Professional Documents
Culture Documents
https://doi.org/10.1007/s11036-019-01288-6
Abstract
Internet-of-Things (IoT) deployment underlaying cellular communication have been drawing increasing attention in recent
years. In this work, we consider Device-to-Device (D2D) communication technology for enabling uplink transmission in
multiple IoT groups/clusters at which the resource block (RB) of each cellular user is reused by multiple IoT devices (IoTDs).
Joint resources and power allocation optimization problem is formulated for maximizing both the IoTDs connectivity and
cell throughput under an interference constraint. A heuristic algorithm, namely greedy iterative matching (GIM), is proposed
as a sub-optimal solution for the resource allocation problem while a fixed-power margin technique is used for the cellular
users’ power allocation. Simulations results show that the proposed GIM algorithm provides enhanced cell throughput gain
and accessibility to IoTDs. The proposed reuse model is applicable in a network at which the number of IoTDs deployed
within multiple groups is larger than the available cellular reuse partners.
cells (or gateways), IoT devices can effectively reduce the in a non-orthogonal way, which provides efficient utiliza-
power consumption and guarantee the throughput require- tion of the available spectrum to support more users in the
ment and quality of service (QoS) of IoT systems. With system. Hence, the spectrum efficiency, as well as the network
the appropriate signaling protocols, the small-cell gateways throughput, can be increased under a controlled interference
could aggregate traffic from a large number of low-power from D2D communication. Moreover, this type of non-
IoT devices (IoTDs) into the existing cellular networks. orthogonal resources sharing offers different types of gains
In D2D technology, communication between nodes [22] such as proximity gain, hop gain, reuse gain, and pair-
in close proximity to each other is performed without ing gain. However, interference management is needed to
passing the messages through the macro base station, or guarantee successful detection of the simultaneously transmit-
the evolved Node Base station (eNB). The direct D2D ted signals at the destinations in this mode [2]. There are
communication introduces benefits for the network such as different resource allocation methods in D2D communica-
lower energy consumption due to short range transmission, tion underlaying cellular networks that are described in [22,
load balancing, traffic offloading, and better coverage for 33]. In this work, we exploit D2D reuse mode for enabling
edge users [3, 14]. In order to support the operation of D2D IoT devices to upload messages to a gateway by sharing the
communication, two modes are defined in LTE networks, resources of a group of cellular users while guaranteeing the
namely the dedicated mode, known as overlay, and the reuse target rate for both UE and IoT devices.
mode [13], known as underlay. In dedicated mode, D2D The success of such heterogeneous architecture assuming
nodes share the available cellular spectrum with orthogonal an interplay between IoT systems and D2D communication
sharing (i.e. a part of the available radio resources is necessitates careful resource management, power alloca-
dedicated for D2D nodes). tion, and transmission scheduling which is generally non-
On the other hand, D2D nodes in the underlay cellular deterministic polynomial-time hard (NP-hard) problems [3,
communication mode reuse the same resources of users’ 10, 11]. It is worth noting that the process of resource shar-
equipment (UEs) under control of the base station [22, 37] ing in D2D and machine to machine (M2M) networks are
Our IoTDs Connectivity, Uplink multi-cluster underlaying Joint channel-power Greedy iterative matching
Throughput cellular net. algorithm
[5] Sum Throughput Uplink multi-cluster Joint channel-power Proportional fair (PF) scheduler
underlaying cellular net. with Kuhn Munkres algorithm
[17] Energy Efficiency HetNet, multi-cell massive MIMO Joint beamforming-power Convex quadratic program
[19] Network Utility Multi-hop cognitive radio net. Joint rate-power Successive convex approx., sub-
optimal distributed algorithm
[31] Delivery Capacity, Downlink cellular net. Joint caching-channel Primal-dual interior point
Average number of method, Exhaustive binary
Replicas search
[32] Average Distortion Downlink cellular in dense 5G net. Rate-number of descriptions Genetic algorithm
[26] Sum Rate D2D uplink underlaying cellular net. channel-power Hypergraph based cluster-
ing,Bisection algorithm
