),
and by the MSIT(Ministry of Science and ICT), Korea, under the Grand massive machine-type communication (mMTC), and enhanced
Information Technology Research Center support program(IITP-2020-2015-
0-00742) supervised by the IITP(Institute for Information & communications mobile broadband (eMBB) categories [3]. Gigabit per second
Technology Planning & Evaluation). The associate editor coordinating the (Gbps) level data rates are required for eMBB users, whereas
review of this article and approving it for publication was D. Marabissi. connection density and energy efficiency are the major con-
(Corresponding author: Choong Seon Hong.)
Anupam Kumar Bairagi is with the Discipline of Computer Science and cern for mMTC, and uRLLC traffic focuses on extremely
Engineering, Khulna University, Khulna 9208, Bangladesh, and also with the high reliability (99.999%) and remarkably low latency
Department of Computer Science and Engineering, Kyung Hee University, (0.25 ∼ 0.30 ms/packet) [4].
Seoul 446-701, South Korea (e-mail: anupam@ku.ac.bd).
Md. Shirajum Munir, Madyan Alsenwi, and Choong Seon Hong are with Generally, the lions’ share of wireless traffic is produced by
the Department of Computer Science and Engineering, Kyung Hee University, eMBB UEs. uRLLC traffic is naturally infrequent and needs
Seoul 446-701, South Korea (e-mail: munir@khu.ac.kr; malsenwi@khu.ac.kr; to be addressed spontaneously. The easiest way to settle this
cshong@khu.ac.kr).
Nguyen H. Tran is with the School of Computer Science, The University of matter is to reserve some resources for uRLLC. However,
Sydney, Sydney, NSW 2006, Australia (e-mail: nguyen.tran@sydney.edu.au). under-utilization of radio resources may emerge from this
Sultan S. Alshamrani is with the Department of Information Technology, approach, and generally, effective multiplexing of traffics is
Taif University, Taif 21944, Saudi Arabia (e-mail: susamash@tu.edu.sa).
Mehedi Masud is with the Department of Computer Science, College of required. For efficient multiplexing of eMBB and uRLLC
Computers and Information Technology, Taif University, Taif 21944, Saudi traffics, 3GPP has recommended a superposition/puncturing
Arabia (e-mail: mmasud@tu.edu.sa). skeleton [4] and the short-TTI/puncturing approaches [5] in 5G
Zhu Han is with the Department of Electrical and Computer Engineering,
University of Houston, Houston, TX 77004 USA, and also with the Depart- cellular systems. Though the short-TTI mechanism is straight-
ment of Computer Science and Engineering, Kyung Hee University, Seoul forward for implementation, it degrades spectral efficiency
446-701, South Korea (e-mail: zhan2@uh.edu). because of the massive overhead in the control channel.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCOMM.2020.3040307. On the contrary, the puncturing strategy decreases the above
Digital Object Identifier 10.1109/TCOMM.2020.3040307 overhead, although it necessitates an adequate mechanism for
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1737

recognizing and healing the punctured case. Slot (1 ms) and • Finally, we perform a comprehensive experimental analy-
mini-slot (0.125 ms) are proposed as time units for meeting sis for the proposed scheduling approach and compare
the latency requirement of uRLLC traffic in the 5G new radio the results, minimum expected achieved rate (MEAR)
(NR). At the outset of a slot, eMBB traffic is scheduled and and fairness [43] of the eMBB UEs, with the punctured
continues unchanged throughout the slot. If the same physical scheduler (PS) [21], multi-user preemptive scheduler
resources are used, uRLLC traffic is overridden upon the (MUPS) [25], random scheduler (RS), equally distributed
scheduled eMBB transmission. scheduler (EDS), and matching based scheduler (MBS)
Currently, much attention has been paid to resource sharing approaches.
for offering quality-of-service(QoS) or quality-of-experience The remainder of the paper is systematized as follows.
(QoE) to the users. Studies [6] and [7] investigate the sharing In Section II, we present the literature review. We explain
of an unlicensed spectrum between LTE and WiFi networks, the system model and present the problem formulation in
however, the study [8] con sider LTE-A and NB-IoT ser- Section III. The proposed solution approach of the above-
vices for sharing the same resources. Study [9] solves user mentioned problem is addressed in Section IV. In Section V,
association and resource allocation problems. The study [9] we provide experimental investigation, discussion, and com-
consider the downlink of fog network to support QoS provi- parison concerning the proposed solution. Finally, we conclude
sions of the uRLLC and eMBB. Some other studies, however, the paper in Section VI. A list of acronyms is provided
investigates and/or analyzes the influence of uRLLC traffic on in Table I.
eMBB [10]–[15] or presents architecture and/or framework
for co-scheduling of eMBB and uRLLC traffic [16]–[19]. II. L ITERATURE R EVIEW
Moreover, some authors consider eMBB and uRLLC traffic in Recently, both industry and academia focus on the study
their coexisting/multiplexing proposals [20]–[27] where they of multiplexing between eMBB traffic and uRLLC traffic on
apply puncturing technique. the same physical resources. Information-theoretic arguments-
As per our knowledge, concrete mathematical models and based performance analysis for eMBB and uRLLC traffic
solutions, however, are lacking in most of these coexistence has performed in [10]. The authors consider both orthogonal
mechanisms. Most of the studies mainly focus on analysis, multiple access (OMA) and non-orthogonal multiple access
system-level design or framework. Thus, efficient coexistence (NOMA) for uplink in cloud radio access network (C-RAN)
proposals between eMBB and uRLLC traffic are needed in framework. An insight into the performance trade-offs among
the literature. So, to enable eMBB and uRLLC services in the eMBB and uRLLC traffic is explained in [10]. In [11],
5G wireless networks, we propose an effective coexistence authors have introduced eMBB influenced minimization prob-
mechanism in this paper. Our preliminary work has been lem to protect the uRLLC traffic from the dominant eMBB
published in [23] where we have used a one-sided matching services. This paper explores their proposal for the mobile
and heuristic algorithm, respectively, for resolving resource front-haul environment. In [12], the authors present an effec-
allocation problems of eMBB and uRLLC users. The major tive solution for multiplexing different traffics on a shared
difference between [23] and current work is the involvement resource. Particularly, they propose an effective radio resource
of penalty successive upper bound minimization (PSUM) and distribution method between the uRLLC and eMBB service
transportation model (TM) for solving similar problems. This classes following trade-offs among the reliability, latency and
paper mainly focuses on the followings: spectral efficiency. Moreover, they investigate the uRLLC and
• First, we formulate an optimization problem for eMBB eMBB performance adopting different conditions.
UEs with some constraints, where the objective is to In order to achieve 5G service provisioning (i.e., eMBB,
maximize the minimum expected rate of eMBB UEs over mMTC and uRLLC services), the authors of [13] have studied
time. radio resources slicing mechanism, where the performance of
• Second, to solve the optimization problem effectively, both orthogonal and non-orthogonal are analyzed. They have
we decompose it into two sub-problems: resource proposed a communication-theoretic model by considering the
scheduling for eMBB UEs, and resource scheduling of heterogeneity of 5G services. They also found that the non-
uRLLC UEs. PSUM is used to solve the first sub- orthogonal slicing is significantly better to perform instead of
problem, whereas the TM is employed to solve the second orthogonal slicing for those 5G service multiplexing. Recently,
one. for 5G NR physical layer challenges and solution mechanisms
• Third, we redefine the first sub-problem into a minimiza- of uRLLC traffic communications has been presented in [14],
tion problem for each slot and provide an algorithm based where they pay attention to the structure of packet and frame.
upon PSUM to obtain near-optimal solutions. Additionally, they focus on the improvement of scheduling
• Fourth, we redefine the second sub-problem as a min- and reliability mechanism for uRLLC traffic communication
imization problem for each mini-slot within every slot such that the coexistence of uRLLC with eMBB is established.
and present the algorithm based upon minimum cell cost In [15], the authors have been analyzed the designing principle
(MCC) and modified distribution (MODI) methods of of the 5G wireless network by employing low-latency and
the transportation model to find an optimal solution of high-reliability for uRLLC traffic. To do this, they consider
the second sub-problem. varying requirements of uRLLC services such as variation of
• Fifth, we also present a cost-effective heuristic algorithm delay, packet size, and reliability. To an extent, they explore
for resolving the first sub-problem. different topology network architecture under the uncertainty.
1738 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

