Enabling Device-to-Device (D2D) Communication for the Next

Research ArticleEnabling Device-to-Device (D2D) Communication for the NextGeneration WLAN

Ying Wang , Jianjun Lei , and Fengjun Shang

School of Computer Science and Technology, Chongqing University of Posts and Telecommunications, Chongqing 400065, China

Correspondence should be addressed to Jianjun Lei; [email protected]

Received 19 May 2021; Revised 9 September 2021; Accepted 4 October 2021; Published 22 October 2021

Academic Editor: Fuhui Zhou

Copyright © 2021 Ying Wang et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Device-to-device (D2D) communication technology is widely acknowledged as an emerging candidate to alleviate the wirelesstraffic explosion problem for the next generation wireless local area network (WLAN) IEEE 802.11ax. In this paper, weintegrate D2D communication into IEEE 802.11ax for maximizing the spectrum efficiency. Due to the spectrum scarcity, thenumber of available resource units (RUs) is typically less than D2D pairs and stations (STAs), which makes the managementof spectrum resources more complex. To tackle this issue, we design an efficient resource management algorithm for uplinkrandom access and resource allocation in D2D-enabled 802.11ax, which provides more channel access and resource reuseopportunities for STAs and D2D pairs. Specifically, we propose an enhanced back-off mechanism and derive the optimalcontention window (CW) by a theoretical model, which improves the uplink random access efficiency. To tackle the complexinterference problem in RU scheduling phase, we develop an efficient and low-complexity resource allocation algorithm basedon the maximal independent set (MIS), which can schedule multiple D2D pairs to share the same RU with the specific STA.Simulation results demonstrate that the proposed algorithm significantly improves the network performance in terms ofsystem throughput, collision rate, complete time, and channel utilization.

1. Introduction

Recently, wireless local area networks (WLANs) have beenwidely deployed in residential, commercial, and public areasto provide the near-ubiquitous wireless coverage. As therapid development of wireless communication technology,the wireless devices are becoming more intensive; mean-while, the reliability of data transmission is more demanding.The legacy IEEE 802.11 faces some fundamental challengesin network performance improvement, e.g., higher spectrumefficiency, lower latency, and less collision, especially in suchhighly dense networks, owing to the fact that it adopts thebind binary exponential back-off (BEB) scheme and onlysupports that the single user monopolizes the channelresource. To address these issues, the latest WLAN standardIEEE 802.11ax is designed [1]. Specifically, the orthogonalfrequency division multiple access (OFDMA) technology,as a key innovation in 802.11ax, divides the frequency-domain channel into several orthogonal subcarrier groupsreferred to as resource units (RUs) [2, 3]. Consequently,

802.11ax can effectively solve the negative impact caused bythe exclusive channel mode in carrier sense multiple accesswith collision avoidance (CSMA/CA) protocol by enablingthe uplink multiuser transmissions [4].

Different from the legacy distributed coordination func-tion (DCF) or enhanced distributed channel access (EDCA)mechanism, the uplink OFDMA-based random access(UORA) mechanism is employed in 802.11ax. In UORA,multiple stations (STAs) can simultaneously access theseorthogonal RUs to transmit data, which significantlyimproves the spectrum efficiency by enabling resource reuse.However, the OFDMA contention window (OCW) inUORA cannot be adaptively adjusted with the dynamic net-work environment, and hence, collisions are still inevitableespecially in the highly dense WLAN [5]. Many works haveproposed corresponding modifications to the UORA byoptimizing MAC access parameters. Although these worksenhance the MAC access efficiency and improve the networkperformance well, the optimal spectrum efficiency is limitedby available RUs especially in highly dense network.

HindawiWireless Communications and Mobile ComputingVolume 2021, Article ID 1949352, 17 pageshttps://doi.org/10.1155/2021/1949352

