1

Performance Analysis of Multi-user NOMA

Wireless-Powered mMTC Networks:
A Stochastic Geometry Approach
Thanh-Luan Nguyen, Tri Nhu Do, and Georges Kaddoum, Senior Member, IEEE.

Abstract—In this paper, we aim to improve the con- Project (3GPP)’s Release 16 New Radio (NR) [1],
arXiv:2201.04784v2 [eess.SP] 8 Oct 2022

nectivity, scalability, and energy efficiency of machine- and will continue to be an essential component of
type communication (MTC) networks with different types the sixth generation (6G) of wireless networks [2].
of MTC devices (MTCDs), namely Type-I and Type-
For instance, the deployment of MTC for low-power
II MTCDs, which have different communication pur-
poses. To this end, we propose two transmission schemes applications enables long-distance communications
called connectivity-oriented machine-type communication with relaxed requirements on data rate and latency
(CoM) and quality-oriented machine-type communication between the core network and a large number of
(QoM), which take into account the stochastic geometry- battery-powered devices [3], [4].
based deployment and the random active/inactive status MTC networks that consist of a large number of
of MTCDs. Specifically, in the proposed schemes, the
MTC devices (MTCDs) are referred to as massive
active Type-I MTCDs operate using a novel Bernoulli ran-
dom process-based simultaneous wireless information and MTC (mMTC) [5]. MTCDs can be categorized into
power transfer (SWIPT) architecture. Next, utilizing multi- two classes based on their demands and character-
user power-domain non-orthogonal multiple access (PD- istics [6], namely one that operates continuously
NOMA), each active Type-I MTCD can simultaneously and has deterministic traffic demands, and another
communicate with another Type-I MTCD and a scalable that operates in a probabilistic manner has an event-
number of Type-II MTCDs. In the performance analysis driven appearance and has sporadic traffic demands.
of the proposed schemes, we prove that the true distri-
bution of the received power at a Type-II MTCD in the
To handle the coexistence of different classes of
QoM scheme can be approximated by the Singh-Maddala MTCDs in the same mMTC network, group-based
distribution. Exploiting this unique statistical finding, we mMTC networks, in which diverse MTCDs are
derive approximate closed-form expressions for the outage divided into groups based on different criteria, such
probability (OP) and sum-throughput of massive MTC as location, delay, and traffic characteristics, have
(mMTC) networks. Through numerical results, we show been proposed [7], [8].
that the proposed schemes provide a considerable sum-
throughput gain over conventional mMTC networks.
In this paper, we focus on five major mMTC
issues. First, as pointed out in [9]–[13], the prac-
Index Terms—Massive machine-type communications tical deployment and implementation of mMTC
(mMTC), non-orthogonal multiple access
networks on a large scale faces a number of chal-
(NOMA), wireless-powered communications, Bernoulli
process, stochastic geometry, outage probability, network lenges due to the intensive communications between
throughput. MTCDs, including congestion, coverage, energy
self-sufficiency, scalability, and modelling. Conse-
quently, advanced multiple access techniques, such
I. I NTRODUCTION as non-orthogonal multiple access (NOMA), have
Machine-type communication (MTC) has been been used to alleviate congestion caused by massive
a key enabler of the Third Generation Partnership signalling overhead when a large number of MTCDs
are attempting to access the spectrum [9], [12],
T.-L. Nguyen is with the Faculty of Electronics Technology,
Industrial University of Ho Chi Minh City, HCMC 700000, Vietnam [13]. Second, because MTCDs rely heavily on short
(e-mail: nguyenthanhluan@iuh.edu.vn). transmissions for instantaneous communication, the
T. N. Do and G. Kaddoum are with the Department of Elec- network sum-throughput can be drastically reduced
trical Engineering, the École de Technologie Supérieure (ÉTS),
Université du Québec, Montréal, QC H3C 1K3, Canada (emails: tri- without efficient resource allocation algorithms.
nhu.do@etsmtl.ca, georges.kaddoum@etsmtl.ca). To overcome this barrier, group-based machine-to-
 2

machine communications, i.e., MTC according to we define the random active and/or inactive sta-
3GPP standards [14], can be used for effective tus of MTCDs as follows: the locations of active
radio resource management and spectrum sharing, or inactive Type-II MTCDs form a homogeneous
resulting in increased network throughput. Third, in Poisson point process (HPPP), and their number
order to establish communication, MTCDs can use within a close vicinity of each Type-I MTCD is
up their limited energy, which can be harmful to Poisson distributed, as will be explained in detail
mMTC networks [13]. Therefore, wireless-powered in the following section. Following that, we propose
MTC networks have been proposed in the litera- Bernoulli-based energy harvesting (BEH) protocols,
ture as a solution to this problem [9], [15]–[17]. in which energy harvesting is performed based on
However, as the number of devices grows expo- the demand of Type-I MTCDs. In addition, we use
nentially, a more powerful wireless-powered system multi-user power-domain NOMA (PD-NOMA) to
designed specifically for MTC is required. Fourth, improve connectivity by allowing a Type-I MTCD
because of the ubiquitous coverage of MTCDs, to serve other Type-I and Type-II MTCDs at the
their locations are typically considered random [11], same time. In particular, we use stochastic geometry
making modelling the spatial distribution of the (SG) to capture the random spatial distribution of a
MTCDs deterministic, unappealing, and unscalable. set of active and inactive MTCDs to group MTCDs
A few recent works on stochastic mMTC networks in this phase.
propose modelling the deployment of mMTC using
stochastic geometry as a solution to this critical
problem [18]–[20]. Fifth, transmissions between Recently, downlink mMTC and downlink
distant MTCDs may suffer from severe data losses mMTC-NOMA have been gaining much attention
due to multipath fading and intra-interference from from the research community [21]–[24]. The
other MTC devices, eventually reducing the network authors in [21] proposed a joint subcarrier
throughput. To address this issue, deploying several and power allocation strategy to maximize the
regenerative connections among the MTCDs in a number of connected devices in IoT network
multi-hop fashion can increase network coverage consisting of MTCDs using NOMA in the
and, as a result, network sum-throughput and con- downlink. Recently, the authors in [23] proposed
nectivity. a wireless power transfer-assisted NOMA (WPT-
NOMA) transmission scheme for mMTC. The
authors of [24] integrated rate-splitting multiple-
A. Novel Communication Strategies for Diverse, access (RSMA) with time-division-duplex Cell-
Group-based mMTC Networks free massive MIMO for downlink mMTC. In
To address the aforementioned issues and im- addition, it is envisioned that envisioned that
prove the scalability of MTC networks and mMTC can be implemented in millimeter Wave
energy efficiency at the MTCDs, we propose (mmWave)/Terahertz (THz) wideband in 6th
two transmission schemes as follows. In the Generation (6G) wireless networks [25]. As a
case of a high power budget, we propose the result, a distributed multi-hop relaying is likely to
connectivity-oriented machine-type communication be practical for such networks. In particular, the
(CoM) scheme, which serves all Type-II MTCDs in authors in [26] proposed routing techniques for
the vicinity, and in the case of a low power budget, mmWave networks, namely minimum hop count
the quality-oriented machine-type communication (MHC) and nearest Line-of-Sight (LoS) relay to the
(QoM) scheme, which opportunistically serves a destination with MHC (NLR-MHC). It is noted that
single Type-II MTCD. Specifically, we consider two [22] has introduced a multi-hop mMTC decode-
types of MTCDs as follows: MTCDs with seamless and-forward (DF) relaying system with the digital
operation in a multi-operator network [2], called enhanced cordless telecommunication new radio
Type-I MTCDs, which are able to store-and-forward (DECT-2020 NR) standard. However, regenerative
messages through a series of regenerative relaying mMTC-NOMA with multi-hop infrastructure has
transmissions; MTCDs with consistent connectivity not been given out in literature. As a result, its
[2], called Type-II MTCDs, which operate under theoretical work on the outage probability (OP)
stringent instantaneous QoS requirements. Notably, and throughput analysis are still open issues.
 3

B. Related Works massive number of devices sending very short pack-

ets. Moreover, the authors in [31] studied the appli-
In this subsection, we elaborate on recent re- cation of NOMA in MTCDs to increase network
lated works on NOMA, simultaneous wireless in- connectivity. Particle swarm optimization (PSO)
formation and power transfer (SWIPT), and their was implemented to improve the sum-throughput of
applications to MTC/mMTC networks. In the lit- the proposed group-based MTCD-NOMA networks.
erature, WPT has been considered for mMTC. To enhance the coverage of MTC networks, the
For instance, the WPT-NOMA offers an alterna- authors in [32] proposed a multi-hop cooperative
tive for energy-constrained devices [23], [27], [28]. transmission where signaling interaction mecha-
The authors in [27], [28] proposed integrating the nisms were developed and the relevant structure
MTCDs with WPT technologies. In particular, en- of frames was modified to be compatible with
ergy harvesting (EH) from ambient radio frequency current Wi-Fi architectures. In [33], the authors used
(RF) sources is utilized for prolonging the lifespan cooperative relaying-based device-to-device (D2D)
of energy-constrained MTCDs. In [23], a WPT- communications to improve the transmission range
NOMA transmission scheme for mMTC networks and coverage of MTC networks. In [34], considering
is proposed for uplink and downlink communica- the coverage and average rate analysis of multi-
tions. The proposed WPT-NOMA enables MTCDs hop MTC networks using stochastic geometry, the
to simultaneous EH and information transmission, authors showed that multi-hop D2D communication
which is crucial for the support of massive number is able to increase the coverage and average rate
of energy-constrained MTCDs. For high connectiv- of MTC networks. Note that the usage of NOMA
ity, yet energy-efficient and self-sustainable mMTC was overlooked in [34] and [33]. In [35], the au-
networks, SWIPT-NOMA has been considered as thors proposed multi-hop transmission to enhance
a key enabling technology [15]. The combination the scalability and energy efficiency of wireless
of SWIPT and NOMA (SWIPT-NOMA) is a po- software-defined-networking (SDN)-based mMTC
tential solution to improve the spectral and energy networks. Recently, in [36], the authors investigated
efficiency (EE) of fifth-generation (5G) networks the transient behavior of multi-hop wireless trans-
and beyond, especially in order to support the mission in MTC networks under static routing and
functionalities of the Internet-of-Things (IoT) and showed that such a multi-hop scheme can be applied
MTC scenarios [16]. Specifically, PD-NOMA has to mMTC.
been shown to be a promising technology for im- For vast numbers of devices in mMTC networks,
proving the connectivity of MTC networks [9]. The deterministic pathloss models fail to capture the
outage performance of relay selection strategies in spatial randomness of the MTCDs. Stochastic ge-
SWIPT-NOMA under the assumption of residual ometry (SG) is an important mathematical tool used
hardware impairments and channel estimation errors to capture the spatial distribution of these devices.
was studied in [17]. In [33] and [18], HPPPs were adopted to model the
In [8], the authors focused on the diverse and network density and the MTCDs’ locations. Specif-
scalable connectivity of MTCDs by proposing two ically, the work in [33] studied the joint spectral
distinctive types of devices in MTC networks. In efficiency and fairness in an underlay hybrid SG-
the case of different types of MTCDs coexisting based network consisting of MTC, D2D-enabled,
in the same network, the authors in [29] pro- and cellular devices. Moreover, the work in [18]
posed sophisticated scheduling algorithms to set up considered an SG-based uplink NOMA ultra-dense
connections for different active MTCDs. Recently, network consisting of human-type communication
random access with layered preambles (RALP) has (HTC) and MTC devices. In this work, the authors
been proposed to support devices with two different aimed to ensure connectivity and maintain quality-
priorities based on the concept of PD-NOMA [30]. of-service for a massive number of MTCDs. The
In addition, the work in [5] introduced PHY and results showed a significant increase in network
MAC layer solutions developed for the mobile and throughput due to the large number of connections
wireless communications enablers for the twenty- from MTCDs. Unlike the above works, the authors
twenty information society (METIS) project to ad- in [19] used Matérn cluster processes to model the
dress the scalability and efficient connectivity of a wireless-powered MTC networks.
 4

