1

Network Slicing for URLLC and eMBB with

Max-Matching Diversity Channel Allocation
Elço João dos Santos Jr, Richard Demo Souza, Senior Member, IEEE,
João Luiz Rebelatto, Senior Member, IEEE, and Hirley Alves, Member, IEEE

Abstract

This work considers the problem of radio resource sharing between enhanced mobile broadband
(eMBB) and ultra-reliable and low latency communications (URLLC), two heterogeneous 5G services.
More specifically, we propose the use of a max-matching diversity (MMD) algorithm to properly allocate
the channels to the eMBB users, considering both heterogeneous orthogonal multiple access (H-OMA)
and heterogeneous non-orthogonal multiple access (H-NOMA) network slicing strategies. Our results
indicate that MMD can simultaneously improve the eMBB achievable rate and the URLLC reliability
regardless the network slicing strategy adopted.

Index Terms

5G, eMBB, network slicing, URLLC, channel allocation, non-orthogonal multiple access.

I. I NTRODUCTION

5G technology aims at three heterogeneous use cases: enhanced mobile broadband (eMBB),
ultra-reliable and low latency communications (URLLC) and massive machine type communica-
tions (mMTC) [1], [2]. The performance target of eMBB is to achieve high data rates with packet
error rates (PER) around 10−3 , while such kind of traffic is stable and tolerates a certain amount
of latency. URLLC is an innovative service supported by 5G, which aims at much lower PER

Elço João dos Santos Jr. and Richard Demo Souza, EEL, UFSC, Florianópolis-SC, Brazil. e.joaojr@gmail.com,
richard.demo@ufsc.br
João Luiz Rebelatto, CPGEI, UTFPR, Curitiba-PR, Brazil. jlrebelatto@utfpr.edu.br
Hirley Alves, CWC, University of Oulu, Oulu, Finland. hirley.alves@oulu.fi
This work has been supported in Brazil by CNPq and Print CAPES-UFSC “Automation 4.0”, and in Finland by Academy of
Finland 6Genesis Flagship (Grant no318927) and EE-IoT (no319008).
 2

with strict latency constraints. The required data rate of URLLC is typically low compared to
eMBB, while traffic is intermittent. Finally, mMTC must provide connectivity to a large number
of devices, sporadically transmitting short packets, where the target PER is much larger than for
eMBB and URLLC.
The diverse requirements of the different 5G use cases pose great challenges for the network
design. One of the enablers of 5G is network slicing [3], a service-oriented point of view to model
a system that can support multiple types of users in a common physical infrastructure, which
adapts the network shape and allocates resources according to the services in use [3]. The original
slicing concept assumes an orthogonal approach, where the network resources of one service are
isolated from others. Recently, focusing on the physical layer (PHY) and in the uplink, Popovski
et al discuss heterogeneous orthogonal multiple access (H-OMA) and introduce the paradigm of
heterogeneous non-orthogonal multiple access (H-NOMA), where the term heterogeneous refers
to the heterogeneity of services (eMBB, URLLC, mMTC) [2]. In H-OMA the slicing of PHY
resources is orthogonal among the services, while in H-NOMA heterogeneous services may
share different PHY slices, making use of successive interference cancellation (SIC) decoding.
Such concept is remarkably different than regular NOMA, in which radio resources are shared
by devices with the same requirements [4].
By means of a communication theoretic approach, the trade-offs between H-OMA and H-
NOMA in the slicing of eMBB and URLLC, as well as of eMBB and mMTC, have been
analyzed in [2]. For instance, results indicate that H-NOMA slicing between eMBB and URLLC
can achieve significant improvements in the system performance, a consequence of exploiting
the concept of reliability diversity. Since URLLC traffic is much more reliable, and therefore
has a high probability of being successfully decoded under the interference of eMBB traffic,
sharing eMBB resources with URLLC brings considerable performance gains specially at high
eMBB target data rates. Nevertheless, depending on the channel conditions and on the URLLC
target data rate, H-OMA can be a better solution than H-NOMA [2]. Interesting trade-offs also
arise in the case of eMBB and mMTC slicing, bringing new optimization opportunities for the
system designer.
In [5], the H-OMA and H-NOMA strategies for eMBB and URLLC slicing from [2] are
applied to a multi-cell scenario, where the URLLC traffic is decoded at the base station (BS) to
meet latency requirements, while eMBB traffic is forwarded to a cloud server. The H-NOMA
approach leads to improvements for both eMBB and URLLC when the activation probability
 3