TABLE I a dynamic mechanism for multiplexing of eMBB and uRLLC

L IST OF A BBREVIATIONS traffic and apply this in both frequency and time domain.
The efficient way of network resource sharing for the
eMBB and uRLLC is studied in [20]. A dynamic puncturing
mechanism is proposed for uRLLC traffic in [20] within
eMBB resources to increase the overall resource utilization
in the network. To enhance the performance for decoding
of eMBB traffic, a joint signal space diversity and dynamic
puncturing schemes have proposed, where they improve the
performance of component interleaving as well as rotation
modulation. For reducing the queuing delay of the uRLLC
traffic, the authors introduce punctured scheduling (PS) in [21].
In case of insufficient radio resource availability, the scheduler
promptly overwrites a portion of the eMBB transmission by
the uRLLC traffic. The scheduler improves the uRLLC latency
performance; however, the performance of the eMBB users
are profoundly deteriorated. The authors of [22] and [23]
manifest the coexistence technique for enabling 5G wireless
services like eMBB and uRLLC based upon a punctured
scheme. The authors present an enhanced PS (EPS) scheduler
to enable an improved ergodic capacity of the eMBB users
in [24]. EPS is capable of recovering the lost information
due to puncturing and partially. eMBB users are supposed
to be cognizant about the corresponding resource that is
being penetrated by uRLLC. Therefore, the victim eMBB users
ignore the punctured resources from the erroneous chase con-
densing HARQ process. The authors of [25] propose a MUPS,
where they discretize the trade-off among network system
capacity and uRLLC performance. MUPS first tries to match
the incoming uRLLC traffic inside an eMBB traffic in a conven-
tional multi-user multiple-input multiple-output (MU-MIMO)
transmission. MUPS serves the uRLLC traffic instantly by
using PS if MU pairing cannot be entertained immediately.
Though MUPS shows improved spectral efficiency, it is not
feasible for uRLLC latency as MU pairing mostly depends on
the rate maximization. Hence, the inter-user interference can
further degrade the signal-to-interference-noise-ratio (SINR)
quality of the uRLLC traffic, which can lead to reliability
concerns. The authors of [26] propose a null-space-based
preemptive scheduler (NSBPS) for jointly serving uRLLC
and eMBB traffic in a densely populated 5G arrangement.
The proposed approach ensures on-the-spot scheduling for
the sporadic uRLLC traffic, while makes a minimal shock
The authors of [16]–[18] present a resilient frame formation on the overall system outcome. The approach employs the
for multiplexing the provisions of different users. In [16], system spatial degrees of freedom (SDoF) for uRLLC traffic
the authors jointly MBB and mission-critical communica- for spontaneously providing a noise-free subspace.
tion traffic by engaging dynamic TDD and TTI. In [17], In [27], a joint scheduling problem is formulated for
the authors represent tractable multiplexing of mobile broad- eMBB and uRLLC traffic in the goal of maximizing eMBB
band (MBB), massive machine communication (MCC), and users’utility while satisfying stochastic demand for the uRLLC
mMTC considering dynamic TTI. The authors of [18] present UEs. Specifically, they measure the loss of eMBB users for
a holistic overview of the agile scheduling for 5G that incorpo- superposition/puncturing by introducing three models, which
rates multiple users. They envision an E2E QoS architecture to include linear, convex and threshold-based schemes. In [28],
offer improved opportunities for application-layer scheduling the authors propose a non-orthogonal coexistence scheme for
functionality that ensures QoE for each user. M/D/m/m uRLLC and eMBB services by processing uRLLC traffic at
queueing model-based system-level design has proposed for the edge nodes, whereas eMBB traffic is controlled centrally
fulfilling uRLLC traffic demand in [19], where they exhibit at the cloud. They analyze both uplink and downlink scenario
that the static bandwidth partitioning is inefficient for eMBB considering the heterogeneous requirements of those traffic.
and uRLLC traffic. Thus, the authors of [19] have illustrated In [29], the authors present a risk-sensitive approach for
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1739

TABLE II
S UMMARY OF N OTATIONS

Fig. 1. System model for coexisting eMBB and uRLLC services in 5G. e1 is
sharing RBs with u1 and u2 , and hence, creating capacity loss for e1 . e2 is
sharing RBs with u3 , and hence, creating capacity loss for e2 .

allocating resource blocks (RBs) to uRLLC traffic in the

goal of minimizing the uncertainty of eMBB transmission.
Particularly, they launch the Conditional Value at Risk (CVaR)
for estimating the uncertainty of eMBB traffic.

III. S YSTEM M ODEL AND P ROBLEM F ORMULATION

In this work, we consider a 5G network scenario with one
next generation base station (gNB) which supports a group
of user equipment (UE) E requiring eMBB service, and a set
of user equipment U demanding uRLLC service. Like most
of the works in literature, e.g. [20]–[27], we considered a
scenario with downlink transmissions from a common gNB
to UEs using different services (i.e., eMBB and URLLC), and
the overall system diagram is shown in Fig. 1. gNB supports
the UEs using licensed RBs K each with equal bandwidth
of B. Every time slot, with a length Δ, is split into M mini-
slots of duration δ for managing low latency services. For
supporting eMBB UEs, we consider Ts LTE time slots and
denoted by T = {1, 2, · · · , Ts }. uRLLC traffic arrive at gNB
(any mini-slot m of time slot t) follows Gaussian distribution,
i.e., U ∼ N (μ, σ 2 ). Here, μ and σ 2 denote the mean and
variance of U . Each uRLLC UE u ∈ U request for a payload
of size Lm,t
u (varying from 32 to 200 Bytes [30]).
gNB allots the RBs to the eMBB UEs at the commencement
of any time slot t ∈ T . The achievable rate of e ∈ E for RB spectral density. eMBB UEs require more than one RB for
k ∈ K is as follows: satisfying their QoS. Therefore, the achievable rate of eMBB
t
re,k = ΔB log2 (1 + γe,k
t
), (1) UE e ∈ E in time slot t as follows:

Pe h2e ret = αte,k re,k
t
, (2)
where γe,kt
= N0 Bpresents signal-to-noise-ratio (SNR).
k∈K
However, in our mathematical modeling of the system, we con-
sidered the residual interference generated by adjacent gNBs where α denotes the resource allocation vector for E at any
to be negligible assuming that an interference avoidance tech- time slot t, and each element is as follows:

nique (e.g. using disjoint sets of sub-channels for neighboring 1, if RB k is allocated for e ∈ E at time slot t,
cells) [31] can keep the inter-cell interference to minimal αe,k =
t
(3)
0, otherwise.
levels. The overview of the generalization process of the
proposed model into multicell model is presented in Appendix. uRLLC traffic can arrive at some moment (i.e. mini-slot)
Pe is the transmission power of gNB for e ∈ E and he denotes inside any time slot t and requires to be attended quickly.
the gain of e ∈ E from the gNB, and N0 represnts the noise Any uRLLC traffic needs to be completed within a mini-slot
1740 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

period for its’ latency and reliability constraints. Normally,

the payload size of uRLLC traffic is really short, and there-
fore, we cannot straightforwardly adopt Shannon’s data rate
formulation [10]. The achievable rate of a uRLLC UE u ∈ U
in RB k ∈ K, when its’ traffic is overlapped with eMBB traffic,
can properly be approximated by employing [32] as follows:
  