C. Contributions of the Paper the performance of the proposed schemes in

mMTC networks.
The main challenges in the performance analysis • Through numerical results1 , we obtain new
of the proposed transmission schemes, which are the findings for the considered mMTC networks
QoM and CoM schemes with EH at Type-I MTCDs, as follows. By using either the CoM or QoM
are highlighted as follows. In the considered multi- schemes, the network sum-throughput of Type-
user PD-NOMA wireless-powered mMTC network, I and Type-II MTCDs increases, whereas the
determining exact formulas for the end-to-end (e2e) outage performance of Type-I MTCDs de-
outage probability (OP) can be an intricate task. creases. Thus, there is a trade-off between
This is mainly because the e2e OP of each individ- the achievable performance of each type of
ual Type-I MTCD transmission is heavily dependent MTCD; i.e., the network performance improve-
on the harvested power and the performance of the ment comes at the cost of some degradation for
previous hops. Specifically, as the number of tra- Type-I MTCDs.
versed regenerative connections increases, the com- • In addition, the maximum reliable transmission
plexity of the analytical derivations also becomes rate (MRTR), which is a function of the PD-
noticeable. Second, evaluating the performance of NOMA power allocation coefficients and the
mMTC networks can be challenging. This is mainly probabilistic EH splitting parameters, is the
because the performance of an MTCD is determined key factor for balancing the performance trade-
by the device’s QoS requirements. As a result, off between different types of MTCDs in the
obtaining the instantaneous QoS requirements of all network.
MTCDs is essential for performance analysis, which • Finally, the implementation of PD-NOMA in
becomes increasingly challenging as the number of multi-hop-based networks can cause an out-
MTCDs grows. The key contributions of our paper age performance degradation by allocating a
can be summarized as follows: portion of the transmission power to serving
• To address the aforementioned problems as- extra MTCDs instead of utilizing full power
sociated to mMTC networks, we propose two for relaying information. On the other hand,
novel transmission schemes, namely the CoM by enabling additional connections, the overall
and QoM schemes, which have not been re- network connectivity and sum-throughput can
ported in the literature. be effectively improved. It is noted that SG-
• Moreover, we consider the random ac- based NOMA networks heavily rely on the
tive/inactive status of Type-II MTCDs and their network density and the spatial separations
stochastic geometry-based locations, which re- among the paired devices. As a result, even
flects a practical realization of mMTC net- when numerous Type-II MTCDs are available,
works. Type-I MTCDs only schedule those that are
• To investigate the performance of the pro- suitably distanced apart to execute NOMA
posed schemes, we provide closed-form ex- transmissions.
pressions for the network OP, sum-throughput, The rest of this paper is organized as follows:
and maximum reliable transmission rate for Section II presents the system model including the
the proposed schemes. Specifically, the total HPPP-based deployment of the MTCDs and the
probability theorem is adopted not only to fading channel model. In Section III, we introduce
relax the recursive formulations of the trans- the proposed probabilistic EH architectures and the
mission power from Type-I MTCDs, but also proposed transmission schemes. The performance
to tackle the joint statistics of multiple non- analysis of the proposed schemes in terms of OP and
mutually-exclusive events. The results are then throughput are presented in Section IV. Section V
expressed in terms of Meijer G-functions and shows the asymptotic analysis. Section VI presents
Fox H-functions, where the Singh-Maddala the numerical results. Finally, Section VII concludes
distributions are adopted to assist the analysis
of QoM-based schemes. The obtained closed- 1
It is noted that our results are reproducible using our Mat-
form expressions are validated through simu- lab code, which is available at https://github.com/thanhluannguyen/
lations, which provide technical insights into mMTC-WPT-NOMA
 5

this work. A. Homogeneous Poisson Point Process-based

Notations: CN(µ, σ 2 ) denotes the circular complex mMTC Network Deployment
Gaussian random variable with mean µ and variance Considering a square area of x × x [m2 ] in a
σ 2 , γ(a, x) denotes the lower incomplete Gamma two-dimensional (2D) Cartesian coordinate system,
function [37, Eq. (8.350.1)]. In addition, 0v and D[I] s are randomly distributed over the network area,
1v denote vectors containing v elements of zeros and their locations are assumed to be fixed over the
and ones, respectively. Moreover, Pr[·] denotes the transmission from one D[I] to another D[I] . Over the
[I]
probability operator, FX (x) = Pr[X < x], F̄X (x) , circular coverage area Ct with radius rt around Dt ,
Pr[X ≥ x], and fX (x) are the cumulative dis- [II]
the active and inactive D s are modeled by HPPPs
tribution function (CDF), the complementary CDF [1] [0] [1]
Φt and Φt with densities per meter square λt
(CCDF), and the probability density function (PDF) [0]
and λt , respectively. Thus, the number of active
of the random variable (RV) X. and inactive Type-II MTCDs within the proximity
[I]
of Dt are modeled by Poisson distributions with
[1] [0]
mean intensities πλt rt2 and πλt rt2 , respectively.
II. S YSTEM M ODEL As a result, the null probability of the 2D Poisson
[i]
process Φt can be expressed as
Let us consider an mMTC network, where a
[I]
h i [i] 2
Type-I MTCD D1 plans to communicate with the [i]
Pr |Φt | = 0 = e−λt πrt , (1)
[I]
Type-I MTCD DM , as depicted in Fig. 1. Due to
spatial separation, D1 requires the help of other where i ∈ {1, 0}. Next, the coverage area Ct
[I]

Type-I MTCDs, namely D2 , D3 , . . . , DM −1 , to can be equally divided into Kt disjoint subareas,

[I] [I] [I]

[I] namely Ct,1 , Ct,2 , . . . , Ct,Kt . Each area represents

store-and-forward the intended message to DM via [I]
an annulus-shaped area around Dt with inner and
multi-hop regenerative relaying transmission. In the
outer radiuses of k−1 Kt t
r and Kkt rt , respectively. Ac-
proposed network, there coexists a massive num- [II]
ber of active and inactive Type-II MTCDs, as de- cordingly,[i]D s within each Ct,k construct a thinned
picted in Fig. 1. Specifically, Type-II MTCDs can HPPP (Φt,k ) from the independent thinning of the
[i]
be viewed as IoT devices that are non-stationary, parent HPPP (Φt ) where the thinning probability is
and stochastically spatially distributed around Type- K1t . Hence, the average number of D[II] s within Ct,k
[i]
I MTCDs; in addition, a Type-I MTCDs can be λ
follows Poisson distribution with intensity Ktt πrt2 .
viewed as an central node that connect other Type- [i]
II MTCDs by means of access via an aggrega- The null probability of Φt,k can then be expressed by
tion node or short-range device-to-device access, h i λ
[i]
[i] − Kt πrt2
as considered in [28]. To enhance the connection Pr |Φt,k | = 0 = e t . (2)
efficiency of the considered network, a portion of
It is noted that the number of active or inactive
the active Type-II MTCDs are served by Type-I
Type-II MTCDs is different from one annulus-
MTCDs. Specifically, during the t-th time slot, the [i]
[I] shaped area to another due to the density λt .
corresponding Type-I MTCD, denoted as Dt , can
employ PD-NOMA to serve up to Kt active Type-
II MTCDs situated nearby, in addition to relaying B. Fading Channel Statistics
[I]
information to Dt+1 , where 1 ≤ t ≤ M − 1. In In order to capture the propagation effects, let
[I]
addition, each Dt+1 can probabilistically perform us employ large-scale and small-scale models as
2 [I]
energy harvesting from the information forwarded follows. Let xm ∈ R denote the locations of Dm ,
by Dt and exploit the harvested power for infor- where 1 ≤ m ≤[I]M , and [I]lq , |xq−1 − xq | be the
[I]

mation transmission during the subsequent time slot. distance from Dq−1 to Dq , where 2 ≤ q ≤ M .
2
The EH behavior of Dt+1 can be characterized by Let xt,k ∈ R be the location of the k-th selected
[I]
[II] [II]
a Bernoulli RV, as detailed in section III. Hereafter, active D within the coverage area Ct , i.e., Dt,k ,
let us use D[I] and D[II] as acronyms for Type-I and thus dt,k , |xt − xt,k | specifies the distance from
[II]
Type-II MTCDs, respectively, and Dt,k to refer to D[I] [II]
t to Dt,k . Accordingly, the corresponding large-
the k-th paired D[II] during the t-th time slot. scale path-loss models are denoted by `(lm ) and
 6

QoM scheme

Probabilistic wireless
powered link

Type-II MTCD
Active: HPPP

Type-II MTCD

Fig. 1: Illustration of the proposed multi-user PD-NOMA probabilistic wireless-powered mMTC networks with stochastic
deployment of MTC devices.

`(dt,k ). Specifically, for the e2e transmissions, we

adopt the non-line-of-sight (NLoS) condition for the 2
( `(rϕt ) )− 
3GPP Urban Micro (UMi) path-loss model [38], 2
Fϕt,k (ϕ) = 1 −
[39] as `(x) = −Gr − Gt + 22.7 + 26 log10 (fc ) −  ( Kkt )2 − ( k−1 )2
   Kt
36.7 log10 (x/d0 ) [dB], where `(x) = L( dx0 ) in
 
2 ϕ k
linear scale with Gr and Gt being the receiving and × γ ,
 `(rt ) Kt
transmitting antenna gains, respectively, fc (GHz)     
2 ϕ k−1
denotes the carrier frequency,  = 3.67, d0 is the −γ , , ϕ > 0. (5)
reference distance in meter and  `(rt ) Kt

(fc )2.6
L = 186.2087 . (3)
100.1(Gr +Gt ) III. T HE P ROPOSED M ULTI - USER NOMA
P ROBABILISTIC WIRELESS - POWERED M MTC
Based on the spatial HPPP models, the Type-II
S CHEMES
MTCDs are uniformly distributed within the an-
nulus Ct,k [33], [18]. Accordingly, the PDF of the We propose Bernoulli random process-based en-
[I] [I]
distance from Dt to each Dt,k within Ct,k , denoted ergy harvesting architectures to power the Type-II
by fdt,k (r), can be expressed as MTCDs. It is worth noting that the data arrive at the
devices via a Bernoulli arrival process [40]. In heavy
traffic, devices use their own energy to process
 
2Kt r  k−1   k 
fdt,k (r) = 2 ζ r − rt − ζ r − rt , heavy data, whereas in light traffic, devices employ
rt Kt Kt
(4) EH to preserve their own energy. Therefore, in
our proposed transmission schemes, we characterize
where r > 0 and ζ(·) is the unit step function. the EH activity as a Bernoulli process. In addi-
For the small-scale propagation, we assume that tion, considering PD-NOMA, we proposed some
all channels experience Rayleigh fading, where transmission schemes for the considered mMTC
[I] [I]
the channel coefficients from Dq−1 to Dq and networks. In our proposed EH architectures, each
[I]
[I] [II]
from Dt to Dt,k are denoted as hq ∼ CN(0, 1) Dt+1 may or may not perform EH during the t-th
[I]
and gt,k ∼ CN(0, 1), respectively. Subsequently, the time slot. Let ρt+1 denote the probability that Dt+1
CDF of the joint large-scale and small-scale gain, performs EH where the Bernoulli trial It+1 ∈ {1, 0}
2 [I]
denoted as φq , |h  q | `(dq ), can
 be expressed as indicates the on/off EH status of Dt+1 . The value of
Fφq (φ) = 1 − exp − φ/`(dq ) , φ > 0. In addition, It+1 characterizes the probabilistic EH operation of
Dt+1 . Specifically, for It+1 = 0, Dt+1 will likely
[I] [I]
since the active Type-II MTCDs within Ct,k are uni-
formly distributed, |gt,k |2 `(dt,k ) , ϕt,k are mutually not perform EH with probability 1 − ρt+1 ; it will
independent and identically distributed following prefer to consume its inner battery to transmit
 with a
the CDF fixed transmission power P0 . Meanwhile, It+1 = 1
 7

[I]
implies that Dt+1 is likely to perform EH with potentially to the downlink D[II] s. Accordingly, the
[I] [I]
probability ρt+1 . It is noted that D1 and DM do harvested energy can be expressed as
not need to perform EH. Such probabilistic EH
Et+1 = ηt+1 βt+1 It+1 Pt φt+1
T
behavior can be modeled via a finite Bernoulli pro- . (8)
cess Ξ(ρ) = {I1 , . . . , It+1 , . . . , IM } with parameter
M −1
ρ , [ρ1 , . . . , ρM ], where ρt+1 , E[It+1 ]. Hence, the harvested power Pt+1 can be formu-
lated as

= βt+1 It+1 ηt+1 φt+1 Pt . (9)

