Performance of a Multiuser Cooperative IoT NOMA Network With Battery-Assisted Energy Harvesting

This article investigates a cooperative nonorthogonal multiple access (Co-NM)-based network consisting of a multiantenna source, a full-duplex energy harvesting (EH) near user (NU) Internet of Things (IoT) node, and multiple distant user (DU) IoT nodes. The source shares a direct link to the NU, while the NU augments the harvested energy by a limited amount of its battery energy to relay the information to the selected DU. Considering time-switching (TS) or power-splitting (PS) protocol, practical nonlinear EH, successive interference cancellation error, and opportunistic Co-NM/orthogonal multiple access (OMA) (OM) switching, closed-form expressions are derived for the outage probability and throughput of both DU and NU. We demonstrate that the proposed opportunistic Co-NM/OM switching can ensure a performance similar to OM at the NU without loss in DU throughput. Also, a joint optimal choice of battery energy and PS/TS parameter helps in attaining a maximum energy efficiency (EE). Moreover, Co-NM/OM switching ensures higher EE as compared with Co-NM and OM.


I. INTRODUCTION
T HE next generation of wireless communication networks is supposed to handle extremely high data rates and facilitate massive connectivity while ensuring an enormous reduction in energy consumption.The inclusion of massive machine-type communication and the Internet of Things (IoT) in future-generation networks has increased the need to prolong the battery lifetimes of low-power machine-type devices (MTDs) [1], [2].To avoid frequent battery replacements, harvesting energy from radio frequency (RF) signals to implement self-sustaining communication nodes is clearly well motivated.The massive connectivity of IoT devices and MTDs can offer higher spectral efficiency but presents challenges of resource scarcity and battery lifetime.To realize these spectral efficiency goals, full-duplex (FD) and nonorthogonal multiple access (NOMA) have emerged as feasible solutions.While the former can approximately double the spectral efficiency by allowing simultaneous transmission and reception of signals in the same frequency band [3], the latter enables multiple users to share the time/frequency/code resources using power domain multiplexing and successive interference cancellation (SIC) [4].
To extend the range and reliability of NOMA networks, the use of cooperative relaying has emerged as a viable solution [5].Cooperative relaying in NOMA can be primarily categorized into cooperative NOMA (Co-NM) and relayed NOMA.In the context of IoT, a near IoT user (NU) in Co-NM helps in information exchange to a distant user (DU) [6].On the other hand, in relayed NOMA, a dedicated relay is used for forwarding the information to the DU, while the source (S) communicates to the NU over a direct link [7].Conventionally, in a Co-NM framework, the relay operation utilizes energy from the NU's battery, which greatly limits its battery life and makes frequent battery replacement necessary [8].To avoid these battery replacements in communication devices, simultaneous wireless information and power transfer (SWIPT)-enabled self-sustainable green communication nodes have emerged as an appealing alternative.The use of SWIPT in cooperative scenarios extends the coverage area and ensures economical use of the battery energy at the relay [9].In the context of SWIPT, two energy harvesting (EH) protocols, namely power splitting (PS) and time switching (TS), have been extensively studied.In TS, the signaling duration is apportioned into EH and information transmission (IT) phases.On the other hand, in PS, the received signal is divided into two, one for EH and the other for IT [10].The TS protocol has the advantage of requiring lower complexity circuits than PS [11].

