A Chance Constrained Based Formulation for

Dynamic Multiplexing of eMBB-URLLC Traffics in

5G New Radio
Madyan Alsenwi, Shashi Raj Pandey, Yan Kyaw Tun, Ki Tae Kim, and Choong Seon Hong
Department of Computer Science and Engineering, Kyung Hee University, 446-701, Republic of Korea
Email: {malsenwi, shashiraj, ykyawtun7, glideslope, cshong}@khu.ac.kr

Abstract—5G New Radio (NR) is envisioned to provide three The allocated resources are orthogonal in frequency-time and
major services: enhanced Mobile Broadband (eMBB), Ultra therefore there is no interference [3]. The main drawback here
Reliable Low Latency Communications (URLLC), and massive is an inefficient use of the resources.
Machine Type Communication (mMTC). URLLC services (i.e.,
autonomous vehicles, industrial Internet of Things (IoT),...) Another coexistence approach is a non-orthogonal slicing.
require strict latency, on-way latency of 1 ms, with 99.999% In this approach, the same resources are shared to all services
reliability. eMBB applications aims extreme data rate while at the same time. Reusing the resources increases the spectral
mMTC is designed to serve a large number of IoT devices that efficiency however it causes an interference among these
send a small data sporadically. In this paper, we address the services [4], [5].
resource scheduling problem of URLLC and eMBB traffics. First,
the Resource Blocks (RBs) are allocated to eMBB users at the
Recently, studies focusing on URLLC are gaining in both
beginning of each time slot based on the channel state of each academia and industry. The authors in [4] discuss the prin-
eMBB user and his previous average data rate up to current time ciples for achieving URLLC and describe several building
slot. The RBs allocation problem modeled as as a 2-Dimensions blocks of wireless communication systems for supporting
Hopfield Neural Networks (2D-HNN) and the energy function URLLC (framing, use of diversity, and access topology).
of 2D-HNN is investigated to solve the RBs allocation problem.
Then, the resource scheduling problem of URLLC and eMBB is
Authors in [6] propose a punctured scheduling approach for
formulated as an optimization problem with chance constraint. transmission of low latency communication (LLC) traffic mul-
The chance constraint based problem aims to maximize the tiplexed on a shared channel with eMBB. Moreover, they pro-
eMBB data rate while satisfying the URLLC critical constraints. pose recovery mechanisms for punctured eMBB transmissions.
The cumulative Distribution Function (CDF) of the stochastic In this paper, we tackle the download resource scheduling of
URLLC traffic is investigated to relax the chance constraint into
a linear constraint. The simulation results show efficiency of the URLLC and eMBB based on the puncturing approach. The
proposed dynamic scheduling approach. time domain is divided into equally spaced time slots with
Index Terms—5G New Radio (NR), Ultra Reliable Low La- duration of 1 ms as that in the current cellular networks. Each
tency Communications (URLLC), enhanced Mobile Broadband time slot is further divided into minislots in order to achieve
(eMBB), Resource Scheduling, Chance Constraints, Hopfield the latency requirements of URLLC [1].
Neural Networks. The resources are allocated to eMBB users at the level of
I. I NTRODUCTION Resource Blocks (RBs)1 at the beginning of each time slot. We
model the RBs allocation problem as a 2-Dimension Hopfield
The Third Generation Partnership Project (3GPP) is final- Neural Network (2D HNN). The optimization problem is
izing the standardization of the 5G New Radio (NR) [1]. The formulated based on the Generalized Proportional Fair (GPF)
first specification of the upcoming 5G NR support three types formulation and then rewritten in the form of the energy
of services: Ultra Reliable Low Latency Communications function of 2D HNN. The decreasing property of HNN energy
(URLLC), massive Machine Type Communications (mMTC), function is investigated to solve the problem. The main moti-
and enhanced Mobile Broad Band (eMBB). The URLLC ac- vation to use the HNNs is that it can give an on-line solution
commodate the applications which require ultra high reliability due to its ability to process in parallel and thus it can save
(1 − 10−5 success probability) and low transmission latency the computation time. However, the incoming URLLC traffic
of 1 ms [1], [2]. Whereas, eMBB is an extension of the LTE- during a time slot is placed immediately to transmit in the
Advanced and its application require high data rate. next minislot, by puncturing the eMBB transmissions, due to
One approach to support the coexistence of this various its hard latency requirements and this impacts the data rate of
services is the orthogonal slicing. In this approach, different re- eMBB traffic. Therefore, we formulate the resource scheduling
sources are allocated to each service to satisfy its requirements. problem of URLLC and eMBB as an optimization problem
This work was supported by Institute for Information and communica- with chance constraint of which objective is to maximize
tions Technology Promotion (IITP) grant funded by the Korea government
(MSIT) (No.2015-0-00567, Development of Access Technology Agnostic 1 Resource Blocks are small frequency-time rectangles. In 4G, the RB is
Next-Generation Networking Technology for Wired-Wireless Converged Net- 180 KHz wide in frequency and one slot long in time. 5G New Radio permits
works). *Dr. CS Hong is the corresponding author. a large number of block shapes varying from 15 kHz to 480 kHz