m,t Vu −1 d
ru,k = δ B log2 (1 + γu ) −
m,t
Q (εu ) , (4)
Nub
h2 P
where γum,t = N0 B+h u u
2P represents the SINR for u ∈ U
u e
at mini-slot m of t. Here, h2u Pe indicates the interference
generated from serving e ∈ E in the same RB, Vu =
h2u Pu
N0 B+h2u (Pu +Pe ) depicts the channel dispersion, and meaning
of other symbols are shown in II. However, the reliability of
uRLLC traffic fall into vulnerability due to the interference.
Hence, superposition mechanism is not a suitable for serving Fig. 2. Example of multiplexing between eMBB and uRLLC traffic.
uRLLC UE [11]. Thus, for serving uRLLC UEs, we con-
centrate on the puncturing technique. In the punctured mini-
slot, gNB allots zero power for eMBB UE, and therefore, control channel through which the scheduler can signal to the
the interference cannot affect the uRLLC traffic. At that time, eMBB receiver indicating the positions of URLLC overlap.
h2 P h2u Pu Indeed such a control channel has been proposed in the 3GPP
γum,t = Nu0 Bu and V = N0 B+h 2 P . The achieved rate of
u u standards [4]. We utilize the linear model of [27] for estimating
u ∈ U, when it uses multiple RBs, is as follows:
 m,t m,t the throughput-losses of eMBB UE. Therefore, the throughput-
rum,t = βe,k ru,k , (5) losses e ∈ E looks like as follows:
k∈K    m,t
t
re,loss = t
re,k I(αte,k = βu,k ). (10)
where β is the resource allocation vector for U at m of t, and k∈K m∈M u∈U
each of its’ element follows:
 So, the actual achievable rate of e ∈ E in any t is as follows:
m,t 1, if RB k is allocated for u ∈ U at m of t,
βe,k = (6) t
re,actual = ret − re,loss
t
. (11)
0, otherwise.
We see that β affects on α, and hence, impact negatively
All the uRLLC request in any m of t needs to be served
to the eMBB throughput in each t ∈ T . At the start of any
for sure, and hence,
 t ∈ T , gNB allocates the RBs K among the E in an orthogonal
P( φm,t
u < U ) ≤ , ∀m, t. (7) fashion as shown in Fig. 2. These characteristics of α are
u∈U shown mathematically as follows:

where φ denotes a vector for the serving uRLLC UEs, and αte,k ≤ 1, ∀k, (12)
thus, e∈E
 
1, if u ∈ U is served by the gNB at m of t, αte,k ≥ 1, ∀e, (13)
φu =
m,t
(8)
0, otherwise. 
k∈K

αte,k ≤ |K|. (14)

Within the stipulated period δ, the payload Lm,t
u of u ∈ U e∈E k∈K
needs to be transferred, and hence, satisfy the following:
Within each t ∈ T , gNB allows uRLLC UEs to get some
φm,t
u Lu
m,t
≤ δrum,t , ∀u, m, t. (9) RBs immediately on a mini-slot basis. Therefore, uRLLC
traffic overlaps with eMBB traffic at m and also shown
Hence, the reliability and latency requirements of uRLLC
in Fig. 2. Accordingly, β satisfy the following conditions on
traffic are respectively considered in (7) and (9). Besides,
each m:
e ∈ E loses some throughput at t if uRLLC traffic is punc-  m,t
tured within its’ RBs. We consider that the eMBB rate loss βu,k ≤ 1, ∀k, (15)
associated with URLLC puncturing is directly proportional to u∈U
the fraction of punctured minislots. This linear proportional  m,t
u βu,k ≥ 1, ∀u,
φm,t (16)
is motivated by basic results for the channel capacity of
k∈K
AWGN channel with erasures, see [44] for more details. Our  m,t
system in a given network state can be approximated as an u βu,k ≤ |K|.
φm,t (17)
u∈U k∈K
AWGN channel with erasures, when the slot sizes are long
enough so that the physical layer error control coding of Finally, our objective is to maximize the actual achievable
eMBB users use long code-words. Further, there is a dedicated rate of each eMBB UE across T while entertaining nearly
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1741

every uRLLC request within its’ speculated latency. We apply

Max-Min fairness doctrine for this mission, and it contributes
stationary service quality, enhances spectral efficiency and
makes UEs more pleasant in the network. Hence, the max-
imization problem is formulated as follows:
⎛ ⎞
|T |

max min E ⎝ t
re,actual ⎠ (18)
α,β e∈E
t=1


s.t. P φm,t
u <U ≤ , ∀m, t, (18a)
u∈U

φm,t
u Lu
m,t
≤ δrum,t , ∀u, m, t, (18b)

αte,k ≤ 1, ∀k, t, (18c)
e∈E
 m,t
βu,k ≤ 1, ∀k, m, t, (18d)
u∈U

αte,k ≥ 1, ∀e, t, (18e)
Fig. 3. Overview of the solution process for (18).
k∈K
 m,t
u βu,k ≥ 1, ∀u, m, t,
φm,t (18f) 
k∈K s.t. αte,k ≤ 1, ∀k, t, (19a)
  m,t e∈E
αte,k + φm,t
u βu,k ≤ |K|, ∀t, (18g) 
e∈E k∈K u∈U k∈K αte,k ≥ 1, ∀e, t, (19b)
m,t
αte,k , βu,k , φm,t ∈ {0, 1}, ∀e, u, k, m, t. (18h) k∈K
u 
In (18), the reliability and latency constraints of the uRLLC αte,k ≤ |K|, ∀t, (19c)
UEs are preserved by (18a) and (18b). Constraints (18c) e∈E k∈K
and (18d) are used to show the orthogonality of RBs among αte,k ∈ {0, 1}, ∀e, k, t. (19d)
eMBB and uRLLC UEs, respectively. At least one RB is
posed by every active UE and is encapsulated by both (18e) On the other hand, the second sub-problem (with αt , ∀t as
and (18f). Resource restriction is presented by constraint (18g). the solution of 19) is manifested as follows:
⎛ ⎞
Constraint (18h) shows that every item of α, β and φ are |T |

binary. The formulation (18) is a Combinatorial Program- max min E ⎝ t
re,actual ⎠ (20)
β e∈E
ming (CP) problem having chance constraint, and NP-hard t=1
due to its nature.

IV. D ECOMPOSITION AS A S OLUTION s.t. P φm,t
u <U ≤ , ∀m, t, (20a)
u∈U
A PPROACH FOR P ROBLEM (18)
We assume that eMBB UEs are data-hungry over the con- φm,t
u Lu
m,t
≤ δrum,t , ∀u, m, t, (20b)
 m,t
sidered period. Thus, at the commencement of a time slot βu,k ≤ 1, ∀k, m, t, (20c)
t ∈ T , gNB schedules all of its’ RBs among the eMBB UEs u∈U
and stay unchanged over t. If uRLLC traffic requests come  m,t
in any m of t, the scheduler tries to serve the requests in u βu,k ≥ 1, ∀u, m, t,
φm,t (20d)
the next m + 1. Hence, the overlapping of uRLLC traffic over k∈K

eMBB traffic happens as shown in Fig. 2. Usually, a portion of m,t
u βu,k ≤ |K|, ∀m, t,
φm,t (20e)
all RBs is required for serving such uRLLC traffic. However, u∈U k∈K
the challenge is to find the victimized eMBB UE(s) following m,t
the aspiration of the problem (18). βu,k , φm,t
u ∈ {0, 1}, ∀u, k, m, t. (20f)
For getting an effective solution to the problem (18), we can Fig. 3 shows the solution overview of the optimization
utilize the concept of a divide-and-conquer strategy. Here, problem (18). We can better understand the philosophy of
we divide (18) into two resource allocation sub-problems, the problem and the solution approach with an illustrative
namely, for eMBB UEs on time slot basis and uRLLC UEs example in Fig. 2. At the beginning of the time slot, t − 1,
on a mini-slot basis. The first sub-problem is as follows: let us assume that there are 3 eMBB UEs, each of whom
⎛ ⎞
|T | owns 4 RBs. Within t − 1, the service request for uRLLC
max min E ⎝ re,actual ⎠
t
(19) UEs came abruptly and the allocation of RBs for that UEs is
α e∈E
t=1 shown in Fig. 2, as overlapped uRLLC traffic in the mini-slots.
1742 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