https://orcid.org/0000-0003-4402-0102

https://orcid.org/0000-0002-1965-3170

https://orcid.org/0000-0001-6730-1199

https://creativecommons.org/licenses/by/4.0/

https://creativecommons.org/licenses/by/4.0/

https://doi.org/10.1155/2021/1949352

More recently, device-to-device (D2D) communicationhas become an emerging technology in cellular networkand WLAN, which can effectively release theinfrastructure-based network resource and greatly improvethe spectrum efficiency by allowing direct communicationsbetween proximal devices [6]. Thus, two STAs in close prox-imity need to compete for channel access only once becausethey transmit the data directly. Meanwhile, the serious datacollision caused by channel contention can be markedlyrelieved by offloading STAs into D2D communicationmode. Besides, the D2D communication with overlay modefurther improves spectrum utilization by scheduling D2Dpairs and the specific STA to share the same RU, simulta-neously. However, the unreasonable resource allocationstrategy could result in the severe cochannel interferenceamong D2D pairs and STA. Most previous studies on D2Doffloading in WLAN focus on coordinating the switchingof the nodes in infrastructure-based communication modeto D2D communication mode, although the resource alloca-tion has attracted attentions in the D2D-enabled WLAN forimproving the spectrum efficiency. However, the resourceallocation algorithm for multiple D2D pairs and RU sce-nario is rarely considered in most existing works due to highcomputational complexity or overheads.

Motivated by these advantages and challenges, this paperintegrates D2D technology into 802.11ax and proposes anefficient channel access and resource allocation algorithmfor improving the spectrum efficiency. Correspondingly,two significant challenges must be tackled: (1) it is necessaryto determine whether or not two STAs can switch from tra-ditional infrastructure-based mode to D2D communicationmode; (2) it is crucial to design proper interference manage-ment and resource allocation algorithms to allocate RUs forD2D pairs and STAs for maximizing the spectrum efficiency.Many works have been dedicated to coordinating adaptivelythe communication modes of nodes in D2D-enabled WLANand achieve the significant performance improvement.Hence, this paper only focuses on the interference manage-ment and resource allocation algorithm in D2D-enabled802.11ax network, where multiple D2D pairs meeting theinterference conditions can share one RU with the specificSTA so as to maximize the spectrum efficiency. In particular,the proposed algorithm, referred to as MISD, combines anenhanced back-off mechanism for STAs in uplink randomaccess phase and an RU allocation algorithm based on max-imum independent sets (MISs) in resource allocation phase.The main contributions of this paper are summarized asfollows:

(i) We integrate the D2D communication technologyinto 802.11ax and develop a spectrum efficiency opti-mization solution for random access and resourceallocation in D2D-enabled 802.11ax network

(ii) We propose an enhanced back-off mechanism infrequency and time domains, where the optimalCW is derived by a theoretical model for maximiz-ing the uplink random access efficiency. Moreover,the proposed algorithm reduces the control over-

heads and the data collisions significantly by cancel-ing the ACK frame and offloading nodes into theD2D communication mode in the uplink randomaccess phase, respectively

(iii) We propose an efficient resource allocation algo-rithm based on the MISs, where AP can schedulemultiple D2D pairs to share the same RU with thespecific STA, simultaneously, which significantlyimproves the spectrum efficiency. Moreover, theproposed algorithm has low complexity, i.e., APonly needs to estimate the interference betweenthe specific STA and the D2D pairs in the largestMIS

(iv) A numerical analysis model is established to verifytheoretically the performance improvement of theMISD algorithm. Furthermore, the extensive simu-lation results clearly demonstrate the benefits ofthe proposed MISD algorithm from the enhancedback-off mechanism and the efficient resource allo-cation algorithm

The remainder of this paper is organized as follows. Wesummarize the related work in Section 2. Section 3 presentsthe system model and algorithm design. Then, Section 4describes the numerical analysis. The performance simula-tions of the proposed algorithm are carried out in Section5. Finally, Section 6 concludes this paper and indicates thefuture research direction.

2. Related Work

This section describes the related work of the previousstudies on OFDMA-based and D2D-based offloading MACprotocols in highly dense WLAN.

IEEE 802.11ax introduces the OFDMA technology forsupporting concurrent data transmissions [6, 7]. The workin [8] presents that enabling OFDMA in WLAN canenhance the channel utilization in dense scenarios. In [9],Qu et al. propose the whole channel physical sensing andfast back-off mechanisms for IEEE 802.11ax. Although theproblems of synchronization and overhead are effectivelysolved, the MAC efficiency is substantially limited due tothe blind CW adjustment strategy. To tackle this issue, Uwaiet al. propose an adaptive back-offmechanism by optimizingMAC parameters in [10], which improves the throughput inuplink random access period. However, the spectrum effi-ciency is not optimal due to the collision in high load net-work scenario. In [11], Shahin et al. propose a cognitiveback-off mechanism that each station adaptively adjustsCW, which solves the critical medium collision problem indense WLAN, but the channel access delay is not consid-ered. In [12], Wang et al. propose a probability complemen-tary transmission scheme (PCTS) to mitigate the delaycaused by retransmission. In PCTS, the stations with unsuc-cessful transmission trials can transmit directly with thecomplementary probability. However, it is unsuitable forhighly dense networks, which increases the severe data colli-sion. In [13], Lin et al. enable multiple RTS intervals in