of the last is small and the edge has sufficient capacity. The authors of [6] recently investigated
the slicing between URLLC and eMBB considering a minimum mean square error receiver with
multiple antennas, comparing the performance with and without SIC. Results reveal that H-
NOMA with SIC brings improvements in high SNR or with low URLLC loads, while H-OMA
can achieve higher URLLC reliability.
It is interesting to note that, as eMBB traffic is scheduled only after a radio access and
contention resolution phase, it is practical to consider that channel state information (CSI) for
eMBB users is known at the BS, as implemented in current wireless standards [7], and assumed
in [2]. As a consequence, supposing the use of orthogonal frequency division multiple access
(OFDMA), different eMBB users can be adequately allocated to the frequency resources as
to maximize the diversity and meet the reliability target. In this sense, in [8], a maximum-
matching diversity (MMD) method based on random bipartite graph theory is considered for
allocating users to channels in an OMA scenario with homogeneous types of users. Through
MMD, frequency diversity for each user is maximized, equaling the number of independent
channels.
In this work, we modify the communication theoretic analysis of [2] in order to investigate
the impact of eMBB channel allocation in the performance of H-OMA and H-NOMA slicing
between eMBB and URLLC uplink traffics. The main contributions of this work are summarized
as follows:
• We evaluate the achievable rate of eMBB when adopting the MMD method from [8] to
allocate the channels;
• We evaluate the performance of the network slicing between URLLC and eMBB with
channel allocation. The increased frequency diversity achieved by the proposed MMD-
aided scheme is beneficial to both H-OMA and H-NOMA, being capable of improving the
eMBB achievable rate and the URLLC reliability simultaneously.
The rest of this paper is organized as follows. Section II presents the system model, while
Section III brings the communication theoretic performance formulation of H-OMA and H-
NOMA between eMBB and URLLC traffics, considering MMD channel allocation. Numerical
results are discussed in Section IV, while Section V concludes the paper.
 4

II. S YSTEM M ODEL

We consider the uplink of a network with eMBB and URLLC devices transmitting to a
common BS. The bandwidth is divided into F channels of index f ∈ {1, . . . , F }, where each
channel is subject to independent and identically distributed (i.i.d.) Rayleigh fading, which is
assumed to be constant during one time slot (TS). The channel coefficient of user i ∈ {B, U }
in channel f is thus Hi,f ∼ CN (0, γ̄i ), where γ̄i corresponds to the average signal-to-noise ratio
(SNR), being Gi,f , |Hi,f |2 the channel gain, and where subscripts B and U refer to eMBB
and URLLC devices, respectively. The number of channels allocated to user i is Fi ≤ F , with
i ∈ {B, U }. Moreover, each TS is divided into S minislots.
In accordance to [2], we adopt the following approaches regarding eMBB and URLLC
transmissions:
• An URLLC device transmits in a pre-assigned minislot1 (to meet latency requirements), in
grant-free fashion, and spreads the transmission over FU channels (to increase reliability).
The activation probability of the device is aU .
• An eMBB user transmits in a single channel f among the FB available channels, but during
the entire TS.
This time-frequency grid is illustrated in Fig. 1, considering H-OMA in Fig. 1(a) and H-NOMA
in Fig. 1(b). It is worthy mentioning that, as in [2], we do not aim at evaluating the influence of
the sharing of wireless resources among devices of the same type. Thus, we assume that radio
access and competition among eMBB devices have been resolved prior to the considered time
slot, i.e., the number of eMBB devices able to transmit in such time slot is equal to the number
of channels FB . We also do not model collisions among URLLC devices, by assuming that a
single URLLC device is active in a given pre-assigned minislot with some probability aU .
Moreover, for eMBB we assume that the BS has CSI before transmission, obtained during
a scheduling phase, as in [2]. However, differently from [2], in this work we consider MMD
channel assignment for eMBB users as follows.