During this time, eMBB users 1, 2 and 3 waste throughput Algorithm 1 Solution of (19) for Each t Based on PSUM
equivalent to 4RBs × 1 mini-slot, 7RBs × 1 mini-slot, and 1: Initialization: ε1 , σ1 , Imax and let i = 0
2RBs × 1 mini-slot, respectively. At the start of the next 2: Solve problem (21) and obtain solution αt,0
time slot, t, gNB acknowledges the resource scheduling of 3: while i < Imax do
uRLLC UEs of t − 1 to allocate and compensate eMBB 4: Set ε = εi+1 and σ = σi+1
UEs. gNB allocates more RBs to eMBB user 2 and less to 5: Solve problem (26) with the initial point being αt,i , and
eMBB user 3 as they lose more and less, respectively, in the obtain a new solution αt,i+1
time slot t − 1. Moreover, EgNB tries to serve uRLLC users 6: if αt,i+1 is binary then
such that the loss of throughput of eMBB users are almost 7: Stop
similar in the time slot t. Therefore, gNB makes a balance 8: else
among the throughput of eMBB users in each time slot, which 9: Set i = i + 1
ultimately serves to reach the goal of (18) on a long-run 10: Update εi+1 = ηε, and σi+1 = ζσ
basis. 11: end if
12: end while
A. PSUM as a Solution of the Sub-Problem (19) It is not easy to solve (22) directly. However, by utilizing the
Problem (19) is still is computationally expensive to reach successive upper bound minimization (SUM) technique [35],
a globally optimal solution due to its’ NP-hardness. In this [36], we can efficiently resolve (22). This method tries to
sub-section, we propose the PSUM algorithm to solve (19) secure the lower bound of the actual objective function by
approximately with low complexity. Relaxation of the binary determining a sequence of approximation of the objective
variable and the addition of a penalty term to the objective functions. As Pε (αt ) is concave in nature and hence,
function is the main philosophy of our proposed PSUM
algorithm. We redefine (19) as follows: Pε (αt ) ≤ Pε (αt,i ) + ∇Pε (αt,i )T (αt − αt,i ), (25)
 where αt,i is the value of current allocation of iteration i.
min Wet (αt ), ∀t, (21)
αt At the (i+1)-th iteration of t, we solve the following problem:
e∈E

s.t. (19a), (19b), (19c), (21a) 

min Wet (αt ) + σi+1 ∇Pε (αt,i )T αt (26)
  t
t−1 α
1  e∈E
Wet (αt ) = ret  ,actual + ret
t|E| s.t. (21a), (21b), (21c). (26a)
e ∈E t =1

t−1  In each iteration, we can get a globally optimal solution
1 
− t
re,actual + ret , (21b) for sub-problem (26) by using the solver. Algorithm 1 shows
t the proposed mechanism for solving (19). In this Algorithm,
t =1
0 < η < 1 < ζ where ζ and η represent two constants defined
αte,k ∈ [0, 1], ∀e, k, t. (21c)
previously.
Now according to Theorem 2 of [33], if |K| is sufficiently
large then original sub-problem (19) and (21) are equivalent. B. Solution of Sub-Problem (20) Through TM
Moreover, we add a penalty term Lp to the objective function
Due to the existence of chance constraint (20a) and also
to get binary soltion of relaxed variable from (21). Let αtk =
the combinatorial variable, β, (20) is still difficult to resolve
{αte,k }e∈E and we can rewrite (19a) as  αtk 1 ≤ 1, ∀t, k. The
by using traditional optimizer. Now, we need to trans-
penalized problem is as follows:
mute (20a) into deterministic 
form for solving (20). Moreover,

min Wet (αt ) + σPε (αt ), ∀t (19) let us assume g(φ, U ) = u∈U φu
m,t
− U , U ∈ R and
2
t
α
e∈E U ∼ N (μ, σ ), ∀m, t and hence,
s.t. (21a), (21b), (21c), (22a) P r{g(φ, U ) ≤ 0}
 
where σ > 0 is the penalty parameter, =Pr φm,t − U ≤ 0 (27)
u

Pε (αt ) = ( αtk + ε1 pp −cε,k ). (23) u∈U

k∈K =Pr φm,t
u ≤U (27a)
with p ∈ (0, 1), and ε is any non-negative constant. Following u∈U
 
the fact of [34] which is further described in [33], the optimal =1 − Pr φm,t ≥U (27b)
u
value is as follows: u∈U
  
cε,k = (1 + ε)p + (|E| − 1)εp . (24) U −μ m,t
u∈U φu −μ
=1 − Pr ≤ (27c)
σ σ
Generally, the parameter σ should big enough to make the 
values of {αte,k } near zero or one. Then, we achieve a feasible = 1 − FU φm,t . (27d)
u
solution of (22) by applying the rounding process. u∈U
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1743

Here, FU is the cumulative distribution function (CDF) and (30c), respectively, which convert them into equality,
of random variable U . Thus, from constraint (20a), we can we have:
rewrite as follows: 
χue + χ|U  |+1,e = se , ∀e ∈ E, (31)
P r{g(φ, U ) ≤ 0} ≥ , (28) u∈U 
   
du + d|U  |+1 = se . (32)
1 − FU φum,t
≤ , (28a)
u∈U  e∈E
 u∈U
  Now the modified problem in (30) is a balanced trans-
φm,t ≥1− ,
 (BTM). Moreover, we have to add d|U  |+1 =
FU (28b) portation model
u

u∈U
  e∈E s e − u∈U  du to the demand vector d as d =
 d ∪ {d|U  |+1 } and a row [0]1×|E| to cost matrix C as
−1
φm,t ≥ F 1 − , (28c)
u U
C = C ∪ {[0]1×|E| }. BTM can be solved by the simplex
u∈U
 method [38]. The solution matrix χ will be in the form of
φm,t
u − FU−1 (1 − ) ≥ 0. (28d) 
Z(|U |+1)×|E| . Northwest corner (NWC) [39], MCC [39], and
u∈U
Vogel’s approximation method (VAM) [39], [40] are some
Now, (28d) and (20a) are identical. Hence, the renewed form of the popular methods for obtaining initial feasible solution
of (20) looks like as follows: of BTM. We can use the stepping-stone [41] or MODI [42]
 method to get an optimal solution of the BTM. In the following
min Vet (αt , βt ), ∀t (29)
βt sub-section, we use the combination of the MCC and MODI
e∈E
 for acquaring the optimal result from the BTM.
s.t. φm,t
u − FU−1 (1 − ) ≥ 0, ∀m, (29a) 1) Determining Initial Feasible Solution by MCC Method:
u∈U
MCC method allots to those cells of χ considering the
(20b), (20c), (20d), (20e), (20f ), ∀u, m, (29b) lowest cost from C. Firstly, the method allows the maximum
1  permissible to the cell with the lowest per RB cost. Secondly,
Vet (αt , βt ) = ret  ,loss − re,loss
t
, ∀e. (29c)
|E|  the amount of quantity and need is synthesized while crossing
e ∈E
out the satisfied row(s) or column(s). Either row or column is
Problem (29) is still NP-hard due to the appearance of ruled out if both of them are satisfied concurrently. Thirdly,
combinatorial variable. In (29), (29a) holds for a particular we inquire into the uncrossed-out cells which have the least
value of when gNB serves a certain portion of uRLLC UE unit cost and continue it till there is specifically one row
U  ≤ U . For a m of t, let us assume U  = {1, 2, . . . ., U  } or column is left uncrossed. The primary steps of the MCC
and φm,tu = 1, ∀u ∈ U  . We can determine the requisite method are compiled as follows:
RBs, ∀u ∈ U  holding δ as the upper-bound in (20b) and let Step 1: Distribute maximum permissible to the worth-
d = [d1 , d2 , · · · , d|U  | ]. As gNB engages OFDMA for uRLLC while cell of χ which have the minimum cost found from
UEs, constraint (20c) holds. Moreover, depending on U  , C, and update the supply (s) and demand (d).
constraints (20d), (20e), and (20f) also hold. Constraint (29c) Step 2: Continue Step 1 till there is any demand that
can be used as a basic block to build a cost matrix C = needs to be satisfied.
(cu,e ), u ∈ U  , e ∈ E. As K are held by eMBB UEs E in any 2) MODI Method for Finding an Optimal Solution: The
time slot t ∈ T , we can find a vector s = [s1 , s2 , · · · , s|E| ]. initial solution found from section IV-B.1 is used as input in
Now redefine problem (29) as follows: the MODI method for finding an optimal solution. We need to
 augment an extra left-hand column and the top row (indicated
min cue χue (30)
χ by xu and ye respectively) with C whose values require to be
u∈U  e∈E
 calculated. The values are measured for all cells which have
s.t. χue = du , ∀u ∈ U  , (30a) the corresponding allocation in χ and shown as follows:
e∈E
 xu + ye = cu,e , ∀χu,e = ∅. (33)
χue ≤ se , ∀e ∈ E, (30b)
u∈U  Now we solve (33) to obtain all xu and ye . If necessary
 
du ≤ se , (30c) then assign zero to one of the unknowns toward finding the
u∈U  e∈E solution. Next, evaluate for all the empty cells of χ as follows:

se = |K|, (30d) ku,e = cu,e − xu − ye , ∀χu,e = ∅. (34)
e∈E
χue ≥ 0, ∀u ∈ U  , e ∈ E. (30e) Now select ku,e corresponding to the most negative value
and determine the stepping-stone path for that cell to know the