[30] System Capacity Uplink-downlink D2D underlaying Joint uplink-downlink resources Modified Hungarian
cellular net. algorithm
[21] Total Throughput Fractional frequency reuse D2D Channel-power Resources control scheme
uplink cellular
[36] D2D Sum Rate Downlink D2D underlaying cellular net. Channel-power Iterative resources allo-
cation algorithm
[1] D2D Sum Rate Uplink D2D underlaying cellular net. Joint channel-power Distributed coalition formation
algorithm/iterative power con-
trol method
[7] Energy-Efficiency D2D multicast transmission Channel-power Heuristic resource and power
undelaying cellular net. allocation algorithm
[15] Sum Throughput Multicast D2D underlay- Joint channel-power Maximum weighted bipartite
ing cellular net. matching based scheme/less
complexity heuristic algorithm
Mobile Netw Appl
investigated from different perspectives in order to optimize D2D communication to guarantee reliable transmission and
various performance metrics and enable different applica- extend the unmanned aerial vehicles (UAV) coverage for
tions [3, 5, 6, 10–12, 17–20, 23, 28, 29, 31, 32]. In the IoT in disasters.
context of this work, we summarize a survey report in In this work, we consider the D2D communication tech-
Table 1, which compares a group of optimization problems nique for enabling a multi-cluster IoT deployment at which
in terms of deployment scenario, objective, optimization multiple groups of IoTDs are uniformly deployed in the vicin-
variables, and the used algorithm. ity of K cluster heads devices that works as a gateway/small-
The authors in [19] considered a successive convex cell for the entire cluster. Each IoTD is allowed to reuse
approximation to solve a joint rate and power control the RB of one UE for uplink transmission while the RB
problem that maximizes the network utility subject to the of each cellular user can be shared by one IoTD from
capacity and the primary user (PU) outage constraints in each IoT group (IoTG) provided that all reuse partners can
multi-hop cognitive radio networks (CRNs). By targeting achieve their pre-defined quality-of-service (QoS) require-
a high-performance video streaming, the authors in [32] ments. The proposed optimization problem is formulated for
exploited D2D helper-requester pairs for achieving a high maximizing both the IoTDs connectivity and cell through-
quality of experience (QoE) at high energy efficiency put under a fixed power-margin interference management
(EE) taking into consideration the co-channel interference scheme that is used for the cellular users’ power allocation.
over D2D communications in 5G networks. Similarly, the By carefully allowing IoT devices to share the cellular
authors in [31] proposed a joint caching and downlink resources, the spectrum efficiency, in terms of the total
resource sharing optimization framework to efficiently network throughput, is maximized while guaranteeing QoS.
deliver multimedia contents to the mobile users in a wireless In spite of the fact that we do not investigate EE in this
multimedia sensor networks (WMSNs). Additionally, the work, the proposed algorithm can intuitively improve the
EE maximization has recently gained a lot of attention EE by limiting the transmit power of the admitted IoTDs
for different networks architectures. In those schemes, in order to satisfy the interference constraints. This means
the EE is targeted alone or as a joint objective with that the admitted IoTDs are the group that can achieve QoS
other performance metrics such as latency, sum rate requirements with a low-power transmission. The major
maximization, etc. [17, 18, 28, 29]. The authors in [17] contributions in this work are listed as follows:
investigated different resource allocation schemes for EE
– Investigate the throughput and access rate for a single-
maximization for various wireless networks. The authors
cell scenario with a multiple-clusters of IoT devices
in [5] investigated a sum-rate maximization problem for
under a fixed power-margin interference management
simultaneous spectrum access of M2M and human-to-
scheme.
human (H2H) communications in uplink multi-cluster
– Propose joint resources and power allocation optimiza-
underlaying cellular communication. In [6], the authors
tion problem for maximizing both the IoTDs connectiv-
considered the dynamic allocation for M2M using a
ity and cell throughput.
proportional integral derivative (PID) controller to mitigate
– A sub-optimal heuristic algorithm, namely greedy itera-
M2M interference on H2H communication under data rate
tive matching (GIM), is proposed as a low-complexity
and fairness constraints.
sub-optimal solution for the resource allocation problem.
