This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
1

On-Device Cognitive Spectrum Allocation for

Coexisting URLLC and eMBB Users in 5G Systems
Shu-Feng Cheng∗ , Li-Chun Wang∗ , F ellow, IEEE , Chien-Hwa Hwang† , Ju-Ya Chen† and Li-Yu Cheng∗

Abstract— Ultra Reliable and Low Latency Communications multiplexing URLLC, eMBB and mMTC packets in a sharing
(URLLC) play a key role in 5G vertical markets, but pose many spectrum environment is a good strategy from the viewpoint of
technical challenges especially when sharing the spectrum with spectrum efficiency, but needs to overcome the packet collision
Enhanced Mobile Broadband (eMBB) customers. This study aims
to overcome the spectrum inefficiency issue of fully separate issue [8], [9]. Because 5G eMBB services are more popular
(FS) approach and the contention issue of the fully overlap (FO) than mMTC services, we focus on the multiplexing issue of
approach. We present a user-initiated probability elastic resource URLLC and eMBB packets [10]–[12].
(UPER) approach by dynamically adjusting the probability of To support time-critical applications with the latency less
using the shared spectrum for eMBB and URLLC traffic based than 1 msec, URLLC adopts the grant-free transmission pro-
on the current success and failure status of packet transmission
status. The probabilities of successful transmission are derived tocol in the medium access control (MAC) layer [13]–[15].
for UPER, FS, and FO and partially overlap (PO) sharing Grant-free transmission protocol can simplify the handshake
spectrum approaches. We find that the successful transmission process between the device and the base station to the two-
probability of UPER approach is 28% and 46% higher than way handshake. The conventional grant-based requires the
FS and FO approaches, respectively. We further evaluate the four-way handshake process [16]. Therefore, grant-free trans-
reliability and throughput performance of URLLC and eMBB.
When the URLLC packet load is low, the UPER method can mission MAC protocol can control the access delay within
almost achieve the best performance of the FS method. When the one msec with much higher probability than the grant-based
URLLC packet load is high, we show that UPER can improve the transmission protocol. Nevertheless, grant-free transmission
reliability performance up to 54% compared with other methods. may face serious packet collisions. When URLLC share the
same spectrum with eMBB, the packet collision issue is
I. I NTRODUCTION worsen.
Ultra-reliable and low latency communications (URLLC) In the literature, power control [13], [17] and spectrum
of the fifth generation (5G) wireless communications aim management [18]–[23] are the two main technical directions to
for providing time-critical machine-to-machine or human-to- solve the packet collision issue in a spectrum sharing environ-
machine vertical applications, such as factory automation, ment for URLLC and eMBB. However, using different power
vehicular communications, and augmented reality, etc. [1]– levels to assist multiple access may degrade the performance of
[4]. Most URLLC services require 99.999% reliability perfor- the URLLC system because high eMBB signals still interferes
mance within 1 msec latency in the data plane [5], [6]. Remote with URLLC [24]. From the spectrum management aspect
monitoring of patient’s the vital signals is an important human- for multiplexing URLLC and eMBB, we need to resolve
to-machine URLLC applications use case. On the other hand, the optimal dynamic spectrum allocation for different service
intelligent transportation systems is an important machine-to- request rate from URLLC and eMBB.
human URLLC use case [7], where vehicles obtain important Therefore, this study develops a dynamic spectrum al-
information from neighboring vehicles or roadside facilities to location method that can reduce the packet collisions of
enhance the security of self-driving cars. multiplexing URLLC and eMBB packets based on grant-
However, serving URLLC together with enhanced mo- free MAC protocol in order to maintain low latency and
bile broadband (eMBB) and machine type communications high reliability for URLLC customers. This paper makes the
(mMTC) users need to design the advanced spectrum man- following contributions.
agement to overcome the packet collision issue. In general, • Propose a novel user-initial probability elastic resource

the service requests for time-critical URLLC applications (UPER) for multiplexing URLLC and eMBB traffic,
happen much less frequent than the non-emergency eMBB and which can flexibly adjust the probability of using ded-
mMTC applications. Reserving too many exclusive spectrum icated spectrum based on the current packet collision
for URLLC will result in the waste of resources, while reserv- status. It will be shown that UPER, an on-device data-
ing insufficient spectrum will cause packet collision. As such, driven cognitive spectrum allocation, can improve the
reliability performance up to 54% compared with other
∗ The authors are with National Chiao Tung University, Hsinchu, Taiwan.
† The authors are with MediaTek Inc., Hsinchu, Taiwan. methods when the URLLC packet load is high. When the
This work is jointly sponsored by the National Chiao Tung University URLLC packet load is low, the reliability of the proposed
and Ministry of Education (MOE), the Ministry of Science and Technol- UPER can reach 99.45%, which approaches to the FS
ogy (MOST) and Pervasive Artificial Intelligence Research (PAIR) Labs, method and is 6.74% better than the PO method.
and MediaTek-NCTU Reseach Lab under the projects “Higher Education
Sprout”, MOST 108-2634-F-009-006, and MTKC-2013-1414, respectively. • Derive the closed-form expressions for the successful
Corresponding author: Li-Chun Wang (Email: [email protected].) transmission probability in terms of the number of