A. Related Works
The integration of self-sustained EH nodes in cooperative NOMA networks with FD relaying can improve spectral efficiency and energy efficiency (EE) to a great extent.In [12], the outage performance of the FD Co-NM network with beamforming and EH was studied.In particular, it was shown that SWIPT in Co-NM not only motivates users to collaborate but also mitigates the impact of self-interference (SI).Further, in [13], the EE of the FD Co-NM network (where the NU cooperates with the DU) was maximized by jointly optimizing the beamforming vector and the PS ratio.Considering a TS EH protocol at the NU, expressions for the outage probability and ergodic capacity were derived for an FD Co-NM network in [14].In [15], a relayed NOMA system was investigated wherein a dedicated EH FD relay enhances the performance of the DU.In [16], multiple antennas were considered at the base station (BS) and the relay, and the EE was maximized by optimizing the beamforming vector.In [17], expressions for the throughput and EE were derived for a SWIPT-enabled uplink NOMA network.The secrecy capacity of a NOMA-based unmanned aerial vehicle (UAV)-assisted mobile edge computing system was maximized in [18].Further, in [19], the total energy consumption of NOMA-assisted device communication was minimized using federated learning.In [20], PS-based SWIPT in relayed NOMA was considered, and a union bound on the error rate was derived.
All of the aforementioned works has considered a somewhat idealistic linear EH model in which the harvested energy has a linear relationship with the input energy of the EH receiver.However, the efficiency of EH circuits is limited due to the use of devicessuch as transistors and diodes, which exhibit nonlinear characteristics [21].Due to this nonlinearity, the harvested energy saturates when the input energy exceeds a threshold level.Recently, in [22], a nonlinear EH-enabled FD Co-NM system was investigated, and its outage probability was minimized by optimally choosing the PS parameter.In [23], the reinforcement learning-based offloading scheme was presented for reducing the energy consumption of EH-enabled IoT devices in a mobile edge computing setup.The energy consumption minimization for wireless sensor networks based on a task-based model has been discussed in [24].Further, in [25] and [26], energy management schemes were studied for IoT-based home automation and microgrids, respectively.The PS-SWIPT-based spectrum sharing in relay-assisted cognitive IoT network was investigated for amplify-and-forward (AF) and decode-and-forward (DF) relaying schemes in [27].

B. Motivation and Contributions
The harvested energy is random due to the fading nature of communication channels and is often quite small, which lowers the reliability of communication links [28].Storing energy in batteries or supercapacitors until there is sufficient energy to transmit is a possible solution [29].However, rechargeable batteries make devices more complex and expensive besides increasing their form factor, whereas supercapacitors cannot store charge for a long time.For Industrial IoT (IIoT) applications, the self-sustainable IoT devices can cooperate with other devices in their vicinity, and a large battery life is desired for such communication devices [30].To satisfy the requirements on performance, depending solely on the harvested energy alone is not feasible.For these reasons, augmenting the harvested energy with as little battery energy as possible is a promising alternative in the immediate future [31].In NOMAbased IIoT networks, cooperation among the devices always helps to achieve the goal of massive connectivity along with improved spectral efficiency (SE) [32].However, in Co-NM mode, NU sometimes loses its performance due to cooperation with the DUs.In such cases, operating NU in hybrid mode (intelligent switching between Co-NM to OMA (OM) mode) can be a viable solution [33].Both spectral efficiency and EE can be further improved by using EH-based FD IoT nodes.However, for very high transmit powers, the residual self-interference (RSI) can cause degradation in performance [34].Therefore, the nodelevel EE is of great importance in such battery-assisted FD EH communication nodes.Motivated by this, this article investigates the performance of an IoT network in which an FD NU assists in information exchange with the DUs using the harvested energy.The major contributions of this article are as follows.
1) We consider a Co-NM framework consisting of a multiantenna S, an FD NU, and multiple DUs.The NU possesses EH capability and assists in IT to the DU.The NU augments the harvested energy with a quantum of battery energy.A novel decoding status-based Co-NM/OM switching scheme is used to ensure enhanced QoS at both NU and DU. 2) Considering TS/PS protocols, nonlinear EH, and imperfect SIC at NU, closed-form expressions are derived for the outage probability and throughput for NU as well as DU.An expression for EE is obtained using the derived throughout expressions.3) We demonstrate that DU losses diversity due to nonlinear EH.However, the increase in the number of DUs and NU's battery energy improves QoS at DU.In addition to this, with Co-NM/OM switching, NU attains a performance similar to the scenario when the DU is absent in the network.4) We then demonstrate that by the joint optimal choice of battery energy and TS/PS parameter, EE can be maximized.For both PS and TS protocols, Co-NM/OM switching ensures efficient utilization of the NU's battery energy and provides high immunity to RSI.