978-1-5386-8350-7/19/$31.00 ©2019 IEEE 108 ICOIN 2019

the data rate of eMBB users while satisfying the URLLC optimization problem has to maximize the total data rate of all
constraints. The cumulative Distribution Function (CDF) of users while maintaining a certain level of long-term fairness
the stochastic URLLC traffic is investigated to relax the chance between eMBB users. Let B = {1, 2, ..., B} denote the set
constraint into a linear constraint. The resulting optimization of all available RBs, U = {1, 2, ..., U } denote the set of all
problem is a convex optimization problem which we can solve eMBB users. Therefore, the maximum data rate of an eMBB
it easily using any convex solver. user u at time slot t can be approximated based on Shannon
The remaining parts of this paper are organized as follows: capacity model as follows [12], [13], [14]:
section II presents the system model and problem formulation.  pu hu (t)
In this section, the problem of RBs allocation to eMBB users is r̂u (t) = fb xu,b (t) log2 (1 + β ) (1)
N0 F
formulated based on the 2D-HNN. Furthermore, the resource b∈B

scheduling of URLLC and eMBB problem is formulated as where xu,b is the RB allocation result, with xu,b = 1 means
an optimization problem with chance constraint and then that RB b is allocated to user u and xu,b = 0 means the
transformed into a deterministic form. Section III introduces opposite case, fb is the bandwidth of RB b, F is the total
the proposed solution approach. The performance evaluation bandwidth, pu is the transmission power of user u, hu is the
is introduced in section IV. Finally, section V concludes the channel gain of user u, N0 denotes the noise power, and β is
paper. a constant related to a targeted bit-error rate (BER) and given
by:
II. S YSTEM M ODEL AND P ROBLEM F ORMULATION β = 1.5/(− ln(5BER)) (2)
The time domain is divided into equally spaced time slots
The average data rate of user u up to time t can be defined
with one millisecond time duration as that in the current
as follows:
cellular networks. The RBs are allocated to eMBB users at
the slot boundary based on their channel states [7]. However, R̄u (t) = η R̄u (t − 1) + (1 − η)r̂u (t) (3)
the stochastic URLLC traffic may arrive during the time slot of
where η ∈ [0, 1].
which the RBs are already allocated to different eMBB users.
Therefore, the optimization problem can be formulated as
We cannot delay the URLLC traffic into the next time slot due
follows:
to its hard latency constraint. Instead, each time slot is further 
divided into minislots 2 and the arriving URLLC traffic is maximize r̂u (t)/[R̄(t)]α (4a)
x
immediately placed to transmit in the next minislot as shown in u∈U