In [11], the authors proposed a two time-slots down-
– Simulations results show that the proposed GIM algorithm
link transmission protocol for ultra-reliable and low latency
provides enhanced cell throughput gain and accessibility
communication (URLLC) between the base station (BS)
for IoTDs compared to single IoT cluster in [3].
and a group of clustered IoT devices (IoTDs) within a fac-
tory system. The proximity service of D2D communication The rest of the paper is organized as follows. In Section 2,
is used as a means for delivering a short-range reliable trans- we describe the system model while illustrating the QoS
mission mechanism from a cluster-head to IoTDs in the requirements. Then, we introduce the joint optimization
second time-slot. The authors in [10] adopted D2D commu- problem formulation and the proposed sub-optimal algo-
nication technique as a mean of route extension of NB-IoT rithm in Section 3. In Section 4, the performance analysis
systems. In [20], the authors proposed a social-aware group of the proposed algorithm is analyzed. Finally, the paper is
formation framework to allocate resource blocks (RBs) concluded in Section 5.
effectively in NB-IoT solution. The authors in [3, 23] pro-
posed resource allocation protocols to provide the required
connectivity between a group of IoTDs and its associated 2 System model
cluster-head using D2D communication underlaying LTE
networks for uplink transmission. In [12], the authors pro- In this work, we focus on a single-cell uplink scenario,
posed a multi-antenna transceiver design and multi-hop as shown in Fig. 1. We consider a group of cellular users
Mobile Netw Appl
1 , if RB of the mth UE is reused by the nth IoTD in the k th IoTG,
am,n,k = (3)
0 , otherwise.
We assume that the destinations (i.e., eNB and the K IoT- In FPM scheme, the average received powers from all UEs
GW) have certain QoS constraints, u and g , that need to at eNB are controlled to the same power level denoted by
u . Therefore, based on the path-loss model, the transmit
be satisfied in order to correctly decode the received signals PeNB
from the UE and the IoTDs. In order to consider sharing power of the mth UE, Pm , is given by:
the same subcarrier between one UE and a K IoTDs (one u
PeNB
IoTD from each IoTG at maximum), a fixed-power margin Pm = b )−α
. (4)
L (dm
(FPM) interference management scheme is adapted [3, 8, 9].
1 The orthogonality among the M UE channels can be achieved in 2 Theassumption that only one IoTD at maximum from each IoTG is
different systems by separating them in time, frequency or code allowed to share the RB of each cellular user is to enforce fairness
domain. between the IoT clusters.
Mobile Netw Appl
A power margin in the signal-to-interference-plus-noise the IoTDs transmit power can be derived using (8) and (10)
ratio (SINR) of the UE is assumed to protect the UE from as follows:
excessive interference from the IoT reuse partners. This No (κ − 1)
power margin (i.e., κ) is used by the eNB to determine:
m
Pn,k = Pd ≤ K N b −α
. (11)
k n am,n,k L (dn,k )
(1) The maximum allowable interference introduced by
all paired IoTDs to avoid harmful interference to the UE.