2 Wireless Communications and Mobile Computing

random access phase and propose a joint grouping andsubchannel allocation algorithm to improve the spectrumefficiency. However, it lacks in further analyzing the numberof RTS intervals, which has direct impacts on the networkperformance.

Many studies have shown that the scheduling-basedMAC protocols in OFDMA WLAN can be exploited toimprove the spectrum efficiency. In [14], Bankov et al. inves-tigate the scheduling problem and transfer the well-knownschedulers in LTE for different application scenarios in802.11ax, i.e., high throughput, fairness, and low latency,respectively. In [15], they propose an uplink transmissionscheduler for the OFDMA network to minimize the averageupload time, which considers various RU configuration. In[16], Dovelos and Bellalta propose another channel resourcescheduling scheme to maximize the long-term average rateunder the constraints of the average rate and power. How-ever, the nonsaturated traffic condition scenarios are not dis-cussed in [14–16]. Consequently, the hybrid accessalgorithms including random access and scheduling accessare proposed. In [17], a theoretical model is established toadjust the number of RUs for random access and schedulingaccess in highly dynamic 802.11ax network. However, thecorresponding optimization mechanisms are not discussed.In [18], Yang et al. propose a hybrid channel access algo-rithm, including a greedy scheduling access mechanismbased on capacity constraints and a random-access mecha-nism based on channel quality perception, where the optimi-zation problem for the joint carrying capacity of thenetworks is a challenge. In [19], Karaca et al. develop twoOFDMA-based resource scheduling algorithms: (i) using afixed scheduling duration and (ii) dynamically adjustingthe scheduling duration by comprehensively consideringfilling overhead, fairness, and user power consumption.Although the previous OFDMA-based literature canimprove the spectrum efficiency by enabling multiusercurrent transmissions, the optimal spectrum efficiency islimited by the available RUs. Besides, they cannot funda-mentally tackle the bottleneck that all traffic still needs tobe forwarded centrally by the AP.

The emerging D2D technology has been widely used indense network scenarios, which can alleviate data collisionsand enhance the spectrum efficiency by offloading nodes toD2D communication mode and enabling spectrum reusetechnology. In [20], Mach et al. provide a survey and discussthe advantages, critical challenges, and research directionsof in-band D2D in OFDMA cellular network. In [21],Shah et al. review the interference mitigation and radioresource management of in-band D2D communicationsand discuss the problems of resource management, modeselection, energy efficiency, and coexistence of D2D com-munications. Correspondingly, the researches on D2Dcommunication technology are mainly divided into D2Dlink management [22–24], resource allocation, and inter-ference management [25–27].

In [22], Cai et al. propose a D2D communication frame-work based on the Software Defined Network (SDN) andNetwork Function Visualization (NFV) in virtual wirelessnetworks, which employs a stochastic approximation algo-

rithm to optimize the network-wide welfare by switchingthe mobile nodes between infrastructure-based and D2D-based communication modes. In [23], Cheng et al. proposeanother switching scheme for nodes based on SDN technol-ogy in D2D-based WLAN. However, the resourceallocation-related issues are not considered both in [22,23]. In [24], a network-assisted D2D technology by utilizingWiFi Direct as the link layer technology for proximal D2Dconnections has been proposed, which mitigates the dispro-portion between user demand and available radio resources.However, the resource management for D2D connection isnot detailed. In [25], Zhao et al. transform the channel allo-cation problem into a graph coloring problem by building aninterference graph and finding an approximate optimal solu-tion in D2D underlying cellular networks. However, multi-ple D2D pairs are not allowed to share the same channelresource. In [26], Wen et al. propose a joint power controland resource allocation algorithm that greatly reduces theinterference and improves the network performance, butthe algorithm is too complex. In [27], Seyedebrahimi et al.propose a novel management framework for D2D commu-nications in dense WiFi network, which proactively estab-lishes and manages D2D connections by considering theavailable radio resource and the effect of the subsequentinterference. However, the overlay D2D communicationmode is not considered.