The goal of (30) is to find a matrix χ ∈ Z|U |×|E| = (χue ), reallocation amount to the cell. Next, allocate the maximum
∀u ∈ U  , e ∈ E that will minimize the cost/loss of eMBB permissible to the empty cell of χ corresponding to the
UEs. This is a linear programming problem equivalent to the selected ku,e . xu and ye values for C and χ must be recom-
Hitchcock problem [37] with inequities, which contributed puted with the help of (33) and a cost change for the empty
to unbalanced transportation model. Introducing slack vari- cells of χ need to be figured out using (34). A corresponding
ables χ|U  |+1,e , ∀e ∈ E and d|U  |+1 in the constraints (30b) reallocation takes place just like the previous step and the
1744 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

process continues till there is a negative ku,e . At the end of Algorithm 2 Heuristic Algorithm for Solving (19)
this repetitive process, we get the optimal allocation (χ). The 1: Initialization: ε1 , σ1 , Imax and let i = 0
MODI method described above can be summed as follows: 2: Solve problem (21) and obtain solution αt,0
Step 1: Develop a preliminary solution (χ) applying the 3: for each t ∈ T do
MCC method. 4: if t = 1 then
Step 2: For every row and column of C, measure xu 5: Calculate NRB = |K| |E|
and ye by applying (33) to each cell of χ that has an 6: for each e ∈ E do
allocation. 7: for each k = 1 · · · NRB do
Step 3: For every corresponding empty cell of χ, calcu- 8: αte,(e−1)∗NRB +k = 1
late ku,e by applying (34). 9: end for
Step 4: Determine the stepping-stone path [41] from χ 10: end for
corresponding to minimum ku,e that found in Step 3. 11: else
Step 5: Based on the stepping-stone path found in Step 12:
t−1
Determine re,loss t−1
and re,actual for all e ∈ E by using
4, allocate the highest possible to the free cell of χ. (10) and (11) respectively
Step 6: Reiterate Step 2 to 5 until all ku,e ≥ 0. 13: Set loc = 0
14: for each e ∈ E do

t−1
re,loss
C. Low-Complexity Heuristic Algorithm for Solving 15: Calculate NRB e
= |K|
r t−1
e ∈E e ,loss
Sub-Problem (19)
16: for each k = 1 · · · NRB
e
do
Though Algorithm 1 can solve the sub-problem (19) opti- 17: αe,loc+k = 1
t
mally, but computation time requires to solve it grows much 18: end for
faster as the size of the problem increase. Besides, the number 19: Set loc = loc + NRB t
of eMBB UEs is large in reality, and we have a short period to 20: end for
resolve this kind of problem. Therefore, we need a faster and 21: end if
efficient heuristic algorithm, which may sacrifice optimality, 22: end for
to solve (19). Thus, we propose Algorithm 2 for solving (19). 23: t
Determine re,actual for all e ∈E by using (11)
At t = 1, Algorithm 2 allocate resources equally to the eMBB 
|T | t
UEs. But, it allocates resources to eMBB UEs in the rest 24: Determine E t=1 re,actual for all e ∈ E
of the time slots depending on the proportional loss of the
previous time slot. In this way, Algorithm 2 can accommodate
the MEAR of eMBB UEs in the long-run. The complexity of   2
|T |
e∈E E
t
Algorithm 2 depends on T and E. t=1 re,actual
Fairness =   2 . (36)
|T |
V. N UMERICAL A NALYSIS AND D ISCUSSIONS |E| · e∈E
t
t=1 re,actual

In this section, we assess the proposed approach using

In our scenario, we consider an area with a radius of 200 m
comprehensive experimental analyses. Here, we compare
and gNB resides in the middle of the considered area. eMBB
our results with the results of the following state-of-the-art
and uRLLC UEs are disseminated randomly in the coverage
schedulers:
space. gNB works on a 10 MHz licensed band for supporting
• PS [21]: PS immediately overwrite part of the continuing
the UEs in downlink mode. Every uRLLC UE needs a single
eMBB transmission with the sporadic uRLLC traffic if PRB for its service. Furthermore, gNB estimates path-loss for
there are not sufficient physical resource blocks (PRBs) both eMBB and uRLLC UEs using a free space propagation
available. It chooses PRBs with the highest MCS that model amidst Rayleigh fading. Table III exhibits the signif-
already been allotted to eMBB UEs. icant parameters for this experiment. We use similar PSUM
• MUPS [25]: In case of insufficient RBs, MUPS allocates
parameters as of [33]. Moreover, the values of important
PRBs to the uRLLC UEs where they endure better simulation parameters of our work follow the 5G NR values as
channel quality depending on the CQI feedback. indicated in [45]. The decoding probability of the preempted
• RS: RS takes the RBs from the eMBB UEs randomly in
eMBB transmission depends on whether the UE is informed
case of inadequate PRBs for supporting uRLLC traffic. about that or not. If the eMBB UE is conscious of the
• EDS: For supporting sporadic uRLLC traffic, EDS offers
preemption then the performance is surely improved. It can
the PRBs to this traffic after preempting PRBs equally be expedited by granting a preemption indication (PI) to the
from the eMBB UEs in case of unavailable PRBs. concerned eMBB UEs, such that they understand which RB(s)
• MBS: gNB uses many to one matching game for snatch-
transmission have been corrupted. The eMBB UE(s) benefit
ing PRBs from eMBB UEs for supporting uRLLC traffic. from PI information by overlooking the corrupted RBs of
The main performance parameters are MEAR and fair- the transmission in its decoding process, including potentially
ness [43] of the eMBB UEs and defined as follows: performing HARQ soft combining, thereby improving the per-
⎛ ⎞
|T | formance. However, it is fair to compare the proposed method

MEAR = min E ⎝ re,actual ⎠ , ∀e ∈ E, (35)
t
with similar methods i.e. other punctured schemes (without
t=1 recovery mechanism), and thus, we have compared our method
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1745

TABLE III
S UMMARY OF THE S IMULATION S ETUP

Fig. 5. Comparison of fairness score when E = 4 and single uRLLC UE in

each mini-slot along with L = 32 bytes.

Fig. 4. Comparison of MEAR during E = 4 and single uRLLC UE in every

mini-slot when L = 32 bytes.

with such punctured schemes [21], [25], along with other

mechanisms. We realize the results of every approaches after
taking 1, 000 runs.
A comparison of MEAR and fairness scores are presented
in Fig. 4 and Fig. 5, respectively, between the proposed
(PSUM+TM) and the optimal value for a small network.
Fig. 4 shows the empirical cumulative distribution function
(ECDF) of MEAR and the probability of MEAR being at
least 20 Mbps are around 0.50 and 0.70, respectively, for the Fig. 6. Comparison of MEAR for (a) σ = 1, (b) σ = 5, and (c) σ = 10,
proposed and optimal methods, consequently. The optimality along with L = 32 Bytes.
gap of average MEAR for the proposed method is 4.20% as
represented in Fig. 4. Fig. 5 shows the ECDF of the fairness 0.889, 0.405, 0.367, 0.653, 0.653, and 0.052 for the proposed,
scores where the probability of the scores being 0.995 at least RS, EDS, MBS, PS, and MUPS methods, respectively, that
is 0.80 in the proposed method in comparison of being 1 in are shown in Fig. 6(a). Fig. 6(b) reveals that the likelihood
the optimal mechanism. The optimality gap of the proposed of MEAR values for obtaining a minimum of 18.0 Mbps
method for the average fairness score is 0.32% as exposed are 0.736, 0.089, 0.050, 0.541, and 0.647 for the proposed,
from Fig. 5. RS, EDS, MBS, and PS methods, respectively, while the
For growing uRLLC arrivals, the ECDF of the MEAR MUPS method can accommodate under 18 Mbps in every
values is exhibited in Fig. 6. Fig. 6 reveals the results that case. Fig. 6(c) shows that the proposed, MBS and PS meth-
are preferred to those of the other considered methods. The ods provide a minimum MEAR value of 18.0 Mbps with a
probability of MEAR values for being at least 18.0 Mbps are probability 0.231, 0.089, and 0.231, respectively, while RS,
1746 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

Fig. 8. Comparison of average MEAR with varying value of σ and

L = 32 Bytes.

