IEEE TRANSACTIONS ON COMMUNICATIONS 1 Joint Access and

IEEE TRANSACTIONS ON COMMUNICATIONS 1

Joint Access and Fronthaul Radio ResourceAllocation in PD-NOMA Based 5G Networks

Enabling Dual Connectivity and CoMPMohammad Moltafet, Roghayeh Joda, Member, IEEE, Nader Mokari, Member, IEEE, Mohammad R. Sabagh, and

Michele Zorzi, Fellow, IEEE

Abstract—In this paper, fifth generation (5G) cellular networksunder three promising technologies, namely, dual connectivity,coordinated multi-point transmission (CoMP) and power domainnon orthogonal multiple access (PD-NOMA) are investigated. Themain aim is to maximize the downlink energy efficiency (EE) byusing both millimeter wave (mmW) and micro wave (µW) links inaccess and fronthaul, while employing CoMP and PD-NOMA. Inthis regard, joint access and fronthaul radio resource allocationfor a downlink heterogeneous cloud radio access network (H-CRAN) is considered. The proposed optimization is a mixed in-teger non-convex problem with a high computational complexitysolution and hence, the alternate search method (ASM) basedon a successive convex approximation (SCA) approach usingfractional programing is exploited. Furthermore, the convergenceof the proposed iterative resource allocation method is proved andits computational complexity is investigated. As the numericalresults show, via dual connectivity through receiving signals fromboth mmW and µW transmitters, the system EE is improved byapproximately 50%, in contrast to using only µW subcarriers(e.g., as in LTE). In addition, by applying both PD-NOMA andCoMP technologies on the µW subcarriers, the EE of the systemincreases by approximately 45%.

Index Terms—5G, dual-connectivity, NOMA, millimeter wave,CoMP, resource allocation, SCA.

I. INTRODUCTION

To fulfill the broadband requirements and satisfy the hugedemand for wireless data rate in fifth generation (5G) mobilenetworks, various new emerging technologies such as ex-ploiting higher spectrum, coordinated multi-point transmission(CoMP) and non-orthogonal multiple access (NOMA) arebeing considered. Using millimeter wave (mmW), becauseof the huge free existing spectrum, can provide the multiplegigabit-per-second data rates to support extreme mobile broadband services [2]. However, despite the significant capacitygains offered by mmW bands, two essential factors, namely,deafness and blockage, may make the mmW links unreliable[3], [4]. Therefore, these challenges cause essential constraintsin system quality of service (QoS) which have to be con-sidered especially for delay-sensitive applications [5]–[7]. In

M. Moltafet is with the Centre for Wireless Communications Ra-dio Technologies, University of Oulu, 90014 Oulu, Finland ([email protected]). R. Joda is with the Iran Telecommunication ResearchCenter, Tehran, Iran ([email protected]). N. Mokari is with the Department ofECE, Tarbiat Modares University, Tehran, Iran ([email protected]).Mohammad R. Sabagh is with the Iran Telecommunication Research Center,Tehran, Iran ([email protected]). M. Zorzi is with the Department of Infor-mation Engineering, University of Padova, Padova, Italy ([email protected]).

This work is a generalization of [1]. The work of M. Zorzi was partiallysupported by NYU Wireless.

particular, in a heterogeneous structure, two scenarios can beconsidered: 1) Each user can be connected to multiple basestations (BSs) such as a macro base station (MBS) and severalsmall base stations (SBSs) simultaneously on the same sub-carrier, a scheme called CoMP; 2) Each receiver can receiveits data from transmitters with micro wave (µW) and mmWlinks simultaneously, in a scenario where it is consideredthat each transmitter and receiver can support µW and mmWsimultaneously, a scheme called dual connectivity. In the firstmodel, heterogeneity comes from using different BSs, whilein the second model, heterogeneity comes from using variousfrequencies. Moreover, 5G radio access techniques play apivotal role in enhancing the system performance and theconnectivity capacity, which recently have received significantattention from both academic and industrial researchers. Asan appropriate candidate for 5G multiple access, power do-main non-orthogonal multiple access (PD-NOMA) has beenproposed to meet the radio access demands [8], [9]. Thistechnique applies superposition coding (SC) and successiveinterference cancellation (SIC) at the transmitter and receiversides, respectively [10]–[12]. By applying SC, each subcarrieris assigned to more than one user at the transmitter side, andby using SIC at each receiver, the corresponding transmittedsignal is detected.

A. Related Works

Related works on this topic can be categorized into fivegroups, namely: 1) µW-based heterogeneous cellular network(HCN) systems with OFDMA; 2) mmW-based systems; 3)dual connectivity-based systems; 4) µW-based systems withCoMP; 5) µW-based systems with PD-NOMA.

1) µW Based HCN Systems with OFDMA: HCN technol-ogy combined with appropriate resource allocation methodsplay a key role to improve the system performance. Therefore,they have been recently studied from different aspects in theliterature, e.g., sum rate maximization, total transmit powerminimization, fairness and energy efficiency [13]–[18].

2) MmW Based Systems: An ultra-dense (UD) wirelessnetwork using TDMA-based mmW fronthaul1 and OFDMA-based access is studied in [19]. In [20], an uplink multi-user OFDMA-based system in mmW bands is studied, wheremobile users communicate with the BS through different relay

1Fronthaul is the link between cloud center and BSs


stations. Furthermore, the authors of [20] use a high signal-to-noise ratio (SNR) regime to solve the proposed optimizationproblem. In [21], an OFDMA-based system including one BS,one relay, and multiple users and supporting a downlink point-to-multi point wireless network is proposed. However, becauseof the propagation characteristics of the mmW links, a singlerelay can not support the multiple users and high data ratedemands of 5G networks. In [22], a measurement approachis proposed to address the challenges of tracking the mmWlink direction. In [23], an end-to-end evaluation of handovermechanisms in mmW cellular is investigated and a novel dualconnectivity protocol for 5G networks is proposed.

3) Dual Connectivity Based Systems: There are various re-source management approaches considering QoS constraints indual connectivity-based systems [24]–[27]. In [24], a context-aware scheduling framework is presented to allocate bothmmW and µW resources to user applications, which allowseach user with different QoS constraints to run multiple appli-cations simultaneously. A scheme of energy-efficient resourceallocation for both µW and mmW bands is proposed in [25].In [26], a scheduling scheme with integrated device-to-devicemmW links is presented in order to transfer the blocked mmWlinks to the 4G system. The work in [28] provides a model tostudy the resource allocation problem for dual connectivity-based networks, where the cell association is decoupled in theuplink for mmW users. However, this work does not considerany QoS constraint in the proposed system.

4) µW Based Systems with CoMP: In [29], a CoMP trans-mission in an OFDMA-based HCN system is studied by con-sidering multiple-input multiple-output (MIMO) technology,in which the main aim is to maximize the weighted sum rate.The work in [30] investigates the symbol timing offset ofCoMP technology in an OFDMA-based system and proposesan algorithm to address the issues related to the symbol timingoffset.

5) µW Based Systems with PD-NOMA: In [31], the ef-fect of user pairing on the performance of single cell PD-NOMA systems is investigated. A sum rate maximization ina PD-NOMA-based system with maximum transmit powerconstraint is studied in [32], where a single BS is taken intoconsideration. In [33], the authors consider a single cell multiuser PD-NOMA-based system and propose a novel problemto maximize the system sum rate with total power and SICconstraints. To solve the proposed problem, they apply anoptimal method with high computational complexity and asub-optimal method with low complexity based on successiveconvex approximation (SCA). The authors of [34] minimizethe total transmit power by considering the various SICordering in a single cell PD-NOMA-based system. In [35], theauthors maximize the energy efficiency (EE) of a PD-NOMAbased system. In order to solve the proposed optimizationproblem they use the SCA based algorithm. In [36], a radioresource allocation approach for heterogeneous traffic in ageneralized frequency division multiplexing (GFDM)-NOMA-based HCN system is studied to maximize the system sum rate.In [37], a comparison study between PD-NOMA and sparsecode multiple access (SCMA) is investigated. Various fairnessmethods for an energy harvesting enabled PD-NOMA based

HCN system are studied in [38].Statistical data show that 8% of the worldwide energy con-

sumption was related to the Information and CommunicationTechnology (ICT) industry in 2008, and is expected to doubleby 2020. In addition, mobile communication systems accountfor 0.5% of the global energy consumption [39]. Therefore,efficient resource allocation has a pivotal role in the nextgeneration of cellular networks.

The aforementioned works do not consider the PD-NOMAapproach and CoMP technology in dual connectivity-basedHCN systems, which are necessary in 5G to meet the demandsof high data rate users. In addition, they do not consider aCRAN system in which fronthaul uses the concept of dualconnectivity and PD-NOMA approaches. We consider threeenabling technologies, namely, dual connectivity, PD-NOMAand CoMP and propose an efficient energy and spectrumresource allocation algorithm for the access and fronthaullinks. Although each of the above mentioned issues can bea topic of research by itself, in a real situation we face acomplicated scenario where they should be considered jointly,and hence our main aim is to address these technologies in aunified framework.

B. Contributions

In this paper, considering CoMP technology, we proposea downlink dual connectivity mode of a PD-NOMA-basedheterogeneous cloud radio access network (H-CRAN) systemin order to provide an energy efficient and reliable commu-nication system. In the considered system model we supposethat both mmW and µW carriers could be used for the linksamong cloud center and BSs (fronthaul links) and the linksamong BSs and users (access links). Moreover, to evaluate theperformance of the proposed system model, a novel resourceallocation problem is formulated to maximize the EE. Themain contributions of our paper are listed as follows:• We consider a downlink H-CRAN, in which each receiver

and transmitter in both fronthaul and access can receiveand transmit data on mmW and µW links simultaneously.