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
2

• Shared spectrum allocation [21], [22], [25], [26]: The

URLLC users and eMBB users who share the whole spec-
trum can have better spectrum efficiency, while causing
more packets collisions.
• Hybrid spectrum allocation [23]: To overcome the short-
comings of dedicated and shared spectrum allocation
methods, a hybrid approach was proposed to periodically
allocate the dedicated and the shared spectrum. In this
way, the advantages of high reliability of the dedicated
spectrum allocation method and the high spectrum uti-
lization of the sharing method can be preserved.
Fig. 1. Spectrum Sharing Approach - Dedicated, Shared and Hybird.
Nevertheless, the quantitative performance comparison of ded-
icated, shared and hybrid, spectrum allocation approaches
URLLC and eMBB packets based on grant-free MAC for multiplexing URLLC and eMBB systems is rare in the
protocol under four different spectrum allocation meth- literature.
ods, including fully separate (FS), fully overlap (FO),
partially overlap (PO) and UPER. Such analytical models B. Load Control
can help verify the feasibility of UPER and provide quan- In the following, we introduce some related works of the
titative performance comparison with other considered load control mechanism to reduce the congestion of the access
spectrum allocation methods. channel. Extended access barring (EAB) [27] and access class
• A simple but effective load control mechanism is pro- barring (ACB) [28], [29] are two load control mechanisms
posed for the co-existing eMBB and URLLC network, proposed by 3GPP. In the former mechanism, different access
which is shown to reduce packet collisions and respec- classes are assigned to the devices based on their service
tively improve the reliability performance of the FO, PO requirements of the device to relieve the load on the network.
and UPER methods by 52%, 16% and 2% compared to In this way, some non-emergency services will be temporarily
the case without load control. blocked to improve access efficiency. In the latter mechanism,
• The URLLC throughput with the packet length of 50 the base station periodically broadcasts an access probability
bytes is 56.25% higher than that of 32 bytes. That is, factor. Each device selects a random value between [0, 1)
the throughput performance of URLLC is proportional before accessing the radio resources. When the selected value
to the packet length in our considered case. is less than the access probability factor, the device is allowed
The rest of the paper is organized as follows: Related to transmit the data. Otherwise, the device will restart the
works are discussed in Section II. We describe our system uplink packet transmission after a random backoff time. The
model in Section III. Section IV introduces the Dedicated, authors of [30] derived the optimal access probability factor
Shared and Hybrid spectrum allocation approaches considered of the ACB mechanism in the uplink transmission interval to
in this paper. Then, the transition probability analysis of calculate the random access channel capacity within average
network model is shown in Section V. We demonstrate some time period (i.e., the expected number of nodes successfully
numerical results with different traffic loads for comparison in transmitting in an average period of time). Additionally, an an-
Section VII. Finally, we conclude this paper in Section VIII. alytical model was proposed in [31] to determine the expected
delay for dynamically changing the ACB access probability
factor.
II. R ELATED W ORKS
A. Spectrum Allocation Methods III. S YSTEM M ODEL
In the literature, some spectrum allocation methods were A. Assumptions
proposed to reduce the collision of data packets in URLLC We consider a scenario where Nu URLLC devices coexist
and eMBB multiplexing systems [18]–[23]. The spectrum with Ne eMBB devices. The arrival processes of URLLC
allocation methods can be classified into three types as shown and eMBB packets are modeled as Poisson processes with
in Fig. 1. arrival rates λu and λe , respectively. Suppose that fu URLLC
• Dedicated spectrum allocation [18]–[20]: Basically, re- requests will contend with fe eMBB requests for the spectrum
taining some exclusive spectrum for URLLC users can resource blocks. Both URLLC users and eMBB users adopt
improve the URLLC performance. In [18], the dedicated grant-free transmission [13], [32]. To ease the analysis, we
spectrum was allocated for periodic URLLC packets to define a grant-free opportunity (GFO) as the total time of
improve the reliability performance. However, it was single grant free transmission until a URLLC device either
mentioned in [19], [20] that the dedicated spectrum receives an acknowledgment (ACK) from a base station or
allocation is inefficient for URLLC in the URLLC and its timer expires. If no packet collision occurs in the GFO,
eMBB coexistence system because URLLC traffic is the base station will successfully receive the uplink data.
sporadic. Preserving too much spectrum will result in Otherwise, the conflicting data will be retransmitted in the
wasted spectrum. next GFO.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
3

Note that parr is the transition probability when the device

݂௨
URLLC requests Collision status changes from WAIT to BUSY. Similarly, pST is the
URLLC probability of the opposite state transition. The device main-
݂ௗ
݂௘ tains the same state in the WAIT and BUSY states with the
Load control
eMBB mechanism probability 1 − parr and 1 − pST , respectively. By using the
eMBB requests
Markov network model, we will derive the transition proba-
Compete sharing resource Uplink processing bilities of the existing and the proposed spectrum allocation
strategies in Sections V and VI, respectively.
‫ܦ‬௔௖௖௘௦௦ ‫ݐ‬௣௥௢௣ ‫ݐ‬௣௥௢௖

‫ீܦ‬ி
D. Delay Model
Fig. 2. System model of the considered URLLC and eMBB coexistence with The overall communication delay from the transmitter to the
load control mechanism.
receiver is shown in Fig. 2. We define the time interval between
a device generating a packet and receiving an feedback ACK
B. Load control Mechanism as grant-free transmission delay (Φ), which can be written as
To reduce the data packet conflicts between URLLC and Φ = Daccess + 2tprop + tproc , (1)
eMBB users, we exploit a load control mechanism in the
system model as shown in Fig. 2. The role of the load where Daccess is the delay caused by the channel access and
control mechanism is to manage the number of eMBB devices spectrum resource competition before packet transmissions.
accessing the network. We use the ACB mechanism because tprop is the round-trip propagation time between the device
all the eMBB users are regarded as the same category in and the base station, and tproc is the processing time of the
our system model. Before sending the data packet, the eMBB data packet received at the base station.
request randomly selects a value between 0 and 1. When the In grant-free setting, Daccess is only affected by the time to
selected value is less than the predefined access probability, compete for the spectrum resource because users transmit the
the packets will be allowed to transmit. Otherwise, the backlog data directly. Besides, we assume that tprop is negligible [30].
of requests continue to wait and re-execute load control in the tproc is assumed to be 4 transmission time intervals (TTIs) for
next GFO. Therefore, the number of eMBB requests changes data decoding at the base station [32]. The TTI length (LT T I )
from fe to fd after using load control. Finally, fu URLLC is defined as
requests and fd eMBB requests will compete for spectrum LT T I = Nsymbol ∗ Lsymbol , (2)
resources based on the different spectrum sharing methods.
When we regulate the number of eMBB requests competing where Nsymbol is the number of symbols and Lsymbol is the
for spectrum resources, the successful transmission probability corresponding symbol length. The number of symbol per TTI
of URLLC request will increase. Therefore, we combine the can be set to 2, 4 or 8, etc. However, we use a GFO of 0.25
design of load control mechanism (e.g., access class barring msec in our simulation. The corresponding symbol length is
[29]) with the spectrum allocation schemes to mitigate the 16.7 µsec. Therefore, the constant tproc is very short compared
packet collisions in the coexisting system of URLLC and to the length of GFO.
eMBB.
E. Performance Metrics
C. Network Model In our system-level simulations, two performance metrics
The Markov network model of URLLC and eMBB device of URLLC are defined in the following. The reliability is as
is shown in Fig. 3. The two states of Markov chain are defined a function of latency requirement, while the throughput is as
as follows. a function of packet length and latency requirement.
• Reliability:
• WAIT state: The device is waiting for uplink data
generation. Pfu (i)
I(Φu ≤ τu )
• BUSY state: The device is trying to send data packets Ru = i=1 (3)
fu
to the base station and competing for spectrum resources
with other devices. • Throughput:
fu (i)
X σu
Mu = (i)
I(Φ(i)
u ≤ τu ) (4)
i=1 Φu
where indicator function is denoted by I(·) and takes the value
(i)
1 or 0 on the condition that Φu ≤ τu is true or not. τu is
the latency requirement of URLLC. σi is the packet length of
(i)
URLLC request i and Φu is the corresponding transmission
delay. When we replace fu with fd and replace the subscript
Fig. 3. The diagram of two-state Markov network model. u with e, the performance metrics are applicable to eMBB.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
4
Policies
Shared resources
eMBB dedicated
URLLC dedicated
Shared resources
URLLC dedicated
1 െ ܲ௥ ܲ௥
eMBB URLLC
eMBB URLLC eMBB URLLC eMBB URLLC
Fig. 5. Illustration of proposed user-initiated probability elastic reservation
(a) Fully overlap (FO) (b) Fully separate (FS) (c) Partially overlap (UPER) scheme.
scheme. scheme [33]. (PO) scheme [33].