(2) The maximum power to be allocated to the IoTDs
which can be derived by taking κ into account when
3 Resources allocation for IoTDs
calculating the signal-to-noise ratio (SNR) at eNB. If the
The resource allocation in such underlay multi-cluster IoT
IoTDs control their introduced interference perfectly, the
network is a crucial process in optimizing the performance
UE achieves the required QoS (i.e., u ) by scaling its own
of the network. Therefore, suitable uplink resources
transmit power, Pm , by a factor of κ. In other words, in
allocation need to be controlled by eNB to manage the
the absence of interference from the IoTDs, the UE link
interference between UEs and IoTDs and between IoTDs
achieves κ u signal-to-interference ratio (SIR), or SINR
themselves to achieve the QoS constraints for the admitted
with zero interference, as follows:
u nodes. The author in [3], formulated an optimization
PeNB
≥ κ u , (5) problem that maximizes the achievable throughput gain in
No a single IoT group network where one IoTD is reusing the
where (PeNBu = κ u No ) represents a lower bound on the subcarrier of each UE. In the following sub-section, we
received power at the BS from each UE. We consider the formulate an optimization problem for supporting multiple
mth UE to be a candidate partner for the K IoTDs only IoT groups and we propose a low complexity suboptimal
if the following QoS interference constraints are satisfied solution using a greedy iterative matching (GIM) algorithm.
(rmu ≥ and r d ≥ ∀ IoTDs paired with the UE) where
u n,k g
rmu and r d are the achievable SINR of the mth UEs and 3.1 Problem formulation
n,k
the nth IoTDs in the k th IoTG which are given as follows,
respectively: The objective of the proposed joint resource and power
u allocation problem is to find the sub-optimal transmit powers
PeNB ∗ | ∀ n ∈ N and k ∈ K} and pairing coefficients
u
rm = , (6) Pnk = {Pn,k
Imu + No ∗
{am,n,k } of all admitted paired IoTDs that maximize the total
L (dn,k,k )−α Pn,k
g cell throughput and achieve the QoS constraints for all nodes.
d
rn,k = d +N
, (7) The problem can be mathematically formulated as follows:
In,k o
m
where Imu and In,kd are the interference from all IoTDs maximize T (am,n,k , Pn,k
m
),
am,n,k , Pn,k
that reuse the mth UE subcarrier from all IoTGs, and the
subject to C1 : rm
u
≥ u , ∀ m ∈ M ,
interference to the nth IoTD on the k th IoTG from the mth
C2 : rn,k
d
≥ g , ∀ n ∈ N and k ∈ K,
UE and all other IoTDs from other groups, which are given
C3 : am,n,k ∈ {0, 1} ∀ m, n, k,
as follows:
N
K
N C4 : am,n,k ≤ 1, ∀ m ∈ M and k ∈ K,
b −α
Imu = am,n,k L (dn,k ) Pn,k , (8) n
k n
M
C5 : am,n,k ≤ 1, ∀ n ∈ N and k ∈ K,
−α
N
K g
dm,k m
d
In,k = am,n,l L (dn,l,k )−α
g
Pn,l +L u
PeNB . C6 : Pn,k
m
≤ Pdmax , ∀ m, n, k,
b
dm
n l =k (12)
(9) where
Now, we would like to find the upper-bound of the
M
M
K
N
m
interference to the cellular user by considering the SINR T (am,n,k , Pn,k )= log2 1 + rm
u
+ am,n,k log2 1 + rn,k
d
.
u m m n
QoS constraint (rm u = PeN B ≥ ) where the upper bound k
Imu +No u (13)
is given as as follows:
Constraints C1 and C2 ensure the QoS of all UEs and
Imu ≤ No (κ − 1). (10) admitted IoTDs. The pairing coefficients in C3 is set to 1
Hence, without losing generality, if we assume an equal if the RB of the mth UE is assigned to the nth IoTD in
transmit power level of all IoTDs assigned to the mth UE, the k th IoTG and 0 otherwise. Constraints C4 are used to
Mobile Netw Appl
make sure that the maximum number of IoTDs in each IoTG (12) in order to propose our solution. The proposed match-
reusing the subcarrier of any UE is one to avoid severe ing algorithm is a many-to-one matching at which multiple
harmful interference between IoTDs within the same IoTG. IoTDs (from different IoTGs) are allowed to be paired with
Constraints C5 are used to make sure that each IoTD is one cellular user. Since there is only one IoTD from each
reusing only one UE. The group of constraints in C6 is IoTG is allowed to reuse the spectrum of each UE, there is
to limit the assigned powers to all IoTDs to the maximum at most N K possible combinations for each UE to consider.