• We consider that the µW subcarriers in both access andfronthaul can be assigned to more than one receiver si-multaneously by applying a PD-NOMA approach, whichimproves EE and system sum rate.

• We also consider two heterogeneous connectivity models:I) dual connectivity in which each user can receive signalsfrom mmW and µW transmitters; and II) CoMP in whicheach user can receive the corresponding signal on thesame subcarrier from multiple BSs simultaneously.

• To evaluate the performance of the considered systemmodel, we propose a novel resource allocation problem tomaximize the system EE by considering dual connectivityin both fronthaul and access, CoMP technology, and PD-NOMA technique.

• To solve the proposed optimization problem, the ASMbased on a successive convex approximation (SCA) ap-proach using fractional programing is applied. Moreover,the convergence of the exploited iterative algorithm isproved and also its computational complexity is studied.


Fig. 1: Frequency bands and subcarriers.

The remainder of this paper is organized as follows: InSection II, the system model and problem formulation areproposed. In Section III, the SCA-based iterative algorithmsare developed. The proof of the convergence of the proposedsolution method is presented in Section IV. In Section V,the computational complexity of the proposed solution isinvestigated. Numerical results are presented in Section VI toevaluate the performance of the proposed algorithms. Finally,the conclusions are given in Section VII.

II. SYSTEM MODEL AND PROBLEM FORMULATION

We consider a downlink H-CRAN system with F −1 SBSsand one MBS. Also, we assume that both mmW and µWsubcarriers are exploited in the fronthaul and access networksto transmit the data, and that each user and BS can supportmmW and µW bands simultaneously. Moreover, we consider aPD-NOMA approach on access and fronthaul links over µWsubcarriers to improve both EE and spectral efficiency. Thenarrow beams, due to the directionality and high attenuationfeatures of mmW links, imply a small number of users ineach beam, which makes the implementation of PD-NOMAquite hard. Therefore, we assume that the mmW subcarriersare assigned to receivers based on an OFDMA approach. Inaddition, we suppose that each user can receive its data onµW by using the CoMP technology. Based on the CoMPtechnology, each user assigned to a BS can also receive its datafrom the other base stations, simultaneously [40]. In addition,in a PD-NOMA based system, in order to calculate the intracell interference, it is important to determine which user isassigned to which BS, because the intra cell interference iscussed by the signals of other users assigned to the sameBS in the same subcarrier with less SNR. It should be notedthat, in order to implement the CoMP technology, we assumedthat the coordinating commands among different base stationsare scheduled in the central processor (CP). Furthermore, weconsider that the mmW links in LOS conditions can transmitdata. Therefore, a random variable for LOS communication isdefined. As seen in Fig. 1, there are several subcarriers in bothmmW and µW spectrum bands where each µW subcarriercan be assigned to more than one user by applying SC. A

Fig. 2: Typical illustration of the considered network

typical illustration of the system model is shown in Fig. 2.Here, for simplicity perfect channel state information (CSI)is considered to get some preliminary results and understandwhether the considered technique is promising, leaving a moredetailed and realistic study as future work. In the consideredsystem model, K1 = {1, . . . ,K} is the set of all µW sub-carriers, K2 = {1, . . . , S} is the set of all mmW subcarriers,F = {1, . . . , F} is the set of BSs and Af is the set of Af usersbelonging to BS f , in which ∪f∈FAf = A. The definitions ofall variables are summarized in Table I. Considering the abovedefinitions, the system parameters in access and fronthaul aredefined as follows:

1) Access link parameters of µW band: pk,fa and hk,farepresent respectively the transmit power and channel gainfrom BS f to user a on subcarrier k. The binary variable xk,faindicates subcarrier assignment as xk,fa = 1 if subcarrier k isassigned to user a over BS f and xk,fa = 0 otherwise.

2) Access link parameters of mmW band: qs,fa indicatesthe transmit power from BS f to user a on subcarrier s, gs,farepresents the channel gain from BS f to user a over subcarriers, and τs,fa is a binary variable which indicates the subcarrierassignment as τs,fa = 1 if subcarrier s is assigned to user aover BS f and τs,fa = 0 otherwise.


3) Fronthaul parameters of µW band: wk,f represents thetransmit power from the CP to BS f on subcarrier k, ηk,f

represents the channel gain from CP to BS f over subcarrierk, and yk,f is a binary variable which indicates the subcarrierassignment as yk,f = 1 if subcarrier k is assigned to BS fand yk,f = 0 otherwise.

4) Fronthaul link parameters of mmW band: zs,f indicatesthe transmit power from CP to BS f on subcarrier s, ζs,f

represents the channel gain from CP to BS f over subcarriers, and us,f is a binary variable which indicates the subcarrierassignment as us,f = 1 if subcarrier s is assigned to BS fand otherwise us,f = 0.

TABLE I: NOTATIONS

Notation Descriptionwmask Maximum power which can be transmitted on a µW link in

fronthaulqmask Maximum power which can be transmitted on a mmW

access linkpmask Maximum power which can be transmitted on a µW

access linkK1 Set of µW subcarriersK2 Set of mmW subcarriersK Number of µW subcarriersS Number of mmW subcarriersramin Minimum rate requirement for user a(σk

a)2 Noise power from µW BS to user a over subcarrier k

pk,fa Transmit power from µW BS f to user a over subcarrier kqs,fa Transmit power from mmW BS f to user a over

subcarrier swk,f Transmit power from µW CP to BS f on subcarrier kzs,f Transmit power from mmW CP to BS f on subcarrier shk,fa µW channel gain from BS f to user a over subcarrier kgs,fa mmW channel gain from BS f to user a over subcarrier sηk,f µW channel gain from CP to BS f over subcarrier kζs,f mmW channel gain from CP to BS f over subcarrier sxk,fa xk,fa = 1 if subcarrier k is assigned to user a over µW

BS f , xk,fa = 0 otherwiseτs,fa τs,fa = 1 if subcarrier s is assigned to user a over mmW

BS f , τs,fa = 0 otherwiseyk,f yk,f = 1 if subcarrier k is assigned to µW BS f ,

yk,f = 0 otherwiseus,f us,f = 1 if subcarrier s is assigned to mmW BS f ,

us,f = 0 otherwisehk,ca µW channel gain from CP to user a over subcarrier k

hf,f′

k µW channel gain from BS f to BS f ′ over subcarrier kPmaxf Maximum available transmit power in downlink

transmission at BS fPmax

CP Maximum available power in downlink transmission at CPAf Set of users attached to the f -th BSF Set of BSsP

[pk,fa

]Q

[qs,fa

]W

[wk,f

]v2 [ ~Q]

v1 [ ~P , ~W]

U[us,f

]Y

[yk,f

]X

[xk,fa

]τ

[τs,fa

]Z

[zs,f

]

A. Signal Model

As mentioned, PD-NOMA technique is used for fronthauland access in µW links, whereas OFDMA is exploited forfronthaul and access mmW links. In the following, the signalmodels for mmW and µW links are explained.

1) µW Signal model: By applying the PD-NOMA ap-proach, µW subcarriers on both access and fronthaul linkscan be assigned to more than one receiver simultaneously,by using SC at the transmitter side. Thus, each user detectsits signals at the receiver side by exploiting SIC. With PD-NOMA, each user successfully removes the signals of theusers with higher SINR using SIC, and considers the signalof other users as noise during the detection process [41].Moreover, each µW access link may contain three interferenceterms, namely I) Intracell interference which comes from SIC;II) Intercell interference which comes from the other cellsusing the same subcarrier; and III) Fronthaul interferencewhich comes from the fronthaul links employing the samesubcarrier. Consequently, the achievable µW rate for user a onsubcarrier k over BS f is expressed as r1k,fa = log(1+γ1k,fa ),where γ1k,fa is the SINR of user a which was assigned to BSf on subcarrier k and obtained as

γ1k,fa =

∑f ′∈F p

k,f ′

a xk,f′

a |hk,f ′

a |2

Ik,fa (Inter) + Ik,fa (Intra) + Ik,fa (Front) + (σka)2,

(1)

in which Ik,fa (Inter), the intercell interference atuser a on subcarrier k of BS f , is given byIk,fa (Inter) =

∑f ′∈F/{f}

∑a′∈Af′ x

k,f ′

a pk,f′

a |hk,f ′

a |2,Ik,fa (Intra), the NOMA interference, is expressedas Ik,fa (Intra) =

∑i∈Af ,|hk,f

i |>|hk,fa | x

k,fi pk,fi |hk,fa |2,

Ik,fa (Front) is the fronthaul interference and is given byIk,fa (Front) =

∑f ′′∈F y

k,f ′′wk,f

′′ |hk,ca |2, and (σka)2 indicates

the noise power for user a on subcarrier k of BS f . It isworth noting that the numerator of γ1k,fa indicates that,because of the use of CoMP technology, user a can receiveits data from different BSs simultaneously. In the PD-NOMAbased systems, in order to calculate the intra cell interferenceit is important to determine each user assigned to which BS.In other words, the intra cell interference of user a assignedto base station f comes from the other users with betterchannel condition assigned to base station f . Moreover,each µW link in fronthaul may contain two interferenceterms, namely I) PD-NOMA interference which comes fromSIC, and II) Access interference which comes from the BSsusing the same subcarrier. Consequently, the maximum rateat which CP can transmit to BS f over µW is given byrh1k,f = log(1 + γh1k,f ), where γh1k,f is the SINR at BSf on subcarrier k and is obtained as

γh1k,f =wk,fyk,f |ηk,f |2

Ik,f (Access) + Ik,f (NOMA) + (σk)2, (2)

in which Ik,f (Access), the interference on subcarrier kfrom the access BSs, is calculated as Ik,f (Access) =∑f ′∈F/{f}

∑a′∈Af′ x

k,f ′

a′ pk,f′

a′ |hf,f ′

k |2, Ik,f (NOMA) is theNOMA interference and is given by Ik,f (NOMA) =


∑i∈F,|ηk,i|>|ηk,f | y

k,iwk,i|ηk,f |2, and (σk)2 indicates thenoise power for BS f on subcarrier k. It is worth noting that,because each BS knows its own transmitted signal, we supposethat it can subtract it from the received one. Therefore, thereis no self interference at the received signal.