The previous studies on OFDMA-based MAC protocolsimprove network performance by enabling concurrenttransmission. However, the optimal spectrum efficiency islimited by the available RUs. Integrating D2D offloadinginto WiFi network can further improve the spectrum effi-ciency but also bring the complex interference managementproblem [28]. The resource allocation and interference man-agement problems in D2D offloading WiFi network are notaddressed well in the aforementioned research studies.Specifically, these problems will be more complicated in802.11ax due to more RUs. Hence, this paper investigatesthe resource allocation and interference management prob-lems in D2D-enabled 802.11ax network to enhance the spec-trum efficiency.

3. The Proposed Algorithm

In this section, we introduce the system model of the D2D-enabled 802.11ax network. Then, we propose an uplinkchannel access and resource allocation algorithm referredto as MISD. In detail, the proposed MISD algorithm isaimed at enhancing the spectrum efficiency and reducingcollision by combining an enhanced back-off mechanismwith optimal CW and an efficient resource allocationmechanism.

3.1. Architecture and Model. We consider an uplink trans-mission scenario in the single Basic Service Set (BSS)governed by IEEE 802.11ax, which supports two communi-cation modes: AP forwarding and D2D offloading, wherethe D2D pair can communicate directly. As shown inFigure 1, AP is deployed in the geometrical center of theBSS, and all nodes (i.e., STAs and D2D pairs) are randomly

3Wireless Communications and Mobile Computing

distributed in the coverage region of AP. Assume n STAsand m D2D pairs in the D2D-enable 802.11ax network. LetS = fS1, S2,⋯, Sng and D = fD1,D2,⋯,Dmg denote the setof STAs and D2D pairs, respectively. AP and all nodes sup-port MU transmissions which is a basic characteristic of802.11ax. The whole channel is divided into r subchannels(referred to as RUs), denoted as R = fR1, R2,⋯, Rrg. To bemore specific, the spectrum resource (i.e., RU) can be sharedby the communication links under the two communicationmodes. Moreover, multiple D2D pairs meeting the interfer-ence conditions can share the same RU. Since the downlinktransmission can be scheduled directly by AP and operatedeasily [29], this paper only focuses on the uplink data trans-mission. We also make an assumption that the network is inthe saturated condition, i.e., all nodes in the network alwayshave packets to send. Specifically, only STAs need to con-tend for channel by sending buffer status report (BSR) framein uplink random access phase. Besides, the RU allocationdecisions for STAs and D2D pairs are made by AP accordingto the interference relationship among nodes.

3.2. MISD Algorithm. This subsection details our proposedMISD algorithm. Firstly, we present an overview of theMISD algorithm. Then, the random access, resource alloca-tion, and data transmission phases are introduced in detail,respectively.

3.2.1. MISD Overview. This paper concentrates on maximiz-ing the spectrum efficiency by optimizing jointly randomchannel access and resource allocation in D2D-enabled802.11ax, where the complex network scenario, i.e., multiple

D2D pairs and RUs, is considered. To this end, we proposean efficient resource management algorithm MISD, whichdecouples the spectrum efficiency optimization problem intotwo subproblem in uplink random access and resourceallocation phases. Correspondingly, the proposed MISDalgorithm separates the channel time into a series of super-frames, where each superframe includes uplink randomaccess, resource allocation phase, and uplink data transmis-sion phases, as shown in Figure 2. Furthermore, theenhanced back-off mechanism and efficient resource allo-cation algorithm in resource allocation phase are developedfor improving the spectrum efficiency in random accessphase and data transmission phase, respectively. In partic-ular, the STA contends for the RU based on the proposedenhanced back-off mechanism, independently; in resourceallocation phase, AP allocates RUs for STAs and D2Dpairs by the proposed efficient resource allocation algo-rithm; hence, the specific STA and D2D pairs can sharethe same RU for concurrent data transmission in datatransmission phase.

3.2.2. Random Access Process. In this subsection, we firstestablish a theoretical analysis model to derive the optimalcontention window CW∗ in order to maximize the systemthroughput in uplink random access phase. Then, we pro-pose an enhanced back-off mechanism, where the STAscan contend for RUs with the CW∗ in both time and fre-quency domains.