A. Bernoulli Random Process-based Time- Et+1
Pt+1 = (M − 1)
Switching Energy Harvesting (BTEH) Architecture T
During the t-th time slot, the corresponding It is noted that both (7) and (9) are expressed in
[I] [I]
Dt+1 harvests energy from Dt for a duration of recursive form as a function of Pt . The extended
T
α
M t+1 t+1
I block time, where 0 ≤ αt+1 ≤ 1 is the expression of the transmission power during the t-
so-called time-switching or EH ratio, and T is th time slot is given in the following Remark.
the block time in which an information block is [I]
[I] [I] Remark 1. The transmission power of Dt , t ∈
transmitted from D1 to DM . Accordingly, the re-
(1−αt+1 It+1 )T [1, M − 1], depends on whether it adopts EH or
maining block time, M
, is then consumed
not. Accordingly,
[I] [I]
a) if It = 0 , Dt likely transmits with a fixed
for information transmission from Dt to Dt+1 . Fur-   [I]
[I]
thermore, by implementing NOMA, Dt can simul-
 thus Pt = P0 , with probability ρt ,
power,
taneously serve D[II] s while forwarding information b) if It = 1 and there exists τ  ∈ [1, t − 1]
that satisfies Iτ = 0 and Ij = 1 ,
[I] [I]
to Dt+1 . Accordingly, the harvested energy at Dt+1 
is expressed as ∀j ∈ [τ + 1, t], the
harvested
power
is
P = P Ω φ · · · Ω φ · · · Ω φ
t t ,
Et+1 = ηt+1 (Pt φt+1 )
αt+1 T
M −1
It+1 , (6)
t 0 τ +1 τ +1 j j
where Ωq = (M − 1)αq ηq /(1 − αq ) for BTEH
and Ωq = βq ηq for BPEH.
where 0 ≤ ηt+1 ≤ 1 denotes the energy conversion
[I] Accordingly, Pt can then be formulated as
efficiency, Pt denotes the transmission power of Dt ,
if It = 0 ,

and P0 is the fixed transmission power. Suppose
 
 P0 ,
I

that all the harvested energy is consumed by each P =   
τ = 0 ,
t
[I] t Q
source node, the harvested power at Dt+1 can be P0 Ωi φi , if 
I
 
expressed as i=τ +1 j = 1 , ∀j ∈ [τ + 1, t],
(10)
Et+1
(1 − αt+1 It+1 )T
Pt+1 = (M − 1) Consider a scenario where ρq = 1, ∀q ∈ [2, M ],
[I]
and therefore all D[I] s, except D1 , perform EH with
φt+1 Pt It+1 .
αt+1 ηt+1
= (M − 1) (7) a probability of one, thus the transmission power at
1 − αt+1
[I]
Dt can be expressed as
B. Bernoulli Random Process-based Power- t
Y t
Y t
Y
2
Splitting Energy Harvesting (BPEH) Architecture Pt = P0 |hi | `(di ) Ωi . (11)
i=1 i=1 i=1
During the t-th time slot, a proportion of the
received power, βt+1 It+1 Pt φt+1 , where 0 ≤ βt+1 ≤ Remark 2. It can be observed from (11) that
1 denotes the power-splitting or the EH ratio, is the harvested power Pt is contaminated by the
utilized for EH. Accordingly, the remaining power, path-loss in every i-th time slot, where i ∈ [1, t].
(1 − βt+1 It+1 )(Pt φt+1 ), is then exploited for infor- This phenomenon is familiar in traditional wireless-
mation transmission. Similar to the BTEH archi- powered regenerative relaying systems, where the
tecture, the (M − 1)-th time slot is the effective path-loss from previous paths accumulates at the
[I]
transmission duration in which the destination DM energy harvesting devices until, for some devices,
does not perform EH, and thus all received power is the harvested power reaches zero, leading to a
[I] [I]
exploited for transmission from DM −1 to DM , and severe outage.
 8

To tackle the above issue, one approach is to and with the BPEH architecture is obtained as
deploy power beacons within the proximity of each (1−βt+1 It+1 )Pt φt+1
 
[I]
D[I] . However, this requires the power beacon within Rt+1 = τt+1 log2 1+ 2
Mct
σ
the last hops to radiate with enormous power in 
order to overcome the accumulation of path-loss, + τt+1 log2 1+
not to mention the high installation cost. Another
[I]
approach is to have some D[I] s, e.g., Dτ where pM (1−βt+1 It+1 )Pt φt+1


(1−pM )(1−βt+1 It+1 )Pt φt+1 + σ 2

Mt , (14)
τ ∈ [1, t−1], transmit with a fixed power so that the
accumulated path-loss in the i-th hop, ∀i ∈ [1, τ ], is
It+1
[I]
eliminated in the harvested power at Dt+1 , as shown where τt+1 , 1−αMt+1−1
for BTEH and τt+1 , M1−1
[1]
in (10). Ideally, each D[I] should switch to energy for BPEH, Mt , [|Φt | ≥ 1] is the event where at
[II]
harvesting mode if the harvested power exceeds an least one active D exists within the proximity of
[I]
acceptable power threshold. This also implies that Dt whereas Mct is the complementary of Mt , and σ 2
there should be a probability of EH imposed on denotes the noise power. In order to improve the net-
each D[I] . work sum-throughput, in what follows, we propose
two possible device pairing techniques employed at
[I]
each Dt .
C. Co-Existing Type-I and Type-II MTCDs Trans- 1) Connectivity-oriented Machine-Type Commu-
mission Schemes nication (CoM) Scheme: In this scenario, a trans-
[I]
mitting Dt can randomly schedule Kt active D[II] s,
During the t-th time slot, if there exists at least where each k-th paired device is randomly dis-
[I]
one D[II] within Ct around the transmitting Dt , tributed within Ct,k . Accordingly, the transmitted
additional D[II] s can be scheduled to jointly perform power-domain multiplexed signal, st , can be ex-
[I]
PD-NOMA. In this case, Dt performs superposi- pressed as
tion coding (SC) to form a mixture of information Kt
[I]
signals dedicated to DM , denoted as sM , and other
X √
st = pt,k st,k , (15)
D[II] s, denoted as st . Accordingly, the information k=1
[I]
signal transmitted by Dt can be expressed as
where st,k is the unit energy information signal
√ p dedicated to the randomly selected user and pt,k
xt = pM sM + 1 − pM st , (12) denotes the proportionPof power allocated to st,k
which satisfies pM + K k=1 pt−1,k = 1 and pt,1 ≤
t

where 0 ≤ pM ≤ 1 denotes the power allocation pt,2 ≤ · · · ≤ pt,Kt since dt,1 < dt,2 < · · · < dt,Kt .
for sM and st specifies the mixture of the D[II] s’ Subsequently, the scheduled D[II] first decodes the
signals. destination signal (sM ), and then applies successive
[I] [I]
In practice, the distance from Dt to Dt+1 is much interference cancellation (SIC) before decoding its
[II]
larger than the distance between the paired D[II] s, own signals. Without loss of generality, Dt,k first
as observed in Fig. 1. Accordingly, to ensure trans- attempts to decode st,Kt while treating the signals
[I]
mission fairness, Dt should allocate more power st,v , where v ∈ [1, Kt − 1], as interference. After
[II]
to sM than to st . In addition, if there aren’t enough st,Kt is successfully decoded, Dt,k directly cancels
[I]
D[II] within the proximity of Dt , it then allocates it from the remaining received signals using SIC,
[II]
all power to sM . As a result, the achievable instan- Dt,k then decodes st,Kt −1 in a similar manner.
[I]
taneous rate at Dt+1 with the BTEH architecture is This procedure is successively iterated until st,k is
obtained as decoded correctly [9]. Accordingly, the achievable
[II]
  instantaneous rates at Dt,k for decoding sM and st,κ ,
[I] pM Pt φt+1 where κ ∈ [k, Kt ], are obtained as
Rt+1 = τt+1 log2 1 + Mt
(1 − pM )Pt φt+1 + σ 2 !
pM Pt

Pt φt+1

[II] 2 ϕ t,k
+ τt+1 log2 1 + Mct , (13) Rt,k = τt+1 log2 1 + (1−p σ)Pt , (16)
M
σ2 σ 2 ϕ t,k + 1
 9

!
[II],[κ] pt,κ σP2t ϕt,k Proof: The proof is provided in Appendix A.
Rt,k = τt+1 log2 Pt
Pκ−1 , (17)
ϕ
σ 2 t,k v=1 pt,v + 1
For brevity, let us denote Xt+1 , σP2t φt+1 , Yt,k ,
respectively. Pt
ϕ , and Zt , σP2t ϕt . It is noted that Xt+1 ,
2) Quality-oriented Machine-Type Communica- σ 2 t,k
Yt,k , and Zt denote the received powers at the t-th
tion (QoM) Scheme: In this case, to serve active
time slot normalized by the noise power observed
critical D[II] s, only the nearest active D[II] is selected [I]
[1] at Dt+1 , the k-th D[II] in the CoM scheme, and
among Φt . Without loss of generality, assuming
the paired D[II] in the QoM scheme, respectively.
that k ∗ specifies the index of the scheduled D[II] [I]
In what follows, the OP at Dt+1 during the t-th
such that
time slot with BTEH and BPEH architectures are
{k ∗ } = argmin dt,k ,
 
(18) given in subsection IV-A. Accordingly, the OPs at
1≤k≤Kt
the scheduled D[II] s for ToM and CoM schemes
the exact t-th hop’s instantaneous rate before are provided in subsection IV-B. Using the results
and after SIC at this D[II] can be expressed as in subsections IV-A and IV-B, we formulate the
[I] [II]
p M Pt
! e2e OPs at the DM , and each Dt,k∗ in the QoM
[II] 2 ϕ t [II]
RM →t = τt+1 log2 1 + (1−p σ)Pt , (19) scheme and each Dt,k in the CoM scheme, where
M
σ2
ϕt + 1 1 ≤ t ≤ M − 1 and 1 ≤ k ≤ Kt in subsection

(1 − pM )Pt ϕt
 IV-C. The sum-throughput of the proposed network
[II]
Rt = τt+1 log2 1 + , (20) is also provided in subsection IV-C.
σ2
2
respectively, where ϕt , gt,k∗ `(dt,k∗ ).
A. Performance Analysis of Type-I MTCDs
IV. P ERFORMANCE A NALYSIS [I]
The OP of each Dt+1 is determined as the
In this section, we study the outage performance probability that the instantaneous rate drops below
of the proposed transmission schemes, which is a the target transmission rate of the destination D[I] M.
key performance metric for delay-limited mMTC Furthermore, the instantaneous rate is expressed
networks. An outage event occurs whenever the in terms of the effective transmission time. For
instantaneous achievable rate drops below a pre- BTEH, the effective transmission time depends on
defined transmission rate threshold [41]. As a result, whether D[I] t+1 adopts EH. It is noted that for a fair
[I] [II]
the OP at D or D is defined as the proba- performance comparison, we consider the amount of
bility that the outage event occurs at that device. power allocated for each D[I] is equivalent in both
Accordingly, the e2e OP at D[I] or D[II] is defined CoM and QoM schemes, the OP of D[I] t+1 during the
as the probability of an outage event occurring t-th time slot is expressed as
at that device or at other devices involved in the [I] [I]
[I]
transmission from D1 to that D[I] or D[II] . In order OPt+1 , Pr[Rt+1 < RM ], (23)
to assist the analysis in the upcoming subsections, [I]
we introduce the following Lemma. where Rt+1 is given by (13) and (14) for BTEH and
BPEH architectures, respectively, and RM denotes
Lemma 1. Let us denote ξτ,t , ti=τ Ωi φi as the the target transmission rate of sM , measured in
Q
product of (t − τ ) independent and not-necessarily bit/s/Hz.
identically distributed (i.n.i.d) exponential RVs, The outage analysis of D[I] s using the BTEH
each with mean Ωi `(li ), the CDF and PDF of ξτ,t architecture is obtained in the following theorems.
is obtained as To provide a consistent comparison of the pro-
 
t−τ +1,0 ξ posed transmission schemes, i.e., the CoM and QoM
Fξτ,t (ξ) = 1 − G0,t−τ 1 t−τ , 0 , (21) schemes, in the following sections, the total power
+1
ξ¯τ,t
  allocated to D[I] s in both schemes is considered to
1 ξ
fξτ,t (ξ) = ¯ Gt−τ +1,0
0,t−τ +1 ¯ 0t−τ +1 , (22) be identical. As a result, the received [I]
information
ξτ,t ξτ,t and interference powers at Dt+1 are maintained
where ξ¯τ,t , t m,n
Q
Ω
i=τ i `(li ) and G p,q denotes the regardless of the transmission schemes adopted,
Meijer G-function [37]. resulting in similar OPs.
 10