2) mmW Signal Model: The achievable mmW rate fromsubcarrier s for user a of BS f is formulated as r2s,fa =log(1+γ2s,fa ), where γ2s,fa is the SINR at BS f on subcarriers and is given by

γ2s,fa =qs,fa ξfaτ

s,fa |gs,fa |2

(σsa)2

, (3)

where (σsa)2 is the noise power for user a on subcarrier s of

BS f and ξfa is a Bernoulli random variable indicating thefeasibility of a LOS link from BS f to user a with probabilityof success ρa [24]. In fact, the feasible and infeasible LOSlink from BS f to user a is shown by ξfa = 1 and ξfa = 0,respectively. In addition, the rate at which CP transmits toBS f over mmW is rh2s,f = log(1 + γ2s,f ), where (σs)2

indicates the noise power for BS f on subcarrier s and γ2s,f

is the SINR at BS f on subcarrier s and is obtained as

γ2s,f =zs,fξ′fus,f |ζs,f |2

(σs)2, (4)

where ξ′f is a Bernoulli random variable indicating thefeasibility of LOS link from CP to BS f with probabilityof success ρ′f . It is worth noting that the attenuation inthe mmW links is very high, therefore, the effect of intercell interference is very low. Moreover, we suppose that themmW links are significantly directional. Therefore, as anapproximation, we do not consider inter cell interference forthe mmW links in (3) and (4). The total rate between BSf and its assigned users over both mmW and µW is givenby rf =

∑a∈Af

∑k∈K1

r1k,fa +∑a∈Af

∑s∈K2

E{r2s,fa },where E{.} indicates the expectation of r2s,fa . Moreover, thetotal rate from CP to BS f over both mmW and µW is obtainedas rhf =

∑k∈K1

rh1k,f +∑s∈K2

E{rh2s,f}. In CRANsystems, the total rate of each BS should be smaller than thetotal rate of the CP that transmits to it. Consequently, thefollowing constraint has to be considered rhf ≥ rf∀f ∈ F .Accordingly, we define a utility function as the total achiev-able rate as follows: UT =

∑f∈F

∑a∈Af

∑k∈K1

(r1k,fa +

rh1k,f)+∑f∈F

∑a∈Af

∑s∈K2

(E{r2s,fa } + E{rh2s,f}

).

Furthermore, the total power dissipation for both mmW andµW is obtained as UP =

∑f∈F

∑a∈Af

∑k∈K1

(xk,fa pk,fa +

wk,fyk,f )+∑f∈F

∑a∈Af

∑s∈K2

(τs,fa qs,fa +zs,fus,f )+Pc,where Pc represents the fixed energy costs related to thecircuitry.

B. Necessary Conditions to Perform SIC

Based on information theory, in the PD-NOMA approach,user a can successfully detect the signals of other users, if theSINR of each of those users at user a is higher than its ownSINR [12], [33]. Consequently, the following constraints haveto be considered to perform perfect SIC in both access andfronthaul.

1) Necessary Conditions to Perform SIC in Access: TheSIC at user a on the µW subcarriers in access is performedsuccessfully if we have (5) [12], [33], where a and j representusers which are assigned to subcarrier k.

2) Necessary Conditions to Perform SIC in Fronthaul: Forthe fronthaul µW subcarriers we have [12], [33]

wk,f |ηk,j |2

Ik,j(Access) + Ik,f (NOMA) + (σk)2≥ (6)

wk,f |ηk,f |2

Ik,f (Access) + Ik,f (NOMA) + (σk)2,

∀ k ∈ K1, f, j ∈ F , |ηk,j |2 > |ηk,f |2,

where f and j denote BSs which are assigned to subcarrierk. By straightforward calculation we can see that (5) and (6)are linear constraints.

C. Problem formulation

Considering the above definitions, our goal is to maximizethe system EE. Therefore, the corresponding optimizationproblem is formulated as follows:

Problem 1:

maxP,Q,W ,Z,τ ,X,Y ,U

UT

UP(7a)

s.t. :∑f∈F

∑k∈K1

r1k,fa +∑f∈F

∑s∈K2

E{r2s,fa } ≥ ramin,∀a ∈ A,

(7b)

0 ≤ pk,fa ≤ pmask, 0 ≤ qs,fa ≤ qmask, (7c)

0 ≤ wk,f ≤ wmask, 0 ≤ zs,f ≤ zmask

xk,fa ∈ {0, 1}, yk,f ∈ {0, 1}, ∀a ∈ A, f ∈ F , k ∈ K1,(7d)∑

a∈Af

xk,fa ≤ LT ,∀f ∈ F , k ∈ K1, (7e)

∑i∈F

yk,i ≤ LT ,∀, k ∈ K1, (7f)

τs,fa ∈ {0, 1}, us,f ∈ {0, 1}, ∀a ∈ A, f ∈ F , s ∈ K2,(7g)∑

a∈Af

τs,fa ≤ 1,∀f ∈ F , s ∈ K2, (7h)

∑i∈F

us,i ≤ 1,∀, s ∈ K2, (7i)

rhf ≥ rf ,∀f ∈ F , (7j)∑k∈K1

∑a∈A

xk,fa pk,fa +∑s∈K2

∑a∈A

τs,fa qs,fa ≤ Pmaxf , (7k)

∀f ∈ F ,∑k∈K1

∑f∈F

yk,fwk,f +∑s∈K2

∑f∈F

us,fzs,f ≤ PmaxCP ,

(7l)(5), (6).

where ramin is the minimum rate requirement for user a, Pmaxf

is the maximum available transmit power in downlink at BSf , Pmax

CP is the maximum available power in downlink at CP.Moreover, the constraints of (7) are explained as follows: (7b)


pk,fa |hk,fj |2

Ik,fj (Inter) + Ik,fa (Intra) + Ik,fj (Front) +∑f ′∈F/f p

k,f ′

j xk,f′

j |hk,fj |2 + (σkj )2≥ (5)

pk,fa |hk,fa |2

Ik,fa (Inter) + Ik,fa (Intra) + Ik,fa (Front) + (σka)2, ∀ k ∈ K1, f ∈ F , a, j ∈ Af , |hk,fj |

2 > |hk,fa |2,

indicates the minimum rate requirement of each user, (7e)and (7f) show the PD-NOMA approach for µW in accessand fronthaul links respectively, where each subcarrier can beassigned to at most LT receivers, (7h) and (7i) represent thatthe mmW links in each BS can be assigned to at most oneuser, (7j) demonstrates that the achievable rate from each BSin access must be smaller than the rate that fronthaul linksgive it, (7k) indicates the power limitation at each BS, and (7l)represents the maximum available transmit power for fronthaullinks. Problem (7) is complex because i) there are integer andcontinues variables, and ii) both the objective function andconstraints (7b) and (7j) are non-convex. Therefore, existingsolutions for convex optimization can not be exploited directly.

III. SOLUTION ALGORITHM

To solve the optimization problem, we use the alternat-ing method in which at each iteration power and subcarrierare assigned separately and the algorithm is continued untilconvergence is achieved [43]. Suppose Pt(P,ρ) shows theoptimization problem which should be solved at iteration t.Based on the alternate method, at iteration t, we have

• • • −→

Iteration t︷︸︸︷Pt(P,ρt−1)→ Pt(Pt,ρ) • • • (8)

where Pt(P,ρt−1) shows the optimization problem of find-ing power allocation variables while subcarrier assignmentparameters were determined in the previous iteration. In otherwords, in the optimization problem Pt(P,ρt−1), subcarrierassignment parameters are not variables, but fixed parameterswhich were determined in iteration t−1. In the following, wefirst present the solution of the power allocation problem, thenexplain the solution of the subcarrier assignment and finallydescribe the overall solution algorithm. The power allocationand subcarrier assignment problems that need to be solved ineach iteration are as follows:• Power allocation problem by considering a fixed subcar-

rier assignmentProblem 2:

maxP,Q,W ,Z

UT

UPs.t. : (7b), (7c), (7j)− (7l), (5), (6).

(9)

• Subcarrier assignment problem by considering a fixedpower allocationProblem 3:

maxτ ,X,Y ,U

UT

UPs.t. : (7b)− (7l), (5), (6). (10)

A. Solution to the Power Allocation Problem

To solve optimization problem 2, we first use the SCAapproach to transform the main problem into the standardform of fractional programming, and then apply the fractionalprogramming theorem [44]. In order to apply the fractionalprogramming theorem to solve an optimization problem withfractional form objective function, the constraints of the op-timization problem should be convex, and the numerator anddenominator of the objective function should be concave andconvex, respectively. In the sequel, we apply the differenceof two concave functions (DC) approximation to transformoptimization problem 2 into the standard form of fractionalprogramming.