C. Organization of This Article
The rest of this article is organized as follows.Section II demonstrates the system model and problem formation for the multiuser cooperative IoT NOMA network.Section III presents the performance analysis in terms of outage probability and throughput.Section IV discusses the EE of the considered framework.Analysis results are compared to the computer simulations in Section V. Finally, Section VI concludes this article.
Notations: The probability density function (PDF) of a random variable Z taking value z is denoted by f Z (z).Vectors are represented by bold letters, and y H denotes the Hermitian of vector y.The cardinality of a set A, Euclidean norm of a vector y, and the lower incomplete Gamma function are denoted by | S|, ||y||, and γ (v, μu) = μ v u 0 x v−1 exp(−μx)dx [35, 3.381.1],respectively.All the acronyms used in this paper are listed in Table I.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.).S directly communicates with U n , whereas U n assists in IT to DUs.U n harvests energy using either the PS or TS protocol as shown in Fig. 2. Since the harvested energy is random in nature and often small [36], a limited amount of NU's battery energy is used to augment the harvested energy [31] and establish reliable U n − U d communication.Due to severe shadowing, the S to DU link is not considered [14].At U n , the FD operation is performed using separate transmit and receive antennas.

Denote by h i
sn the channel coefficient from the ith (i ∈ (1, M)) transmit antenna of the S to the receive antenna of NU.The channel vector between S and U n is denoted by The channel coefficient between the transmit antenna of NU and jth DU is denoted by Note that λ pq = d θ pq with θ denoting the path-loss exponent and d pq denoting the distance from node p to node q with p ∈ {s, n} and q ∈ {n, d}.SI cancellation up to 110 dB [37] can be achieved by using analog and digital cancellation.The RSI strength, therefore, becomes extremely small, and |h nn | 2 can be replaced by λ nn [31], [36].
Denote the signaling duration, PS EH parameter, and TS EH parameter by T , ρ, and α, respectively.In PS, ρ portion of the received signal power at U n is allocated for EH, and the other 1 − ρ portion is used for IT such that 0 < ρ < 1.On the other hand, with TS, each signaling interval T is partitioned into two phases, one of duration αT for EH and the other of duration (1 − α)T for IT (0 < α < 1).The amount of energy required for operations at R (like threshold activation and information decoding) is almost negligible compared with that required for IT [31].
In this network, it is assumed that the NU assists DU, but its own performance is required to be the same as in a network without the DU.It is shown in this chapter that this can be accomplished by allowing the network to switch to OM mode, in which the DU is ignored when the NU cannot meet its QoS with NOMA signaling.In all other cases, the network is operated in the Co-NM mode.Prior to IT, both S and NU first transmit pilot symbols and use a timer-based mechanism to jointly perform operating mode selection (Co-NM or OM) as well as far-user selection in a manner described later in this section.The Co-NM/OM switching mode is referred to as the hybrid mode.Signaling with PS and TS for both Co-NM and OM modes of operation is presented next, assuming that the jth DU (denoted by U ĵ d ) is selected.

A. Power Splitting SWIPT
As noted already, with PS at NU, ρ and (1 − ρ) fractions of the received signal power are utilized for EH and IT, respectively.S transmits a superposition of symbols x n (k) and x d (k) intended for U n and U ĵ d , respectively.The superimposed signal is , where E s denotes the transmit energy per symbol, which is apportioned for symbols x n (k) and x d (k) in the ratio a n : a d , and a n = 1 − a d .Further, with symbol rate R s , the total transmit power at S is given by S uses MRT to transmit information symbols to the NU.The received signal at U n is where E n,PS denotes the energy at U n with PS, Φ =  3(a)].U n first decodes x d (k) and x n (k), and then transmits x d (k − δ) to U ĵ d , where δ ≥ 1 is the delay incurred in processing.
The received signal at U ĵ d is where w ĵ d represents the noise at U ĵ d .