1
One could relax this restriction and allow a URLLC device to use more than one minislot, establishing a trade-off between
delay and performance. However, this is beyond the scope of this paper and is left as future work.
 5

in G1 G3 a in
G1 G3
a l G
n el G G2 ne G5 G2
an han
Ch C
G4
min
GB,f min
GB,f
u
B,3
u u
B,1
B,3
u U,2 u U,2 u
B,1
cy u
tim u U,1 en BB u B,5
e fre
q u
eM
u U,1 cy
tim B,4
en
e qu
LLC fre
UR

(a) H-OMA. (b) H-NOMA.

Fig. 1: System model with F = 4 channels and S = 4 subslots.

A. Max-Matching Diversity (MMD) Channel Allocation

The MMD channel allocation from [8] increases the frequency diversity by means of a proper
allocation of users to channels. Through the random graph theory, herein eMBB users and
channels are seen as opposite parts of the vertex set of a bipartite graph, while a user is connected
to a channel if it is not in outage. The MMD algorithm then aims at minimizing the number of
users in outage, by means of channel allocation.
The outage probability of the MMD scheme, in a scenario with F channels subjected to
independent fading and with average SNR γ̄ per channel, is [8]

PMMD (γ̄, F ) = 2Ps (γ̄)F + O(Ps (γ̄)F ) ≈ 2Ps (γ̄)F , (1)

where Ps (γ̄) is the outage probability of a single channel and O(·) is the higher order infinites-
imal. Note in (1) that the MMD scheme from [8] achieves optimal frequency diversity, which
equals the number of independent channels.

III. S LICING FOR URLLC AND E MBB WITH MMD

A. eMBB with MMD

Let us consider a scenario where a radio resource f ∈ {1, . . . , F } is allocated exclusively

to an eMBB user, following the MMD algorithm from [8]. As for eMBB it is reasonable to
 6

assume CSI before data transmission, then transmit power can be adapted. The main objective
of eMBB is to maximize its data rate, subject to the reliability requirement B and the average
power constraint PB = 1.

Theorem 1. The MMD-aided eMBB rate is

MMD
= log2 1 + Gtar


rB B,f , [bits/symbol] (2)
0
where γ̄B
Gtar
B,f =P
 kGmin , (3)
FB FB
k=1 (−1)
k−1
k
kΓ 0, γ̄B,f
0
B
 
0 1/FB 0
Gmin
B,f = −γ̄B ln 1 − B and γ̄B , 2−(1/FB ) γ̄B .

Proof: Please refer to Appendix A.

B. URLLC

Differently from eMBB users, it is assumed that URLLC devices do not have CSI, due to
latency requirements. The URLLC device transmits data in all the FU i.i.d. channels of a minislot,
such that the outage probability, in the absence of interference from other services, is2 [2]
FU
!
1 X
PU (GU,f ) = Pr log (1 + GU,f ) < rU . (4)
FU f =1 2

The target rate rU is obtained by imposing the requirement PU (GU,f ) ≤ U to (4), which we
refer to as rUOMA .

C. H-OMA

In H-OMA, orthogonal slicing is achieved by allocating FB ≤ F channels exclusively to the

eMBB users, while the remaining FU = F − FB are allocated to URLLC. Thus, the rate region
OMA-MMD OMA OMA-MMD
(rB , rU ) is obtained by considering rB as the sum-rate of the active eMBB users,

OMA-MMD MMD
rB = FB rB , (5)

MMD
where rB comes from (2) and rUOMA is computed from (4).