Fig. 4. Illustrations of existing allocation schemes.

the PO scheme, respectively. Therefore, the UPER scheme has
the advantages of these two schemes by utilizing the spectrum
IV. R ESOURCE A LLOCATION A PPROACHES resources flexibly.
Here, we will introduce the policy of the proposed UPER
In URLLC and eMBB multiplexing, the spectrum allocation method. The dynamic update algorithm of the proposed
is important to reduce the collision probability of grant- method is shown in Algorithm 1. The URLLC users can
free URLLC packets. To improve the latency performance of dynamically update Pr based on the status of the network
URLLC, some spectrum resources are reserved for URLLC congestion at each GFO until receiving an ACK from the base
transmission. In the following, we will introduce three ex- station. When a collision occurs and the last resources selected
isting network-centric spectrum allocation schemes, including for uplink transmission is on the dedicated spectrum, the value
dedicated, shared, and hybrid approaches. We further propose of Pr will be decreased to alleviate the congestion problem
a user-centric hybrid scheme that can flexibly use spectrum of URLLC packers in the next transmission. In contrast, the
resources according to the designed strategy. probability of using proprietary spectrum for URLLC packets
will be increased to reduce the contention with eMBB packets.
A. Existing Resource Allocation Schemes [33] To stabilize the algorithm, we define |Pr (t+1)−Pr (t)| ≤  as
As shown in Fig. 4, the network-centric spectrum allocation a condition. Note that  is the difference of P r between time
methods of dedicated, shared and hybrid are defined as fully t and time t + 1. If  is very small, Pr will not be updated.
separate (FS), fully overlap (FO) and partially overlap (PO). In Otherwise, we update the new Pr to adjust the probability of
the FO method, eMBB devices and URLLC devices share the using the dedicated spectrum.
same entire spectrum resources together. In the FS method,
Algorithm 1 Dynamic update of the proposed user-initiated
the entire spectrum resource is divided into two dedicated
probability elastic reservation (UPER) scheme
parts, which are exclusive used for URLLC devices and eMBB
1: initialize: WAIT state with Pr (0) ← 1;
devices, respectively. The difference between the PO and the
2: if Packet arrives then
FS is that one part of the spectrum in the PO method is only
reserved for URLLC devices. Another part of the spectrum 3: Transit from WAIT state to BUSY state
is shared by URLLC devices and eMBB devices at the same 4: repeat
time. 5: Perform grant-free transmission
6: if Receive ACK then
7: Transit from BUSY state to WAIT state return
B. Proposed User-initiated Probability Elastic Reservation 8: else
(UPER) Scheme 9: if Use the dedicated
 resources then

r
However, the network-centric spectrum allocation method 10: Pr (t + 1) ← Pr (t) + S+r /2
may not be feasible for URLLC users with stringent latency. 11: else
Therefore, we proposed a user-initiated probability elastic 12: Pr (t + 1) ← (Pr (t) + 1) /2
reservation (UPER) scheme to further improve the latency 13: end if
performance of URLLC when coexisting with eMBB users. 14: end if
The proposed UPER scheme is a hybrid spectrum allocation 15: Check stabilization
approach as shown in Fig. 5. The entire spectrum resources 16: if |Pr (t + 1) − Pr (t)| ≤  then
(S + r) are divided into two parts. Resources S is shared 17: Pr (t + 1) ← Pr (t)
by eMBB devices and URLLC devices. On the other hand, 18: else
resource r is proprietary to URLLC devices. To implement the 19: Update Pr (t + 1)
user-centric approach, we design a resource decision parameter 20: end if
to determine that URLLC devices will transmit the data on the 21: until latency requirement is reached
shared or the reserved spectrum resources. The decision pa- 22: Transit from BUSY state to WAIT state
rameter Pr is a selection probability of the reserved resources 23: end if
when the URLLC packet performs a grant-free transmission.
In other words, the shared part of the spectrum can be selected In Fig. 6, we illustrate a flow chart of the proposed algorithm
with a probability of 1 − Pr . When Pr is equal to 1 and Sr , between the URLLC user and the base station. The value of
the proposed UPER scheme can become the FS scheme and Pr is initially set to 1 when the URLLC device is idle and no

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
5

URLLC User Equipment Base Station

ͳ െ ܲ௥ ܲ௥
(Probability of using
dedicated resources) UL grant-free
Initial ܲ௥ ሺ‫ݐ‬ሻ ൌ ͳ transmission Receive
Packet arrival Uplink request
Enter
WAIT state No
Yes

Enter
BUSY state Yes (successful)
Decoding successfully ?
Increase ࡼ࢘(t+1)
ܲ௥ ‫ ݐ‬൅ ͳ = ܲ௥ ‫ݐ‬ Update ܲ௥ ‫ ݐ‬൅ ͳ Response message
(Increase the probability of
using dedicated resources)
2 1. Shared resources No (failed or collision)
or
ܲ௥ ‫ ݐ‬൅ ͳ െ ܲ௥ ሺ‫ݐ‬ሻ ൑ ߝ Decrease ࡼ࢘(t+1) 2. Dedicated resources
(Increase the probability of 1
using shared resources)

Fig. 6. Dynamic update flow chart of the proposed UPER scheme.