CW optimization. Let τ be the probability that a STAattempts to transmit data packet in the considered RU,which can be expressed as

D2D pair D2D pairD2D pair

STAs

D2D pair

RU

RU: Resource unitD2D: Device to device

1…

RUr

AP Resource allocation

RUs

Multiple concurrentcommunication links

RU2

RU3

RU4

RU5

RU6

Figure 1: The network scenario.


τ = 21 + CW , ð1Þ

where CW denotes the contention window for eachuplink random access phase, which is calculated and broad-cast by AP in the considered system model. Assume that theback-off value of AP is equal to CW.

Let Ptr denote the transmission probability, i.e., at leastone transmission occurs in the considered RU. Therefore,Ptr can be expressed as

Ptr = 1 − 1 − τð Þn, ð2Þ

where n is the number of contending STAs. Let PIdenote the idle probability, i.e., no transmission in the con-sidered RU, which can be calculated by

PI = 1 − τð Þn: ð3Þ

Let Ps denote the probability of successful transmissionin considered RU, i.e., only one STA transmits the packetin the considered RU. Accordingly, Ps can be obtained by

Ps =n 1 − τð Þn−1

Ptr= n 1 − τð Þn−11 − 1 − τð Þn : ð4Þ

Analogously, the collision probability Pc in the consid-ered RU is expressed as Pc = Ptr − Ps. Therefore, thethroughput for considered RU is

SRU = PsPtrE P½ �1 − Ptrð ÞTi + PsPtrTs + Ptr 1 − Psð ÞTc

, ð5Þ

where E½P� denotes the average packet payload size. Tidenotes the duration of an empty slot. Ts and Tc are theaverage length of time periods for a successful packet and acollision packet, respectively. To reduce unnecessary controloverheads, there is no ACK frame for each BSR frame in theproposed enhanced back-off mechanism. Consequently,assume that Ts = Tc as

Ts = Tc = TBSR + 2TSIFS, ð6Þ

where TBSR denotes the average length of time period fortransmitting successfully a BSR frame. Analogously, TSIFS isshort interframe space (SIFS). By substituting the above cor-relation formulas in (5), the SRU can be rewritten as

SRU = E P½ �n 1 − τð Þn−11 − τð ÞnTi + 1 − 1 − τð Þnð ÞTc

, ð7Þ

where the E½P� is a fixed value for all nodes. To achievethe maximum theoretical throughput limitation, as seen in(7), we only need to maximize the following equation:

θ = n 1 − τð Þn−11 − τð ÞnTi + 1 − 1 − τð Þnð ÞTc

: ð8Þ

For a given n, there exists an optimal CW∗ for maximiz-ing the system throughput SRU in considered RU. Thus,taking the partial derivative in (8) with respect to τ, thederivative of (8) is given as

dθdτ

= n 1 − τð Þn−2 1 − nτð ÞTs + Ti − Tsð Þ 1 − τð Þn + n2τ Ti − Tsð Þ 1 − τð Þ2n−2

Ts + Ti − Tsð Þ 1 − τð Þnð Þ2= 0:

ð9Þ

Using the simplification rule in [30], the expression of τcan be approximately derived from the following equation:

τ =ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffin + 2 n − 1ð Þ Ts′ − 1ð Þð Þ/ n − 1ð Þp

n − 1ð Þð Þ/ Ts′ − 1ð Þð Þ ≈1

nffiffiffiffiffiffiffiffiffiffiTs′/2

q , ð10Þ

where Ts′ = Ts/Ti. By simplifying and substituting the τvalue in (1) into (10), we can obtain an approximate rela-tionship between CW∗ and n from the following:

DIFS

BSR-STA1

BSR-STA2

C

C

BSR-STA5

I

BSR-STA3

BSR-STA1

BSR-STA2

BSR-STA5

BSR-STA3. . . TF

Downlink transmission andresource allocation phase

K slot

... ......

Uplink random access phase

SIFS

SIFS

TF-R

RU1

RU3RU4

RU5RU6RU7

RUr

...

DL_MU_data_transmission

SIFS

ACK

ACK

ACK

ACK

ACK

ACK

ACK

ACK

SIFS

STA1-D1

STA2-D3

STA3

STA5-D7

STA4

STA6

STA8-D2

...