2
We follow [2] and assume that the block length utilized in the URLLC protocol is long enough so that the finite block length
formulation can be well approximated by the asymptotic outage formulation [9].
 7

D. H-NOMA

In H-NOMA, all the F channels are available simultaneously for both eMBB and URLLC
(FB = FU = F ), such that the interference from URLLC transmissions into eMBB (and vice-
versa) needs to be considered. To this end, we consider that the BS performs SIC3 . Due to
latency and reliability constraints, we assume that the SIC decoder always attempts to decode
the URLLC transmission first, while treating the eMBB traffic as interference. In case of a
successful decoding, the URLLC signal is then removed from the superimposed signal before
attempting to decode the eMBB traffic. As a consequence, URLLC would only interfere with
eMBB when the SIC decoder is not able to decode the URLLC messages.
In this scenario, we have the following Lemma regarding the achievable rate of the eMBB
device.

Lemma 1. The achievable rate of a MMD-aided eMBB device in H-NOMA is

NOMA-MMD
= F log2 1 + Gtar


rB B,f , (6)

Gtar
B,f is upper bounded by (3) and the threshold SNR is
!
1/F
0 1 − B B
Gmin
B,f ≤ −γ̄B ln . (7)
1 − U (1 − (1 − aU )S )

Proof: Following [2], the eMBB outage probability under H-NOMA can be bounded by
the law of total probability as
min 1 − B
Ps (γ̄B ) = 1 − e−GB,f /γ̄B ≤ , (8)
1 − U (1 − (1 − aU )S )
which accounts for the fact that the eMBB user is in outage when the SIC decoder does not
0
decode the URLLC signal. Hence, by resorting to the fact that PB = Ps (γ̄B )FB = B due
to MMD channel allocation, one can isolate the threshold SNR from (8) as presented in (7),
concluding the proof.
The threshold from (7) indicates that the impact of URLLC transmissions in the eMBB
decoding should be minimal, due to the fact that, by definition, U << B . On the other hand,

3
Note that, as presented in [2], SIC outperforms other techniques of multi-user detection, such as puncturing and erasure
decoders.
 8

the eMBB interference in the URLLC traffic is supposed to be more critical, since URLLC is
decoded prior to eMBB. As in [2] the outage probability of URLLC under H-NOMA is
FU
! !
1 X GU,f
PUNOMA (GU,f ) = Pr log 1 + tar < rU , (9)
FU f =1 2 GB,f

where we assume that the interference of eMMB is always present in the URLLC decoding, due
to their long period activation. The URLLC achievable rate rUNOMA is then obtained by imposing
the reliability constraint PUNOMA (GU,f ) ≤ U .
For comparison purposes, the eMBB achievable rate from [2], without channel allocation, can
be obtained by setting FB = 1 in (3) (OMA) and in (7) (NOMA).