Theorem 1. The closed-form expression for the OP Denoting Xt+1 , Pr[Xt+1 ≥ x, It = 0] and Xt+1 ,
[0] [1]

at Dt+1 , using the BTEH-powered QoM or CoM Pr[Xt+1 ≥ x, It = 1], their CCDFs are obtained as
[I]

scheme, denoted as the TQoM or TCoM scheme, [0]


x

respectively, can be written as F̄X [0] (x) = ρt exp − , (29)
t+1 γ̄0 `(lt+1 )
OPTQoM = OPTCoM t−1
" t
#
[I] [I] X Y
Dt+1 D [1]
t+1
X X [i] [j] F̄X [1] (x) = ρ[0]
τ ρj
[ij] t+1
=1− ρt+1 ιt F̄Xt+1 (τM ), (24) τ =1 j=τ +1
" #
i j
x
M −1 R × Gt−τ +1,0
0,t−τ +1 ¯τ +1,t 1t−τ , 0 , (30)
[00]
where for τM ,2 (M −1)R M −1 and τ
[10]
M ,2 1−α t+1 M
−1 , we γ̄0 `(lt+1 )ξ
[i0]
[i1] τM
have τM , [i0] . where x > 0 and γ̄0 , Pσ20 . The derivation of (29)
1−(τM +1)(1−pM )
[I]
and (30) is provided in Appendix B.
Proof: Considering the event where Dt adopts By invoking (29) and (30), we obtain the exact
EH in conjunction with a successful decoding event analytical expression of P in (27). The second
[I] [I] [I]
at Dt , i.e., Rt+1 ≥ RM , the OP of Dt during the probability in (25) can then be derived in an analo-
[00] [01] [10]
t-th time slot is expressed as gous manner, i.e., replacing τM and τM with τM
[11]
Pr[It+1 = i]
X and τM , respectively. This completes the proof of
OPTQoM [I] = OPTCoM[I] = 1 −
Dt+1 D t+1 Theorem 1.
i
It should be noted that F̄Xt+1 (x) is defined for
≥ RM | It+1 = i]. (25)
[I]
× Pr[Rt+1 x > 0, since received powers are always positive.
[01] [11]
It is worth mentioning that the second probability Hence, F̄Xt+1 (τM ) > 0 and F̄Xt+1 (τM ) > 0
[I] [01] [11]
considers that BTEH is implemented at Dt+1 ’s re- when both τM and τM are positive, which is
ceiver, and also represents the outage performance equivalent to
gain from performing EH during the t-th time [i1] pM
τM < , ∀i ∈ {0, 1}. (31)
slot. Meanwhile, the second probability considers 1 − pM
[I]
no EH is executed at Dt+1 . Taking into account This also implies that the target transmission rate
[I] [01] [11]
the number of D[II] s within the proximity of Dt , of sM so that F̄Xt+1 (τM ) and F̄Xt+1 (τM ) are
the first probability, denoted as P, can be further theoretically positive should satisfy
expressed as 1 − αt+1

pM

RM < log2 1 + , (32)
P = Pr It+1 = 0 Pr Xt+1 ≥ τM , Mct
   [01] 
M −1 1 − pM
  which is obtained by solving (31) for RM . Hence-
+ Pr It+1 = 0 Pr
  pM Xt+1 [01]
≥ τM , Mt . forth, let the right-hand side of (32) be depicted as
(1 − pM )Xt+1 + 1 [I]
(26) the MRTR of the DM ’s desired information signal
using the TCoM or TQoM scheme. The MRTR
It is noted that Xt+1 is independent of Mct and Mt , of sM in (32) is undefined when pM = 1, which
thus one can separate their joint probability into a implies a) D[I] s only focus on relaying information
product of the corresponding probabilities such that [I]
to DM and neglect the demand of D[II] s, and b) the
[0] [1] [01] [0]
P = ρt+1 {ιt Pr[Xt+1 ≥ τM ]+ιt Pr[Xt+1 ≥ τM ]}.
[00] interference power from the mixture of D[II] s while
(27) decoding sM is neglected.
In general, the MRTR of an information signal
As shown in (27), both probabilities are expressed is defined for NOMA-aided transmission in delay-
via F̄Xt+1 (x), i.e., the CCDF of Xt+1 . limited systems and is the minimum target trans-
[I]
Next, the CDF of the received power at Dt+1 mission rate that cannot be satisfied. As a result,
normalized by the noise power in the t-th time slot the MRTR can offer effective upper-bounds for the
can be expressed as target transmission rate, ensuring that the perfor-
F̄Xt+1 (x) = Pr[Xt+1 ≥ x, It = 0]
mance gains from NOMA operations are theoreti-
cally greater than zero, thus improving transmission
+ Pr[Xt+1 ≥ x, It = 1]. (28) reliability.
 11

probability Pr[Rt+1 ≥ RM , It+1 = 1], denoted as

[I]
From (24), one can observe that the OP of D[I]
during the t-th time slot relies on the following key P0 , is rewritten as
factors:  
P = Pr[It+1 = 1]Pr (1−βt+1 )Xt+1 ≥ τM , Mt
[I] 0 [00] c
a) Density of D[II] s around each dedicated Dt
[I]
where 1 ≤ t ≤ M − 1: As λt increases,  
[1] pM (1−βt+1 )Xt+1 [00]
ιt → 0. In this case, the power allocation has + Pr ≥ τM , Mt
a significant impact on the outage performance. (1−pM )(1−βt+1 )Xt+1 +1
[1]
In contrast, at ιt = 0, the network gains no × Pr[It+1 = 1]. (34)
benefit from performing NOMA and the power Using the identity Pr[V ] Pr[U | V ] = Pr[U V ]
[I]
allocation has no impact on the outage at Dt . for two RVs U and V , and some mathematical
b) The EH ratio (αt+1 ): As αt+1 increases, manipulations, the above equation can be further
[10] [1] [0] [10]
τM also increases until ρt+1 ιt F̄Xt+1 (τM ) rewritten as
in (24) reaches zero and there is no bene- [0] [10] [1] [11]
fit from performing EH. Specially, by solv- P0 = ρt+1 [ιt Pr[Xt+1 ≥ τ̃M ]+ιt Pr[Xt+1 ≥ τ̃M ]].
ing τM
[10]
= 1−p pM
for αt+1 , we find (35)
M
log2 (1−pM )+(M −1)RM
that at αt+1 = log (1−pM )
, the Invoking (29) and (30) in Theorem 2, we can ob-
2
[1] [1] [1]
ρt+1 ιt F̄Xt+1 (τM ) component suddenly drops tain the CCDF of Xt+1 . In addition, the probability
Pr[Rt+1 ≥ RM , It+1 = 0] in (25) can be derived in
[I]
to zero, which in turn leads to a drastic drop
in the performance. an analogous manner. This completes the proof of
[I]
c) Power allocation at Dt : as pointed out in a) Theorem 2.
this factor only affects the OP when ιt < 1
[1] In (33), the target transmission rate RM should
[1] satisfy
since at ιt = 1, no D[II] is active within the
[I] [I]  
proximity of Dt . Hence, Dt should allocate 1 pM
[I] RM < log2 1 + , (36)
all its power for relaying information to Dt+1 . M −1 1 − pM
Note that OPTQoM[I] and OPTCoM[I] also depend on [01] [11]
Dt+1 D t+1 so that F̄Xt+1 (τ̃M ) and F̄Xt+1 (τ̃M ) are positive.
[1]
other factors in of ιt and the derived CDF of Xt+1 , Accordingly, the MRTR of sM using the PCoM
such as the coverage area, the distance between two and PQoM schemes, defined as the right-hand side
consecutive D[I] s, ηt , , etc. However, the above key of (36), is higher than using TCoM and TQoM
factors are obvious just by observing the expression schemes.
derived in (24).
Theorem 2. The closed-form expression for the OP B. Performance Analysis of Type-II MTCDs
[I]
at Dt+1 , using the BPEH-powered QoM or CoM The paired D[II] s require the correct decoding
scheme, denoted as the PQoM or PCoM scheme, of their own information signals, which depends
respectively, can be written as on the correct decoding of sM . Although each
paired D[II] does not utilize EH, its performance
OPPQoM [I] = OP PCoM
[I]
Dt+1 can be affected by whether its own dedicated D[I]
Dt+1
X X [i] [j] [ij] adopts EH. It is recalled that both the TCoM and
=1− ρt+1 ιt F̄Xt+1 (τ̃M ), (33)
TQoM schemes use the BTEH architecture to power
i j
the transmission from D[I] to D[II] s, where TCoM
[00] [00] [10] [10] [10]
[00]
τM applies the CoM scheme to schedule more than one
where τ̃M = τM , τ̃M = τM , τ̃M , 1−β , and
[11]
t+1 D[II] s whereas TQoM only serves an additional D[II] .
[11] τM
τ̃M , 1−βt+1 . Meanwhile, both the PCoM and PQoM schemes
[I]
adopt the BPEH architecture for regenerative trans-
Proof: Using (14), the OP of Dt , in the t- mission with a similar use of the CoM and QoM
th time slot, can be expressed as in (25), with schemes.
[I]
Rt+1 is given by (14). Next, considering the event Considering that the CoM scheme is adopted
[I] [II]
where there exists at least one D near Dt , the at the D[I] s, the OP at each Dt,k , given that the
[II]
 12

[I]
dedicated Dt adopts BTEH with probability ρt , can It is worth noting that F̄Yt,k (y) is only defined for
be formulated as [01] [11]
y > 0, thus F̄Yt,k (τt,k ) = 0 and F̄Yt,k (τt,k ) = 0
[01] [11]
" Kt
#
TCoM [II]
\ [II],[n]
when τt,k < 0 and τt,k < 0, or, equivalently,
OPD[II] = 1−Pr Rt,k ≥ RM , Rt,k ≥ Rt,n ,
t,k [i0]
n=k τt,k
(37) [i0] Pn−1
. (42)
pt,n − τt,k q=1 pt,q < 0
which can be derived from the following Theorem.
This also implies that the target transmission rates
Theorem 3. The exact closed-form expression of the of every s so that F̄ (τ [01] ) and F̄ (τ [11] ) are
[II] t,n Yt,k t,k Yt,k t,k
OP at Dt,k , using the TCoM scheme, is given by theoretically positive should satisfy
X [i] [i1]
OPTCoM
!
[II]
Dt,k
= 1 − ρt+1 F̄Yt,k (τt,k ), (38) 1 − αt+1 pt,n
i Rt,n < log2 1 + Pn−1 , (43)
M −1 q=1 pt,q
[00] [10]
where for τt,k , 2(M −1)Rt,k − 1 and τt,k ,
M −1
R
which is obtained by solving (42) for Rt,n . The
2 1−αt+1 t,k − 1, we have right-hand side of (43) is the MRTR of st,n using
" [i0]
τt,k
# the TCoM scheme. It is noted that the MRTR of
[i1] [i0] [II]
τt,k , max τM , [i0]
, the Dt,1 ’s information signal is undefined since
P n−1
k≤n≤Kt pt,n − τt,k q=1 pt,q the intra-interference power Pn−1while decoding st,1 is
Proof: Substituting (17) and (16) into (37) with perfectly cancelled, i.e., q=1 pt,q = 0 when n = 1.
P ϕ
a note that tσ2t,k = Yt,k , we obtain Using the BPEH architecture, the decoding
[I]
 threshold of st,k is not affected by whether Dt+1
pM Yt,k
≥ τM , adopts EH. Specifically, the outage probability of
X [i0]
OPTCoM [II] = 1 − Pr [II]
Dt,k
i
(1 − pM )Yt,k + 1 Dt,k can be formulated analogously to (37), in
Kt  which τt+1 = M1−1 , which is derived in the fol-
≥ τt,k , It+1 = i . (39) lowing Theorem.
\ pt,n Yt,k [i0]  
Pn−1
n=k t,k
Y q=1 pt,q + 1
Theorem 4. The exact closed-form expression of the
After some mathematical manipulations, the OP at D[II] , using the PCoM scheme, is given by
t,k
above equation can be rewritten as
[01]
X [i] [i1] OPPCoM[II] = 1 − F̄Yt,k (τt,k ). (44)
OPTCoMD
[II] = 1 − ρ t+1 Pr[Y t,k ≥ τ t,k ]. (40) Dt,k
t,k
i [II]
Proof: From (39) and (40), the OP at Dt,k for
The CDF of the received power normalized by the PCoM scheme can be derived as
[II]
the noise power at each paired Dt,k using the CoM [0] [01]
scheme is given by OPPCoM [II]
Dt,k
= 1 − ρt+1 Pr[Yt,k ≥ τt,k ]
t−1  t [1] [01]
− ρt+1 Pr[Yt,k ≥ τt,k ]. (45)
  