1) DC Approximation to Convert Optimization Problem2 to Standard Form of Fractional Programming: The DCapproximation, by utilizing a linear approximation, replacesa non-convex function by a convex one [45]. To apply the DCmethod, we rewrite (9) with fixed subcarrier assignment:

Problem 4:

maxP,Q,W ,Z

(∑f∈F

∑a∈Af

∑k∈K1

({log(ψ1)− log(ψ′

1)}+

{log(φ1)− log(φ′

1)}) +∑f∈F

∑a∈Af

∑s∈K2

(E{log(ψ2)− log(σsa)2}+ E{log(φ2)

− log(σs)2}))/UP,

s.t. : (7c)(7l), (5), (6), (11a)∑f∈F

∑k∈K1

{log(ψ1)− log(ψ′

1)}+ (11b)∑f∈F

∑s∈K2

{E{ log(ψ2)− log(σsa)2} ≥ ramin,

∀a ∈ A,

(∑k∈K1

{log(φ1)− log(φ′

1)}+∑s∈K2

E{log(φ2)− log(σs)2})−

(∑a∈Af

∑k∈K1

{log(ψ1)− log(ψ′

1)}+ (11c)

∑a∈Af

∑s∈K2

E{log(ψ2)− log(σsa)2}) ≥ 0,

∀f ∈ F ,

where

ψ1 = Ik,fa (Inter) + Ik,fa (Intra) + Ik,fa (Front) + (σka)2+


∑f∈F

pk,fa xk,fa |hk,fa |2,

ψ′

1 = Ik,fa (Inter) + Ik,fa (Intra) + Ik,fa (Front) + (σka)2,

φ1 = Ik,f (Access) + Ik,f (NOMA) + (σk)2+

wk,fyk,f |ηk,f |2,φ

′

1 = Ik,f (Access) + Ik,f (NOMA) + (σk)2,

ψ2 = (σsa)2 + qs,fa ξs,fa τs,fa |gs,fa |2,

φ2 = (σs)2 + zs,fξ′s,fus,f |ζs,f |2.

The basic idea behind the DC method is to exploit a firstorder Taylor approximation. For example, if the main functioncan be written as the difference of two concave functionsMF = CF1 − CF2, by applying the DC method the secondterm (CF2) is approximated by a linear function. In order toapproximate (11) by a convex optimization problem with theDC method, the following functions should be replaced bytheir first-order Taylor approximations

G1 =∑i∈F

∑a∈Af

∑k∈K1

log(ψ′

1), G2 =∑i∈F

∑a∈Af

∑k∈K1

log(φ′

1),

(12)

G′1 =

∑i∈F

∑k∈K1

log(ψ′1), G′′1 =

∑a∈Ai

∑k∈K1

log(ψ1),

G′′2 =

∑k∈K1

log(φ′1), G′′3 =

∑a∈Ai

∑s∈K2

log(ψ2).

Consequently, by applying the DC approximation,G1, G

′1, G

′′1, G

′′2 and G

′′3 at iteration j are given by

G1(j) ≈ G1(v1(j−1))+∇TG1(v1(j−1))(v1(j)−v1(j−1)),in which v1 = [ ~P , ~W] where ~P represents the vectorizedversion2 of matrix P and ∇TG1 is a vector of lengthKF (A+ 1) and its k, f entry is given by

∂G1

∂pk,fa= (13)

∑i∈F/{f ′}

∑a′∈Ai

xk,fa |hk,fa |2

ψ′1

+∑

a′∈Ai,|hk,i

a′,t|<|hk,ia |

xk,fa |hk,fa |2

ψ′1

,

∂G1

∂wk,f=∑i∈F

∑a′∈Ai

yk,f |hk,ca |2

ψ′1

.

G2(j) ≈ G2(v1(j−1))+∇TG2(v1(j−1))(v1(j)−v1(j−1)),where

∂G2

∂pk,fa=

∑i∈F/{f ′}

∑a′∈Ai

xk,fa |hf,f ′

k |2

φ′1

,

∂G2

∂wk,f=

∑i∈F,|ηk,i|<|ηk,f

t |

∑a′∈Ai

yk,f |ηk,f |2

φ′1

.

G′1(j) ≈ G′

1(v1(j−1))+∇TG′1(v1(j−1))(v1(j)−v1(j−

1)), where

∂G′1

∂pk,fa=

∑i∈F/{f}

xk,fa |hk,fa |2

ψ′1+

∑a′∈Ai,|hk,i

a′,t|<|hk,ia |

xk,fa |hk,fa |2

ψ′1,

2The vectorized version of a matrix is obtained by concatenating its rowsinto a vector

∂G′1

∂wk,f=∑i∈F

yk,f |hk,ca |2

ψ′1, G

′′1(j) ≈ G

′′1(v1(j − 1)) +

∇TG′′1(v1(j − 1))(v1(j) − v1(j − 1)), where

∂G′′1

∂pk,faand

∂G′′1

∂wk,fare given by (14).

G′′2(j) ≈ G

′′2(v1(j − 1)) +∇TG′′

2(v1(j − 1))(v1(j) −v1(j − 1)), where

∂G′′2

∂pk,fa=

xk,fa |h

f,f ′

k |2

φ′1, if : f 6= f ′

0, if : f = f ′

∂G′′2

∂wk,f=

∑i∈F,|ηk,i|<|ηk,f′

t |yk,f |ηk,f |2

φ′1,G

′′3(j) ≈

G′′3(v2(j − 1)) + ∇TG′′

3(v2(j − 1))(v2(j) − v2(j − 1)),in which v2 = [ ~Q] and

∂G′′3

∂qs,fa=

0, if : f 6= f ′

ξs,fa τs,fa |gs,fa |2

ψ2, if : f = f ′.

Consequently, by applying the DC approximation, a stan-dard form of the fractional programming is achieved asfollows:

Problem 4:

maxP,Q,W ,Z

UT

UPs.t. : ˆ(7b), (7c), ˆ(7j), (7k), (7l). (15)

where ˆ(7b) and ˆ(7j) are convex approximations of (7b) and(7j), respectively.

By applying the DC approximation, the first term of thefunctions which were approximated are in form log(

∑), and

therefore are concave. The second terms of these functions,by applying the DC approximation are transformed into linearfunctions. Finally, summation of a concave and a linearfunction is a concave function.

2) Dinkelbach Method to Solve the Power Allocation Prob-lem: To solve problem 4, an iterative algorithm, namely,Dinkelbach method by using the fractional programming the-orem (Theorem 1) [44], is applied.

Theorem 1. Fractional programming Theorem [44]: The

maximum value of E∗ = UT

UP(where UT is an approximation

of UT ) is achievable if and only if

max UT (P,Q,W ,Z)− E∗UP (P,Q,W ,Z) (16)

= UT (P∗,Q∗,W ∗,Z∗)− E∗UP (P∗,Q∗,W ∗,Z∗) = 0,

for concave UT and convex UP .

The proof is given in [44]. By applying Theorem 1, anequivalent optimization problem with the objective functionin subtractive form, e.g., UT − E∗ × UP , can be achieved.Based on the fractional programming in each iteration, thefollowing problem with constant E must be solved.

Problem 5:


∂G′′1

∂pk,fa=

∑a′∈Ai,|hk,f

a′,t|<|hk,fa |

xk,fa |hk,fa |2

ψ1+xk,fa |hk,fa |2

ψ1+∑a∈Ai

xk,fa |hf,f ′

k |2

ψ1, f ′ 6= f∑

a′∈Ai,|hk,f

a′,t|<|hk,fa |

xk,fa |hk,fa |2

ψ1+xk,fa |hk,fa |2

ψ1, f ′ = f,

(14)

∂G′′1

∂wk,f=∑a∈Ai

yk,f |hk,ca |2

ψ1,

maxP,Q,W ,Z

UT − E × UP, s.t. : ˆ(7b), (7c), ˆ(7j), (7k), (7l).

(17)

The steps of the Dinkelbach method are presented in Algo-rithm 1. As seen in this algorithm, the first step considers thatE = 0, then with a given E the power allocation problem insubtractive form (Problem 5) is solved, and finally, with theachieved power allocation, E is updated. The power allocationand updating E is continued until convergence is achieved. Theoutput of the described algorithm is {P∗,Q∗,W ∗,Z∗} whichmaximizes the objective function of problem 5. In Algorithm1, the optimization problem (17) in step 1-a is solved by usingpublicly available software such as CVX [46].

Algorithm 1 Dinkelbach algorithm to solve power allocationInitialization: Set i = 0 (i is the iteration number) andE = 0 (initialization of EE), initialize the maximum numberof iterations Imax and maximum tolerance ε1,Repeat step 1-a and step 1-b until convergence:1-a: Solve problem 5 for a given E .1-b: When i 6= 0 and UT

i− E i−1 × UP i < ε1 or i = Imax,

stop and return {P,Q,W ,Z} = {P∗,Q∗,W ∗,Z∗} andE∗ = E i−1.

Otherwise, Set E i = UTi

UP i, i = i+1 and go back to step 1-a.

Output: {P,Q,W ,Z}.

B. Subcarrier Assignment

The subcarrier assignment problem given a fixed powerallocation is formulated as

maxτ ,X,Y ,U

UT

UPs.t. : (7b)− (7l), (5), (6). (18)

In order to reduce the computational complexity, we relaxthe subcarrier assignment to real values between zero andone [47]. Then, xk,fa is interpreted as the portion of timethat subcarrier k is assigned to user a in BS f [47], [48].In order to solve the subcarrier assignment, at first we applythe DC approximation to transform the subcarrier assignmentproblem into the standard form of fractional programming.After applying the DC approximation method, the subcarrierassignment problem is written as

maxτ ,X,Y ,U

UT

UP(19)

s.t. : (7b), (7j), (7h), (7i), (7k), (7l), (7e), (7f), (5), (6),

0 ≤ xk,fa ≤ 1, 0 ≤ yk,f ≤ 1, 0 ≤ us,f ≤ 1, 0 ≤ τs,fa ≤ 1,

where (7b) and (7j) are convex approximations of (7b) and(7j), respectively. To solve problem (19), the Dinkelbachmethod by using the fractional programming theorem is ap-plied. Based on the fractional programming in each iteration,the following problem with constant E must be solved

maxτ ,X,Y ,U

UT − E × UP (20)

s.t. : ˆ(7b), ˆ(7j), (7h), (7i), (7k), (7l), (7e), (7f), (5), (6),

0 ≤ xk,fa ≤ 1, 0 ≤ yk,f ≤ 1, 0 ≤ us,f ≤ 1, 0 ≤ τs,fa ≤ 1.