IV. N UMERICAL R ESULTS

We resort to Monte Carlo simulations to evaluate the performance of the MMD-aided proposed
schemes. Figs. 2(a) and 2(b) present the rate region (rB , rU ), respectively for the cases γ̄U > γ̄B
and γ̄B > γ̄U . As expected, H-NOMA tends to achieve higher rates when γ̄U > γ̄B (due to the
increased probability of recovering the URLLC in the SIC process), while H-OMA may be more
advantageous when γ̄B > γ̄U . The beneficial impact of MMD is more evident in Fig. 2(a), where
its additional frequency diversity compensates the worst average channel condition of eMBB. In
NOMA NOMA-MMD
this scenario, the maximum rB goes from ≈ 13.7 bits/symbol to rB ≈ 46 bits/symbol.
Significant improvements are also noticed in Fig. 2(b), mainly to OMA.
Tab. I evaluates rB for U ∈ {10−5 , 10−6 , 10−7 }, rU ∈ {1, 2} bits/symbol, γ̄B = 10 dB, and γ̄U
= 20 dB. One can see that the MMD-aided schemes achieve larger rB , while operating at more
stringent values of U , simultaneously improving the rate of eMBB and reliability of URLLC.
OMA-MMD
For example, while rB = 17 bits/symbol for  = 10−6 , the rate achieved at a higher target
outage probability  = 10−5 is only rB
OMA
= 7.5 bits/symbol, for a fixed rU = 2 bits/symbol.
This shows that the gains provided by MMD do not benefit only the eMBB user: the URLLC
user is indirectly benefited from the eMBB channel allocation, since eMBB can operate at lower
transmission power levels, achieving the same (or even improved) performance, while decreasing
the interference on URLLC devices. Moreover, in general, NOMA-MMD outperforms OMA-
MMD for small values of rU , and OMA-MMD becomes more advantageous when rU increases.

Finally, since in practice eMBB CSI is already available at the BS for scheduling purposes,
the main drawback of the proposed scheme resides on the complexity increase to execute the
 9

3.5 1.2
H-OMA H-OMA
3 H-NOMA 1 H-NOMA
H-OMA-MMD H-OMA-MMD
2.5 H-NOMA-MMD H-NOMA-MMD
0.8
2
0.6
1.5
0.4
1

0.5 0.2

0 0
0 10 20 30 40 50 0 20 40 60 80

(a) γ̄B = 10 dB, γ̄U = 20 dB. (b) γ̄B = 20 dB, γ̄U = 10 dB.

Fig. 2: S = 5, aU = 0.1, F = 10, B = 10−3 , U = 10−5 .

TABLE I: rB vs U , for rU ∈ {1, 2} bits/symbol.

U 10−5 10−6 10−7

rU 1 2 1 2 1 2

OMA
rB 9.6 7.5 8.8 6.1 7.3 4.0
NOMA
rB 13.7 13.7 13.7 13.7 13.7 10.0
OMA-MMD
rB 30.8 21.5 26.8 17.0 25.2 9.0
NOMA-MMD
rB 36.9 18.8 34.1 15.0 30.0 10.0

channel allocation, which scales with O(FB2.5 ) [8], and a very slight increase (few bits) in control
traffic to eMBB users.

V. F INAL C OMMENTS

We considered network radio resource slicing between eMBB and URLLC users in a 5G
system. By resorting to the MMD approach to properly allocate channels to the eMBB users,
we showed that the eMBB achievable rate and the URLLC reliability can be improved simulta-
neously, under both H-OMA and H-NOMA network slicing strategies.
 10

A PPENDIX A
P ROOF OF T HEOREM 1

Assuming that the eMBB devices have CSI [2], a transmission only occurs when GB,f , the
instantaneous channel gain, is greater than a threshold SNR Gmin
B,f . The outage probability of a

point-to-point (single channel) communication is then

Z Gmin
B,f
min
Ps (γ̄B ) = Pr[GB,f < GB,f ] = pGB,f (x)dx
0
Gmin
e−x/γ̄B
Z B,f min
= dx = 1 − e−GB,f /γ̄B . (10)
0 γ̄B
When considering MMD channel allocation from [8] with FB independent channels, the outage
probability of an eMBB user, in a scenario where the number of users is equal to FB , can be
approximated following (1) as
PB ≈ 2Ps (γ̄B )FB . (11)
0 0
For Rayleigh fading, it can be shown from (10) that 2Ps (γ̄B )FB ≈ Ps (γ̄B )FB , with γ̄B ,
2−(1/FB ) γ̄B . After imposing the reliability constraint PB = B , we obtain the threshold SNR
from (10) and (11) as
0
 
1/FB
Gmin
B,f = −γ̄B ln 1 − B . (12)