[0] y 2 X [0] Y [1]
F̄Yt,k (y) = ρt F̄ϕt,k + ρτ ρj
γ̄0  τ =1 j=τ +1 Invoking (41) in Theorem 3 to obtain the CCDF of
(  
k
   Yt,k , we rewrite the above equation as in (44). This
y 2
[0] K t 1−  completes the proof of Theorem 4.
× χk Gt−τ +1,1 
1,t−τ +2 ¯ 2

γ̄0 `(rt ) ξτ +1,t 1 t−τ , 0, − 
In the BTEH-powered QoM scheme, the outage
    event is avoided by correctly[II]decoding both sM and
k−1
Kt y 1 − 2  the information signal
 of Dt,k∗ , i.e., st,k∗ . Thus, the
[1] t−τ +1,1 
− χk G1,t−τ +2 [II]
γ̄0 `(rt ) ξ¯τ +1,t 1t−τ , 0, −   OP of the paired Dt,k∗ during the t-th time slot can
2
be expressed as
, y > 0. (41) [II] [II]
OPTQoM[II] = 1−Pr[RM →t ≥ RM , Rt ≥ Rt,k∗ ],
Dt,k∗
The derivation of (41) is provided in Appendix C.
(46)
Substituting (41) into (40), we obtain (38). This
completes the proof of Theorem 3. which is derived in the following Theorem.
 13

Theorem 5. The exact closed-form expression of the Substituting the above equation into (49), we
[II]
OP at Dt,k∗ , using the TQoM scheme, is given by obtain (47). This completes the proof of Theorem 5.
X [i] [i1]
OPTQoM [II] = 1 − ρt+1 F̄Zt (τt,k∗ ), (47) In (47), the target transmission rate of Rt,k∗
Dt,k∗
i [01] [11]
should satisfy τt,k∗ > 0 and τt,k∗ > 0 otherwise
where for τt,k[00] (M −1)Rt,k∗ [10]
M−1 R
∗ the corresponding OP becomes one.
∗ ,2
 −1 and [i0]
 τt,k∗ ,2 1−αt+1 t,k −1,
[i1] [i1] τt,k∗ Theorem 6. The exact closed-form expression of the
we have τt,k∗ , max τM , 1−p . [II]
M OP at Dt,k∗ , using the PQoM scheme, is given by
Proof: Substituting (19) and (20) into (46), and [01]
OPPQoM
[II] = 1 − F̄Zt (τt,k∗ ). (52)
after some mathematical simplifications, we obtain D t,k∗

TQoM
X  pM Zt [i0] Proof: From (19) and (20), the outage proba-
OP [II] = 1 − Pr ≥ τM , [II]
Dt,k∗ (1 − pM )Zt + 1 bility of Dt,k∗ can be expressed as
i

(1 − pM )Zt ≥ τt,k∗ , It+1 = i . (48)

[i0] PQoM pM Zt [00]
OP [II] = 1 − Pr ≥ τM ,
Dt,k∗ (1 − pM )Zt + 1

After some mathematical manipulations, the [00] [01]
(1 − pM )Zt ≥ τt,k∗ = 1 − Pr[Zt ≥ τt,k∗ ]. (53)
above equation can be rewritten as
X [i] [i1]
OPTQoM[II] =1− ρt+1 Pr[Zt ≥ τt,k∗ ]. (49) By invoking (51), we obtain (52). This completes
Dt,k∗
i the proof of Theorem 6.
[II]
From (52) and (47), the MRTR of Dt,k∗ ’s infor-
which is equivalent to (47) as Pr[Zt ≥ z] = F̄Zt (z),
mation signal in the TQoM and PQoM schemes is
i.e., the CCDF of Zt . Recalling that Zt , Pt ϕt /σ 2 ,
undefined since the intra-interference power while
thus it is required to obtain the distribution of ϕt
decoding st,k∗ is perfectly canceled.
before deriving the CCDF of Zt . Accordingly, the
[II]
CDF of the received power observed at each Dt,k∗
during the t-th time slot for the QoM scheme can C. Sum-Throughput Analysis of the Proposed
be expressed as mMTC Network
 −θt  In this subsection, the e2e OPs of the MTCDs
1 1,1 ϕ (1, 1)
Fϕt (ϕ) = 1 − H , and the network sum-throughput are studied. First,
Γ(mt ) 1,1 µt (mt , 1) the network sum-throughput is obtained in the fol-
(50) lowing Theorem.
m,n
where ϕ > 0, Hp,q [·] is the Fox H-function [42]. Theorem 7. The closed-form expressions for the
Proof: The proof is provided in Appendix D. sum-throughput can be expressed as

The CCDF of the received power normalized by RM [I],e2e

TΣ = (1 − OPM )
the Gaussian noise at each paired D[II] in the QoS M −1
M −1 X
schemes is expressed as X Rt,κ [II],e2e
+ (1 − OPt,κ ), (54)
t−1 [0] Qt
ρτ
[1] M − 1
j=τ +1 ρj
  X
[0] z t=1 κ∈Kt
F̄Zt (z) = ρt F̄ϕt +
γ̄0 τ =1
Γ(mt ) where Kt is the set of D[II] s that are scheduled to
" θt (1 − mt , 1)
# perform NOMA. Herein, Kt , {1, 2, . . . , Kt } for
×H1,t−τ
t−τ +1,1
+1
z
ξ̄τ +1,t µt γ̄0
(0, 1), (1, θt ), · · · , (1, θt ) , TCoM and PCoM schemes, and Kt , {κ : κ = k ∗ }
| {z
t−τ terms
} for TQoM and PQoM schemes. Note that the factor
1
(51) M −1
accounts for the (M − 1) time slots consumed
[I] [I]
in the transmission from D1 to DM .
for z > 0.
Proof: The proof is provided in Appendix E. Proof: In order to characterize the outage per-
formance of the proposed network, it is important
 14

[I]
to obtain and inspect the e2e OP of DM . In the A. Asymptotic OP of D[I]
[I]
proposed schemes, the outage at DM depends on Lemma 2. In the high transmission power regime,
[I]
whether an outage occurs at other Dt s, 1 ≤ t ≤ M . γ̄0 → ∞, FXt+1 (x) can be approximated as
[I]
In other words, the e2e OP at DM can be ex-  
pressed as x x
FXt+1 (x) → Ft , (60)
M
[ −1 γ̄0 `(lt+1 ) γ̄0 `(lt+1 )
[I],e2e [I]
OPM = Pr[ (Rt < RM )], (55)
where
t=1
However, it is intricate and even impossible to t−1 [0] Qt [1] t−τ (n)
[0]
X ρτ j=τ +1 ρj
X ψt−τ +1
obtain the exact analytical expression for the above F t (x) , ρ t + ¯
ξτ +1,t n!
equation. In the low probability of EH (LPEH) τ =1 n=0
[I] t−τ −n r
 r
regime, the e2e OP at DM using scheme S can be X (−1) x
given by × log ¯ . (61)
r=0
r! ξτ +1,t
M −1
OPM ∼
[I],e2e
Y
=1− [1 − OPSD[I] ], (56) Proof: The proof is provided in Appendix F.
t+1
t=1

where S ∈ {TQoM, TCoM, PQoM, PCoM}.

[II]
In addition, the performance of Dt relies on
[I] B. The Asymptotic OP at D[II]
whether the previous Dτ , ∀τ ∈ [1, t − 1], and a a+1
[I]
its dedicated transmitter Dt can operate without Given that γ(a; x) → xa − xa+1 and e−x → 1 − x
outage. Hence, the e2e OP at the paired D[II] in as x → 0, we have
TCoM and PCoM schemes can be formulated as ϕ ϕ
[II],e2e [II]
Fϕt,k (ϕ) → ϕ̄t,k , Fϕt (ϕ) → ϕ̄t , ϕ → 0,
OPt,k = 1 − Pr[(Rt,k ≥ RM ), `(rt ) `(rt )
Kt t
(62)
\ [II],[n]
\ [I]
(Rt,k ≥ Rt,n ), (Ri ≥ RM )], (57) 2 
[0] k
 [1] k−1
 
n=k i=1
where ϕ̄ t,k , 2

+1
χ k K t
− χ k Kt
and ϕ̄t ,
 
(πλt rt2 )− 2
γ 2 + 1; πλt rt2 .


and the e2e OP at the paired D[II] in TQoM and −πλt rt2
1−e
PQoM schemes can be formulated as
[II],e2e [II]
Lemma 3. In the high transmission power regime,
OPt,k∗ = 1 − Pr[(RM →t ≥ RM ), FYt,k (y) and FZt (z) can be approximated as
t
\ [II] [I]  
(Rt,k∗ ≥ Rt,k∗ ), (Ri ≥ RM )]. (58) ϕ̄t,k y ϕ̄t,k
FYt,k (y) → Ft y, γ̄0 → ∞, (63)
i=1 γ̄0 `(rt ) γ̄0 `(rt )
In the LPEH regime, the above e2e OPs can be
 
ϕ̄t z ϕ̄t
asymptotically expressed as FZt (z) → Ft z, γ̄0 → ∞. (64)
γ̄0 `(rt ) γ̄0 `(rt )
t
[II],e2e ∼
Y
S
OPt,κ = 1−[1−OPD[II] ] [1−OPSD[I] ]. (59) Proof: In Lemma 3, we obtain (63) and (64) by
t,κ t+1
i=1 adopting mathematical steps shown in Appendix F.
The throughput of the signals sM and st,k are Hence, we omit them from the paper.
defined via the corresponding e2e OPs. Adding the
sum of (59) over t ∈ [1, M − 1] and k ∈ Kt we
C. Diversity order
obtain (54). This completes the proof of Theorem 7.
Using the asymptotic results of FXt+1 (x) in
Lemma 2, FYt,k (y), and FZt (z) in Lemma 3, we
V. A SYMPTOTIC P ERFORMANCE A NALYSIS [I],e2e [II],e2e [II],e2e
find that the OPM , OPt,k , and OPt,k∗ have
In this section, we provide approximations for the no error floors, respectively, when increasing γ̄0
e2e OP, and consequently the sum-throughput of the (i.e., increasing P0 ). In this context, the diversity
proposed schemes. order which is important to evaluate the system
 15

TABLE I: Simulation Parameters

Parameter Value Parameter Value
[1]
Number of D[I] , M , [device] 4 Active D[II] density, λm [device/m2 ] 1e-2
[0]
Antenna gains of D[I] , [dBi] 5 [43] Inactive D density, λm [device/m2 ]
[II]
1e-3
Antenna gains of D[II] , [dBi] 5 [43] Carrier frequency, fc [GHz] 3 [41]
Bandwidth, BW [MHz] 10 [43] Fixed transmission power, P0 [dBm] 0
Noise power density, σ 2 [dBm/Hz] -174 [44] Energy conversion efficiency, ηm 1 [33]

performance in the high transmission regime, is probabilities are ρ2 = · · · = ρM −1 = ρ = 10−1 .