Payload (STA1 and D2D1)


Payload (STA3)


Payload (STA4)

Payload (STA6)


MBA

...

SIFS

Uplink data transmission

SIFS

SIFS

TF

RU2

I: idle channelC: collision

Figure 2: The channel access and data transmission process of MISD.


CW∗ ≈ n

ffiffiffiffiffiffiffi2Ts

Ti

s: ð11Þ

Based on the proposed theoretical analysis model, theoptimal CW∗ can be calculated according to the variables,i.e., n, Ts, and Ti. In practical 802.11ax networks, these var-iables might be different and changed with respect to time.Compared to the STAs, AP can track these variables timelyand adjust adaptively the CW∗ for optimizing the MACaccess efficiency.

Enhanced back-off mechanism. Different from the legacyuplink random access mechanism, the proposed enhancedback-off mechanism executes the back-off process in fre-quency and time domains, simultaneously. As shown inFigure 3, the uplink random access period includes multipleback-off substages and r available RUs in time and frequencydomains, respectively. Let CWi = CW∗ denote the CW ofSTA i, i ∈ S. Accordingly, let boi ∈ ½0, CW∗� denote theback-off value of STA i.

Let Si denote the STA i, i ∈ S. In each back-off substage, Siupdates its boi, as boi = boi − r. Specifically, if boi ≤ 0, Si ran-domly chooses a RU and sends its BSR to AP. After that, theback-off value of Si will be reset as boi = CW∗. Note that themaximum back-off stages are limited by ∣CW∗/r ∣ ; thus,each STA can contend for channel resource once duringthe current uplink random access phase. Let TSi ∈ f0, 1gdenote the transmission result of BSR of Si, where TSi = 1denotes that AP has received successfully the BSR framefrom Si; otherwise, TSi = 0. Correspondingly, AP adds Si intothe available scheduling stations set denoted as NS, whileTSi = 1. Repeat the back-off process until the number ofSTAs in NS is larger than r or uplink random access phaseis over. Since the resource allocation results for data trans-mission will be announced by AP with the trigger frame(TF) frame, AP will not respond with ACK for each BSRframe in the proposed algorithm, which can reduce controloverheads. More specifically, we summarize the enhancedback-off mechanism with optimal CW in Algorithm 1.

3.2.3. Resource Allocation. The uplink resource allocation isexecuted while AP successfully sends the TF on the wholechannel. To avoid the waste of channel resources, AP canschedule the downlink transmission during this phase. Sincethis paper only focuses on uplink resource allocation anddata transmission, the downlink data scheduling and trans-mission processes are skipped. Note that, the resource reusetechnology is employed in D2D-enabled 802.11ax network.Hence, AP can schedule multiple D2D pairs to share thesame RU with the specific STA based on the efficientresource allocation algorithm, which significantly enhancesthe spectrum efficiency in data transmission. However, theresource allocation algorithm faces two key challenges: (1)identifying whether or not the STA and D2D link can sharethe same RU; (2) determining which D2D pairs can reusethe same RU. To tackle these challenges, the complex inter-ference problems among D2D pairs and STAs have beeneffectively solved in the proposed resource allocation algo-rithm. Specifically, we introduce the interference graph and

the conception of MIS to group the D2D pairs withoutinterference into the same MIS. In addition, the interfer-ence matrix for D2D pairs and STAs, denoted as L, isestablished. Based on MIS and L, we develop an efficientresource allocation algorithm, where multiple D2D pairsand the STA are allowed to share the same RU. Todescribe the resource allocation mechanism in detail, wedecouple it into three parts: formulating D2D interferencegraph and generating maximum independent set andresource allocation decision, and give the correspondingintroduction as follows.

Formulating D2D interference graph. The formulatingalgorithm of D2D interference graph is executed whenD2D pairs are established or D2D pairs change. Let G = <V , E > denote the interference graph of D2D pairs. Notethat G is constructed according to the interference relation-ship of D2D pairs. The vertexes in set V represent D2D pairscorresponding to D = fD1,D2,⋯,Dmg defined in Section3.1. Besides, the edges in set E represent the interferencerelationship between D2D pairs defined by the SINR ofreceiver. It is considered that the communication links willbe interfered, when the SINR of receiver of any link has lessthan the threshold. The instantaneous SINR of receiver forD2D pair is as

SINRD = PDαDPNαN ,D +No

, ð12Þ

where PD is the transmit power of D2D communicationdevice. αD is the path loss of the D2D pair, and αN ,D can berepresented as

αN ,D = λ

4πρ

� �2, ð13Þ

where ρ is the distance of the two D2D pairs and λ is thewavelength of the signal.