Fig. 7. Comparison of fairness scores (a) σ = 1, (b) σ = 5, and (c) σ = 10,

along with L = 32 Bytes.

EDS, and MUPS can produce less than 18 Mbps for sure.
Moreover, the MEAR value decreases with the growing rate of
σ for all the methods because of the requirement of more RBs
for the uRLLC UEs as shown in Fig. 6. But, the increasing
arrivals of uRLLC traffic affect the MUPS method more as
they require extra RBs from the distant eMBB UEs. However,
the performance gap between the proposed and PS method
reduces with the increased arrival of uRLLC traffic, as the PS Fig. 9. Comparison of fairness score with varying value of σ and
scheme gets more chance to adjust the users with the higher L = 32 Bytes.
expected achieved rate.
We compare the fairness scores among various methods
with different values of σ which is shown in Fig. 7. The scores the case of average MEAR. The figure also explicates that the
originating from the proposed method are greater than or average MEAR is declining with the growing value of σ due
similar to that of others as indicated in Fig. 7. Fig. 7(a) reveals to the additional requirement of PRBs for extra uRLLC traffic.
that the median of the scores for the proposed, RS, EDS, Particularly, our method results 10.20%, 10.87%, 5.77%,
MBS, PS, and MUPS methods are 0.9977, 0.9897, 0.9897, 5.77%, and 18.55% higher on average MEAR than those of
0.9975, 0.9972, and 0.9789, respectively. The similar scores RS, EDS, MBS, PS, and MUPS, respectively, for σ = 1.
are 0.9998, 0.9902, 0.9902, 0.9987, 0.9995, 0.9488, and 1.00, Moreover, similar values are 15.22%, 16.43%, 6.22%, 3.75%,
0.9891, 0.9891, 0.9985, 0.9998, 0.8784 for the corresponding and 70.20% for σ = 10. The average fairness score emerging
methods and are presented in Fig. 7(b) and 7(c), respectively. from our method is bigger than or similar to other comparing
Moreover, the fairness scores increase for the Proposed, MBS methods for different values of σ and shown in Fig. 9. Fig. 9
and PS methods with the increasing value of σ as it gets more also reveals that the σ value has a negligible impact on the
chance to maximize the minimum achieved rate, whereas the average score of the fairness in the Proposed, RS, EDS, MBS,
same scores decrease with the increasing value of σ for RS, PS methods, but it impacts inversely to the MUPS method
EDS and MUPS as eMBB UEs have more opportunity to be more and more uRLLC traffic choose same eMBB UE for the
affected by the uRLLC UEs. PRBs. Moreover, the average fairness scores of the proposed
Fig. 8 and 9, respectively, show the average MEAR and method are similar to both MBS and PS methods. However,
fairness score for varying value of σ. In Fig. 8, we find that the proposed method treats eMBB UEs 0.92%, 0.92%, and
our method overpasses other schemes for different rates of σ in 1.92% fairly than RS, EDS, and MUPS methods, respectively,
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1747

Fig. 10. Comparison of average MEAR with varying uRLLC load and σ. Fig. 12. Comparison of average uRLLC latency with varying uRLLC load
and σ.

Moreover, for the MUPS method, these values decrease with

the increasing value of σ and L.
In Fig. 12, we compare the average latency for uRLLC
traffic with varying value of σ and uRLLC load (L). The
Fig. 12 reveals that the average uRLLC latency is below
0.25 ms, which is the requirement for uRLLC traffic, for all
considered cases. Moreover, this average uRLLC latency has
no relation with the value of σ and L as all of the uRLLC
traffics are served in one mini-slot for all the considered
scenarios. However, the small differences of average latency
values are due to the arrival period of uRLLC traffics within
a mini-slot.
VI. C ONCLUSION
In this paper, we have introduced a novel approach for
coexisting uRLLC and eMBB traffic in the same radio resource
for enabling 5G wireless systems. We have expressed the
Fig. 11. Comparison of average fairness score with varying uRLLC load coexisting dilemma as a maximizing problem of the MEAR
and σ.
value of eMBB UEs meanwhile attending the uRLLC traffic.
when σ = 1, whereas, the similar scores are 1.23%, 1.23%, We handle the problem with the help of the decomposi-
and 12.21%, respectively, during σ = 10. tion strategy. In every time slot, we resolve the resource
In Fig. 10, we compare the average MEAR of eMBB UEs for scheduling sub-problem of eMBB UEs using a PSUM based
considering varying uRLLC load (L) and uRLLC traffic (σ). algorithm, whereas the similar sub-problem of uRLLC UEs is
The MEAR value of our method surpasses other concerned unraveled through optimal transportation model, namely MCC
methods in every circumstance as revealed from Fig. 10. The and MODI methods. For the efficient scheduling of PRBs
same figure also explicates that these values degrade when L among eMBB UEs, we also present a heuristic algorithm.
increases for varying σ as the system needs to allocate more Our extensive simulation outcomes demonstrate a notable
PRBs to the uRLLC UEs. Moreover, these values decrease performance gain of the proposed approach over the baseline
with the increasing value of σ for a fixed L, and also the approaches in the considered indicators.
same for increasing the value of L with a fixed σ. In Fig. 11,
we compare the average fairness score of eMBB UEs for A PPENDIX
the different methods for changing the uRLLC load (L) and G ENERALIZATION OF THE P ROPOSED
uRLLC traffic (σ). Fig. 11 exposes that the fairness scores M ODEL I NTO M ULTICELL M ODEL
of our method are better than or at least similar to that of Considering multiple gNBs onto our system model would
its’ rivals. The figure also reveals that these scores decrease result in an interference Considering multiple gNBs onto our
with an increasing L for the lower value of σ. However, these system model would result in an interference component on
scores increase with the increasing L when σ value is high. the SINR expression, and this will help us to generalize
1748 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 3, MARCH 2021