allowed power level Pdmax . This resource allocation problem This leads us to define an M ×N K cost matrix E K = [em,c(v)
K ]
is a non-convex, mixed integer non-linear program. The K
whose element em,c(v) represents the throughput achieved
complexity of an exhaustive search algorithm to solve this when the subcarrier of the mth UE is shared with K IoTDs
problem is O((ML)KN ) and hence it is NP-hard for the in the combination c(v). c(v) represents one possible com-
large values of M, L, N, and L, where L is a finite set of bination out of the N K possible combinations that consists
available power levels. of only one device from each IoTG and v is the index of
this combination where v ∈ {1, 2, ..., N K }. The throughput
3.2 Matching algorithm K
element, em,c(v) , is given as follows:
where
u the well-known weighted bipartite matching technique is
PeN
K
T Em,c(v) = log2 1+ N
B
b −α P
not applicable as IoTDs will be paired with multiple UEs
No + L K
k n fn,k (dn,k ) n,k
⎛ ⎞ which violates C5 . Therefore, we propose a Greedy Iterative
⎜ ⎟
Matching algorithm (GIM) as shown in algorithm 1 which
K
L (dn,k,k )−α Pn,k
N g m
⎜ ⎟ is introduced in the following sub-section.
+ fn,k log2 ⎜1+ g −α ⎟, (15)
⎝ dm,k ⎠
k n N o + I1 + L d b u
PeN B
m
3.3 Greedy iterative matching algorithm (GIM)
N
K
fn,l L (dn,l,k )−α Pn,l ,
g
I1 = The GIM algorithm is proposed to provide a low complexity
n l =k suboptimal solution of problem (12) iteratively by pairing
and the maximum allowable number of IoTDs with each UE
(i.e., K IoTDs) in step 2 and 3 using the initial cost
1 , if n, k exists in combination c(v),
fn,k = (16) matrix E k with k = K. In steps 4 to 7, we propose a
0 , otherwise.
greedy matching to get a sub-optimal solution that satisfies
where fn,k is a selector function that equals 1 when the the constraint C5 and guarantee the uniqueness of the
nth IoTD in the k th IoT-GW is paired (i.e., exists in the set of IoTDs. We iteratively check the IoTDs used in
combination c(v) for which the sum throughput is being best combination in E k and match the UE in this best
calculated). For example, if we have K = 2 IoTGs with combination to the corresponding IoTDs. Then, we set all
N = 2 IoTDs per group, the possible combinations of other elements in E k to zero that use the same UE and
any UE will be c = {(1, 1), (1, 2), (2, 1), (2, 2)}, where IoTDs in Step 8. We also remove those nodes from the lists
K
em,c(1) = em,(1,1)
K K
, em,c(2) = em,(1,2)
K , and so on. Notice of unpaired nodes. If any UEs are still unpaired, this means
that all IoTDs are using the same transmit power level, that those UEs cannot afford the interference introduced by
Pn,k = Pn,l = Pn,k m , given in (11). By inspecting
K IoTDs, i.e., the QoS constraints are violated. If any IoTDs
the the elements of E K , it is obvious that they are not are still unpaired, we iteratively repeat the process and try
unique in terms of the individual IoTDs since each IoTD to pair a less number of IoTDs to the unpaired UEs until all
appears in N combinations in each row. Consequently, UEs are paired or all IoTDs are paired.