Figure 4 shows an illustrative example of generating theinterference graph of D2D pairs, where Figure 4(a) gives thenetwork topology of D2D pairs. Based on this, thecorresponding interference graph G is formulated inFigure 4(b). Referring to Figure 4(c), when the D1 and D2transmit data packets in the same RU in parallel, both datalinks will be interfered, i.e., the communication links fromSender 1 to Receiver 1 and Sender 2 to Receiver 2 will causethe interference for Receiver 1 and Receiver 2, respectively.Accordingly, both Receiver 1 and Receiver 2 cannot receivedata correctly. To avoid this issue, the proposed resourceallocation algorithm introduces the MIS to avoid allocatingthe same RU for these interference links.

Maximum independent sets. In accordance with thegraph theory, there is no adjacency between any two nodesof the same MIS. Inspired by this, we introduce the conceptof MIS and divide the D2D pairs into multiple MISs accord-ing to their interference relationship in an iterative manner.Correspondingly, the D2D pairs without interference will bedivided into the same MIS, which can share the same RU forparallel transmission. Note that, the process of generating


the MISs is triggered when the interference graph is estab-lished or updated. Let IS = fIS1, IS2,⋯, ISKg denote theinformation of MISs, where the ISk, k ∈ ½1, K� stores theD2D pairs’ IDs in the kth MIS. Note that the K varies with

the interference graph. Let I = ½Ii,j�ðm×mÞ, i, j ∈D denote theinterference relationship between D2D pairs. Specifically,Ii,j can be expressed as

Ii,j =1, mutual interference,0, otherwise:

(ð14Þ

Consequently, the D2D pairs i and j, where i, j ∈ ISk, canshare the same RU to transmit data, simultaneously due toIi,j = 0. Algorithm 2 describes the procedure of generatingMISs in detail. Once the D2D interference graph G is createdor changed, AP will run Algorithm 2 to generate or updatethe IS. For a given interference graph G, it is easy to findthe nodes which belongs to the MIS by judging the interfer-ence relationship between the D2D pairs in current MIS andthe D2D pairs in V . If there is no interference relationshipbetween the D2D pair i ∈ V and all D2D pairs in the kthMIS ISk, i.e., ∑j∈ISk Ii,j = 0, the D2D pair i will be added intoISk and removed from V . Repeat the above process to find

RUr…

RU5RU4RU3RU2RU1TF-R

randomaccessstart

boi = CW⁎

Uplink random access phase

Random access END

bo_AP ≤ 0 || n_STAs ≥ r

boi = boi – r

if boi ≤ 0

boi = boi – r

if boi ≤ 0

boi = CW⁎

boi: Back off counter of STAi

Multiple back-off sub-stages

n_STAs: Total number of STAs having contended RUsbo_AP: Back off counter of AP

RU in successful state

RU in idle stateRU in collision state

Freq

uenc

y do

mai

n

Time domain

Rand

om ly

sele

cta

RU

Rand

om ly

sele

cta

RU

RUr…

RU5RU4RU3RU2RU1

Figure 3: The uplink random access mechanism of MISD.

1: Initialize: Node set S, TSi = 0, i ∈ S2: AP computes CW∗ by (11) and broadcasts it and r3: for i in S do:4: Si randomly select a value from ½0, CW∗� as boi5: end for6: while t ≤ ∣CW∗/r ∣ do:7: for i in S do:8: boi = boi − r9: if boi ≤ 0 do:10: Si randomly choose a RU to send BSR11: if AP received BSR from Si do:12: TSi = 1 and NS =NS ∪ Si13: end if14: boi = CW∗

15: end if16: end for17: if ∣NS ∣ ≥r do:18: break19: end if20: t + = 121: end while

Algorithm 1: The enhanced back-off mechanism


the next MIS ISk+1 until the node set V is empty. Finally, APsort IS in descending order, i.e., the number of D2D pairs inIS1 is maximum, which helps to reduce the computationalcomplexity of resource allocation decision process andimprove the decision efficiency.