data is being transmitted. If the packet arrives, the status of the The approximation result motivates the use of a load control
device will change from WAIT state to BUSY state. Then, the mechanism to limit the number of eMBB requests that com-
URLLC device performs grant-free transmission with Pr = 1 pete for the spectrum resources. By substituting (7) for n in
in the next GFO. According to the response message from the (6), we have
base station, the device will re-enter the waiting state after
1 Q−1 Q 1
the successful transmission or adjust the value of Pr after the NST (Q) = Q(1 − ) ' + . (9)
failed transmission. When the packet conflicts or fails, Pr will Q e 2e
be dynamically updated by the aforementioned policy.
B. Probability of Request Arrival
V. P ROBABILISTIC A NALYSIS FOR EXISTING RESOURCE The probability of the request arrival is derived under the
ALLOCATION SCHEMES
condition that at least one request arrives in each GFO. As
In this section, we first calculate the approximate number shown in Fig. 2, the device will transit from WAIT to BUSY
of competing request to explain the motivation for using load state when a request arrives. Hence, the transition probability
control. Then, we derive the transition probabilities of Markov parr is obtained in Lemma 1.
network model as mentioned in Section III.
Lemma 1 (Transition probability from WAIT state to BUSY
state). If the packet arrivals of devices follow Poisson process
A. The approximate number of competing request
with arrival rate λ, the probability mass function (PMF) of
In [30], the typical probability that radio access network Poisson distribution is P (.). The transmission probability from
(RAN) can successfully transmit at each time slot can be WAIT state to BUSY state parr can be written as
written as
1 parr = 1 − e−λ .
pST (n) = (1 − )n−1 , (5) (10)
Q
where n is the number of competing requests and Q is
the number of total spectrum resources for competition. The The PMF of the Poisson random variable k is
average number of requests successfully served by the base
station at each GFO is e−λ λk
P (X = k) = , (11)
1 k!
NST (n) = n(1 − )n−1 . (6)
Q where k is the number of the packet arrivals. The probability
When Q is large, we calculate that of at least one request arrives at a certain time interval can be
1 expressed as
n∗ = lim ∼Q .
= (7)
Q→∞ log( Q ) parr = 1 − P (0) . (12)
Q−1
Hence, we know that the approximate number of the compet- By replacing (12) with (11), we have
ing requests is same as the total number of available spectrum
resources. Based on the above derivation, fd can be written as e−λ λ0
parr = 1 −
0! (13)
fd = min(fe , Q) . (8) = 1 − e−λ .

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
6

C. Probability of Successful Transmission

In the following, we will analyze the probability of suc-
cessful transmission for the three existing spectrum allocation
schemes as shown in Fig. 4. Note that the number of URLLC
requests and the number of eMBB requests are fu and fe
in uplink grant-free transmission, respectively. After we adopt
the load control mechanism, the number of eMBB requests
changes from fe to fd .
• Fully overlap (FO) Scheme
In FO scheme, both URLLC UEs and eMBB UEs share
all spectrum resources (S + r) to perform grant-free
transmission. The probability that URLLC and eMBB
requests can successfully perform radio access can be
expressed as
(u) (e) 1 fu +fd −1
pST = pST = (1 − ) , (14) Fig. 7. Probability of successful transmission for URLLC UE with various
S +r number of URLLC in FO scheme, FS scheme, PO scheme, and the proposed
(u) (e) UPER scheme in the case of r = 10, S = 40 and Pr = 0.7.
where pST and pST are the probability of successful
transmission for the URLLC UE user and the eMBB user,
respectively. In this case, URLLC requests and the eMBB
requests have the same priority. Thus, the probability of
successful transmissions for two services have the same
formula.
• Fully Separate (FS) Scheme
In FS scheme, the whole spectrum resources are divided
into two parts to different service priorities. One part of
the spectrum resource defined as S is for eMBB users.
The other part of the spectrum resources defined as r
is for URLLC users. That is, the URLLC users and
the eMBB users each have their exclusive spectrum to
transmit their packets and do not affect each other. The
successful transmission probability of URLLC devices
can be expressed as
(u) 1
pST = (1 − )fu −1 . (15) Fig. 8. Probability of successful transmission for URLLC UE with various
r number of eMBB in FO scheme, FS scheme, PO scheme, and the proposed
Similarly, the successful transmission probability of UPER scheme in the case of r = 10, S = 40 and Pr = 0.7.
eMBB devices can be expressed as
(e)1 fd −1 The successful transmission probability of eMBB devices
pST = (1 −
) . (16)
S can be expressed as
Obviously, the probability of successful transmission
(e) 1 fd −1 1 fu
for URLLC traffic is only affected by the number of pST = (1 −) (1 − ) . (18)
URLLC requests in BUSY state. When the amount of S S +r
the spectrum resources reserved for URLLC increases, Note that in the FO scheme, because the number of
the successful transmission probability of the URLLC eMBB uplink requests is greater than the number of
request increases. URLLC uplink requests, the probability of successful
• Partially Overlap (PO) Scheme transmission for the URLLC UE is mainly affected by
The PO scheme is a hybrid spectrum resource allocation the number of eMBB requests. Therefore, some spectrum
approach that combines the benefits of the FO and the FS resources are provided exclusively for URLLC requests to
schemes. URLLC users can use the entire radio spectrum avoid packet collisions with eMBB requests. In this way,
resources S + r. However, the eMBB UEs exploit only only certain URLLC requests will conflict with eMBB
a portion of the total resources defined as S to transmit requests, thereby improving the reliability of URLLC.
the packets. The successful transmission probability of
URLLC devices can be expressed as VI. P ROBABILITY ANALYSIS OF PROPOSED UPER
S CHEME
(u) 1 fu −1 S 1 fd r
pST = (1 − ) [ (1 − ) + ] . To provide timely URLLC services, we propose a technique
S +r S +r S S +r
(17) for dynamically adapting the probability on the reserved