defined as [43] The power allocation pM , pt,k , ∀t ∈ [1, M − 1],
[I],e2e ∀k ∈ [1, Kt ], is adopted from [45]. In addition,
[I] logr OPM we consider 50% of the MRTR with a maximum
DM = − lim ,
γ̄0 →∞ logr γ̄0 of 0.75 [bits/s/Hz] of the target transmission rates
[II],e2e [I] [I]
[II] logr OPt,k during the transmission from D1 to DM . It is
Dt,k = − lim , noted that Figs. 3-5 adopt the network topology
γ̄0 →∞ logr γ̄0
[II],e2e T1 in Fig. 2, where the number of D[II] s joining
logr OPt,k∗
(65) the NOMA transmission is Kt = M − t devices
[II]
Dt,k∗ = − lim ,
γ̄0 →∞ logr γ̄0 where t ∈ [1, M − 1]. Moreover, Fig. 6-8 use
the network topology T2 , while Fig. 9 uses T3 ,
in which the coefficient r > 0 does not affect
T4 , and T5 for the results of M = 3, M = 4,
the diversity order. It is noted that when γ̄0 is
and M = 5, respectively. From Fig. 3 to Fig.
relatively large, increasing γ̄0 by r [dB] results in
[I],e2e [II],e2e [II],e2e 6, we compare the results of QoM and CoM
a decrease in OPM , OPt,k , and OPt,k∗ by
[I] [II] [II] schemes with and without being powered by the
DM · r, Dt,k · r, and Dt,k∗ · r [dB], respectively. In probabilistic EH architectures to the results of
the proposed system, with FXt+1 (x), FY,t,k (y), and conventional non-regenerative relaying (CNRR)
FZt (z) given in (90), (63), and (64), respectively, scheme, which consists only non-regenerative
[I] [II] [II]
we obtain DM = Dt,k = Dt,k∗ = 1. This result relaying transmission from D[I] to D[I] while
1 M
[I] [II]
implies that the e2e OPs at DM and Dt,k for the neglecting the D[II] s.
[II] Fig. 3 provides a comparison between the e2e OP
CoM scheme, and Dt,k∗ for the QoM scheme can
be reduced by x times when γ̄0 increases by x times of regenerative and non-regenerative relaying under
in the high transmission power regime. In addition, different values of the fixed transmission power P0
when the diversity order is non-zero, the sum- [dBm]. It is observed that the analytical results and
throughput of mMTC-NOMA the simulation results match, which validates our
PM −1 Pcan be approximated
as TΣ → M −1 (RM + t=1
1
κ∈Kt Rt,κ ).
analysis. By increasing P0 [dBm], the e2e OP at
[I]
DM eventually decreases, which improves network
VI. N UMERICAL R ESULTS reliability. In the case of no EH, the proposed
  transmission schemes, which are represented by the
200 200 200
As shown in Fig. 2, where T1 = 100 100 100 , yellow curves (QoM/CoM w/o EH), exhibit higher
  3 2 1 
200 100 100 50 50 OP than the CNRR scheme. The reason behind this
T2 = 100 50 50 , T3 = 25 25 ,
 3 2 1  2 2 is that the proposed transmission schemes utilize a
200 200 200 200 200 200 200
T4 = 100 100 100 , and T5 = 100 100 100 100 , we portion of P0 to serve the D[II] s, whereas the CNRR
2 2 2 2 2 2 2
consider different network topologies in this section. scheme uses all the power to relay information to the
In addition, numerical examples are provided to D[I] s and neglects the D[II] s. In the context of EH-
validate the analysis and provide additional insight enabled transmissions, the TCoM/TQoM schemes
into the performance of the proposed multi-user outperform the PCoM/PQoM schemes in terms of
PD-NOMA wireless-powered mMTC network. reliability. When jointly adopting the proposed QoM
Hereafter, unless otherwise stated, we set the and CoM schemes, the traditional TS-based and
simulation parameters as shown in Table I. In PS-based EH architectures in [46] requires more
addition, we set the EH ratios as αm = 0.2 and transmission power to reach the same outage per-
βm = 0.8. Moreover, we consider that the EH formance in comparison to that of the proposed
 16

150 100 150

100 50 100
50 0 50
0 0
-50 -50 -50

0 200 400 600 0 100 200 300 0 100 200 300

(a) T1 . (b) T2 . (c) T3 .

200
100
100
0
0
-100
0 200 400 0 200 400 600
(d) T4 . (e) T5 .
 
l1 l2 ··· lM −1
Fig. 2: Network topology, where Ti = r1 r1 ··· rM −1 . The solid blue triangles are D[I] s, active and inactive D[II] s are
K1 K2 ··· KM −1
the solid green and red circles, respectively.

TQoM/TCoM (ana.) QoM/CoM w/ TS-based EH TQoM (ana.) PCoM (ana.) CNRR

PQoM/PCoM (ana.) QoM/CoM w/ PS-based EH TCoM (ana.) QoM w/o EH (ana.) sim
QoM/CoM w/o EH (ana.) TS-based EH w/o QoM/CoM
PQoM (ana.) CoM w/o EH (ana.)
CNRR (ana) PS-based EH w/o QoM/CoM
asymp. sim.
1.2
100

1
Throughput [bits/s/Hz]

0.8
10-2
e2e OP

0.6

0.4
10-4

0.2

0
10-6 -50 0 50
-50 0 50 100

[I]
Fig. 3: The e2e outage probability of DM as a function of the Fig. 4: Network sum-throughput [bits/s/Hz] versus the fixed
fixed transmission power, P0 , [dBm]. transmission power, P0 [dBm].

of one, i.e., OM1 = OM2 = 1, at P0 = 0 [dBm].

schemes. This is because the CNRR system utilizes all of the
[I]
Fig. 4 presents the sum-throughput of the CoM power to relay information to DM while ignoring
and QoM schemes with and without EH operations, the D[II] s, and thus it does not gain additional
in which the dotted
PM −1 upper
P bounds in the throughput throughput by serving extra devices, unlike the
1 proposed schemes. As P0 increases beyond 0 [dBm],
is M −1 (RM + t=1 κ∈Kt Rt,κ ). The solid black
curve shows an upper limit of 0.11 [bits/s/Hz] in the more power is wasted for relaying operations, i.e.,
[00]
sum-throughput as P0 reaches 0 [dBm]. Compared increasing Xt+1 −τM , rather than serving additional
to the CNRR scheme, the proposed schemes can in- D[II] s. Meanwhile, the proposed schemes can utilize
crease the sum-throughput by an order of magnitude that wasted power to serve the D[II] s, thus increasing
 17

both connectivity and sum-throughput. 0.5

0.58 0.42
1

0.67 0.33
0.5 TQoM (sim)
TCoM (sim)
PQoM (sim)
0.75 0 0.25PCoM (sim)
QoM (sim)
Fig. 5: Energy Efficiency Difference (EED) among proposed CoM (sim)
transmission schemes, namely TCoM, TQoM, PCoM, and CNRR (sim)
PQoM, as a function of ρ and P0 [dBm].

EH Prob.,
0.83 0.17
In Fig. 5, we compare the EE of our proposed

or
schemes with the traditional non-regenerative re-

,
io
0.92 0.08

at
R
laying multihop-based mMTC. The EE is given

EH
0
×TΣ
by EE = BWPtol [Mbits/Joule] [47], where Ptol Fig. 6: Throughput [bits/s/Hz] versus EH ratios (i.e., α or β),
[W] is the average total power consumed in the and probability of EH (i.e., ρ). Solid lines denote analytical
[I] [I]
transmission from D1 to DM . In the proposed results.
probabilistic EH architectures, we obtain
M
X −1 M
X −1
[0]
Ptol = E[P0 Ij ] = P0 ρt . (66)
j=1 j=1

The EED in Fig. 5 is determined as EED =

EE − EE0 [Mbits/Joule], where EE0 is the EE
of the non-regenerative relaying with CoM/QoM
scheme. The higher the EED is, the more energy
efficient the BTEH/BPEH architecture is compared
to non-regenerative relaying. Among the proposed
schemes, the TQoM has the highest EED, which can
reach to 2.5 [Gbit/Joule]. This behavior is because
the e2e OP at the MTCDs is relatively low when P0
is relatively high, e.g., when P0 = 0 [dBm] in Fig.
4, and the sum-throughput P approaches an upper-
−1 P
bound value of M1−1 (RM M t=1 κ∈Kt t,κ ) where
R
continuing to increase P0 lowers the EE, which in Fig. 7: Network sum-throughput versus time-switching ratio
turn reduces the EED. Since the EE is a decreasing α and probability of EH (i.e., ρ). The lower figure represents
the projection of the upper figure to the 2D Cartesian plane.
function of Ptol , as we further increase P0 while RΣ
remains constant, the EED decreases to zero, which
depicts that EE = EE0 .
Fig. 6 illustrates the sum-throughput [bits/s/Hz] α and β, does not affect the sum-throughput of non-
versus the EH ratios, where αi = α and βi = β, regenerative relaying using QoM and CoM schemes.
∀i ∈ [2, M − 1]. In this figure, as the plotted The reason behind such performance gaps is that the
curves move further away from the center point, proposed network benefits from serving D[II] s while
[I]
the sum-throughput of the illustrated schemes in- maintaining relaying information to DM , whereas
creases and vice versa. Compared to the CNRR the CNRR scheme represents the sum-throughput
scheme presented by the solid black curve, we when D[II] s are ignored.
observe a gain in the sum-throughput of nearly To further examine the BTEH-powered CoM and
0.61 [bits/s/Hz] for the PQoM scheme and 0.75 QoM schemes, Fig. 7 illustrates the sum-throughput
[bits/s/Hz] for the non-generative CoM scheme, i.e., versus EH ratio α and the probability of EH (ρ).
the cyan curve. Note that varying the EH ratios, i.e., When ρ increases from ρ = 0 to ρ = 0.4, the
 18

sum-throughput of the TQoM scheme drops from 0.8 1.1

0.86 [bits/s/Hz] to about 0.45 [bits/s/Hz], whereas
1
that of the TCoM scheme is hardly affected and 0.75

remains around 1.12 [bits/s/Hz]. However, as ρ 0.9

0.7
increases beyond 0.8, the TQoM scheme eventually 0.8
outperforms the TCoM scheme due to less block 0.65
0.7
time consumed for EH, such as α < 0.4. When more
0.6 0.6
time is consumed for EH, such as α ≥ 0.75, the 10 -5 10 -5
TQoM scheme tends to achieve a higher throughput
than the TCoM scheme. Fig. 8: Sum-throughput [bits/s/Hz] versus device density
In Fig. 8, we observe the network sum-throughput where P0 = 0 [dBm].
for different device density [device/m2 ] values. The
left-hand side figure shows that the QoM schemes
1.5 TQoM

Throughput [bits/s/Hz]
outperform CoM schemes when the network only 1.5 TCoM
serves one D[II] during each time slot. The fun- 1.11
PQoM
PCoM
damental reason for this is that QoM schemes 1 0.88
schedule the nearest D[II] , but the CoM schemes 0.64
choose one D[II] without considering its channel 0.5
condition. As a result, the received power at the
selected D[II] using QoM schemes is substantially 0
3 4 5
higher than using CoM schemes, and hence the sum- Number of Type-I MTCDs [device]
throughput of the QoM schemes is larger than that
Fig. 9: Maximum sum-throughput [bits/s/Hz] where up to 3
of the CoM schemes. However, as shown in the D[II] s can be served simultaneously.
right-hand side figure, when there are more devices
served per transmission, e.g., Kt = 2, the sum-
throughput using the CoM schemes further increases [I]
shown in Fig. 9. In general, D1 eventually requires
and even outperforms that of the QoM schemes. more time slots to relay information to DM as
[I]