Fig. 1. Transmitting the incoming URLLC traffic immediately subject to xu,b ≤ 1, ∀u ∈ U (4b)
in minislots overlaps onto previously allocated eMBB trans- b∈B
missions. The gNB chooses either zero transmission power for xu,b ∈ {0, 1}, ∀u ∈ U , b ∈ B (4c)
eMBB, which is referred as Puncturing, or non-zero power for
both eMBB and URLLC, which is referred as superposition The constraint (4b) is to ensure that each RB is allocated
[1], [8], [9]. In this paper, we first allocate the RBs to eMBB to one user only at the same time. The objective is to find
users at the boundary of each time slot. Then, we allocate the allocation matrix x that maximizes the total data rate of
resources to the incoming URLLC traffic (i.e., which already eMBB users while ensuring a fair allocation. Each element of
allocated to eMBB users) based on an optimization problem the allocation matrix x is defined as follows:

that aims to maximize the total data rate of all eMBB users 1, if b ∈ Bu
xu,b = (5)
while satisfying the URLLC reliability constraint. 0, Otherwise
A. Resource Blocks Allocation to eMBB Users where Bu is the set of all RBs allocated to eMBB user u.
The objective here is to allocate the RBs to eMBB users B. Scheduling of URLLC and eMBB Transmissions
such that the total data rate of all eMBB users is maximized The loss of eMBB user data rate at each time slot is directly
while ensuring a certain level of fairness between the users. proportional to the punctured resources of that user [5]. Let
The RBs are allocated to eMBB users at the slot boundary γu (t) define the punctured resources of eMBB user u by
based on their channel states and the average previous data rate URLLC traffic at time slot t. Therefore, the impacted data
of each user up to time slot t in order to consider the eMBB rate of an eMBB user u with punctured resources γu at time
users with bad channel conditions and the punctured eMBB slot t can be approximated as follows:
users by URLLC traffic in the previous time slots. Therefore,   
we use channel aware based Generalized Proportional Fair pu hu (t) 
ru (t) = fb xub (t) − γu (t) log2 1 + β (6)
(GPF) scheduling that considers the multi-user diversity [10], N0 F
b∈B
[11]. GPF based formulation gives different levels of trade-off Let ϕu (t) define the fraction of all resources
between total data rate and fairness by using different values of  allocated to
eMBB user u at time slot t, where ϕu (t) = b∈B fb xub (t).
the GPF parameter α. The data rate and user fairness trade-off Therefore, equation (6) can be written as follows:
2 The formal term for a ‘slot’ in 3GPP is eMBB TTI, and a ‘minislot’ is a    pu hu (t) 
URLLC TTI, where TTI expands to Transmit Time Interval. ru (t) = ϕu (t) − γu (t) log2 1 + β (7)
N0 F

109
 Arriving The URLLC Traffic URLLC Transmissions
Frequency

Frequency
1 2 3 4 5 6 1 2 3 4 5 6
eMBB User 1
eMBB User 2
eMBB User 3
eMBB User 4
URLLC Traffic
.. ..

Time Time
One eMBB Time Slot Mini Slot One eMBB Time Slot Mini Slot

(a) (b)

Fig. 1: Multiplexing of URLLC/eMBB traffics in 5G NR

Let the random variable D(t) denotes the incoming URLLC solve specific problems. According to its structure, the ANNs
traffic at time slot t. Therefore, the outage probability of are classified into Feed-forward Networks and Feedback Net-
URLLC can be given as follows: works or RNNs. Both types have to be configured, one way
   by training the neural network and letting its weights change
purllc hurllc 
P (E) = P γu (t) log2 1+β < D(t) (8) according to learning rule. The other way is to set the weights
N0 F
u∈U explicitly by using a prior knowledge. HNNs, which are a type
The URLLC/eMBB scheduler aims to maximize the total of RNNs, belong to the non-training model.
data rate of eMBB users while satisfying the URLLC con- In HNNs, the output of each neuron is either ’1’ or ’0’
straints. This can be formulated as the follows: depending on the neuron input (i.e., smaller or larger than
 its threshold). Every pair of neurons, neuron ij and neuron
maximize ru (t) (9a) kl, are connected with the weight wijkl . In HNNs, The self
γ
u∈U connections of neurons are set to zero (i.e., wijij = 0) and
subject to P (E) ≤  (9b) the connections between any two neurons are symmetric (i.e.,
γu ≤ ϕu , ∀u ∈ U (9c) wijkl = wklij ). The updating rule of neuron ij is [15]:
  