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
7

spectrum resources of URLLC users. In the UPER scheme,

spectrum resource allocation is same as the PO scheme. We
design a resource selection parameter Pr in the proposed
method to avoid the severe packet collisions. The parameter
Pr will dynamically change the probability of using the
dedicated spectrum for URLLC user. To support the sporadic
traffic of URLLC, the user-centric Pr can be updated based
on the proposed algorithm mentioned in Section IV. The
probability of successful transmission for URLLC devices can
be expressed as
(u) 1 fd 1 − Pr fu −1
pST =(1 − Pr ){(1 − ) (1 − ) }
S S (19)
Pr fu −1
+ Pr (1 −
) .
r
Similarly, the probability of successful transmission for eMBB (a)
devices can be expressed as
(e) 1 fd −1 1 − Pr fu
pST = (1 −
) (1 − ) . (20)
S S
Proof. To analyze the probability of successful transmission,
we first need to obtain the non-collision probability of both
URLLC and eMBB packets. Under the condition that a given
URLLC user uses the shared spectrum, the probability that at
least one user among other URLLC users does not select the
same resource is represented as
1 − Pr
p(u)
ns = 1 −. (21)
S
Given that a certain URLLC user uses the dedicated spectrum,
the probability that at least one of other URLLC UEs does not
select the same resource is calculated as (b)
Pr
p(u)
nr = 1 − . (22) Fig. 9. The impact of Pr on the probability of successful transmission for
r URLLC UE with various number of URLLC in FO scheme, FS scheme, PO
scheme, and the proposed UPER scheme in the case of r = 10, S = 40: (a)
On the other hand, eMBB users share the shared spectrum Pr = 0.5; (b) Pr = 0.9.
with the URLLC users. Therefore, only under the condition
that the given URLLC UEs use the shared spectrum resources,
we calculate the probability that at least one of eMBB users We use (21) and (23) instead of (25) to obtain
does not select the specific shared resource is expressed as (e) 1 fd −1 1 − Pr fu
pST = (1 −
) (1 − ) . (26)
1 S S
p(e)
ns = 1 − . (23)
S The packet collision probabilities of URLLC and eMBB can
The probability of successful transmission for the URLLC user be respectively written as
(u)
(pST ) is equal to the sum of the conditional probabilities for (u)
p(u)
c = 1 − pST , (27)
a specific URLLC user transmitting the packet on the shared
(u)
and dedicated spectrum resource. Hence, we can obtain pST and
(e)
as follows. p(e)
c = 1 − pST . (28)
(u) fd (u) fu −1 fu −1
pST = (1 − Pr )((p(e)
ns ) (pns ) ) + Pr (p(u)
nr ) . (24)
By substituting (21), (22) and (23) with (24), we have
(u) 1 fd 1 − Pr fu −1 VII. N UMERICAL R ESULTS
pST =(1 − Pr )((1 − ) (1 − ) )
S S Here, we first show the theoretical results in Subsec-
Pr fu −1 tion VII-A. Then, we discuss the system-level simulations in
+ Pr (1 − ) .
r Subsection VII-B to VII-D to evaluate a variety of spectrum
Besides, the probability of successful transmission for the allocation methods. In our simulations, we show the average
(e) reliability and throughput performance of URLLC users and
eMBB user pST can be written as
eMBB users for various spectrum allocation methods. We
(e) fd −1 (u) fu
pST = (p(e)
ns ) (pns ) . (25) compare the performance impact of using and not using load

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
8

control. We further show that when the URLLC packet load is TABLE I
high, the different reserved widths of the dedicated spectrum S IMULATION PARAMETERS
can cause changes in reliability performance.
Parameter Value
Nu 10
A. Analytical Results Ne 10
In the following, we show the analytical results derived in λe 8 (packets/msec)
Section V and VI. For ease of analysis, we assume that the λu {0.5, 1, 3} (packets/msec)
size of the spectrum resources is the same. We divide the
σu 32 bytes
entire spectrum into 50 segments in the frequency domain. The
σe 160 bytes
amount of the reserved spectrum and the shared are r = 10 and
S = 40, respectively. Fig. 7 shows the successful transmission GFO 0.25 msec
(u)
probability of URLLC (pST ) for different number of URLLC *
GF O: time unit of grant-free transmission time.
requests (fu ) when Pr = 0.7 and fd = 30. Compared with
the PO scheme, serious data packet conflicts will occur in the
TABLE II
FO scheme because URLLC users and eMBB users share the I MPROVEMENT RATIO OF URLLC RELIABILITY AFTER USING LOAD
entire spectrum resources at the same time. When the number CONTROL IN THE CASE OF λe = 8 AND λu = 1
of URLLC packets competing for spectrum resources is small,
the probability of successful transmission of the FS scheme is Spectrum Partition Methods Improvement Ratio
higher than that of the PO scheme. This is because URLLC UPER 2.39 %
user only sends data packets in dedicated resources, and the Fully overlap (FO) 51.75 %
packets will not conflict with eMBB. On the contrary, when Partially overlap (PO) 16.44 %
the number of URLLC data packets increases, the dedicated Fully separate (FS) 0%
spectrum resources are insufficient to accommodate URLLC
requests, resulting in serious collisions. However, the above
problem can be solved by the proposed UPER. The UPER and 4 msec latency requirements in system-level simulations,
(u)
method has the better pST than 27.89% of the FS method, respectively. The system-level simulation parameters are listed
especially when the URLLC traffic load is high. On the other in TABLE I. The numbers of URLLC users and eMBB users
hand, the successful transmission probability of the proposed are Nu = 10 and Ne = 10, respectively. The packet arrival
scheme is 27.89% higher than that of the FS scheme when rates of URLLC users (λu ) and eMBB users (λe ) are set to
(u)
URLLC traffic load is low. If Pr < 0.7, the curve of pST different values because URLLC packets arrive less frequently
is closer to the PO method as shown in Fig 9(a). URLLC that eMBB packets. We consider that the corresponding packet
users tend to send packets on the shared spectrum resources. sizes of URLLC and eMBB are 32 and 160 bytes, respectively.
(u)
If Pr > 0.7, the curve of pST of the UPER is approximate The slot length of GFO is defined in Section III. We assume
to the FS method as shown in Fig 9(b), thereby reducing the that all the users transmit their packets to the base station by
collisions with the packets requested by eMBB. grant-free transmission. When no packet collision occurs in
(u)
In Fig. 8, we show the change of pST under different the GFO, the base station will successfully receive the uplink
number of eMBB requests (fd ) when Pr = 0.7 and fu = 5. data. Otherwise, the data will be retransmitted in the next
(u)
When fd increases, the pST in the FS method does not GFO. For URLLC with the 1 msec latency requirement, the
change. The reason is that the URLLC users have their own packets can be transmitted up to four times, including the first
the dedicated spectrum without being affected by the number transmission and three retransmission. Similarly, the packets
(u) can be transmitted up to 16 times for eMBB with the 4 msec
of eMBB requests. In the FO, PO and UPER scheme, pST
gradually decreases as fd increases because URLLC users delay requirement.
share all or part of the spectrum resources with eMBB users. Fig. 10 shows the average reliability and average throughput
For the PO scheme, the URLLC user will not share the entire of URLLC and eMBB, respectively. We consider that the time-
spectrum with the eMBB user. The URLLC UEs will not critical URLLC applications happen much less frequently than
share the entire spectrum in the PO scheme with eMBB UEs. eMBB in the case of λe = 8 and λu = 0.5. The average
(u) reliability of URLLC within latency requirement of 1 msec is
Thus, the pST of the PO scheme is higher than that of the
FO scheme in the case of high eMBB traffic load (e.g., fe is shown in Fig. 10(a). We observe that the FS method is superior
greater than 30). Compared to the PO and the FO methods, to other methods. The reason is that URLLC users send data
the proposed UPER has the designed the resource selection packets on the dedicated spectrum without conflicting with
parameter, which can effectively reduce the impact of fd on eMBB packets. When the URLLC traffic load is highly low,
the successful transmission probability of URLLC user. the FS method has the better performance than other spectrum
allocation methods. The reliability of the FO method is better
than that of the PO method because URLLC users compete
B. System Performance of URLLC and eMBB with eMBB for resources in the entire spectrum rather than
Next, we will consider the average reliability and throughput in the shared spectrum. Thus, the probability that URLLC
performance of URLLC users and eMBB users within 1 msec packets collide with eMBB packets is small. Nevertheless, the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
9

(a) (b)

(c) (d)

Fig. 10. System performance comparisons between URLLC and eMBB with load control in the case of λe = 8 and λu = 0.5: (a) URLLC reliability with
latency requirement of 1 msec; (b) eMBB reliability with latency requirement of 4 msec; (c) URLLC throughput with latency requirement of 1 msec; (d)
eMBB throughput with latency requirement of 4 msec.