Mobile Netw Appl
Parameter Value
M 50
N 50
K 1:2
u 10:25 dB
g 5:20 dB
α 4
κ 3 dB
L 10−2
d
Pmax 23 dBm
u
PeN B
No 20 dB
500
400
10 log D
300
200
100
0
1 2 3 4 5 6 7 8 9 10
Number of IoTGs
each UE and hence, increases the computation complexity. Notice that the worst case computational complexity of
Figure 2 shows the variation of D as a function of the the proposed algorithm can be written as:
number of IoTGs for different values of the permitted
number of IoTDs to be paired with each UE from each
M−2
M NK + M N K − i N K −[(M −i) (2 (N −i) + 1)]
IoTG (i.e., j). On the other hand, pairing only one IoTD
i=1
from each IoTG for each UE (i.e., j = 1) has a significant
(18)
increase in the throughput compared with the single cluster
scenario in [3] with much less complexity than higher values which can be reduced to O(M N k ),
where M is the number
of j . Therefore, we assume a pairing scheme where at most of cellular users, N is the number of IoT devices in each
one IoTD from each IoTG is paired with each UE if the cluster, and K is the number of clusters.
QoS constraints are satisfied. Notice that the dimension D In order to reduce the computation complexity, we could
increases also with the increasing number of IoTGs per cell. assume that IoTDs that are closer than a certain threshold
IoTDs/Cluster is 50
180
d
P = 7 dB m
160 IoT-GW eNB Distance=800m
d IoT-GW eNB Distance=600m
P = 0 dB IoT-GW eNB Distance=400m
m
140
120
0 5 10 15 20
Power (dB m)
Mobile Netw Appl
80
60
40
20
0
1 5 10 15 20 25 30 35 40 45 50
Number of IoTDs/IoTG
distance to the BS are not allowed to be reuse partners. This all IoTDs. therefore, any IoTD with a distance larger than
assumption gives a two-fold advantage, first, it helps in avoid- this threshold is allowed to be reuse candidate which can be
ing severe interference at the BS and reduce the number of recast as follows:
combinations. This threshold distance can be lower-bounded
u LP 1/α
using the high-SNR approximation of (6) as follows: K rm d
b
dn,k ≥ dth = u . (20)
PeNB
u
PeNB
u
rm ≤ , (19) If we assume a threshold distance equal to dth , 1.1 dth and
K L Pd (dth )−α
1.2 dth , the matching algorithm execution time is reduced
by 48%, 55% and 61%, while the throughput gain and the
where the maximum number of allowed IoT inter- access rate for a threshold distance dth are almost the same
feres to
each UE from all groups are assumed without taking a threshold (this will be explained later in
(i.e., k=1 N
K
n=1 am,n,k = K) and equal transmit power for Figs. 4 to 9).
80 %
70 %
Access Rate
60 %
50 %
40 %
60
40
20
0
1 5 10 15 20 25 30 35 40 45 50
Number of IoTDs/IoTG
On the other hand, since we assume an equal power when the IoTGs radii are 100 m. Both figures compare
allocation of all IoTDs reusing the same UE, we need to the single IoT group ORA algorithm, the proposed GIM
m = P ) that achieves
determine a sub-optimal value (Pn,k d algorithm for single IoT group scenario (K =1 GIM), and
the maximum throughput. In Fig. 3, we adopted numerical the proposed GIM algorithm for double IoT groups (K = 2
simulation to find the power level that maximizes the GIM). The simulation results show that the cell throughput
throughput gain within the allowed range of values (i.e., Pd gain and IoTDs accessibility increase with increasing the
from 0 : 23 dBm ) for different distances between eNB and distance between eNB and IoT-GWs due to the reduced
IoTD-GWs. The simulation results show that increasing Pd interference from IoTDs. Figure 4 reveals that throughput
improves the throughput gain until a certain value where the gain improves with the increases in the number of IoTDs
gain saturates due to the introduced interference. due to the availability of much more combination of
Figures 4 and 5 illustrate the effect of changing the IoTDs that to select from. On the other hand, the results
distance between any IoT-GW and eNB (400, 600, and 800 in Fig. 5 show that the access rate drops with the
m) on the cell throughput gain and accessibility of IoTDs increase of the number of deployed IoTDs due to the
40 %
30 %
20 %
10 %
0
1 5 10 15 20 25 30 35 40 45 50
Number of IoTDs/IoTG
Mobile Netw Appl
100
80
60
40
20
0
1 5 10 15 20 25 30 35 40 45 50
Number of IoTDs/IoTG
assumption of a constant number of available UEs where eNB (600 m). The simulation results show that the cell
each UE is limited to be paired with at most K IoTDs. throughput gain and IoTDs accessibility increase with
Additionally, the proposed GIM algorithm with K=2 has a decreasing the radii of the IoTGs. Decreasing the radius
better performance in terms of both of the cell throughput reduces the interference from IoTDs to eNB which increases
gain and accessibility of IoTDs compared to the single the number of paired IoTDs and the total cell throughput.