Where  denotes the maximum allowed outage probability 1, if l wijkl vij (t) ≥ θij
vij (t + 1) = k
(10)
(i.e. reliability level of URLLC traffic). Constraint (9b) rep- 0, Otherwise
resents the reliability of URLLC traffic and constraint (9c)
ensures that the allocated resources to URLLC traffic is less where vij (t) is the state of neuron (i, j), wijkl is the connec-
than the total available resources. We need first to relax the tion weight between neuron (i, j) and neuron (k, l), and θij is
chance constraint (9b) in order to simplify the optimization the threshold of neuron (i, j). HNNs have a value associated
problem and achieve a closed form. with each state of the network called the energy of the network
E(v):
III. P ROPOSED A PPROACH −1     
In this section, the 2D-HNNs is described and then modified E(v) = wijkl vij vkl + θij vij (11)
2 i j i j
k l
to solve the problem of RBs allocation to eMBB users. Then,
we relax the chance constraint of optimization problem 9 in The value of the energy function decreases when the neurons
order to achieve a closed form for it. are updated randomly based on the updating rule and con-
verges to stable state which is the local minimum of the energy
A. Modified 2-Dimensions Hopfield Neural Networks for function [16][17]. In this paper, this minimization property is
eMBB Users Scheduler investigated by defining and expressing the objective function
Artificial Neural Networks (ANNs) are a promising ap- in terms of neuron states xi . Then, the connection weights wij
proach to solve optimization problems because it can save and thresholds θi can be calculated by comparing the energy
computational time due to the parallel processing and give an function with the formulated objective function.
on-line solution. The ANNs consist of interconnected process- We model the problem of RBs allocation as a 2D HNN.
ing elements called neurons. These neurons work together to Here, we consider that there are B neurons for each eMBB

110
user (i.e., number of neurons of each user equals to the total problem (9) into a deterministic form. Therefore, the reliability
number of RBs). Therefore, we can express the state of RBs constraint of URLLC traffic can be written as follows:
allocation (X) of each each user by the firing pattern of the    purllc hurllc 
neural network (i.e., Firing neuron (u, b) means that the RB P (E) = P D(t) > γu (t) log2 1 + β ≤
N0 F
b is assigned to the user u and thus xub = 1). Fig. 2 (a)  u∈U
 
shows the neural network firing pattern and Fig. 2 (b) shows purllc hurllc 
= 1 − FD γu (t) log2 1 + β ≤
the corresponding RBs allocation. u∈U
N0 F
The optimization problem (4) can be modified and written   purllc hurllc  −1
in a minimization form as follows: = γu (t) log2 1 + β ≥ FD (1 − )
N0 F
u∈U

minimize − r̂u (t)/[R̄(t)]α (12a) (19)
x

u∈U Where FD (x) is the CDF of D. Therefore, the optimization
subject to xu,b ≤ 1, ∀u ∈ U (12b) problem (9) can be written as follows:
b∈B 
maximize ru (t) (20a)
xu,b ∈ {0, 1}, ∀u ∈ U , b ∈ B (12c) γ
u∈U
  purllc hurllc 
Then the objective function can be written in the same form subject to γu (t) log2 1 + −1
≤ FD (1 − )
of HNN energy function as follows: N0 F
u∈U
 (20b)
f (x) = − ¯ α
r̂u (t)/[R(t)] (13) γu ≤ ϕu , ∀u ∈ U (20c)
u∈U
  fb xu,b pu hu (t) We consider that the incoming URLLC traffic D follows the
=− ¯ log2 (1 + β ) Pareto Distribution. The CDF of Pareto distribution is given
[R(t)] α N0 w u
u∈U b∈B
    fb xu,b δubij by:
pu hu (t)
= xm β
¯ α log2 (1 + β N0 wu )
[R(t)] F (x) = P (X ≤ x) = 1 − ( ) , f orx ≥ xm (21)
u∈U b∈B i∈U j∈B x
Where xm is the minimum value of X, and β is a positive
where δubij is defined as follows:
number. Accordingly, we achieve the following optimization