B. Time Switching SWIPT
With TS, S transmits energy symbols for αT duration to enable EH at U n .The received signal at U n for EH is where w n,a ∼ CN (0, σ2 n,a ) is the antenna noise.For (1 − α)T duration, S and U n transmit simultaneously1 [see Fig. 3(b)].At U n , the signal received for IT is given by where w n ∼ CN (0, σ 2 ) is the overall noise at U n .The received signal at U ĵ d can be expressed as

C. EH Model
To bring out the effect of saturation characteristics of the practical EH circuit, a nonlinear EH model is considered in this work [38]. 2 For both PS and TS, the input energy per symbol for EH is expressed as With the energy conversion efficiency η, the harvested energy is Here, E sat represents the saturation threshold. 3In what follows, we write expressions for TS and PS in a unified fashion.After augmenting the harvested energy with battery energy E b the transmit power at U n i.e., E n, can be expressed as where ∈ {TS, PS}, and C 1 , C 2 , and C 3 are as listed in Table II.Since a d > a n , x d (k) will be decoded first at U n , considering x n (k) as interference.After performing SIC, x n (k) is decoded at U n .Therefore, the signal to the interference-plus-noise ratio (SINR) to decode x d (k) is expressed as4 The signal received at OM mode: If during the initial training phase the symbols x n and x d are not successfully decoded at U n , then U n sends one-bit feedback to S, which then switches to OM and transmits only x n with energy E s (ignoring the x d ).The received symbol at NU is given by Thus, the SNR in the OM case can be expressed as

D. Pilot-Based User and Mode Selection
Before data transmission, a very small training phase occurs in which the S first sends a pilot consisting of a superposition of NU and DU symbols [39].Based on the pilot transmission, NU sends one-bit feedback (the feedback bit is denoted by f b ) to the S to indicate its decoding status.If NU decodes both the symbols correctly, it sends f b = 1; otherwise, it sends f b = 0.The feedback-based mode selection is illustrated in Table III.In the case of f b = 0, the system will operate in OM mode, and there will be no requirement of DU selection, whereas if f b = 1, the system will operate in Co-NM mode, and NU to DU cooperation takes place, which necessitates DU selection.For DU selection, all the DUs estimate their channels h j nd with j ∈ {1, L} to NU using this pilot.The DU having the best channel gain to the NU gets selected by the timer-based mechanism, which implies that