TABLE III to transmit data packets and are not prone to collide with
T HE RELIABILITY IMPROVEMENT RATE OF THE DEDICATED other URLLC and eMBB users. In comparison, the average
SPECTRUM INCREMENT IS LISTED IN THE CASE OF λe = 8 AND λu = 3
throughput of eMBB in our method is only 3.37% lower than
Spectrum Partition Methods Improvement Ratio the FO method and 21.76% higher than the PO with the worst
Fully separate (FS) 118.39 % reliability. The corresponding average throughput of eMBB is
shown in Fig. 10(d). We observe that although the average
UPER 49.39 %
throughput of the proposed method is only higher than that of
Partially overlap (PO) 31.15 %
the PO method, it is similar to the FO and the FS methods.
Fully overlap (FO) 1.09 % Fig. 10 shows the average reliability and average throughput
of URLLC and eMBB, respectively. We consider that the time-
critical URLLC applications happen much less frequently than
reliability of the proposed UPER is 99.45%, which is slightly eMBB in the case of λe = 8 and λu = 0.5. On the other hand,
worse than the FS method, and is 6.74% higher than the PO we show the reliability and throughput performance when
method with the worst reliability. The corresponding average eMBB and URLLC packets arrive at a rate of λe = 8 and
throughput of URLLC is shown in Fig. 10(c). Due to the λu = 1, respectively . In Fig. 11, the trend of the curve is
low URLLC traffic load, the average throughput of different similar to the trend in Fig. 10. When the packet arrival rate of
spectrum allocation methods is not significantly different. URLLC increases, the reliability and throughput performance
The average reliability of eMBB within latency requirement of URLLC and eMBB decreases in all methods. In terms of
of 4 msec is shown in Fig. 10(b). The FO is superior to the URLLC reliability, the performance of the proposed UPER
other methods because eMBB users use the entire spectrum is 98.73%, which is almost the same as the FS method, and

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
10

(a) (b)

(c) (d)

Fig. 11. System performance comparisons between URLLC and eMBB with load control in the case of λe = 8 and λu = 1: (a) URLLC reliability with
latency requirement of 1 msec; (b) eMBB reliability with latency requirement of 4 msec; (c) URLLC throughput with latency requirement of 1 msec; (d)
eMBB throughput with latency requirement of 4 msec.

superior to other methods. At this time, the eMBB reliability

of the UPER is 6.57% lower than the PO method and 79.36%
higher than the PO method. The performance degradation of
PO method is greater than that in the Fig. 10. The reason is
that the probability of collision between eMBB packets sent
in the dedicated spectrum and URLLC packets is higher than
the probability of collision between eMBB packets sent in the
entire spectrum and URLLC packets.

C. Comparison of URLLC Reliability with and without Load

Control Mechanism
Fig. 12 shows the URLLC reliability performance when
eMBB has no load control mechanism under the conditions
of λe = 8 and λu = 1. Compared with the performance
of Fig. 11(a), the reliability performance of URLLC under
uncontrolled eMBB traffic load is worse than that under
Fig. 12. URLLC reliability performance without eMBB load control mech- controlled eMBB traffic load. TABLE II lists the improve-
anism in the case of λe = 8 and λu = 1. ment ratio of URLLC reliability after we use load control
mechanism for eMBB traffic. We found that the use of the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
11

load control has the greatest impact on the reliability of the

FO method. This is because most eMBB users who transmit
data packets in the whole spectrum will have serious conflicts
with URLLC or eMBB, when the number of eMBB access
networks is not limited. Therefore, if we do not control a large
number of eMBB accessing networks, URLLC performance
will decrease, especially when the number of URLLC packets
increases.

D. Performance Variation for High URLLC Packet Load

As shown in Fig. 13, we present the changes in the
reliability performance of URLLC under different reserved
widths of the dedicated spectrum when eMBB traffic has a
load control. We consider the case where URLLC packet load
is high, for example, λe = 8 and λu = 3. In Fig. 13(a), we
show the URLLC reliability performance when the reserved
(a)
widths of the dedicated spectrum is 10. We observe that the
URLLC reliability in the FS method is the worst. Reserving
insufficient dedicated spectrum cause URLLC users to have
severe packet collisions on the dedicated spectrum, thereby
greatly degrading the performance of URLLC. However, the
reliability performance of the proposed UPER method is
54.22% higher compared with the FS method and even better
than other methods. Comparing Figs. 13(a) and 13(b), the
reliability performance of all the methods is improved by
increasing the reserved widths of the dedicated spectrum from
r = 10 to r = 15. The reliability improvement ratio of
the dedicated spectrum increment is listed in TABLE III.
When the amount of dedicated spectrum resources increases,
the reliability of URLLC in the FS method is significantly
improved. In addition, the UPER method also outperform than
other spectrum partition methods.
In Fig. 14, we show the URLLC throughput as per the
impact of packet length. The URLLC throughput with packet (b)
length of 50 bytes is 56.25% higher than that of 32 bytes. That
Fig. 13. Comparisons of reliability performance with eMBB load control
is, the throughput performance of URLLC is proportional to mechanism under different dedicated spectrum width in the case of λe = 8
the packet length in our considered case. From the figure, one and λu = 3: (a) The dedicated spectrum resource r = 10; (b) The dedicated
can also observe that the performance of the proposed UPER spectrum resource r = 15.
is superior to other spectrum allocation methods, including the
FS, FO and PO.
• The reliability of the UPER method performed the best,
VIII. C ONCLUSION and is 54%, 32% and 11% higher than the FS, PO and
FO spectrum allocation methods when the URLLC traffic
In this paper, we proposed an on-device data-driven spec-
load is high.
trum allocation algorithm, called the user-initial probability
• The UPER method can almost reach the reliability and
elastic resource (UPER) scheme, to dynamically adjust the
throughput performance as the FS method when the
probability of selecting shared and dedicated spectrum for
URLLC traffic load is low.
URLLC based on the packet collision status coexisting with
• The proposed method with the load control mechanism
eMBB customers. The feasibility of UPER was evaluated first
can respectively improve the reliability performance of
analytically according to the derived closed-form expressions
the FO, PO and UPER methods by 52%, 16% and 2%
for the successful transmission probability of the UPER,
compared to the case without load control.
fully separated (FS), fully overlapped (FO), and partially
• The URLLC throughput with the packet length of 50
overlapped (PO) four spectrum allocation methods considered
bytes is 56% higher than that of 32 bytes. That is, the
in this paper. Further, the proposed UPER spectrum allocation
throughput performance of URLLC is proportional to the
method was compared to other three considered approaches
packet length in our considered case.
by extensive simulations under different conditions, including
URLLC traffic load, eMBB load control and URLLC packet To sum up, we believe that the proposed on-device data-driven
length. We observed the following phenomena: UPER spectrum management framework with load control

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
12

different access protocols in the MAC layer. Hence, designing

the optimal URLLC/eMBB/mMTC coexistence system from
a PHY/MAC cross-layer perspective is still an open issue.