1, if u = i and b = j problem:
δubij = (14) 
0, Otherwise maximize ru (t) (22a)
γ
u∈U
The connection weights and thresholds can be calculated by   β
purllc hurllc  xβ
comparing f (x) in equation (13) with the energy function of subject to γu (t) log2 1 + ≥ m
N0 F 
2D HNN E(x) in equation (11) as follows: u∈U
(22b)
fb δubij pu hu (t) γu ≤ ϕu , ∀u ∈ U (22c)
wubij = ¯ α log2 (1 + β ) (15)
[R(t)] N0 w u
Setting β = 1 gives a convex structure of problem (22) which
θub = 0, ∀ u ∈ U and b ∈ B (16) is easy to solve for optimal γ∗ using any convex solver.
IV. P ERFORMANCE E VALUATION
Each RB can be allocated to one user only at the same
time, to satisfy that we modify the updating rule of 2D-HNN The performance of the proposed approach is evaluated
in equation (2) to become as follows: in this section in terms of achieved data rate and fairness.
 Our system consists of 10 eMBB users with different channel
1, when yub (t+1)=max[y1b (t+1),...,yU b (t+1)] states. We consider that 100 RBs are available at each time
xub (t + 1) = slot and the bandwidth of RB is 180 KHz. We evaluate the
0, Otherwise
performance of the proposed approach (Algorithm 1) with
(17)
different values of the GPF parameter α and URLLC reliability
where
 level . We calculate the long-term data rate of all eMBB users
yub (t + 1) = wubkl xkl (t) − θub (18) and the fairness among them with different values of α an and
k∈B l∈U . For the fairness, Jains fairness index [18] is used which is
B. Solution Approach for the Scheduling Problem of URLLC given by:

m
and eMBB Based on Relaxing The Chance Constraint ( Ti ) 2
We use the Cumulative Distribution Function (CDF) of f (T1 , ..., Tm ) = i=1

m (23)
the random variable D to transform the chance constraint of m Ti2
i=1

111
 Fig. 2: The relation between firing pattern of neural network and RBs allocation

Algorithm 1: LTE-U users Scheduler

Input: U, SBS parameters, α.
1.0
Output: The final state of all neurons v.
1 Start:
0.8
2 Calculate ru using equation (1)
l