IoTG scenario. The simulation results show the effect of Figures 8 and 9 show the effect of the QoS constraints
setting a distance threshold for the multiple IoT group of the IoTDs on the throughput gain and the access rate for
scenario where the throughput gain and the access rates a fixed distance between IoT-GW and eNB (600 m) and
drop slightly when we increase the threshold distance. The IoTGs radii of 100 m. the simulation results show that a
effect appears when there is a large number of IoTDs in lower SIR increases cell throughput gain and accessibility of
the system. IoTDs, since decreasing SIR constrains for IoTDs increases
Figures 6 and 7 show the effect of the IoTGs radii (100, the probability of pairing larger number of IoTDs while
200, and 300 m) on the throughput gain and the access reducing interference of IoTDs on eNB and increasing the
rate for a fixed distance between the IoT-GW and the cell throughput gain.
60 %
50 %
40 %
30 %
20 %
10 %
0
1 5 10 15 20 25 30 35 40 45 50
Number of IoTDs/IoTG
Mobile Netw Appl
100
50
0
20 40 60 80 100 120 140 160 180 200
Number of UEs
Based on the previously presented results, the performance throughput gain and IoTDs access rate compared to the
of the proposed GIM algorithm outperforms the optimal single IoT group deployment scenarios. The proposed
resource allocation (ORA) in [3] as a baseline in terms of analysis and simulation results show that setting a lower-
both of the cell throughput gain and accessibility of IoTDs bound on how close the allowed IoT reuse partners reduces
in the cases of a single cluster and two clusters as well. the computation complexity while preserving sufficient
In Fig. 10, the performance of the GIM algorithm is performance gain. As future research directions, we would
further compared to two multi-cluster based algorithms, like to investigate the joint EE and number of admitted
namely the fixed-radio resource sharing algorithm (F- IoT devices maximization in addition to the investigation of
RRSA) [5] and the adaptive radio resource sharing other matching algorithms to improve performance.
algorithm (A-RRSA) [6]. F-RRSA aims to maximize the
sum rate with a computational complexity of O(M 3 +KM 3 ) Acknowledgements The authors would like to thank Prof. Mostafa M.
while A-RRSA is built upon F-RRSA to consider the trade-off Fouda for his support and help in preparing the reply letter and the
between fairness and data rate with a higher computational revised manuscript.
complexity due to the adoption of proportional integral
control (PID). The simulation parameters in Fig. 10 are Funding Information The authors would like to thank Prof. Mostafa
adopted from [5, 6] for fair comparison which are given as M. Fouda for his support and advice. This research was supported
by Benha University research fund (Project 2/1/14). This work is
follows: the radius of eNB coverage is 500 m, the cluster partially supported by the China NSFC Grant (61872248, U1736207),
radius is 70 m, the distance between eNB and IoT-GW is Guangdong Natural Science Foundation 2017A030312008, Fok Ying-
100 m, the maximum transmission power of UEs is 24 dBm , Tong Education Foundation for Young Teachers in the Higher
and the D2D transmission power is 5:15 dBm . Moreover, Education Institutions of China (Grant No.161064) and GDUPS
(2015). This work is partially supported by the project “PCL
the number of devices per cluster is 50 % of the number of Future Regional Network Facilities for Large-scale Experiments and
UEs. The simulation results show that the proposed GIM Applications (PCL2018KP001).