The steps of the Dinkelbach method are presented in Al-gorithm 2. To solve the subcarrier assignment problem withsubtractive form in each iteration, the CVX solver is exploited.

Algorithm 2 Dinkelbach algorithm to solve subcarrier assign-mentInitialization: Set i = 0 (i is the iteration number) andE = 0 (initialization of EE), initialize the maximum numberof iterations Imax and maximum tolerance ε3,Repeat step 2-a and step 2-b until convergence:2-a: Solve the approximated subcarrier assignment problemwith subtractive form for a given E .2-b: When i 6= 0 and UT

i− E i−1 × UP i < ε3 or i = Imax,

stop and return {τ ,X,Y ,U} = {τ ∗,X∗,Y ∗,U∗} andE∗ = E i−1.

Otherwise, Set E i = UTi

UP i, i = i+1 and go back to step 2-a.

Output: {τ ,X,Y ,U}.

C. Overall Solution AlgorithmAs denoted, optimization Problem 1 can be solved by

applying the alternating method. The overall solution is sum-marized in Algorithm 3. As shown in this algorithm, the set ofpower variables are determined by considering fixed subcarrierassignment, then the set of subcarrier variables are calculatedby considering fixed power allocation, and this process iscontinued until convergence is achieved.

IV. CONVERGENCE OF THE ITERATIVE ALGORITHM WITHDC APPROXIMATION

Theorem 2. The DC approach creates a sequence of enhancedsolutions for the optimization problem (7) and converges to alocal optimum.


Algorithm 3 Iterative resource allocation to solve the mainproblemInitialization: Set ii = 0 (ii is the iteration number),initialize the maximum number of iterations IImax andmaximum tolerance ε0, and initialize {τ 0,X0,Y 0,U0},Repeat step 3-a to step 3-d until convergence:3-a: Set {τ ii,Xii,Y ii,U ii}, approximate the non-convexconstraints by convex constraints, and approximate thenumerator of the objective function by a concave function byapplying the DC method presented in Subsection III.B.1,3-b: Solve Problem 2 with the Dinkelbach algorithm(Algorithm 1) to achieve {P,Q,W ,Z},3-c: Set {P,Q,W ,Z} = {Pii+1,Qii+1,W ii+1,Zii+1},approximate the non-convex constraints by convex constraints,and approximate the numerator of the objective function by aconcave function by applying the DC method,3-d: Solve Problem 3 with the Dinkelbach algorithm toachieve {τ ,X,Y ,U},3-f: When convergence or ii = IImax,stop and return {P,Q,W ,Z, τ ,X,Y ,U} ={P∗,Q∗,W ∗,Z∗, τ ∗,X∗,Y ∗,U∗}.Otherwise, go back to step 3-a.Output: {P,Q,W ,Z, τ ,X,Y ,U} and maximum EE.

Proof. To solve optimization Problem 2, in each iterationwe solve a sequence of instances of optimization Problem 5via the DC approximation. For the sake of simplicity in thenotation, we consider the objective function of Problem 2 as

F(P,Q,W ,Z, τ ,X,Y ,U) =

F1(P,Q,W ,Z, τ ,X,Y ,U)−F2(P,Q,W ,Z, τ ,X,Y ,U),

in which F2(P,Q,W ,Z, τ ,X,Y ,U) = G1 + G2. Byexploiting the DC approximation, function F2 is approximatedby a linear function in each iteration. Since F2 is a concavefunction, its gradient ∇(F2(P,Q,W ,Z)) is also its super-gradient [51]. Therefore, in iteration j we have

∇(F2(Pj ,Qj ,W j ,Zj)) ≤ F2(Pj−1,Qj−1,W j−1,Zj−1)(21)

+∇(F2(Pj−1,Qj−1,W j−1,Zj−1))(PST j − PST j−1),

where PST = [~P ~Q ~W ~Z]. Then, from (21), we have

F(Pj ,Qj ,W j ,Zj) = F1(Pj ,Qj ,W j ,Zj)− (22)

F2(Pj ,Qj ,W j ,Zj , τ j) ≥ F1(Pj ,Qj ,W j ,Zj)−(F2(Pj−1,Qj−1,W j−1,Zj−1)+

∇(F2(Pj−1,Qj−1,W j−1,Zj−1))(PST j − PST j−1)) =max

P,Q,W ,ZF1(P,Q,W ,Z)−

(F2(Pj−1,Qj−1,W j−1,Zj−1)+

∇(F2(Pj−1,Qj−1,W j−1,Zj−1))(PST j − PST j−1))≥ F1(Pj−1,Qj−1,W j−1,Zj−1)−F2(Pj−1,Qj−1,W j−1,Zj−1).

Therefore, after iteration j, the objective function of Problem2 either improves or stays unaltered as the value at iteration

j − 1. Consequently, as presented in [45], the SCA approachwith DC approximation is guaranteed to converge to a localoptimum.

Theorem 3. By applying the iterative algorithm presented inAlgorithm 1, at each iteration the objective value of (17)is improved. The improvement of the objective function iscontinued until convergence.

Proof. By considering fixed subcarrier assignment,{τ j ,Xj ,Y j ,U j} and calculating the power allocationof iteration j + 1, {Pj+1,Qj+1,W j+1,Zj+1}, we have

UTT (τ j ,Xj ,Y j ,U j ,Pj ,Qj ,W j ,Zj) ≤ (23)

UTT (τ j ,Xj ,Y j ,U j ,Pj+1,Qj+1,W j+1,Zj+1),

where UTT represents the objective function. Inequality(23) follows from the fact that the worst solution forthe optimization problem with variables {P,Q,W ,Z} is{Pj ,Qj ,W j ,Zj}. Similarly, by considering the fixed powerallocation {P,Q,W ,Z} = {Pj+1,Qj+1,W j+1,Zj+1},and finding the subcarrier allocation of iteration j + 1,{τ j+1,Xj+1,Y j+1,U j+1}, we have (24). From (23) and(24), we can see that after each iteration, the objective functionincreases or stays unaltered compared to the previous iteration.Since for a finite set of transmit powers and channel gains,the optimal sum rate is bounded above, the procedure mustconverge [43].

A. Approximation Accuracy

In order to evaluate the approximation accuracy, we considera logarithmic function and, by applying the DC approxima-tion under different parameters, we depict the original andapproximated function. The considered logarithmic function is

Y = log(1+h1x

1 + h2x), where h1 and h2 are constant param-

eters. The approximated function using the DC approximationis given by Y ≈ log(1 + (h1 + h2)x) −

(log(1 + h2x0) +

(h2

1 + h2x0)(x−x0)

), where x0 is a constant parameter. Figs.

3 and 4 illustrates the error and relative error for differentvalues of h1, h2, respectively.

From these figures, we can see that for small values ofh1 and h2 the DC approximation has acceptable accuracy.In the proposed solution method, due to the fact that theentries of the channel gain matrices have small values, theDC approximation has good accuracy.

V. COMPUTATIONAL COMPLEXITY

To solve the main optimization problem, the alternatingmethod is used [43], [49]. By exploiting this method, in eachiteration, two subproblems, namely, I) Power allocation, andII) Subcarrier allocation with user association are solved. Thecomplexity of the alternate method is a linear function ofthe total number of iterations needed to converge, and ofthe solution complexity of each subproblem. In the proposedsolution method, in each iteration for both subproblems, theDinkelbach method is used, which is an iterative algorithm. Ineach iteration of the Dinkelbach method, the CVX toolbox is


UTT (τ j ,Xj ,Y j ,U j ,Pj+1,Qj+1,W j+1,Zj+1) ≤ UTT (τ j+1,Xj+1,Y j+1,U j+1,Pj+1,Qj+1,W j+1,Zj+1). (24)

Fig. 3: Error.

0 0.2 0.4 0.6 0.8 1

X

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Rela

tive E

rror

X0=0.01, h

1=h

2=10

-3

X0=0.01, h

1=h

2=10

-2

X0=0.01, h

1=h

2=10

-1

X0=0.01,h

1=h

2=1

Fig. 4: Relative error.

used to solve problems. This toolbox exploits the interior pointmethod to solve an optimization problem [46]. Therefore, thecomputational complexity of each subproblem is given by [46],

[50] C =log(NCν1ν2

)

log(ν3), where NC is the number of constraints

in each subproblem, ν1 is the initial point to approximatethe accuracy of the interior point method, ν2 is the stoppingcriterion of the interior point method, and ν3 is used to updatethe accuracy of the interior point method [46], [50].

VI. NUMERICAL RESULTS AND DISCUSSION

To assess the performance of the proposed resource allo-cation algorithm, we perform extensive numerical evaluationsunder different system parameters. To this end, we consider afew users randomly located in the coverage area of the BSs,and based on the explained channel model we derived thechannel gains of all users. Finally, based on the solution ofthe proposed optimization problem, we plot the EE and sumrate versus the different parameters.

A. Parameters

The different parameters for simulation are considered asfollows: the diameters of the coverage areas of MBS andSBSs are 200 m and 10 m, respectively. The noise power

TABLE II: Path loss exponent and frequency dependency factor [52],[42].

Name Frequency region α1 α2

UHF .6-1.8 (GHz) 2 2V band 50-70 (GHz) 9.4 2.2E band 70-90 (GHz) 2 2.2

TABLE III: Simulation settings.