the model. This is because multiple gNBs consider physical [21] K. I. Pedersen, G. Pocovi, J. Steiner, and S. R. Khosravirad, “Punctured
resource reuse, where frequencies are reused at spatially sepa- scheduling for critical low latency data on a shared channel with
mobile broadband,” in Proc. IEEE 86th Veh. Technol. Conf. (VTC-Fall),
rated locations to increase spectral efficiency. To achieve that, Sep. 2017, pp. 1–6.
the SNR term will be replaced by SINR. Let G = {1, . . . , G} [22] A. K. Bairagi, M. S. Munir, S. F. Abedin, and C. S. Hong, “Coexistence
be the set of gNBs; thus, the SINR can be calculated as of eMBB and uRLLC in 5G wireless networks,” in Proc. Korea Comp.
Cong., Jun. 2018, pp. 1377–1379.
follows: [23] A. K. Bairagi, M. S. Munir, M. Alsenwi, N. H. Tran, and C. S. Hong,
Pe h2e “A matching based coexistence mechanism between eMBB and uRLLC
t,g
γe,k =  , ∀g ∈ G, g  = g, (37) in 5G wireless networks,” in Proc. 34th ACM/SIGAPP Symp. Appl.
Ig + N0 B Comput., Apr. 2019, pp. 2377–2384.
[24] K. I. Pedersen, G. Pocovi, and J. Steiner, “Preemptive scheduling of
where Ig represents the corresponding interference at gNB latency critical traffic and its impact on mobile broadband performance,”
g ∈ G from all other gNBs g  ∈ G. in Proc. IEEE 87th Veh. Technol. Conf. (VTC), Jun. 2018, pp. 1–6.
[25] A. A. Esswie and K. I. Pedersen, “Multi-user preemptive scheduling for
R EFERENCES critical low latency communications in 5G networks,” in Proc. IEEE
Symp. Comput. Commun. (ISCC), Jun. 2018, pp. 136–141.
[1] Cisco Visual Networking Index: Global Mobile Data Traffic Forecast [26] A. A. Esswie and K. I. Pedersen, “Opportunistic spatial preemptive
Update, 2016—2021, Cisco, San Jose, CA, USA, Jun. 2017. scheduling for URLLC and eMBB coexistence in multi-user 5G net-
[2] 5G Service Roadmap 2022, 5G Forum, Bangalore, India, Mar. 2016. works,” IEEE Access, vol. 6, pp. 38451–38463, Jul. 2018.
[3] IMT Vision-Framework and Overall Objectives of the Future Devel-
[27] A. Anand, G. de Veciana, and S. Shakkottai, “Joint scheduling of
opment of IMT for 2020 and Beyond, document ITU-R M.2083-0,
URLLC and eMBB traffic in 5G wireless networks,” IEEE/ACM Trans.
Sep. 2015.
Netw., vol. 28, no. 2, pp. 477–490, Apr. 2020.
[4] 3GPP TSG RAN WG1 Meeting #87, document R1-1612306, 3GPP,
[28] R. Kassab, O. Simeone, P. Popovski, and T. Islam,
Nov. 2016.
“Non-orthogonal multiplexing of ultra-reliable and broadband services
[5] Downlink Multiplexing of eMBB and URLLC Transmission,
in fog-radio architectures,” IEEE Access, vol. 7, pp. 13035–13049,
document R1-1700374, 3GPP, Jan. 2017.
[6] A. K. Bairagi, S. F. Abedin, N. H. Tran, D. Niyato, and C. S. Hong, Jan. 2019.
“QoE-enabled unlicensed spectrum sharing in 5G: A game-theoretic [29] M. Alsenwi, N. H. Tran, M. Bennis, A. Kumar Bairagi, and C. S. Hong,
approach,” IEEE Access, vol. 6, pp. 50538–50554, Sep. 2018. “EMBB-URLLC resource slicing: A risk-sensitive approach,” IEEE
[7] A. K. Bairagi, N. H. Tran, W. Saad, and C. S. Hong, “Bargaining game Commun. Lett., vol. 23, no. 4, pp. 740–743, Apr. 2019.
for effective coexistence between LTE-U and Wi-Fi systems,” in Proc. [30] Study on New Radio Access Technology Physical Layer Aspects, docu-
NOMS–IEEE/IFIP Netw. Oper. Manage. Symp., Apr. 2018, pp. 1–8. ment 3GPP RT 38.802v14.0.0, 3GPP, Mar. 2017.
[8] S. Liu, F. Yang, J. Song, and Z. Han, “Block sparse Bayesian learning- [31] J. Zhang and J. G. Andrews, “Adaptive spatial intercell interference can-
based NB-IoT interference elimination in LTE-advanced systems,” IEEE cellation in multicell wireless networks,” IEEE J. Sel. Areas Commun.,
Trans. Commun., vol. 65, no. 10, pp. 4559–4571, Oct. 2017. vol. 28, no. 9, pp. 1455–1468, Dec. 2010.
[9] S. F. Abedin, M. G. R. Alam, S. M. A. Kazmi, N. H. Tran, D. Niyato, and [32] J. Scarlett, V. Y. F. Tan, and G. Durisi, “The dispersion of nearest-
C. S. Hong, “Resource allocation for ultra-reliable and enhanced mobile neighbor decoding for additive non-Gaussian channels,” IEEE Trans.
broadband IoT applications in fog network,” IEEE Trans. Commun., Inf. Theory, vol. 63, no. 1, pp. 81–92, Jan. 2017.
vol. 67, no. 1, pp. 489–502, Jan. 2019. [33] N. Zhang, Y.-F. Liu, H. Farmanbar, T.-H. Chang, M. Hong, and
[10] R. Kassab, O. Simeone, and P. Popovski, “Coexistence of URLLC Z.-Q. Luo, “Network slicing for service-oriented networks under
and eMBB services in the C-RAN uplink: An information-theoretic resource constraints,” IEEE J. Sel. Areas Commun., vol. 35, no. 11,
study,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2018, pp. 2512–2521, Nov. 2017.
pp. 1–6. [34] P. Liu, Y.-F. Liu, and J. Li, “An iterative reweighted minimization
[11] K. Ying, J. M. Kowalski, T. Nogami, Z. Yin, and J. Sheng, “Coexistence framework for joint channel and power allocation in the OFDMA
of enhanced mobile broadband communications and ultra-reliable low- system,” in Proc. IEEE Int. Conf. Acoust., Speech Signal Process.
latency communications in mobile front-haul,” Proc. SPIE, vol. 10559, (ICASSP), Apr. 2015, pp. 3068–3072.
Jan. 2018, Art. no. 105590C. [35] D. R. Hunter and K. Lange, “A tutorial on MM algorithms,” Amer.
[12] G. Pocovi, K. I. Pedersen, and P. Mogensen, “Joint link adaptation Statistician, vol. 58, no. 1, pp. 30–37, Feb. 2004.
and scheduling for 5G ultra-reliable low-latency communications,” IEEE [36] M. Razaviyayn, M. Hong, and Z.-Q. Luo, “A unified conver-
Access, vol. 6, pp. 28912–28922, May 2018. gence analysis of block successive minimization methods for non-
[13] P. Popovski, K. F. Trillingsgaard, O. Simeone, and G. Durisi, “5G wire- smooth optimization,” SIAM J. Optim., vol. 23, no. 2, pp. 1126–1153,
less network slicing for eMBB, URLLC, and mMTC: A communication- Jan. 2013.
theoretic view,” IEEE Access, vol. 6, pp. 55765–55779, Sep. 2018. [37] F. L. Hitchcock, “The distribution of a product from several sources to
[14] H. Ji, S. Park, J. Yeo, Y. Kim, J. Lee, and B. Shim, “Ultra-reliable and numerous localities,” J. Math. Phys., vol. 20, nos. 1–4, pp. 224–230,
low-latency communications in 5G downlink: Physical layer aspects,” Apr. 1941.
IEEE Wireless Commun., vol. 25, no. 3, pp. 124–130, Jun. 2018. [38] G. B. Dantzig, Linear Programming and Extensions. Princeton, NJ,
[15] M. Bennis, M. Debbah, and H. V. Poor, “Ultra-reliable and low-latency USA: Princeton Univ. Press, 1963.
wireless communication: Tail, risk and scale,” Proc. IEEE, vol. 106, [39] N. V. Reinfeld and W. R. Vogel, Mathematical Programming.
no. 10, pp. 1834–1853, 2018. Englewood Gliffs, NJ, USA: Prentice-Hall, 1958.
[16] Q. Liao, P. Baracca, D. Lopez-Perez, and L. G. Giordano, “Resource [40] D. Shimshak, J. Alan Kaslik, and T. Barclay, “A modification of Vogel’S
scheduling for mixed traffic types with scalable TTI in dynamic TDD approximation method through the use of heuristics,” Inf. Syst. Oper.
systems,” in Proc. IEEE Globecom Workshops (GC Wkshps), Dec. 2016, Res., vol. 19, no. 3, pp. 259–263, Aug. 1981.
pp. 1–7. [41] A. Charnes and W. W. Cooper, “The stepping stone method of explaining
[17] K. I. Pedersen, G. Berardinelli, F. Frederiksen, P. Mogensen, and linear programming calculations in transportation problems,” Manag.
A. Szufarska, “A flexible 5G frame structure design for frequency- Sci., vol. 1, no. 1, pp. 49–69, Oct. 1954.
division duplex cases,” IEEE Commun. Mag., vol. 54, no. 3, pp. 53–59, [42] H. A. Taha, Operations Research: An Introduction. Upper Saddle River,
Mar. 2016. NJ, USA: Pearson, 2007.
[18] K. Pedersen, G. Pocovi, J. Steiner, and A. Maeder, “Agile 5G scheduler [43] R. Jain, D. M. Chiu, and W. R. Hawe, “A quantitative measure of
for improved E2E performance and flexibility for different network fairness and discrimination for resource allocation in shared computer
implementations,” IEEE Commun. Mag., vol. 56, no. 3, pp. 210–217, system,” Eastern Res. Lab., Digit. Equip. Corp., Maynard, MA, USA,
Mar. 2018. Tech. Rep. TR 301, Sep. 1984, vol. 38.
[19] C.-P. Li, J. Jiang, W. Chen, T. Ji, and J. Smee, “5G ultra-reliable and low- [44] D. Julian, “Erasure networks,” in Proc. IEEE Int. Symp. Inf. Theory,
latency systems design,” in Proc. Eur. Conf. Netw. Commun. (EuCNC), Jun. 2003, p. 138.
Jun. 2017, pp. 1–5. [45] Y. Huang, S. Li, C. Li, Y. T. Hou, and W. Lou, “A deep-reinforcement-
[20] Z. Wu, F. Zhao, and X. Liu, “Signal space diversity aided dynamic learning-based approach to dynamic eMBB/URLLC multiplexing in 5G
multiplexing for eMBB and URLLC traffics,” in Proc. 3rd IEEE Int. NR,” IEEE Internet Things J., vol. 7, no. 7, pp. 6439–6456, Jul. 2020,
Conf. Comput. Commun. (ICCC), Dec. 2017, pp. 1396–1400. doi: 10.1109/JIOT.2020.2978692.
BAIRAGI et al.: COEXISTENCE MECHANISM BETWEEN eMBB AND uRLLC IN 5G WIRELESS NETWORKS 1749