R EFERENCES
[1] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia,
O. Queseth, M. Schellmann, H. Schotten, H. Taoka et al., “Scenarios for
5G Mobile and Wireless Communications: The Vision of the METIS
Project,” IEEE Communications Magazine, vol. 52, no. 5, pp. 26–35,
2014.
[2] P. Popovski, “Ultra-reliable Communication in 5G Wireless Systems,”
in IEEE International Conference on 5G for Ubiquitous Connectivity,
2014, pp. 146–151.
[3] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski,
“Five Disruptive Technology Directions for 5G,” IEEE Communications
Magazine, vol. 52, no. 2, pp. 74–80, 2014.
[4] N. A. Johansson, Y.-P. E. Wang, E. Eriksson, and M. Hessler, “Radio
Access for Ultra-reliable and Low-latency 5G Communications,” in
IEEE International Conference on Communication Workshop, 2015, pp.
1184–1189.
[5] V. Hytönen, Z. Li, B. Soret, and V. Nurmela, “Coordinated Multi-cell
(a) Resource Allocation for 5G Ultra-reliable Low Latency Communica-
tions,” IEEE European Conference on Networks and Communications,
pp. 1–5, 2017.
[6] G. Pocovi, B. Soret, K. I. Pedersen, and P. Mogensen, “MAC Layer
Enhancements for Ultra-reliable Low-latency Communications in Cel-
lular Networks,” IEEE International Conference on Communications
Workshops, pp. 1005–1010, 2017.
[7] I. Parvez, A. Rahmati, I. Guvenc, A. I. Sarwat, and H. Dai, “A Survey on
Low Latency Towards 5G: RAN, Core Network and Caching Solutions,”
arXiv preprint arXiv:1708.02562, 2017.
[8] H. Ji, S. Park, J. Yeo, Y. Kim, J. Lee, and B. Shim, “Introduction to
Ultra Reliable and Low Latency Communications in 5G,” arXiv preprint
arXiv:1704.05565, 2017.
[9] TR37.868, “Study on RAN Improvements for Machine-type Communi-
cations,” 3rd Generation Partnership Project (3GPP), Tech. Rep., Sep.
2011.
[10] R. Kassab, O. Simeone, P. Popovski, and T. Islam, “Non-Orthogonal
Multiplexing of Ultra-Reliable and Broadband Services in Fog-Radio
Architectures,” IEEE Access, vol. 7, pp. 13 035–13 049, 2019.
[11] A. A. Esswie and K. I. Pedersen, “Opportunistic Spatial Preemptive
Scheduling for URLLC and eMBB Coexistence in Multi-User 5G
Networks,” IEEE Access, vol. 6, pp. 38 451–38 463, 2018.
[12] A. Bairagi, M. S. Munir, M. Alsenwi, N. H. Tran, and C. S. Hong, “A
Matching Based Coexistence Mechanism between eMBB and URLLC
in 5G Wireless Networks,” in Proceedings of the 34th ACM/SIGAPP
(b) Symposium on Applied Computing, New York, USA, pp. 2377–2384,
2019.
Fig. 14. The impacts of URLLC packet length on throughput performance [13] A. T. Abebe and C. G. Kang, “Comprehensive Grant-free Random Ac-
with eMBB load control mechanism under different dedicated spectrum width cess for Massive & Low Latency Communication,” IEEE International
in the case of λe = 8 and λu = 3: (a) The dedicated spectrum resource Conference on Communications (ICC), pp. 1–6, 2017.
r = 10; (b) The dedicated spectrum resource r = 15. [14] G. Wunder, P. Jung, and M. Ramadan, “Compressive Random Access
Using a Common Overloaded Control Channel,” IEEE Global Commu-
nications Conference Workshops, pp. 1–6, 2015.
[15] R. B. Abreu, T. Jacobsen, G. Berardinelli, K. I. Pedersen, N. H.
mechanism can satisfy the stringent latency and reliability Mahmood, I. Kovacs, and P. E. Mogensen, “On the Multiplexing
requirement for URLLC in a sharing spectrum environment of Broadband Traffic and Grant-Free Ultra-Reliable Communication
in Uplink,” IEEE Vehicular Technology Conference, Kuala Lumpur,
with eMBB customers. Malaysia, pp. 1–6, 2019.
In the future, many interesting research topics can be [16] R1-1612250, “Basic grant-free transmission procedure for URLLC,”
Nokia, Alcatel-Lucent Shanghai Bell, 3GPP RAN1#87, Reno, USA,
extended from the current study. First, how to design an 14th - 18th, Nov. 2016.
method to quickly calculate probability of using dedicated [17] Z. Ding, R. Schober, P. Fan, and H. V. Poor, “Simple Semi-Grant-Free
spectrum resources is an important and interesting research Transmission Strategies Assisted by Non-Orthogonal Multiple Access,”
IEEE Transactions on Communications, pp. 4464–4478, 2019.
topic. Secondly, it would be worthwhile further investigating [18] Y. Han, S. E. Elayoubi, A. Galindo-Serranoa, V. S. Varma, and
the optimal bandwidth of the reserved exclusive spectrum M. Messai, “Periodic Radio Resource Allocation to Meet Latency and
for URLLC by using machine learning techniques based on Reliability Requirements in 5G Networks,” IEEE Vehicular Technology
Conference (VTC Spring), pp. 1–6, Jun. 2018.
the long-term network status. Last, the system parameters [19] Chih-Ping Li, Jing Jiang, W. Chen, Tingfang Ji, and J. Smee, “5G Ultra-
affecting the performance of the URLLC/eMBB/mMTC co- reliable and Low-latency Systems Design,” European Conference on
existing system are from different protocol layers, including Networks and Communications (EuCNC), pp. 1–5, Jun. 2017.
[20] R1-1812392, “Issues on URLLC resource allocation,” 3rd Generation
the modulation and coding scheme in the physical layer and Partnership Project (3GPP), ZTE, 3GPP TSG RAN WG1 Meeting#95,
the number of requested resource blocks, packet length and Spokane, USA, 12th - 16th, Nov. 2018.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCCN.2020.3007890, IEEE
Transactions on Cognitive Communications and Networking
13