Parameter Value Parameter ValueK1 32 Pmax

1 45 WK2 32 |F| 5ramin 0.5 bps/Hz Pmax

i , ∀i ∈ F/{1} 5 Wε1 0.1 ε2 0.05|A| 14 |Ai|, ∀i ∈ F/{1} 2K2 32 Pmax

CP 50 WK2 32

spectral density is −174 dBm/Hz, K1 = 32, K2 = 32,ramin = 0.5 bps/Hz, Pmax

1 = 45 W, PmaxCP = 50 W, Pmax

i = 5 W∀i ∈ F/{1}. The µW and mmW bandwidths are respectively4 MHz and 0.7 GHz. ε1 = 0.1, ε2 = 0.05, |F| = 5,number of users is |A| = 14 (total number of users in theconsidered system), and |Ai| = 2,∀i ∈ F/{1} (numberof users belonging to the ith SBS where they are randomlylocated). In addition, we consider the path loss model definedin [52] and [42] that depends on the distance and frequency

of operation as PL = α1×10× log(f

f0)+α2×10× log(

d

d0),

where α1 and α2 are the path loss exponent and the frequency

dependency factor, respectively.f

f0and

d

d0are the ratios of the

frequency and distance deviation about the center carrier fre-quency and reference distance, respectively. We also considerthree different bandwidths to evaluate the performance of theproposed dual connectivity scheme, whose characteristics aresummarized in Table II. Moreover, for the sake of simplicityand complexity reduction, we consider that the values of ρaand ρ′f are ρa = 1 and ρ′f = 1 where ρa is the probabilityof success of Bernoulli random variable ξfa indicating thefeasibility of a LOS link from BS f to user a and ρ′f isthe probability of success of a Bernoulli random variable ξ′f

indicating the feasibility of a LOS link from CP to BS f . Notethat, to evaluate the performance of each used technology,we consider the performance of a system with OFDMAaccess technique and µW subcarriers as a benchmark. In thefollowing, the system sum rate and EE of different scenariosare investigated. In the numerical results, maximum transmitpower indicates the maximum available transmit power in CPand BSs (Pmax

CP +∑i∈F P

maxi ). The details of the simulation

results parameters are shown in Table III.

B. Results

Fig. 5 and Fig. 6 represent the system sum rate and EEas a function of the maximum transmit power for variousscenarios. From these figures, we can see that the use of E


0 20 40 60 80 100

Maximum Transmit Power (Watts)

0

0.5

1

1.5

2

2.5

Sum

Rat

e (b

it/s

ec)

108

Dual connectivity (E band + W), PD-NOMA and CoMP

W and OFDMADual connectivity (E band + W)

W, PD-NOMA and CoMP W and CoMP

Fig. 5: System sum rate versus maximum transmit power (PmaxCP +∑

i∈F Pmaxi ).

30 40 50 60 70 80 90 100 110


0

2

4

6

8

10

12

EE

(b

its/

Jou

le)

106


W and OFDMADual connectivity (E band + W)

W, PD-NOMA and CoMP W and CoMP

Fig. 6: System EE versus maximum transmit power (PmaxCP +∑

i∈F Pmaxi ).

band dual connectivity (E band + µW), CoMP and PD-NOMAimproves the system sum rate and EE by approximately 90%and 120%, respectively, compared to using only µW (forexample at 60 Watts the EE is increased by 130% in contrast tousing only µW). Furthermore, in these figures the effectivenessof E band dual connectivity (E band + µW) and of thePD-NOMA approach with the CoMP technology are shownseparately. As seen, the E band dual connectivity improvesthe system sum rate by approximately 50% compared to usingonly µW, and the PD-NOMA approach along with the CoMPtechnology improves the system sum rate by approximately40% compared to using the OFDMA technique.

Fig. 7 shows the system sum rate as a function of themaximum transmit power for various maximum number ofusers which can be assigned to a subcarrier in the PD-NOMAapproach (LT ). As can be seen, the system sum rate withLT = 2 is more than the system sum rate with LT = 1by approximately 13% and is less than the system sum ratewith LT = 3 by approximately 8%. Fig. 8 and Fig. 9show the system sum rate and EE as a function of the totaltransmit power when respectively V band dual connectivity(µW + V band) and E band dual connectivity (µW + Eband) are employed. As can be seen, the system sum rate ofE band dual connectivity in terms of system sum rate andEE is approximately 25% more than that of V band dualconnectivity. This is because in the V band the frequencydependency factor is higher than in the E band. Fig. 10 andFig. 11 show the percentage of total access power consumption


0 20 40 60 80 100

Sum

Rat

e (b

ps)

×108

0

0.5

1

1.5

2

2.5

Dual connectivity (E band + µW), PD-NOMA and CoMP (LT=3)

Dual connectivity (E band + µW), PD-NOMA and CoMP (LT=2)

Dual connectivity (E band + µW), PD-NOMA and CoMP

(LT=1)


i∈F Pmaxi ) for various maximum number of users which can be

assigned to a subcarrier in the PD-NOMA approach, LT .


0 20 40 60 80 100

Sum

Rat

e (b

ps)

×108

0

0.5

1

1.5

2

2.5

Dual connectivity (E band + µW), PD-NOMA (LT=3) and CoMP

Dual connectivity (V band + µW), PD-NOMA (LT=3) and CoMP


i∈F Pmaxi ) when V band dual connectivity (µW + V band) or E

band dual connectivity (µW + E band) is used.


30 40 50 60 70 80 90 100 110

EE

×106

2

4

6

8

10

12

Dual connectivity (E band + µW), PD-NOMA (LT=3) and CoMP

Dual connectivity (V band + µW), PD-NOMA (LT=3) and CoMP

Fig. 9: EE versus maximum transmit power (PmaxCP +

∑i∈F P

maxi )

when V band dual connectivity (µW + V band) or E band dualconnectivity (µW + E band) is used.


Fig. 10: Percentage of total access power versus different cell sizetypes, when V band dual connectivity (µW + V band) and CoMPtechnology are used.

Fig. 11: Percentage of total access power versus different cell sizetypes, when E band dual connectivity (µW + E band) and CoMPtechnology are used.

and sum rate for different bandwidths versus different radius ofmacro cell and small cells. In these figures, on the X axis, cellsize type A represents R1 = 50 m (macro BS radius), Ri = 5m for i = 2, . . . , F, (small BSs radius), type B representsR1 = 200 m (macro BS radius), Ri = 10 m for i = 2, . . . , F,(small BSs radius) and type C represents R1 = 450 m (macroBS radius), Ri = 20 m for i = 2, . . . , F, (small BSs radius).As seen, the bigger portion of the access power is assigned

to E-band subcarriers in E-band dual connectivity, while in V-band dual connectivity the power assignment between V-bandand µW subcarriers depends on the radius of the BSs. This isbecause in E-band dual connectivity and for the radius of theconsidered cells, the bandwidth is so much that by increasingthe power we can improve the system efficiency. On the otherhand, the attenuation is very strong for cell size type C inV-band and hence increasing the power in this band can notbe beneficial in spite of the high bandwidth. Therefore, in thiscase the power is allocated to µW subscribers to improve thesystem efficiency. In fact, in V-band dual connectivity whenthe cell size increases from type B to type C, the total powerbecomes larger. In this condition, V-band subcarriers are not

Fig. 12: Access sum rate versus different radius of macro cell andsmall cells, when dual connectivity and CoMP technology are used.

appropriate to transmit data due to the high attenuation, andBSs have to transmit more power on µW subcarriers to meetthe minimum rate requirement of the users.

Fig. 12 shows the access sum rate in both V-band and E-band. As observed from this figure the access sum rate inboth V-band and E-band decreases when the radius of the BSsincreases, and the access sum rate in V-band dual connectivitydecreases more than that in E-band dual connectivity. This isbecause the attenuation in V-band is more than in E-band.Figs. 13 and 14 depict the sum rate and the energy efficiencyversus the number of users, respectively, for a total availablepower of 60 Watts. From this figure, we can see that thesum rate and EE increase as the number of users increasesfor all the solution methods. This is due to the fact that byincreasing the number of users, multiuser diversity increases.In Fig. 15, the EE of the system with dual connectivity(µW + E band), PD-NOMA, and CoMP for three methodsnamely, I) Proposed solution, II) Uniform power allocation,and III) Random subcarrier assignment are compared. Basedon the uniform power allocation method, at first power isassigned uniformly and then subcarriers are allocated basedon the proposed solution for subcarrier assignment problemin Section III-B. Based on the random subcarrier method, atfirst subcarriers are assigned randomly and then powers areallocated based on the proposed solution algorithm in SectionIII-A. From this figure, we can see that the proposed solutionmethod outperforms the other solutions. Fig. 16 illustrates thenumber of required iterations to achieve convergence. In thisfigure, the maximum available power is 60 Watts, the numberof users is 20, and the number of small cells is 4.

C. Discussion

From the numerical results, we can conclude that where dualconnectivity, PD-NOMA and CoMP technologies are jointlyused, the system performance can improve significantly. Withthe PD-NOMA technology, a subcarrier can be assigned tomore than one user simultaneously, and this enhances thesystem performance. Moreover, as the maximum number ofusers which can be assigned to a subcarrier increases, theoverall system performance improves. CoMP is the other


14 16 18 20 22 24 26 28 30 32

Number of Users

4

5

6

7

8

9

10

11

12E

E (

bit

s/Jo

ule

)10

6

W and OFDMA W and CoMP

W, PD-NOMA and CoMPDual connectivity (E band + W)


Fig. 13: EE versus the number of users.

14 16 18 20 22 24 26 28 30 32

Number of Users

0.5

1

1.5

2

2.5

3

Sum

Rat

e (b

it/s

ec)

108

W and OFDMA

W and CoMP

W, PD-NOMA and CoMP

Dual connectivity (E band + W)


Fig. 14: Sum rate versus the number of users.