III. OUTAGE PROBABILITY AND THROUGHPUT ANALYSIS
In this section, the expressions for user outage probabilities of U n and U d are derived with and without mode switching.Let R n and R d denote the respective target information rates of the symbols x n (k) and x d (k) so that the threshold SNRs at U n and U d are γ n = 2 R n − 1 and γ d = 2 R d − 1.For ease of exposition, we consider ||h sn || 2 = X and |h ĵ nd | 2 = Y .
Lemma 1: The closed-form expression for the outage probability of U n with Co-NM/OM switching and nonlinear EH is given by 5 where I 1 , I 2 , I 3 , I 4 , I 5 , and I 6 are expressed as follows: where 5 Superscript Co-NM, OM, and hybrid represent the case of cooperative NOMA mode, OMA mode, and cooperative NOMA/OMA switching, respectively.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
Proof: We substitute E in, , Γ nd, , and Γ nn, from ( 7), (9), and ( 10) into (15).After some mathematical rearrangements, I 1 can be expressed as where Since X is exponentially distributed, I 1 can be evaluated as Solving the aforementioned integral yields I 1 as in (16).
Similarly, the expressions for I 2 -I 6 are obtained and given by ( 17)-( 21).Observation 1: From Lemma 1, it can be inferred that the S-NU link remains in nonoutage when the NU successfully decodes the DU's symbol, followed by its own symbol after SIC.In such conditions, it will assist the information transfer to DU; this mode of operation is called the Co-NM mode.However, if NU is not able to decode the DU's symbol or its own symbol, it will operate in OM mode, in which S forward only NU's symbol to NU to ensure better QoS at NU. Thus, operating NU in hybrid mode ensures that NU does not lose its performance due to cooperation.
Remark 1: From (15), the outage probability at U n in Co-NM mode without Co-NM/OM switching can be expressed as Remark 2: At U n , the outage probability in the linear EH case can be expressed in a closed form by substituting E sat = ∞ into (15).
Remark 3: In OM mode, an expression for outage probability at U n can be expressed as As aforementioned, X is chi-square distributed with PDF Solving for (25) and performing the averaging over the PDF of X yields The diversity of the OM mode is derived by D = − lim γ s →∞ log 2 P OM n log 2 (γ s ) .At higher values of γ s using [35, 8.354.4]approximating γ(M, ) M .The diversity for OM mode can be obtained as D = − lim Remark 4: At U n , with Co-NM/OM switching, a diversity of the order M can be achieved.This is because when NU fails to decode DU symbols successfully, it switches to OM.For this reason NU attains the same performance as in the OM mode, as if the DUs are not present in the network.
Lemma 2: A closed-form expression for outage probability of U d with nonlinear EH is given by where I 7 and I 8 are expressed in a closed form as where β = We first substitute E in , Γ nd, , and Γ nn, from ( 7), (9), and ( 10) into (27).After rearranging, I 7 and I 8 are expressed as Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
We first solve for I 7 as follows: To solve the aforementioned, we use the linear approximation to the exponential term.After simplifying, we obtain (31) where β = . Solving for I 71 , we obtain To solve I 72 , we apply the binomial approximation (x + β) L x L (1 + Lβ x ) in the aforementioned.After rearranging, we obtain We can rewrite the aforementioned as Solving the aforementioned using using [35, 3.381.1],we obtain Substituting ( 32) and ( 34) into (31), an expression for I 7 is obtained as in (28).In a similar way, I 8 is derived in (29).
Observation 2: From Lemma 2, it can be inferred that DU remains in nonoutage only when NU operates in Co-NM mode, and then DU successfully decodes the incoming symbols from NU.Thus, the outage probability at DU remains the same in Co-NM and hybrid modes of operation.
Remark 5: At U d , in the linear EH case, the closed-form expression for the outage probability can be obtained by substituting E sat = ∞ in (27).

IV. ENERGY EFFICIENCY
For the considered Co-NM network, EE is defined as the ratio of sum-throughput (sum of NU and DU throughputs) to the total energy used and is given by where τ m n, represents the throughput in hybrid or Co-NM modes m ∈ {Hybrid, Co − NM}, and τ d, denotes the DU throughput.Here, the DU throughput is expressed as 6and the throughput at NU in the hybrid mode is expressed as where the second τ Co−NM n, represents the NU throughput in Co-NM, and the second term is due to Co-NM/OM switching.
Note that U n, first harvests the energy from S and utilizes this harvested energy for forwarding the information symbols to DU.Sometimes the harvested energy is not enough (based on values of TS and PS parameters), and U n -U d information exchange is not possible.In those instances, the supplementary battery energy assists in NU-DU transmission.However, for high S transmit powers, when harvested energy is large, this additional supplementary battery energy only results in an increase in RSI level.In such cases, the NU cannot decode the DU's symbols in the Co-NM mode, and the throughput of NU as well as DU decrease.With the hybrid mode, the network switches to OM, and only NU is served.Thus, using high S transmit powers can degrade EE.When the battery energy drawn is very high, the SI is large, and the NU cannot decode the DU symbols once again.Both DU as well as NU throughputs decrease when the Co-NM is used exclusively (only the NU is served).Thus, transmission with exceedingly high battery energy at NU can further degrade EE.Therefore, a careful choice of the PS or TS parameter and battery energy is crucial to maximizing the EE.Denote by E b max and E b min the maximum and minimum values of the battery energy that can be drawn.Now the optimization problem can be formulated as where ∈ {PS, TS} and m ∈ {Hybrid, Co − NM}.Here, A * 1 = α * denotes the optimal TS parameter for TS EH protocol and A * 1 = ρ * denotes the optimal parameter for PS EH protocol, respectively.Due to the highly involved mathematical expressions for NU and DU outage probabilities in (1) and ( 2), a closed-form expression for the (E * b , A * 1 ) cannot be determined.However, the optimal point (E * b , A * 1 ) can be determined using a standard numerical search. 7