Based on power inversion, the instantaneous power is chosen as PB (GMMD tar MMD
B,f ) = GB,f /GB,f

when GMMD min MMD tar

B,f ≥ GB,f . Otherwise, PB (GB,f ) is set to zero. The target SNR GB,f is then obtained

by imposing the average power constraint

Z ∞
1 = E PB (GMMD
 
B,f ) = pGMMD
B,f
(x)PB (x)dx, (13)
Gmin
B,f

where pGMMD
B,f
(x) is the probability density function (pdf) of the instantaneous channel gain after
FB 0
MMD, GMMD
B,f , obtained from the cumulative density function (cdf) Ps (γ̄B ) as
d h −Gmin
0
B,f /γ̄B )FB
i
pGMMD (x) = (1 − e
B,f dt
FB h 0
iFB −1 0
−x/γ̄B
= 0 1−e e−x/γ̄B
γ̄B
FB   −kx/γ̄ 0
k−1 FB ke
X B
= (−1) 0 . (14)
k=1
k γ̄B
 11

The summation in (14) is obtained applying the binomial theorem [10], so that the pdf of the
SNR of FB channels can be expressed as a linear combination of FB exponential pdfs. After
replacing (14) in (13), we have:
Z ∞ X FB   −kx/γ̄ 0 tar
B G
k−1 FB ke B,f
1= (−1) 0 dx
min
GB,f k=1 k γ̄B x
FB
!
kGmin
 
k−1 FB k
X B,f
tar
=GB,f (−1) 0 Γ 0, 0 , (15)
k=1
k γ̄B γ̄B

where Γ(·, ·) is the upper incomplete gamma function. Then, one can isolate Gtar
B,f from (15),

concluding the proof.

R EFERENCES

[1] M. Shafi et al., “5G: A tutorial overview of standards, trials, challenges, deployment, and practice,” IEEE J. Sel. Areas
Commun., vol. 35, no. 6, pp. 1201–1221, June 2017.
[2] P. Popovski, K. F. Trillingsgaard, O. Simeone, and G. Durisi, “5G wireless network slicing for eMBB, URLLC, and
mMTC: A communication-theoretic view,” IEEE Access, vol. 6, pp. 55 765–55 779, 2018.
[3] X. Foukas, G. Patounas, A. Elmokashfi, and M. K. Marina, “Network slicing in 5G: Survey and challenges,” IEEE Commun.
Mag., vol. 55, no. 5, pp. 94–100, May 2017.
[4] Z. Ding et al., “A survey on non-orthogonal multiple access for 5G networks: Research challenges and future trends,”
IEEE J. Sel. Areas Commun., vol. 35, no. 10, pp. 2181–2195, Oct 2017.
[5] 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.
[6] R. Abreu et al., “On the multiplexing of broadband traffic and grant-free ultra-reliable communication in uplink,” in IEEE
Vehicular Technology Conference (VTC), 2019, pp. 1–6.
[7] A. A. Zaidi et al., “Designing for the future: the 5G NR physical layer,” Ericsson, Tech. Rep., Jul. 2017. [Online]. Available:
https://www.ericsson.com/en/ericsson-technology-review/archive/2017/designing-for-the-future-the-5g-nr-physical-layer
[8] B. Bai, W. Chen, Z. Cao, and K. B. Letaief, “Max-matching diversity in OFDMA systems,” IEEE Trans. Commun., vol. 58,
no. 4, pp. 1161–1171, April 2010.
[9] W. Yang, G. Durisi, T. Koch, and Y. Polyanskiy, “Quasi-static multiple-antenna fading channels at finite blocklength,”
IEEE Trans. Inf. Theory, vol. 60, no. 7, pp. 4232–4265, July 2014.
[10] F. Rosas, R. D. Souza, M. Verhelst, and S. Pollin, “Energy-efficient MIMO multihop communications using the antenna
selection scheme,” in IEEE Intern. Symp. on Wireless Commun. Systems (ISWCS), 2015, pp. 686–690.