Under non-regenerative relaying, as the network M increases, thus decreasing the network sum-
becomes much more populated with MTC devices, throughput. By increasing M from 3 to 4, the
the network tends to promote NOMA transmission maximum sum-throughput using TCoM and PCoM
more frequently, which can further improve the schemes decreases from 1.50 [bits/s/Hz] to 1.11
sum-throughput. We observe that as the network is [bits/s/Hz], which corresponds to a 26% reduction in
more populated by D[II] s, i.e., larger values of λ, the sum-throughput. The maximum sum-throughput
the sum-throughput of the PCoM scheme decreases using the TQoM and PQoM schemes also decrease
until reaching a floor level. The reason for this as M increases to 4, from nearly 0.88 [bits/s/Hz]
performance loss while adopting the BPEH archi- to 0.64 [bits/s/Hz], which corresponds to a 27%
tecture is the reduction in the portion of the power reduction in the sum-throughput. One reason for
utilized for information processing to serve the D[II] s [I]
this behavior is that as M increases, D1 takes
without lowering the target transmission rates. [I]
additional time slots to relay information to DM ,
In Fig. 9, we illustrate the maximum sum-
thus reducing the sum-throughput of every network
throughput versus the number of D[I] s under i.i.d.
configuration and eventually decreasing the maxi-
setting. Specifically, we consider that, when using
mum sum-throughput.
the TCoM or PCoM schemes, each D[I] can serve
from Kt = 1 to Kt = 3 D[II] s during each time slot.
[I] [I]
With M = 3, both D1 and D2 serving K1 = 1 VII. C ONCLUSIONS
and K2 = 1 D[II] , respectively, is considered as In this paper, we proposed two transmis-
one network configuration. When Kt ≤ 3, there sion schemes, namely CoM and QoM schemes,
would be a total of 3M −1 network configurations. for multi-user PD-NOMA probabilistic wireless-
Then, we consider the maximum sum-throughput powered mMTC networks. Specifically, the pro-
obtained from all 3M −1 configurations, which is posed schemes can serve two types of MTCDs
 19

with different communication priorities. Based on a Repeating similar steps for the subsequent inte-
Bernoulli random process-based EH, the proposed grals, we then obtain Fξτ,t (ξ) in (21).
schemes are self-sustaining. We considered that the
active and inactive MTCDs appear following a ho-
mogeneous Poisson point process over the network B. Derivation of the PDF of ξτ,t in (22)
area. We evaluated the performance of the pro- The proof of (17) is quite similar to that of (16).
posed schemes in terms of sum-throughput and de- It should be noted that by taking the derivatives of
rived closed-form expressions for the e2e OP, sum- (68), the PDF of ξτ,t can be rewritten as
throughput, and EE. Our results showed that signif-
t
Y Z ∞ Z ∞ t
icant gains in the network sum-throughput and EE 1 Y fφi (yi )
fξτ,t (ξ) = ···
can be obtained by operating the proposed schemes. Ωi 0 0 yi
i=τ i=τ +1
The results also revealed that employing NOMA in | {z
t−τ folds
}
low device density networks hinders a slight sum- t
 Y t−1 −1 
throughput loss using the BTEH protocol, whereas
Y ξ
× f φτ Ωi yi dyi . (69)
a slight improvement in the sum-throughput is ob- i=τ i=τ +1
yt
served for the BPEH protocol. We also pointed out
the drawbacks of the proposed schemes, which is f (y )
Rewriting fφτ (x) and φiyi i in terms of the
their maximum achievable throughput deteriorates Meijer G-function, we obtain
as the number of relaying devices increases.  
fφi (yi ) 1 1,0 φ
= G − 1 , yi > 0, (70)
A PPENDIX A yi `(li ) 0,1 `(li )
P ROOF OF L EMMA 1
then the first fold of the above integral, denoted by
A. Derivation of the CDF of ξτ,t in (21) J1 (ξ), can be obtained as
The CDF, Fξτ,t (ξ), can be expressed as  t t−1 
 t  1 0,2 `(lτ )`(lt )
Y Y
Y J1 (ξ) = G Ωi yi 1, 1 .
Fξτ,t (ξ) = Pr Ωτ φτ Ωi φi < ξ `(lτ )`(lt ) 2,0 ξ i=τ i=τ +1
i=τ +1 (71)
Z ∞ Z ∞ t
 Y t−1 −1 
Y ξ
=1− ··· F̄φτ Ωi yi Substituting (71) into (69) and repeating the steps
0 0 i=τ i=τ +1
yt in (70) and (71) for the remaining integrals, we then
| {z }
t−τ folds obtain fξτ,t (ξ) as shown in (69), which concludes
t
Y the proof.
× fφi (yi )dyi . (67)
i=τ +1
A PPENDIX B
Using the identity e = −x
[x|0] = G1,0
0,1 P ROOFS OF (29) AND (30)
0,1 −1
G1,0 [x |1], we rewrite the PDF and CCDF of φτ Using the representation of the transmission
in terms of Meijer-G functions. The first fold of power P provided by (10) and then applying the
t
(67) in terms of yt , denoted by I1 (ξ, yt−1 ), can be total probability theorem, the CDF of X , ∀t ∈
t+1
expressed as [1, M − 1], can be expressed as
h i
Z∞ G1,0 φ 0 " #
0,1 `(lt )
≥ x, It = 0
I1 (ξ, yt−1 ) = P0 φt+1  
`(lt ) FXt+1 (x) = 1 − Pr
0
σ2
t t−1 t−1
"
φt+1 ti=τ +1 Ωi φi
 Y   Q
`(lτ )
I
0,1
Y X  
× G1,0 Ωi yi yt 1 dyt − Pr ≥ x, τ = 0 ,
i=τ i=τ +1
ξ
τ =1
σ2
t
#
t t−1
Ij = 1 . (72)
 
0,2 `(lτ )`(lt )
Y Y \  
= G2,0 Ωi yi 0, 1 . (68)
ξ i=τ i=τ +1 j=τ +1
 20

The first probability can be derived as The derivation of the first probability in the
Z above equation is straightforward. For the second
[0] (a) [1]
F̄X [0] (x) = ρt fφt+1 (φ)dφ = (29) (73) probability, denoted by F̄t,k , it is noted that ξτ +1,t =
i=τ +1 Ωi φi . In addition, since Iτ , Iτ +1 , . . . , It are
t+1 Qt

φ>x γ̄0
mutually independent, the above expression can be
where (a) adopts fφt+1 (φ) = e `(lt+1 ) and some rewritten as
−x/`(lt+1 )

t−1  t 
straightforward mathematical manipulations.
Since I0 , I1 , . . . , and IM are Q
X Y [1]
[0]
mutually inde- F̄Y [1] (y) = ρτ ρj
t,k
t τ =1 j=τ +1
pendent Bernoulli trials, and since i=τ +1 Ωi φi = Z ∞  
ξτ +1,t , whose PDF is provided in Lemma 1, the y
× F̄ϕt,k fξτ +1,t (ξ)dξ . (77)
complementary CDF F̄X [1] (x) can be derived in an 0 γ̄0 ξ
t+1 | {z }
integral-form as Jt,k
" #
t−1
X Y t In order to solve the above integral, we rewrite
[1]
F̄X [1] (x) = ρ[0]
τ ρ j the CCDF F̄ϕt,k using the Meijer G-function as
t+1
τ =1 j=τ +1 " k
 #
Z∞ 2 [0] 1,1 Kt 1 − 2
F̄ϕt,k (ϕ) = χk G1,2 ϕ
 
x 0, − 2
× F̄φt+1 fξτ +1,t (ξ)dξ. (74)  `(rt )
γ̄0 ξ " k−1
 #
0 2
2 [1] 1− 
− χk G1,1 Kt
ϕ , (78)
By substituting the CDF of φt+1 and the PDF  1,2
`(rt ) 0, − 2
of ξτ +1,t provided in Lemma 1, the probability
[i] 2
F̄X [1] (x) can be further expressed as where ϕ > 0 and χk , k2(k−i) −(k−1) 2.
t+1
Substituting the above equation and fξτ +1,t (ξ)
t−1 [0] Qt [1]
X ρτ j=τ +1 jρ in (22) into (77) and applying the identity [37,
F̄X [1] (x) = ¯ Eq. (7.811.1)], we obtain the result in (41). This
t+1
τ =1
ξτ +1,t
complete the proof of (41).
Z∞ " #  
0,1 γ̄0 `(lt+1 ) ξ
× G1,0 ξ 1 Gt−τ,0 0t−τ dξ.
x 0,t−τ
ξ¯τ +1,t A PPENDIX D
0 P ROOF OF (50)
(75)
The PDF of the nearest distance, denoted by
Since the integration over a product of two Meijer fdt (r), can be obtained as
G-functions is also a Meijer G-function, we then
2πλt −πλt r2
obtain F̄X [1] (x) in (30). Substituting (73) and the fdt (r) = −πλ r 2 re , 0 ≤ r ≤ rt . (79)
t+1 1−e t t
result of (75) into (28), we obtain the CDF of Xt+1 .
This completes the proof of (29) and (30). Subsequently, the CDF Fϕt (ϕ) can be expressed
in an integral-form as
Z rt
r
 
A PPENDIX C
Fϕt (ϕ) = 1 − exp −  ϕ fdt (r)dr.
P ROOF OF (41) 0 Ld0
2 Z 1
Using Pt in (10), the CDF of Yt,k can be ex- (a) πλt rt 2 2−
=1− 2 u 
pressed as −πλ
1 − e tt  0 r

FYt,k (y) = 1 − Pr γ̄0 ϕt,k ≥ y, It = 0


   ϕ 2 2
× exp −u − πλt rt u  du, (80)
t−1  `(rt )
X y
− Pr ϕt,k ≥ Qt , in which (a) is obtained by substituting (79) into
τ =1
γ̄0 i=τ +1 Ωi φi the first integral and then applying the change of
t  variable u ← r .
Iτ = 0 , Ij = 1 . (76) To the best of the authors knowledge, the above
  \  
j=τ +1 equation is intractable to derive in closed-form. In
 21

order to provide a tractable expression for Fϕt (ϕ), Substituting (50) and the above equation into
we adopt the curve fitting method that matches the (84), Jt0 can then be expressed as
CDF of ϕt to that of a Singh-Maddala distributed Z∞ " θt #
RV with scale µt , and shapes θt and mt . In other 1 1,1 µ ¯γ
t 0 (1, 1)
Jt0 = H1,1 ξ (m
Γ(m ) z t , 1)
words, (80) can be fitted to t
0
−mt
ϕθt
 " #
Fϕt (ϕ) = 1 − 1 + θt , ϕ > 0. 1 ξ
(81) × ¯ H t−τ +1,0 (0, 1)t−τ dξ. (86)
µt ξτ +1,t 0,t−τ +1 ξ¯τ +1,t
Finally, we use the identities [42, Eq. (2.9.6)] and Finally, we use the identity [42, Eq. (2.8.4)]
[42, Eq. (2.1.3)] to rewrite (81) in terms of the Fox to solve the above integral and substitute the re-
H-function. This completes the proof of (50). sulted expression into (83). This completes the proof
of (51).
A PPENDIX E
P ROOF OF (51) A PPENDIX F
P ROOF OF L EMMA 2
Recalling that Zt = Pt ϕt /σ 2 is a function of
ϕt . Consequently, it is mandatory to examine the When γ̄0 → ∞, we can approximate F̄X [0] (x) as
t+1
CDF of ϕt before that of Zt . Thus, we introduce [0] [0] x
the following Lemma. F̄X [0] (x) → ρt − ρt . (87)
t+1 γ̄0 `(lt+1 )
Invoking Pt in (10), the CDF of Zt where 1 ≤ Let us denote the Meijer G-function in (30)
t ≤ M − 1 can be expressed as as Iτ +1,t (x; `), using [37,
1
H Eq. (9.301)], we obtain
that Iτ +1,t (x; `) = 2πi γ F (s)ds, where F (s) ,
 
FZt (z) = 1 − Pr ϕt ≥ , It = 0
z  
γ̄0 Γt−τ (1 + s)Γ(s)( γ̄0 `ξ̄xτ +1,t )−s and γ is a suitable line
contour, in which all the poles of Γt−τ (1+s)Γ(s) lie
t−1
" t
#
, Iτ = 0 , [Ij = 1] . on the left of γ. As γ̄0 → ∞, we apply the Cauchy’s
X z   \
− Pr ϕt ≥
τ =1
γ̄ 0 ξ τ +1,t j=τ +1 Residual theorem to approximate Iτ +1,t (x; `) by
(82) finding the residuals of F (s) at the simple pole
Since ϕt and It are independent, we obtain s = 0 and the (t − τ + 1)-th order pole s = −1 as
1
Iτ +1,t (x; `) → [sF (s)]|s→0 +
 
[0] z
FZt (z) = 1 − ρt Pr ϕt ≥ (t − τ )!
γ̄0  t−τ
d

t−τ +1
t−1  t  
z
 × t−τ
(s + 1) F (s) , γ̄0 → ∞.
−
X
[0]
ρτ
Y [1]
ρj Pr ϕt ≥ . (83) ds s→−1