V. NUMERICAL RESULTS
This section presents numerical results to validate the correctness of the analysis, to investigate the behavior of the performance with varying system parameters, and to draw useful insights regarding the EH protocols in the Co-NM framework.Unless mentioned otherwise, the used parameters are as follows: and power saturation threshold P sat = E sat R s = 9.2 μW [38].
Figs. 4 and 5 depict the DU outage probability versus P s (dBm) for PS and TS EH protocols, respectively.Performance variations are seen by varying the battery energy E b , the number of transmit antenna M , and the number of DUs L. The accuracy 7 Due to the coupling of the variables A 1 and γ b , it is extremely difficult to solve the optimization problem (37) analytically using standard optimization techniques.Therefore, in order to solve (37), we use a 2-D search to find a jointly optimal solution (E * b , A * 1 ).In particular, a 2-D search is used exhaustively across the range of E b and α.The complexity of the search is O(A), where O(•) is the notation for the big O, and A = A l,b A l,A 1 with A l,b denoting the number values of E b and A l,A 1 denoting the number of values of A 1 .Further, we would like to mention that this 2-D search is the most commonly used search method to determine the optimal point in a complex but low-dimensional optimization problem, and the accuracy of this optimal point depends on the choice of the step size. of the derived analytical expression for the DU outage probability in 2 is clearly evident through simulations.For both PS and TS EH protocols, the outage probability initially improves significantly with increasing E b for the lower range of transmit powers.In contrast, at higher powers, the effect of E b is limited by the saturation threshold γ sat .For low transmit powers, both linear and nonlinear EH provide almost similar performance.At high transmit powers, however, the outage exhibits a floor with nonlinear EH, thus causing a diversity loss to DU (irrespective of an increase in L).For a fixed number of DUs, with an increase in the number M of S transmit antennas, the DU performance improves significantly (in the linear region, i.e., for E in, ≤ E sat ) while it exhibits a floor at high values of P s .However, with an increase in L, i.e., the number of DUs, a significant gain can be observed in DU outage probability with both PS and TS EH protocols.Using Figs. 4 and 5, it can also be seen that the outage performance of the TS EH protocol is superior to that of the PS EH protocol.This is mainly due to the fact that with the TS EH protocol, during EH duration (αT ), R does not transmit-the RSI is present only for the IT duration ((1 − α)T ).In contrast, NU performance degrades with the PS EH protocol since it requires the DU symbol to be decoded in the presence of RSI in the complete signaling duration, resulting in outage performance loss.
For PS protocol in Fig. 6 and for TS protocols in Fig. 7, the variation of NU outage probability versus P s is illustrated for various number of S transmit antennas.For NU, the outage probability is shown for two cases: 1) with hybrid mode (Co-NM/OM switching) as in (15), and 2) without Co-NM/OM switching as in (24).For both the EH protocols at NU, with hybrid mode, NU attains performance similar to OM.On the other hand, in the Co-NM mode, NU performance is inferior to OM.Further, with increasing M , the gain in the outage performance can be clearly seen for both PS and TS EH protocols.Increasing the Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.= 2 × 10 −10 at 0 dBm.Thus, for M = 4, the outage probability is proportional to 1/γ 4 s , and the diversity is 4. Similarly, the diversity for M = 1 and M = 2 is seen to be M .Hence, with M transmit antennas, NU attains a diversity of M for all three transmission modes.
For PS protocol in Fig. 8 and for TS protocols in Fig. 9, the variation of NU and DU throughputs versus P s is illustrated for different settings of SIC errors.It can be clearly seen that the SIC error causes a degradation in the NU as well as DU throughput.This is because the imperfect SIC at NU affects the decoding of its own symbol.As the SIC error Ξ increases, the NU symbol decoding probability decreases.In the Co-NM mode, NU assists  the information transfer to DU only if its symbol is successfully decoded.Thus, both NU and SU throughputs degrade with increasing SIC error.However, operating NU in hybrid mode gives immunity against SIC error as the NU switches to OM mode whenever NU is not able to decode its own symbol.Fig. 10 illustrates the DU versus P s (dBm) to compare the TS and PS protocols in the considered Co-NM framework.For a fair comparison of both the EH protocols, the curves are taken at optimum EH parameters settings, i.e., α = α * and ρ = ρ * , for TS and PS, respectively.For U n , both the EH protocols result in a similar performance for hybrid as well as the Co-NM mode.On the other hand, for U d , PS results in huge throughput gains  compared with TS (this is because with the TS protocol, IT can occur for (1 − α) fraction of the signaling interval T , whereas PS allows IT throughout the signaling interval T ).In the case of both TS and PS EH protocols, the throughput remains almost the same for low values of P s , whereas at high values of P s , PS performs significantly better than TS.However, the performance of both PS and TS is limited by the energy saturation threshold E sat .On the other hand, for L-EH, the throughput linearly increases with increasing power.For both the EH protocols, the effect of the supplementary battery energy on DU throughput is such that a small increase in battery energy helps DU to achieve dramatic gains in the throughput.for R n = R d = 2 bpcu and P s = 0 dBm, whereas the TS EH parameter and PS EH parameter are set at α = α * and ρ = ρ * , respectively.Initially, when no battery energy is applied, the EE is mainly contributed by NU (the battery energy helps in throughput gains at DU).With increasing E b , the EE of the system increases due to an increase in DU throughput.However, a unique value of E b exists beyond which the EE starts decreasing.This is because of the presence of RSI at the NU.With increasing E b , the RSI also increases, resulting in the unsuccessful decoding of DU symbols at the NU.Consequently, DU throughput decreases.Moreover, in the hybrid mode, the EE always remains better than in the Co-NM mode.It can also be seen that, with an increase in the number of S transmit antennas and the number of DUs, the Co-NM mode results in performance similar to that of the hybrid mode of operation (for large M and L, it is very likely that decoding of both DU symbols and NU symbols does not fail at NU).However, when the achievable target rate is high, then more S transmit antennas and DUs will be required for hybrid and Co-NM modes to result in the same performance.and ρ = ρ * , respectively.For both TS and PS protocols, it is observed that for low RSI levels, the RSI does not affect the EE, whereas, at higher levels of RSI, the EE decreases dramatically in Co-NM mode.In contrast, with the hybrid mode, the EE is not affected by the RSI.This is because the NU switches to OM mode if the decoding of DU and NU symbols fails at NU due to increased RSI.With this switching to OM mode, the NU attains a maximum performance as the S only serves the NU, and the EE, therefore, does not degrade at high levels of RSI.
In Figs.For both PS and TS EH protocols, operating NU in the FD relaying mode almost doubles the EE in comparison to the HD relaying mode.This is because FD relaying allows simultaneous reception and transmission of the symbols.On the other hand, two time slots are required when NU operates in HD relaying mode.The performance of FD relaying is significantly affected by the RSI.In the case of Co-NM mode, large RSI causes a significant decrease in EE.In contrast, with the hybrid mode, RSI does not affect the EE significantly.