Anupam Kumar Bairagi (Member, IEEE) received Mehedi Masud (Senior Member, IEEE) received
the B.Sc. and M.Sc. degrees in computer science the Ph.D. degree in computer science from the
and engineering from Khulna University (KU), University of Ottawa, Canada. He is currently a
Bangladesh, and the Ph.D. degree in computer engi- Professor with the Department of Computer Science,
neering from Kyung Hee University, South Korea. Taif University, Taif, Saudi Arabia. He has authored
He is currently an Associate Professor with the and coauthored around 70 publications, including
Discipline of Computer Science and Engineering, refereed IEEE/ACM/Springer/Elsevier journals, con-
Khulna University. He has authored and coau- ference papers, books, and book chapters. He has
thored around 40 publications, including refereed served as a Technical Program Committee member
IEEE/ACM journals and conference papers. He has for different international conferences. His research
served as a Technical Program Committee member interests include machine learning, distributed algo-
for different international conferences. His research interests include wire- rithms, data security, formal methods, and health analytics. He was a recipient
less resource management in 5G and beyond, healthcare, IIoT, cooperative of a number of awards, including the Research in Excellence Award from Taif
communication, and game theory. University. He is on the Associate Editorial Board of IEEE A CCESS , Inter-
national Journal of Knowledge Society Research (IJKSR), and an Editorial
Board member of Journal of Software. He has also served as a Guest Editor
Md. Shirajum Munir (Graduate Student Member, for Computer Science and Information Systems (ComSIS) Journal and Journal
IEEE) received the B.S. degree in computer science of Universal Computer Science (JUCS). He is also a member of ACM.
and engineering from Khulna University, Khulna,
Bangladesh, in 2010. He is currently pursuing the
Ph.D. degree in computer science and engineering
with Kyung Hee University, Seoul, South Korea. Zhu Han (Fellow, IEEE) received the B.S. degree
He has served as a Lead Engineer for the Solution in electronic engineering from Tsinghua University,
Laboratory, Samsung Research and Development in 1997, and the M.S. and Ph.D. degrees in electrical
Institute, Dhaka, Bangladesh, from 2010 to 2016. and computer engineering from the University of
His current research interests include IoT network Maryland, College Park, in 1999 and 2003, respec-
management, fog computing, mobile edge comput- tively. From 2000 to 2002, he was a Research and
ing, software-defined networking, smart grid, and machine learning. Development Engineer with JDSU, Germantown,
MD, USA. From 2003 to 2006, he was a Research
Associate with the University of Maryland. From
Madyan Alsenwi received the B.E. and M.Sc. 2006 to 2008, he was an Assistant Professor with
degrees in electronics and communications engineer- Boise State University, Idaho. He is currently a John
ing from Cairo University, Egypt, in 2011 and 2016, and Rebecca Moores Professor with the Department of Electrical and Com-
respectively. He is currently pursuing the Ph.D. puter Engineering, and the Department of Computer Science, University of
degree in computer science and engineering with Houston, Houston, TX, USA. His research interests include wireless resource
Kyung Hee University, South Korea. He is also allocation and management, wireless communications and networking, game
working as a Leading Researcher with the Intelli- theory, big data analysis, security, and smart grid. He received an NSF Career
gent Networking Laboratory under a project jointly Award in 2010, the Fred W. Ellersick Prize of the IEEE Communication
funded by the prestigious Brain Korea 21st Century Society in 2011, the EURASIP Best Paper Award for the Journal on Advances
Plus and Ministry of Science and ICT, South Korea. in Signal Processing in 2015, IEEE Leonard G. Abraham Prize in the field
Prior to this, he worked as a Research Assistant of Communications Systems [Best Paper Award in IEEE J OURNAL ON
under several research projects funded by the Egyptian Government. His S ELECTED A REAS IN C OMMUNICATIONS (JSAC)] in 2016, and several best
research interests include wireless communications and networking, resource paper awards in IEEE conferences. He was has been an IEEE Communications
slicing in 5G wireless networks, ultra reliable low latency communications Society Distinguished Lecturer from 2015 to 2018, the AAAS Fellow since
(URLLC), UAV-assisted wireless networks, and machine learning. 2019, and the ACM distinguished Member since 2019. He has also been 1%
highly cited researcher since 2017 according to Web of Science. He is also
the winner of 2021 IEEE Kiyo Tomiyasu Award, for outstanding early to
Nguyen H. Tran (Senior Member, IEEE) received mid-career contributions to technologies holding the promise of innovative
the B.S. degree in electrical and computer engineer- applications, with the following citation: “for contributions to game theory
ing from the Hochiminh City University of Technol- and distributed management of autonomous communication networks.”
ogy in 2005 and the Ph.D. degree in electrical and
computer engineering from Kyung Hee University
in 2011. He was an Assistant Professor with the
Department of Computer Science and Engineering, Choong Seon Hong (Senior Member, IEEE)
Kyung Hee University, from 2012 to 2017. Since received the B.S. and M.S. degrees in electronic
2018, he has been with the School of Computer engineering from Kyung Hee University, Seoul,
Science, The University of Sydney, where he is South Korea, in 1983 and 1985, respectively, and the
currently a Senior Lecturer. His research interests Ph.D. degree from Keio University, Japan, in 1997.
include distributed computing and learning over networks. He received the In 1988, he joined KT, where he was involved
Best KHU Thesis Award in engineering in 2011, and several best paper in broadband networks as a Member of Technical
awards, including the IEEE ICC 2016, APNOMS 2016, and IEEE ICCS 2016. Staff. Since 1993, he has been with Keio Univer-
He receives the Korea NRF Funding for Basic Science and Research from sity. He was with the Telecommunications Network
2016 to 2023. He has been an Editor of IEEE T RANSACTIONS ON G REEN Laboratory, KT, as a Senior Member of Technical
C OMMUNICATIONS AND N ETWORKING since 2016. Staff and as the Director of the Networking Research
Team until 1999. Since 1999, he has also been a Professor with the Department
of Computer Science and Engineering, Kyung Hee University. His research
Sultan S. Alshamrani received the bachelor’s interests include future Internet, ad hoc networks, network management,
degree (Hons.) in computer science from Taif Uni- and network security. He is also a member of ACM, the IEICE, the IPSJ,
versity, Saudi Arabia, in 2007, the master’s degree the KIISE, the KICS, the KIPS, and the OSIA. He has served as the General
in information technology (computer networks) from Chair, the TPC Chair/Member, or an Organizing Committee member for
The University of Sydney, Sydney, NSW, Australia, international conferences, such as NOMS, IM, APNOMS, E2EMON, CCNC,
and the Ph.D. degree from the University of Liver- ADSN, ICPP, DIM, WISA, BcN, TINA, SAINT, and ICOIN. He was an
pool, U.K. He is currently working as an Associate Associate Editor of IEEE T RANSACTIONS ON N ETWORK AND S ERVICE
Professor with Taif University. He is also the Head M ANAGEMENT and IEEE Journal of Communications and Networks. He cur-
of the Department of Information Technology. rently serves as an Associate Editor for International Journal of Network
Management and an Associate Technical Editor for IEEE Communications
Magazine.