τ =1 j=τ +1
γ̄0 ξτ +1,t (88)
By applying the generalized Leibnitz’s rule of
The second probability, denoted by Jt0 , can be derivative [48] and after some mathematical manip-
expressed as ulations, we obtain
Z∞ t−τ (n) t−τ −n
ψt−τ X (−1)r
 
z x X
Jt0 = F̄ϕt fξτ +1,t (ξ)dξ. (84) Iτ +1,t (x; `) → 1 −
γ̄0 ξ `γ̄0 ξ¯τ +1,t n=0 n! r=0 r!
0  r
x
It should be noted that the PDF fξτ +1,t (ξ) can be × log ¯τ +1,t . (89)
rewritten using the Fox H-function as `γ̄0 ξ
Plugging this approximation and (87) into (28),
1
fξτ +1,t (ξ) = ¯ we obtain
ξτ +1,t Xt−1  Y t 
  [1] [0] [1]
FXt+1 (x) → ρt − ρτ ρj
t−τ +1,0  ξ
× H0,t−τ (0, 1), . . . , (0, 1)  , ξ > 0. τ =1 j=τ +1
+1
ξ¯τ +1,t | {z } 
1

x
t−τ terms
+ Ft x; , γ̄0 → ∞. (90)
(85) γ̄0 `(lt+1 ) γ̄0 `(lt+1 )
 22

[1] [1]
Note that with ρ1 = 0, we have ρt − [16] J. Tang, J. Luo, M. Liu, D. K. C. So, E. Alsusa, G. Chen, K.-
Pt−1 [0] Qt [1] K. Wong, and J. A. Chambers, “Energy efficiency optimization
τ =1 [ρτ j=τ +1 ρj ] = 0, ∀t. This completes the for NOMA with SWIPT,” IEEE J. Sel. Topics Signal Process.,
proof of Lemma 2. vol. 13, no. 3, pp. 452–466, 2019.
[17] X. Li, Q. Wang, M. Liu, J. Li, H. Peng, M. J. Piran, and L. Li,
“Cooperative wireless-powered NOMA relaying for B5G IoT
R EFERENCES networks with hardware impairments and channel estimation
[1] S. Baek, D. Kim, M. Tesanovic, and A. Agiwal, “3GPP new errors,” IEEE Internet of Things Journal, vol. 8, no. 7, pp. 5453–
radio release 16: Evolution of 5G for industrial internet of 5467, 2021.
things,” IEEE Communications Magazine, vol. 59, no. 1, pp. [18] M. Kamel, W. Hamouda, and A. Youssef, “Uplink performance
41–47, Feb. 2021. of NOMA-based combined HTC and MTC in ultradense net-
[2] N. H. Mahmood, H. Alves, O. A. López, M. Shehab, D. P. M. works,” IEEE Internet of Things Journal, vol. 7, no. 8, pp.
Osorio, and M. Latva-Aho, “Six key features of machine type 7319–7333, 2020.
communication in 6G,” in 2020 2nd 6G Wireless Summit (6G [19] S. Liesegang, O. Muñoz-Medina, and A. Pascual-Iserte,
SUMMIT), 2020, pp. 1–5. “Stochastic geometry analysis and design of wireless powered
[3] J. Ding, D. Qu, M. Feng, J. Choi, and T. Jiang, “Dynamic MTC networks,” in 2020 IEEE 21st International Workshop
preamble-resource partitioning for critical MTC in massive on Signal Processing Advances in Wireless Communications
MIMO systems,” IEEE Internet Things J, vol. 8, no. 20, pp. (SPAWC), 2020, pp. 1–5.
15 361–15 371, Oct 2021. [20] J. Liu, G. Wu, X. Zhang, S. Fang, and S. Li, “Modeling, analy-
[4] W. Yang, M. Wang, J. Zhang, J. Zou, M. Hua, T. Xia, and sis, and optimization of grant-free NOMA in massive MTC via
X. You, “Narrowband wireless access for low-power massive stochastic geometry,” IEEE Internet of Things Journal, vol. 8,
internet of things: A bandwidth perspective,” IEEE Wireless no. 6, pp. 4389–4402, 2021.
Communications, vol. 24, no. 3, pp. 138–145, 2017. [21] S. Mishra, L. Salaün, C. W. Sung, and C. S. Chen, “Downlink
[5] C. Bockelmann, N. Pratas, H. Nikopour, K. Au, T. Svensson, connection density maximization for NB-IoT networks using
C. Stefanovic, P. Popovski, and A. Dekorsy, “Massive machine- NOMA with perfect and partial CSI,” IEEE Things J., vol. 8,
type communications in 5G: Physical and MAC-layer solu- no. 14, pp. 11 305–11 319, July 2021.
tions,” IEEE Communications Magazine, vol. 54, no. 9, pp. [22] S. Soni, R. Makkar, T. Singhwi, D. Rawal, N. Sharma, and
59–65, 2016. L. Minz, “On performance of multi-hop assisted mMTC for
[6] S. Liesegang, O. Muñoz, and A. Pascual-Iserte, “Sensor se- DECT-2020 new radio system,” in 2021 International Confer-
lection and distributed quantization for energy efficiency in ence on Information and Communication Technology Conver-
massive MTC,” IEEE Trans. Commun., 2021. gence (ICTC), Oct 2021, pp. 984–988.
[7] T. Wang, Y. Wang, C. Wang, Z. Yang, and J. Cheng, “Group- [23] M. B. Goktas and Z. Ding, “A wireless power transfer assisted
based random access and data transmission scheme for massive NOMA transmission scheme for 5G and beyond mMTC,” IEEE
MTC networks,” IEEE Trans. Commun., 2021. Wirel. Commun. Lett., vol. 11, no. 6, pp. 1239–1242, 2022.
[8] J. Choi, “Fast retrial for low-latency connectivity in MTC with [24] A. Mishra, Y. Mao, L. Sanguinetti, and B. Clerckx, “Rate-
two different types of devices,” IEEE Wireless Communications splitting assisted massive machine-type communications in cell-
Letters, vol. 9, no. 10, pp. 1786–1789, 2020. free massive MIMO,” IEEE Commun. Lett., vol. 26, no. 6, pp.
[9] M. Mohammadkarimi, M. A. Raza, and O. A. Dobre, 1358–1362, June 2022.
“Signature-based non-orthogonal massive multiple access for [25] X. Shao, X. Chen, C. Zhong, and Z. Zhang, “Exploiting
future wireless networks: Uplink massive connectivity for simultaneous low-rank and sparsity in delay-angular domain
machine-type communications,” IEEE Vehicular Technology for millimeter-wave/Terahertz wideband massive access,” IEEE
Magazine, vol. 13, no. 4, pp. 40–50, Oct. 2018. Trans. Wirel. Commun., vol. 21, no. 4, pp. 2336–2351, April
[10] K. Zheng, S. Ou, J. Alonso-Zarate, M. Dohler, F. Liu, and 2022.
H. Zhu, “Challenges of massive access in highly dense LTE- [26] M. Ibrahim and W. Hamouda, “Performance analysis of min-
advanced networks with machine-to-machine communications,” imum hop count-based routing techniques in millimeter wave
IEEE Wireless Communications, vol. 21, no. 3, pp. 12–18, 2014. networks: A stochastic geometry approach,” IEEE Trans. Com-
[11] S. Chen, R. Ma, H.-H. Chen, H. Zhang, W. Meng, and J. Liu, mun., vol. 69, no. 12, pp. 8304–8318, Dec 2021.
“Machine-to-machine communications in ultra-dense networks: [27] F. Azmat, Y. Chen, and N. Stocks, “Predictive modelling of RF
A survey,” IEEE Communications Surveys & Tutorials, vol. 19, energy for wireless powered communications,” IEEE Commun.
no. 3, pp. 1478–1503, 2017. Lett., vol. 20, no. 1, pp. 173–176, Jan 2016.
[12] U. Singh, A. Dua, S. Tanwar, N. Kumar, and M. Alazab, “A [28] M. E. Morocho-Cayamcela, H. Lee, and W. Lim, “Machine
survey on LTE/LTE-A radio resource allocation techniques for learning for 5G/B5G mobile and wireless communications: Po-
machine-to-machine communication for B5G networks,” IEEE tential, limitations, and future directions,” IEEE Access, vol. 7,
Access, vol. 9, pp. 107 976–107 997, 2021. pp. 137 184–137 206, 2019.
[13] M. Elbayoumi, M. Kamel, W. Hamouda, and A. Youssef, [29] L. Qiao, J. Zhang, Z. Gao, D. W. K. Ng, M. D. Renzo, and M.-
“NOMA-assisted machine-type communications in UDN: S. Alouini, “Massive access in media modulation based massive
State-of-the-art and challenges,” IEEE Commun. Surveys & machine-type communications,” IEEE Trans. Wirel. Commun.,
Tut., vol. 22, no. 2, pp. 1276–1304, 2020. vol. 21, no. 1, pp. 339–356, Jan 2022.
[14] Cisco, “Cisco visual networking index: Global mobile data [30] J. Choi, “Random access with layered preambles based on
traffic forecast update 2015-2020 white paper,” Feb. 2016. NOMA for two different types of devices in MTC,” IEEE Trans.
[15] Q. Qi, X. Chen, and D. W. K. Ng, “Robust beamforming Wirel. Commun., vol. 20, no. 2, pp. 871–881, 2021.
for NOMA-based cellular massive IoT with SWIPT,” IEEE [31] Y. Wang, T. Wang, Z. Yang, D. Wang, and J. Cheng,
Transactions on Signal Processing, vol. 68, pp. 211–224, Dec. “Throughput-oriented non-orthogonal random access scheme
2020. for massive MTC networks,” IEEE Trans. Commun., vol. 68,
no. 3, pp. 1777–1793, 2020.
 23

[32] B. Ji, B. Xing, K. Song, C. Li, H. Wen, and L. Yang, “The ing Theory and Performance Evaluation. New York, NY:
efficient backfi transmission design in ambient backscatter USA:Springer-Verlag, 2000.
communication systems for IoT,” IEEE Access, vol. 7, pp. [41] T. N. Do, G. Kaddoum, T. L. Nguyen, D. B. da Costa, and
31 397–31 408, 2019. Z. J. Haas, “Multi-RIS-aided wireless systems: Statistical char-
[33] R. Atat, L. Liu, N. Mastronarde, and Y. Yi, “Energy harvesting- acterization and performance analysis,” IEEE Trans. Commun.,
based D2D-assisted machine-type communications,” IEEE vol. 69, no. 12, pp. 8641–8658, Dec 2021.
Trans. Commun., vol. 65, no. 3, pp. 1289–1302, 2017. [42] A. A. Kilbas and M. Saigo, H-transforms: Theory and Appli-
[34] S. N. Swain, R. Thakur, and S. R. M. Chebiyyam, “Coverage cations. Boca Raton, FL, USA: CRC Press, 2004.
and rate analysis for facilitating machine-to-machine communi- [43] T. N. Do, D. B. da Costa, T. Q. Duong, and B. An, “Improving
cation in LTE-A networks using device-to-device communica- the performance of cell-edge users in NOMA systems using
tion,” IEEE Transactions on Mobile Computing, vol. 16, no. 11, cooperative relaying,” IEEE Trans. Commun., vol. 66, no. 5,
pp. 3014–3027, 2017. pp. 1883–1901, May 2018.
[35] B. Wang, Y. Sun, and X. Xu, “Loose game theory based [44] K. Feng and M. Haenggi, “The SINR meta distribution in
anomaly detection scheme for SDN-based mMTC services,” Poisson cellular networks,” IEEE Wirel. Commun. Lett., vol. 10,
IEEE Access, vol. 7, pp. 139 350–139 357, 2019. no. 7, pp. 1385–1389, July 2021.
[36] J. P. Champati, H. Al-Zubaidy, and J. Gross, “Transient anal- [45] D. Wan, M. Wen, F. Ji, Y. Liu, and Y. Huang, “Cooperative
ysis for multihop wireless networks under static routing,” NOMA systems with partial channel state information over
IEEE/ACM Transactions on Networking, vol. 28, no. 2, pp. nakagami-m fading channels,” IEEE Trans. Commun., vol. 66,
722–735, 2020. no. 3, pp. 947–958, March 2018.
[37] I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series, [46] A. A. Nasir, X. Zhou, S. Durrani, and R. A. Kennedy, “Relay-
and Products, 7th ed. New York, NY, USA: Academic Press, ing protocols for wireless energy harvesting and information
2007. processing,” IEEE Trans. Wirel. Commun., vol. 12, no. 7, pp.
[38] 3GPP, “Evolved universal terrestrial radio access (E-UTRA); 3622–3636, July 2013.
further advancements for E-UTRA physical layer aspects (Re- [47] R. Hashemi, M. Kazemi, and A. Mohammadi, “Analysis and
lease 9),” Document 3GPP-TR-36.814 V9.0.0, Mar. 2010. optimization of spectral and energy efficiency in underlaid D2D
[39] E. Bjornson, O. Ozdogan, and E. G. Larsson, “Intelligent re- multi-cell massive MISO over Rician fading,” IEEE Trans.
flecting surface versus decode-and-forward: How large surfaces Green Commun. Netw., pp. 1–1, 2022.
are needed to beat relaying?” IEEE Wireless Commun. Lett., [48] M. Silberstein, “Application of a generalized Leibniz rule
vol. 9, no. 2, pp. 244–248, Feb. 2020. for calculating electromagnetic fields within continuous source
[40] T. G. Robertazzi, Computer Networks and Systems: Queue- regions,” Radio Science, vol. 26, no. 01, pp. 183–190, Jan 1991.