VI. CONCLUSION
In this article, the performance of a Co-NM-based IIoT network consisting of multiple DU IoT and a battery-assisted NU with EH capabilities was analyzed.Considering Co-NM/OM switching guarantees that the quality of service at NU is not lower due to cooperation.Considering the effect of imperfect SIC, the closed-form expressions were derived for the DU and NU outage probabilities and throughput for a practical nonlinear EH model at NU. Further, it was demonstrated that with M transmit antennas at the S, NU can achieve a full diversity of order M , while a loss in diversity at DU was observed due to the nonlinear EH.Moreover, it was established that a judicious choice of TS/PS parameter could help to maximize the EE of the considered IIoT network.Finally, it was demonstrated that using Co-NM/OM switching results in higher EE, and RSI degrades the system performance when Co-NM/OM switching is not used at NU.This article demonstrated how node-level energy considerations determine the overall network's performance.Overall, this article provided several useful insights for the efficient utilization of battery-assisted EH nodes, which will be helpful in realizing the goal of the Internet of Everything and IIoT networks.
For efficient battery energy utilization at NU, the availability of channel state information can be exploited in the future.Further, efficient energy management schemes can be devised for dense IIoT networks.
Performance of a Multiuser Cooperative IoT NOMA Network With Battery-Assisted Energy Harvesting Kamal Agrawal , Member, IEEE, Anand Jee , Graduate Student Member, IEEE, and Shankar Prakriya , Senior Member, IEEE Abstract-This article investigates a cooperative nonorthogonal multiple access (Co-NM)-based network consisting of a multiantenna source, a full-duplex energy harvesting (EH) near user (NU) Internet of Things (IoT) node, and multiple distant user (DU) IoT nodes.The source shares a direct link to the NU, while the NU augments the harvested energy by a limited amount of its battery energy to relay the information to the selected DU.Considering time-switching (TS) or power-splitting (PS) protocol, practical nonlinear EH, successive interference cancellation error, and opportunistic Co-NM/orthogonal multiple access (OMA) (OM) switching, closed-form expressions are derived for the outage probability and throughput of both DU and NU.We demonstrate that the proposed opportunistic Co-NM/OM switching can ensure a performance similar to OM at the NU without loss in DU throughput.Also, a joint optimal choice of battery energy and PS/TS parameter helps in attaining a maximum energy efficiency (EE).Moreover, Co-NM/OM switching ensures higher EE as compared with Co-NM and OM.Index Terms-Energy efficiency (EE), energy harvesting (EH), full duplex (FD), Internet of Things (IoT), nonorthogonal multiple access (NOMA), power splitting (PS), throughput, time switching (TS), user selection.