[21] R1-1701666, “On UL multiplexing of URLLC and eMBB transmis- Li-Chun Wang (M’96-SM’06-F’11) received the
sions,” 3rd Generation Partnership Project (3GPP), Huawei, HiSilicon, B.S. degree from National Chiao Tung University,
3GPP TSG RAN WG1 Meeting#88, Athens, Greece, 13th - 17th, Feb. Taiwan, R. O. C. in 1986, the M.S. degree from
2017. National Taiwan University in 1988, and the Ms.
[22] R1-1801390, “LTE URLLC and eMBB multiplexing,” 3rd Generation Sci. and Ph.D. degrees from the Georgia Institute of
Partnership Project (3GPP), Huawei, HiSilicon 3GPP TSG RAN WG1 Technology, Atlanta, in 1995, and 1996, respectively,
Meeting#92, Athens, Greece, Feb. 2018. all in electrical engineering. From 1990 to 1992,
[23] Z. Zhou, R. Ratasuk, N. Mangalvedhe, and A. Ghosh, “Resource he was with the Telecommunications Laboratories
Allocation for Uplink Grant-Free Ultra-Reliable and Low Latency Com- of the Ministry of Transportations and Communi-
munications,” IEEE Vehicular Technology Conference (VTC Spring), pp. cations in Taiwan. In 1995, he was affiliated with
2577–2465, Jun. 2018. Bell Northern Research of Northern Telecom, Inc.,
[24] R1-1804820, “eMBB and URLLC dynamic multiplexing and preemp- Richardson, TX. From 1996 to 2000, he was with AT&T Laboratories, where
tion indication on the uplink,” 3rd Generation Partnership Project he was a Senior Technical Staff Member in the Wireless Communications
(3GPP), Qualcomm Incorporated, 3GPP TSG RAN WG1 Meeting#92b, Research Department. Since August 2000, he has joined National Chiao Tung
Sanya, China, Feb. 2018. University in Taiwan. Currently he is a chair professor of the Department of
[25] L.-C. Wang and C.-W. Wang, “Spectrum Management Techniques with Electrical and Computer Engineering and is jointly appointed by the Depart-
QoS Provisioning in Cognitive Radio Networks,” IEEE 5th International ment of Computer Science. His recent research interests are focused on cross-
Symposium on Wireless Pervasive Computing, pp. 116–121, 2010. layer optimization and data-driven learning techniques for 5G ultra-reliable
[26] L.-C. Wang, C.-W. Wang, and K.-T. Feng, “A Queueing-Theoretical and ultra-low latency communications (URLLC), edge computing, unmanned
Framework for QoS-Enhanced Spectrum Management in Cognitive aerial vehicle (UAV) communications networks, and AI-empowered mobile
Radio Networks,” IEEE Wireless Communications, vol. 18, pp. 18–26, networks.
2011. Dr. Wang was elected to the IEEE Fellow for his contributions in cellular
[27] R.-G. Cheng, J. Chen, D.-W. Chen, and C.-H. Wei, “Modeling and architectures and radio resource management in wireless networks. He won the
Analysis of an Extended Access Barring Algorithm for Machine-type 1997 IEEE Jack Neubauer Best Paper Award (1997), Distinguished Research
Communications in LTE-A Networks,” IEEE Transactions on Wireless Award of Ministry of Science and Technology, Taiwan (2012 and 2018), and
Communications, vol. 14, no. 6, pp. 2956–2968, 2015. IEEE Communications Society Asia-Pacific Board Best Paper Award (2015).
[28] A. Larmo and R. Susitaival, “RAN Overload Control for Machine Type He has published over 90 journal papers and 180 conference papers, 19 US
Communications in LTE,” IEEE Global Communications Conference patents and co-edited a book Key Technologies for 5G Wireless Systems
Workshops, pp. 1626–1631, 2012. (Cambridge 2016).
[29] TR23.898, “Access Class Barring and Overload Protection (ACBOP),”
3rd Generation Partnership Project (3GPP), Tech. Rep., Apr. 2005.
[30] M. Koseoglu, “Lower Bounds on the LTE-A Average Random Access
Delay under Massive M2M Arrivals,” IEEE Transactions on Communi-
cations, vol. 64, no. 5, pp. 2104–2115, 2016. Chien-Hwa Hwang received his B.S. and M.S
[31] S. Duan, V. Shah-Mansouri, and V. W. Wong, “Dynamic Access Class degrees from National Taiwan University, Taipei,
Barring for M2M Communications in LTE Networks,” IEEE Global Taiwan and the Ph.D. degree from the University
Communications Conference Workshops, pp. 4747–4752, 2013. of Southern California, Los Angeles, USA, all in
[32] R1-1611399, “UL grant free transmission for eMBB,” CATT, 3GPP TSG electrical engineering. He is currently with Medi-
RAN WG1 Meeting #87, Reno, USA, 14th - 18th, Nov. 2016. aTek working on LTE and 5G system design and
[33] K.-D. Lee, S. Kim, and B. Yi, “Throughput Comparison of Random advanced receivers. His research interests include
Access Methods for M2M Service over LTE Networks,” IEEE Global statistical signal processing, MIMO communication,
Communications Conference Workshops, pp. 373–377, 2011. and interference management of cellular systems.

Ju-Ya Chen received the B.S., M.S., and Ph.D. de-

grees in communication engineering from National
Chiao Tung University, Hsinchu, Taiwan, R.O.C..
He is currently with MediaTek working on 5G sys-
tem design and product development. His research
interests include statistical signal processing and
communications theory.

Li-Yu Cheng received the M.S. degree from Na-

tional Chiao Tung University, Taiwan, R.O.C., in
Shu-Feng Cheng received the BS degree in elec- 2018. His research interests are in the field of Mobile
trical engineering from National Taiwan Normal Network and 5G URLLC.
University, Taipei, Taiwan, R. O. C. in 2014. She is
currently working toward the PhD degree with the
electrical and computer Engineering, National Chiao
Tung University, Hsinchu, Taiwan. Her current re-
search interests include yield modeling for multi-
band RF-Interconnected through-silicon-via (MRFI-
TSV), scheduling design for narrowband internet
of things (NB-IoT), and cross-layer analysis for
ultra-reliable and ultra-low latency (URLLC) and
enhanced mobile broadband (eMBB) multiplexing in 5G wireless commu-
nications.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.