3 Generate initial state X for all neurons randomly

0.6
wb ξubij pu |hu |2
ECDF

4 Wubij = ¯ α log2 (1 + β N0 wu )
[R(t)]
5 θub = 0 0.4
−1
 
6 E(x) = 2 wubij xub xij + θub xub
u b i j u b 0.2
7 repeat ǫ = 5 × 10−5
ǫ = 4 × 10−5
8 for b ∈ B do 0.0 ǫ = 3 × 10−5
9 for u ∈ U do
  15 16 17 18 19 20 21 22
10 yu,b = wubkl vkl (t) − θub Total Data Rate of eMBB Users (Mbps)
k∈B l∈U
11 end Fig. 3: ECDF of the total data rate of all eMBB users (Mbps) with different
values of  at α = 1
12 Update xu,b based on equation (17)
13 end   decreases the sum data rate of eMBB users.
14 E(x) = −1
2 wubij xub xij + θub xub
u b i j u b Fig. 4 shows the fairness of eMBB users with different
15 until (E(x) converges or maximum nubmer of values of α. Increasing the value of α leads to higher fairness
iterations
 is reached) among the eMBB users since the RBs allocation algorithm
16 ϕu = fb xub aims to maximizes the total data rate at each time slot while
b∈B considering the average data rate of each user over the time.
17 Solve equation (22) to get γ ∗
Therefore, the RBs allocation algorithm gives more resources
18 Calculate r(t + 1) and R̄(t + 1)
to the users with bad channel conditions and whose punctured
19 Wait for time t
by URLLC traffic in the previous time slots. However, setting
20 Goto Start.
the value of α to small values decreases the fairness. In this
case, the RBs allocation algorithm aims to maximize the total
data rate at each time slot without considering the average data
The values of Jains fairness index lies in the interval [0, 1] rate of each user over the time and then the users with good
(i.e, a system with a fairness of 0.9 means that it is 90% fare). channel conditions get more resources. As shown in Fig. 4,
Fig. 3 shows the Empirical Cumulative Distribution Func- the fairness among eMBB users resides between 85% and 95%
tion (ECDF) of total data rate of all eMBB users with different and the median is almost 92% in case of α = 1. However, the
values of reliability metrics at α = 1. for higher upon median of the fairness values decreases to 84% and its values
URLLC traffic, less resources are punctured by URLLC traffic belongs between 74% and 9% when setting α = 0.2.
scheduled, thus increasing the sum rate. However, decreasing Finally, Fig. 5 shows the ECDF of total data rate of all
the values of gives higher reliability to URLLC traffic, eMBB users with different values of α, at = 5 × 10−5 . As
thus more resources are punctured by URLLC traffic and this shown in this figure, decreasing the value of α leads to higher

112
 directly affects the data rate of eMBB users. Moreover, the
1.00 proposed algorithm gives a fair allocation of the RBs to eMBB
users. As a future extension, we intend to consider both uplink
0.95
and downlink scenario while scheduling for URLLC in 5G
0.90 NR.
R EFERENCES
Jains Fairness

0.85
[1] 3GPP TSG RAN WG1 88, Tech. Rep., February 2017.
0.80 [2] A. Mukherjee, “Energy Efficiency and Delay in 5G Ultra-Reliable Low-
Latency Communications System Architectures,” IEEE Network, vol. 32,
0.75 no. 2, pp. 55–61, 2018.
[3] K. Pedersen, G. Pocovi, J. Steiner, and A. Maeder, “Agile 5G Scheduler
0.70 for Improved E2E Performance and Flexibility for Different Network
Implementations,” IEEE Communications Magazine, vol. 56, no. 3, pp.
210–217, 2018.
0.65
α=1 α = 0.5 α = 0.2 [4] P. Popovski, J. J. Nielsen, C. Stefanovic, E. de Carvalho, E. Strom,
K. F. Trillingsgaard, A.-S. Bana, D. M. Kim, R. Kotaba, J. Park,
and R. B. Sorensen, “Wireless access for ultra-reliable low-latency
Fig. 4: Fairness among eMBB users with different values of α
communication: Principles and building blocks,” IEEE Network, vol. 32,
no. 2, pp. 16–23, mar 2018.
[5] A. Anand, G. de Veciana, and S. Shakkottai, “Joint Scheduling of
1.0 URLLC and eMBB Traffic in 5G Wireless Networks,” pp. 1–12, 2017.
[Online]. Available: http://arxiv.org/abs/1712.05344
[6] K. I. Pedersen, G. Pocovi, J. Steiner, and S. R. Khosravirad, “Punctured
0.8
scheduling for critical low latency data on a shared channel with mobile
broadband,” in 2017 IEEE 86th Vehicular Technology Conference (VTC-
0.6 Fall). IEEE, sep 2017.
[7] A. K. Bairagi, S. F. Abedin, N. H. Tran, D. Niyato, and C. S. Hong,
ECDF