Fig. 1
Fig. 1 illustrates an IIoT downlink network comprising of a multiantenna S with M transmit antennas, an EH FD NU (denoted by U n ), and a cluster C D of L single-antenna DUs (denoted by U 1 d , U 2 d , . .., U L d). S directly communicates with U n , whereas U n assists in IT to DUs.U n harvests energy using either the PS or TS protocol as shown in Fig.2.Since the harvested energy is random in nature and often small[36], a limited amount of NU's battery energy is used to augment the harvested energy[31] and establish reliable U n − U d communication.Due to severe shadowing, the S to DU link is not considered[14].At U n , the FD operation is performed using separate transmit and receive antennas.

Fig. 4 .
Fig. 4. DU outage probability versus P s for PS EH protocol.

Fig. 5 .
Fig. 5. DU outage probability versus P s for TS EH protocol.

8 .
Throughput versus P s for PS EH protocol with SIC error.

Fig. 9 .
Fig. 9. Throughput versus P s for TS EH protocol with SIC error.

Fig. 10 .
Fig. 10.Throughput versus P s for PS and TS EH protocols.

Figs. 11
and 12 illustrate the variation of EE versus E b for PS and TS SWIPT, respectively.All the curves are plotted

Fig. 13
plots the EE versus RSI level to compare the hybrid and Co-NM modes of operation.All the curves are plotted for R n = R d = 3 bpcu, M = 2, L = 2, and E b = 5 μJ, whereas the TS EH parameter and PS EH parameter are set at α = α * 14 and 15, we compare the HD and FD relaying operations by plotting EE versus E b for PS and TS SWIPT, Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

Fig. EE
Fig. EE versus E b for PS SWIPT: between HD and FD relaying protocols.

Fig. 15 .
Fig. 15.EE versus E b for TS SWIPT: comparison between HD and FD relaying protocols.

TABLE II PARAMETERS
AND VALUES FOR TS AND PS PROTOCOLS TABLE III DECODING-BASED SWITCHING STRATEGY where γ s = E s σ 2 , γ b = E b σ 2 , and γ sat = E sat σ 2 .After SIC, the SINR to decode x n (k) is