30 40 50 60 70 80 90 100 110 120

Power (watts)

2

4

6

8

10

12

EE

(bit

s/Jo

ule

)

106

Proposed Solution, Dual connectivity (E band + W), PD-NOMA and CoMPRandom Subcarrier, Assignment Dual connectivity (E band + W), PD-NOMA and CoMP

Equal Power Allocation, Dual connectivity (E band + W), PD-NOMA and CoMP

Fig. 15: EE versus the maximum available power for differentsolution methods.

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

Number of Iterations

4

5

6

7

8

9

10

11

12

13

Sum

Rate

(bit

/sec)

107

Fig. 16: Convergence of the proposed solution.

technology studied in this paper. As observed, both the EEand the system sum rate can be improved by applying theCoMP technology. With the dual connectivity technology eachreceiver can receive its data from transmitters with µW andmmW links simultaneously. From the results, we can concludethat exploiting E-band dual connectivity is more efficient thanV-band dual connectivity. We would like to point out thatthe concept of multi-connectivity (including mixed mmW/LTEsolutions as in Non-Standalone 5G scenarios [53]) has beenrecently getting a lot of traction in 3GPP and most vendorsand operators are working under the assumption that this willbe how mmWave in 5G will be first implemented (i.e., beforeStandalone mmW systems).

VII. CONCLUSION

In this paper, we have considered three promising tech-nologies for 5G mobile networks, namely, dual connectivity,PD-NOMA and CoMP. We proposed a new downlink re-source allocation method to enhance the system EE. In thisregard, we formulated a novel resource allocation methodthat selects the powers over mmW and µW subcarriers inboth access and fronthaul to maximize the EE. The proposedoptimization problem is non-convex with high computationalcomplexity, and is solved approximately by using the SCA-based iterative approach with low complexity. Via numericalresults, we investigated the performance of each technologyand compared them with each other. From the numericalresults, we showed that via dual connectivity the systemsum rate is improved approximately by 50% in contrast tousing only µW subcarriers. In addition, by applying bothPD-NOMA and CoMP technologies on the µW subcarriers,the sum rate of a dual connectivity-based system increasesby approximately 40%. As a future work, we will study theperformance of the proposed system model with imperfectchannel state information. Another direction of future researchwould be to study the optimality gap between the proposedsolution and the optimal one 3.

REFERENCES

[1] M. Moltafet, N. Mokari, R. Joda, M. R Sabagh, and M. Zorzi, “Jointaccess and fronthaul resource allocation in dual connectivity and CoMPbased networks,” IEEE ICC 2018.

[2] ITU-R, “ITU-R M.[IMT-2020.TECH PERF REQ] - Minimum require-ments related to technical performance for IMT2020 radio interface(s),”Report ITU-R M.2410-0, November 2017

[3] F. Boccardi, R. Heath, A. Lozano, T. Marzetta, and P. Popovski, “Fivedisruptive technology directions for 5G,” IEEE Commun. Magazine, vol.52, no. 2, pp. 74-80, February 2014.

[4] A. Ghosh, R. Ratasuk, P. Moorut, T. S. Rappaport, and S. Sun,“Millimeter-Wave enhanced local area systems: A high data-rate approachfor future wireless networks,” IEEE J. Sel. Areas Commun, vol. 32, no.6, pp. 1152 -1163, June 2014.

[5] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellularwireless networks: Potentials and challenges,” in Proc. of the IEEE, vol.102, no. 3, pp. 366-385, March 2014.

3In order to achieve the optimal solution of the proposed optimizationproblem we could use the monotonic optimization approach [54], [55]. Whilethis would lead to an exact formulation, we decided not to include it in thepaper due to lack of space and to the fact that this model cannot be solvedexcept in trivial scenarios, and has therefore no practical utility.


[6] H. Shokri-Ghadikolaei, C. Fischione, G. Fodor, P. Popovski, and M. Zorzi,“Millimeter wave cellular networks: A MAC layer perspective,” IEEETrans. Commun, vol. 63, no. 10, pp. 3437-3458, October 2015.

[7] D. Wu, J. Wang, Y. Cai, and M. Guizani, “Millimeter-wave multimediacommunications: challenges, methodology, and applications,” IEEECommun. Mag, vol. 53, no. 1, pp. 232-238, January 2015.

[8] G. Fettweis and E. Zimmermann, “ICT energy consumption trends andchallenges,” in Proc. International Symposium on Wireless PersonalMultimedia Communications (WPMC 2008), 2008, pp. 1-3.

[9] Staff, Qualcomm. “Rising to Meet the 1000x Mobile Data Challenge,”QUALCOMM Incorporated, 2012).

[10] DOCOMO, “5G radio access: requirements, concept and technologies,”(NTT DOCOMO, 2014)

[11] G. Caire, S. Shamai, “On the achievable throughput of a multi antennaGaussian broadcast channel,” IEEE Trans. Inf. Theory, vol. 49, no. 7, pp.1692-1706, July 2003.

[12] Tse. D, Viswanath. P, “Fundamentals of wireless communications,”Cambridge University Press, Cambridge, U.K, 2005.

[13] D. Ng and R. Schober, “Resource allocation and scheduling in multicellOFDMA systems with decode-and-forward relaying,” IEEE Trans.Wireless Commun, vol. 10, no. 7, pp. 2246-2258, July 2011.

[14] H. Zhang, L. Venturino, N. Prasad, P. Li, S. Rangarajan, and X. Wang,“Weighted sum-rate maximization in multi-cell networks via coordinatedscheduling and discrete power control,” IEEE J. Sel. Areas Commun, vol.29, no. 6, pp. 1214-1224, May 2011.

[15] G. Dartmann, W. Afzal, X. Gong, and G. Ascheid, “Joint optimizationof beamforming, user scheduling, and multiple base station assignmentin a multicell network,” in Proc. IEEE Wireless Communications andNetworking Conference (WCNC), 2011, pp. 209-214.

[16] L. B. Le, D. Niyato, E. Hossain, D. I. Kim, and D. T. Hoang, “QoS-aware and energy-efficient resource management in OFDMA femtocells,”IEEE Trans. Wireless Commun, vol. 12, no. 1, pp. 180-194, January 2013.

[17] W. Yu, T. Kwon, and C. Shin, “Multicell coordination via joint schedul-ing, beamforming, and power spectrum adaptation,” IEEE Trans. WirelessCommun, vol. 12, no. 7, pp. 3300-3313, July 2013.

[18] A. Abdelnasser and E. Hossain, “Subchannel and power allocationschemes for clustered femtocells in two-tier OFDMA hetnets,” in Proc.IEEE International Conference on Communications Workshops (ICC),Budapest, 2013, pp. 1129-1133.

[19] R. G. Stephen and R. Zhang, “Joint millimeter-wave fronthaul andOFDMA resource allocation in ultra-dense CRAN,” IEEE Trans. Com-mun, vol. 65, no. 3, pp. 1411-1423, March 2017.

[20] G. A. S. Siding, F. Gao, and A. Nallanathan, “A joint resource allocationscheme for multi-relay aided uplink multi-user OFDMA system,” inProc. International Conference on Wireless Communications and SignalProcessing (WCSP), Suzhou, 2010, pp. 1-6.

[21] Z. Hasan, E. Hossain, and V. K. Bhargava, “Resource allocation for mul-tiuser OFDMA-based amplify-and-forward relay networks with selectiverelaying,” in Proc. IEEE International Conference on Communications(ICC), Kyoto, 2011, pp. 1-6.

[22] M. Giordani, M. Mezzavilla, S. Rangan and M. Zorzi, “Multi-connectivity in 5G mmWave cellular networks,” in Proc. 2016 Mediter-ranean Ad Hoc Networking Workshop (Med-Hoc-Net), Vilanova, 2016,pp. 1-7.

[23] M. Polese, M. Giordani, M. Mezzavilla, S. Rangan, and M. Zorzi,“Improved handover through dual connectivity in 5G mmWave mobilenetworks,” IEEE J. Sel. Areas Commun, vol. 35, no. 9, pp. 2069-2084,September 2017.

[24] O. Semiari, W. Saad, and M. Bennis, “Joint millimeter wave and µWaveresources allocation in cellular networks with dual-mode base stations,”IEEE Trans. Wireless Commun, vol. 16, no. 7, pp. 4802-4816, July 2017.

[25] D. W. K. Ng, M. Breiling, C. Rohde, F. Burkhardt, and R. Schober,“Energy-efficient 5G outdoor-to-indoor communication: SUDAS overlicensed and unlicensed spectrum,” IEEE Trans. Wireless Commun, vol.15, no. 5, pp. 3170-3186, May 2016.

[26] J. Qiao, X. Shen, J. Mark, Q. Shen, Y. He, and L. Lei, “Enabling device-to-device communications in millimeter-wave 5G cellular networks,”IEEE Commun. Mag, vol. 53, no. 1, pp. 209-215, January 2015.

[27] H. Singh, J. Oh, C. Kweon, X. Qin, H. Shao, and C. Ngo, “A 60GHz wireless network for enabling uncompressed video communication,”IEEE Commun. Magazine, vol. 46, no. 12, pp. 71-78, December 2008.

[28] J. Park, S. L. Kim and J. Zander, “Tractable resource managementwith uplink decoupled millimeter-wave overlay in ultra-dense cellularnetworks,” IEEE Trans. Wireless Commun, vol. 15, no. 6, pp. 4362-4379, June 2016.

[29] T. Akbudak, and A. Czylwik, “CoMP in heterogeneous networks: Ajoint cooperative resource allocation approach,” in Proc. InternationalITG Workshop on Smart Antennas (WSA), Dresden, 2012, pp. 12-19.

[30] T. A. Le, T. Gherkar, M. R. Nakhai and K. Navaie, “Symbol timing offsetmitigation in OFDMA-based CoMP utilizing position aware transmis-sion,” in Proc. Vehicular Technology Conference (VTC Spring),Glasgow,2015, pp. 1-5.