“QoE-enabled unlicensed spectrum sharing in 5g: A game-theoretic

0.4 approach,” IEEE Access, vol. 6, pp. 50 538–50 554, 2018.
[8] M. Alsenwi and C. S. Hong, “Resource scheduling of urllc/embb traffics
in 5g new radio,” in Korean Computer Congress (KCC), Jeju, South
0.2
Korea, 2018.
α=1
α = 0.5 [9] A. K. Bairagi, M. S. Munir, S. F. Abedin, and C. S. Hong, “Coexistence
0.0 α = 0.2 of embb and urllc in 5g wireless networks,” in Korean Computer
Congress (KCC), Jeju, South Korea, 2018.
18 19 20 21 22 23 24
Total Data Rate of eMBB Users (Mbps) [10] I. S. Comsa, M. Aydin, S. Zhang, P. Kuonen, J.-F. Wagen, and Y. Lu,
“Scheduling policies based on dynamic throughput and fairness tradeoff
Fig. 5: ECDF of the total data rate of all eMBB users (Mbps) with different control in LTE-a networks,” in 39th Annual IEEE Conference on Local
values of α, at  = 5 × 10−5 Computer Networks. IEEE, sep 2014.
[11] Y. K. Tun, C. W. Zaw, and C. S. Hong, “DownlinK power allocation
total data rate and vice versa. The importance of the average in virtualized wireless networks,” in 2017 19th Asia-Pacific Network
Operations and Management Symposium (APNOMS). IEEE, sep 2017.
data rate of each user over the time increases when increasing [12] A. K. Bairagi, N. H. Tran, W. Saad, and C. S. Hong, “Bargaining game
the values of α and this gives higher probability to the users for effective coexistence between lte-u and wi-fi systems,” in IEEE/IFIP
with bad channel conditions and the users whose punctured by Network Operations and Management Symposium (NOMS 2018), 2018.
[13] A. K. Bairagi, N. H. Tran, and C. S. Hong, “A multi-game approach
URLLC traffic in the previous time slots to get more resources for effective co-existence in unlicensed spectrum between LTE-u system
causing decreasing in the total data rate. Consequently, the and wi-fi access point,” in 2018 International Conference on Information
RBs allocation algorithm with low values of α allocates more Networking (ICOIN). IEEE, jan 2018.
[14] M. Alsenwi, A. K. Bairagi, and C. S. Hong, “Qoe based coexistence
resources to users with good channel conditions and this leads mechanism for lte in unlicensed bands,” in Korean Software Congress
to increase the total data rate of all users. (KSC), Jeju, South Korea, 2017.
[15] M. Alsenwi and C. S. Hong, “A decentralized scheme for load balancing
V. C ONCLUSION in ieee 802.11 wireless networks,” in Korean Computer Congress (KCC),
Jeju, South Korea, 2017.
In this paper, we have considered that the URLLC traffic [16] H. B. Demuth, M. H. Beale, O. De Jess, and M. T. Hagan, Neural
is dynamically multiplexed through puncturing the eMBB Network Design, 2nd ed. USA: Martin Hagan, 2014.
resources. The problem of resource scheduling of URLLC [17] M. Alsenwi, H. Elsayed, and M. Elghoneimy, “Optimization of channel
selection in cognitive heterogeneous wireless networks,” in 2015 11th
and eMBB traffics is formulated as an optimization problem International Computer Engineering Conference (ICENCO). IEEE, dec
with chance constraint where the objective is to maximize 2015.
the sum rate of eMBB users while satisfying the URLLC [18] R. Jain, D.-M. Chiu, and W. R. Hawe, A quantitative measure of fairness
and discrimination for resource allocation in shared computer system.
constraints. The Cumulative Distribution Function (CDF) is Eastern Research Laboratory, Digital Equipment Corporation Hudson,
used in order to transform the problem into an optimization MA, 1984, vol. 38.
problem with a deterministic constraint. Furthermore, the
problem of allocating the resources to eMBB users is modeled
as a 2D-HNN. Then, the minimization property of energy
function of 2D-HNN is investigated to solve these problem.
The results shown that the reliability level of URLLC traffic

113