algorithm outer performs the two other algorithms, while
the F-RRSA achieves better performance compared to the
A-RRSA algorithm. Appendix: Numerical Example of the GIM
Algorithm
5 Conclusion
22. Phunchongharn P, Hossain E, Kim DI (2013) Resource allocation 31. Vo N, Duong TQ, Guizani M, Kortun A (2018) 5g optimized
for device-to-device communications underlaying LTE-advanced caching and downlink resource sharing for smart cities. IEEE
networks. IEEE Wireless Commun 20(4):91–100 Access 6:31,457–31,468
23. Pradhan A, Basu S, Sarkar S, Mitra S, Roy SD (2018) 32. Vo N, Duong TQ, Tuan HD, Kortun A (2018) Optimal video
Implementation of relay hopper model for reliable communication streaming in dense 5g networks with D2D communications. IEEE
of IoT devices in LTE environment through D2D link. In: Proc Access 6:209–223
COMSNETS, pp 569–572 33. Wang B, Chen L, Chen X, Zhang X, Yang D (2011) Resource allo-
24. Ratasuk R, Vejlgaard B, Mangalvedhe N, Ghosh A (2016) NB-Iot cation optimization for device-to-device communication underlay-
system for M2M communication. In: Proc IEEE WCNC, pp 1–5 ing cellular networks. In: Proc IEEE VTC spring, pp 1–6
25. Rebecchi F, de Amorim MD, Conan V, Passarella A, Bruno R, 34. Wu G, Talwar S, Johnsson K, Himayat N, Johnson KD (2011)
Conti M (2015) Data offloading techniques in cellular networks: M2m: from mobile to embedded internet. IEEE Commun Mag
a survey. IEEE Commun Surveys Tuts 17(2):580–603 49(4):36–43
26. Ren Y, Chuai G (2017) Low-complexity uplink resource 35. Yang X, Wang X, Wu Y, Qian LP, Lu W, Zhou H (2018) Small-
allocation algorithm based on hypergraph clustering for d2d cell assisted secure traffic offloading for narrowband internet of
communications. In: Proc IEEE PIMRC, pp 1–6 thing (nb-iot) systems. IEEE Internet Things J 5(3):1516–1526.
27. Shah SH, Yaqoob I (2016) A survey: internet of things (IoT) https://doi.org/10.1109/JIOT.2017.2779820
technologies, applications and challenges. In: Proc IEEE SEGE, 36. Yang Z, Huang N, Xu H, Pan Y, Li Y, Chen M (2016)
pp 381–385 Downlink resource allocation and power control for device-
28. Sheng Z, Tuan HD, Duong TQ, Poor HV (2017) Joint to-device communication underlaying cellular networks. IEEE
power allocation and beamforming for energy-efficient two- Commun Lett 20(7):1449–1452
way multi-relay communications. IEEE Trans Wireless Commun 37. Yu C, Tirkkonen O, Doppler K, Ribeiro C (2009) Power
16(10):6660–6671 optimization of device-to-device communication underlaying
29. Sheng Z, Tuan HD, Duong TQ, Poor HV, Fang Y (2018) Low- cellular communication. In: Proc IEEE ICC, pp 1–5
latency multiuser two-way wireless relaying for spectral and
energy efficiencies. IEEE Trans Signal Process 66(16):4362–4376
30. Song X, Han X, Ni Y, Dong L, Qin L (2019) Joint uplink and
downlink resource allocation for D2D communications system. Publisher’s Note Springer Nature remains neutral with regard to
Future Internet 11(1):12 jurisdictional claims in published maps and institutional affiliations.