[31] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5Gnonorthogonal multiple access downlink transmissions,” IEEE Trans. Veh.Tech, vol. 65, no. 8, pp. 6010-6014, August 2016.

[32] P. Parida and S. S. Das, “Power allocation in OFDM based NOMA sys-tems: A DC programming approach,” in Proc. IEEE Globecom Workshops(GC Wkshps), 2014, pp. 1026-1031.

[33] Y. Sun, D. W. K. Ng, Z. Ding, R. Schober, “Optimal joint power andsubcarrier allocation for full-duplex multicarrier non-Orthogonal multipleaccess systems,” IEEE Trans. Commun, vol. 65, no. 3, pp. 1077-1091,March 2017.

[34] J. Cui, Z. Ding and P. Fan, “A Novel power allocation scheme underoutage constraints in NOMA systems,” IEEE Signal Process. Lett, vol.23, no. 9, pp. 1226-1230, September 2016.

[35] F. Fang, H. Zhang, J. Cheng and V. C. M. Leung, “Energy-efficient re-source allocation for downlink non-orthogonal multiple access network,”IEEE Trans. Commun, vol. 64, no. 9, pp. 3722-3732, September 2016.

[36] A. Mokdad, P. Azmi and N. Mokari, “Radio resource allocation forheterogeneous traffic in GFDM-NOMA heterogeneous cellular networks,”IET Communications, vol. 10, no. 12, pp. 1444-1455, September 2016.

[37] M. Moltafet, N. Mokari, M. R. Javan, and P. Azmi, “Comparison studybetween PD-NOMA and SCMA,” IEEE Trans. Veh. Technol, vol. 67, no.2, pp. 1830-1834, February 2018.

[38] M. Moltafet, P. Azmi, N. Mokari, M. R. Javan and A. Mokdad, “Optimaland fair energy efficient resource allocation for energy harvesting enabled-PD-NOMA based HetNets,” IEEE Trans. Wireless Commun, vol.17, no.3, pp. 2054-2067, March 2018.

[39] M. Pickavet, W. Vereecken, S. Demeyer, P. Audenaert, B. Vermeulen,C. Develder, D. Colle, B. Dhoedt, and P. Demeester, “Worldwide energyneeds for ICT: The rise of power-aware networking,” in Proc. 2ndInternational Symposium on Advanced Networks and TelecommunicationSystems, Mumbai, 2008, pp. 1-3.

[40] D. Lee, B. Clerckx, E. Hardouin, D. Mazzarese, S. Nagata, K. Sayana,“Coordinated multipoint transmission and reception in LTE-advanced:deployment scenarios and operational challenges,” IEEE CommunicationsMagazine, vol. 50, no. 2, pp. 148-155, February 2012.

[41] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of non-orthogonal multiple access in 5G systems with randomly deployed users,”IEEE Signal Process. Lett, vol. 21, no. 12, pp. 1501-1505, December2014.

[42] S. Niknam, A. A. Nasir, H. Mehrpouyan and B. Natarajan, “A multi-band OFDMA heterogeneous network for millimeter wave 5G wirelessapplications,” IEEE Access, vol. 4, pp. 5640-5648, December 2016.

[43] L. Venturino, N. Prasad and X. Wang, “Coordinated scheduling andpower allocation in downlink multicell OFDMA networks,” IEEE Tran.Veh. Technol, vol. 58, no. 6, pp. 2835-2848, July 2009.

[44] W. Dinkelbach,“On nonlinear fractional programming,” Manag. Science,vol. 13, pp. 492-498, March 1967.

[45] D. Trong Ngo, S. Khakurel, and T. Le-Ngoc, “Joint subchannel assign-ment and power allocation for OFDMA femtocell networks,” IEEE Trans.Wireless Commun, vol. 13, no. 1, pp. 342-355, January 2014.

[46] ‘CVX Research - CVX: Matlab software for disciplined convex pro-gramming, version 2.1’, http://cvxr.com/cvx, accessed 4 Oct. 2015.

[47] W. Dang, M. Tao, H. Mu, and J. Huang, “Subcarrier-pair based resourceallocation for cooperative multi-relay OFDM systems,” IEEE Trans.Wireless Commun, vol. 9, no. 5, pp. 1640-1649, May 2010.

[48] Y. Cheong, R. Cheng, K. Lataief, and R. Murch, “Multiuser OFDMwith adaptive subcarrier, bit, and power allocation,” IEEE J. Sel. AreasCommun, vol. 17, no. 10, pp. 1747-1758, October 1999.

[49] R. C. de Lamare and R. Sampaio-Neto, “Adaptive reduced-rank equal-ization algorithms based on alternating optimization design techniques formimo systems,” IEEE Trans. Veh. Tech, vol. 60, no. 6, pp. 2482-2494,July 2011.

[50] N. Mokari, F. Alavi, S. Parsaeefard, and T. Le-Ngoc, “Limited-feedbackresource allocation in heterogeneous cellular networks,” IEEE Trans. Veh.Tech, vol. 65, no. 4, pp. 2509-2521, April 2016.

[51] H. H. Kha, H. D. Tuan, H. H. Nguyen, “Fast global optimal powerallocation in wireless networks by local D.C. programming”, IEEE Trans.Wireless Commun, vol. 11, no. 2, pp. 510-515, February 2012.

[52] T. S. Rappaport, R. W. Heath, R. C. Daniels, and N. Murdock,“Millimeter-wave wireless communications,” Prentice Hall, 2014.


[53] “TS 38.104-BS-Release 15,” 3rd Generation Partnership Project (3GPP),2017.

[54] A. Zappone, E. Bjornson, L. Sanguinetti and E. Jorswieck, “A frame-work for globally optimal energy-efficient resource allocation in wirelessnetworks,” in Proc. IEEE International Conference on Acoustics, Speechand Signal Processing (ICASSP), Shanghai, 2016, pp. 3616-3620.

[55] A. Zappone, E. Bjornson, L. Sanguinetti and E. Jorswieck, “Globallyoptimal energy-efficient power control and receiver design in wirelessnetworks,” IEEE Trans. Signal Process, vol. 65, no. 11, pp. 2844-2859,Jun 2017.

Mohammad Moltafet received the M.Sc. degreein electrical and computer engineering from Tar-biat Modares University, Tehran, Iran, in 2016. Heis currently pursuing the Ph.D. degree with theCentre for Wireless Communications, University ofOulu, Finland. His current research interests includewireless communication networks with emphasis onradio resource allocation and optimization, networkformation theory, machine learning, and compressivesensing.

Roghayeh Joda (M14) received the B.Sc. degreein electrical engineering from Sharif University ofTechnology, Tehran, Iran, in 1998, and the M.Sc.and Ph.D. degrees in electrical engineering fromthe University of Tehran, Tehran, Iran, in 2001 and2012, respectively. During her Ph.D. studies, she hasbeen a Visiting Scholar at the Polytechnic Instituteof NYU, Brooklyn, NY, USA. She has worked as aResearch Associate at University of Tehran, Tehran,Iran, from November 2012 to July 2013 and thenas a Postdoctoral Fellow at University of Padova,

Padova, Italy, from September 2013 to July 2014. She is currently anassistant professor at Iran Telecommunication Research Center, Tehran, Iran.Her current research interests include network information theory, source,channel and network coding, interference management, resource allocationand optimization, cognitive radio networks, machine learning, internet ofthings and 5G networks.

Nader Mokari received the Ph.D. degree in elec-trical engineering from Tarbiat Modares University,Tehran, Iran, in 2014. He joined the Departmentof Electrical and Computer Engineering, TarbiatModares University, as an Assistant Professor, in2015. He has been involved in a number of largescale network design and consulting projects inthe telecom industry. His research interests includedesign, analysis, and optimization of communicationnetworks.

Mohammad Reza Sabagh received the B.Sc.degree in electrical engineering from Amirkabir Uni-versity of Technology, Tehran, Iran, in 1993, theM.Sc. degree in communication engineering fromSharif University of Technology, Tehran, Iran, in1998, and the Ph.D. degree from University ofSurrey, Guildford, UK, in 2017. He is currentlyworking at Iran Telecommunication Research Center(ITRC), Tehran, Iran on 5G projects. His currentresearch interests include radio resource allocationand optimization, scheduling, energy efficiency and

5G network.

Michele Zorzi [F07] received his Laurea andPhD degrees in electrical engineering from theUniversity of Padova in 1990 and 1994, respec-tively. During academic year 1992-1993 he wason leave at the University of California at SanDiego (UCSD). In 1993 he joined the faculty ofthe Dipartimento di Elettronica e Informazione,Politecnico di Milano, Italy. After spending threeyears with the Center for Wireless Communi-cations at UCSD, in 1998 he joined the Schoolof Engineering of the University of Ferrara,

Italy, where he became a professor in 2000. Since November 2003 hehas been on the faculty of the Information Engineering Departmentat the University of Padova. His present research interests includeperformance evaluation in mobile communications systems, WSN andInternet of Things, cognitive communications and networking, vehicularnetworks, 5G mmWave cellular systems, and underwater communicationsand networks. He is the recipient of several awards from the IEEECommunications Society, including the Best Tutorial Paper Award (2008),the Education Award (2016) and the Stephen O. Rice Best Paper Awards(2018). He is currently the Editor in Chief of the IEEE Transactionson Cognitive Communications and Networking, and was the Editor-In-Chief of IEEE Wireless Communications from 2003 to 2005 andof the IEEE Transactions on Communications from 2008 to 2011. Heserved as a Member-at-Large of the Board of Governors of the IEEECommunications Society from 2009 to 2011, and as its Director ofEducation in 2014-15.

Documents

IEEE TRANSACTIONS ON COMMUNICATIONS 1 Joint Access and