Resource allocation decision. This phase is triggered bythe TF frame for maximizing the spectrum efficiency. Let

the interference matrix L = ½Lsi ,Di�ðn×mÞ, where Si ∈ S and Di

∈D denote the interference relationship of Si and Di. TheLi,j can be expressed as

Li,j =1, Si andDi interfere with each other,0, otherwise:

(ð15Þ

In the proposed resource allocation algorithm, AP firstallocates RUs for STA in NS. Note that NS is obtained in

uplink random access phase. Subsequently, based on the ISand L, AP only needs to do match-action operation onceto allocate the specific RU for D2D pairs. Accordingly, thetime complexity of each match-action equals to the numberof D2D pairs in IS1, denoted as m1ðm1 ≤mÞ, and the m1decreases gradually during RU allocation phase. Thus, theworst time complexity of the proposed RU allocation mech-anism is OðrmÞ in r available RU scenario. Let Si denote thespecific STA that can use the ith RU in data transmissionphase. In each match-action operation for making the spe-cific RU allocation decision, AP estimates the interferencerelationship among the Si and Dj ∈ IS1 only. In particular,AP will allocate the RU for Dj, only when LSi ,Dj

= 0, ðDj ∈IS1Þ, i.e., adding the Dj into set DRi and removing Dj fromIS1, accordingly. Exceptionally, if no D2D pair in IS1 canshare the RU with Si, AP will do the same match-action

D3

D4

D6

D8D9

D11

D12

D1

D10

D2

D7

D5

AP

(a) Network topology

D7

D6

D4

D5

D3D2

D1

D12

D11

D10D9

D8

AP

(b) Interference graph

Sender 1

Sender 2

Receiver1

Receiver2

D1

D2

Transmission link

Interference link

(c) Interference link

Figure 4: Illustrative example of the interference graph of D2D pairs.


operation for the next MIS IS2 with the Si. After that, APsorts IS in descending order. Since IS1 always includes max-imum noninterference D2D pairs, AP always try to schedulemore D2D pairs to share the same RU. Hence, the proposedresource allocation algorithm in MISD can significantlyimprove decision-making efficiency by enabling the match-action operation based on MISs, meanwhile maximizingspectrum efficiency by providing more spectrum reuseopportunities for nodes. Algorithm 3 depicts the pseudocodeof the proposed resource allocation algorithm.

We present an example to illustrate the detail of theresource allocation procedure in the proposed algorithm,as described in Figure 5. Firstly, AP will allocate a uniqueRU for each STA in NS. Consequently, these STAs cantransmit simultaneously data in the specific RU in the nextdata transmission phase. To maximize the spectrum effi-ciency, the multiple D2D pairs are scheduled to share thespecific RU based on the proposed resource allocation algo-rithm. More specifically, AP allocates the specific RU for Si∈NS, where IS1 = ½D1,D2,D3,D5� and the interference rela-tionship matrix of Si is L½Si� = ½1, 1, 0, 0, 0, 1, 1�. The numberof elements in L½Si� equals to m. Based on this, AP estimatesthe interference relationships by matching L½Si� and IS1 tomake the RU allocation decision for the D2D pairs in IS1.As shown in Figure 5, the blue box denotes that Si and thecorresponding D2D pair exist interference relationship, i.e.,LSi ,D1

= 1 and LSi ,D2= 1. Accordingly, the D1 and D2 cannot

reuse the RU. Otherwise, the red box denotes the corre-sponding D2D pair, (i.e., D3 and D5) can share the currentRU with Si due to LSi ,D3

= 0 and LSi ,D5= 0. The resource allo-

cation results for D3 and D5 are stored in DR½i� for ith RU;meanwhile, AP will update IS to make the number of D2Dpairs in IS1 maximum.

3.2.4. Uplink Data Transmission Process. The uplink datatransmission phase is performed when AP sends the TFframe. All nodes (i.e., STAs and D2D pairs) sense and parsethe TF frame, where the TF frame contains all RU resourceallocation results for STAs and D2D pairs. Based on this,each node can identify whether the AP has assigned RUfor it or not. Assume that the uplink transmission period isfixed as the transmission opportunity (TXOP). Therefore,the corresponding nodes can share the same RU and trans-mit data in parallel. Moreover, AP will respond with themultiple block ACK (MBA) for the STAs which has success-fully sent the data in uplink data transmission phase.

4. Numerical Analysis

In this section, we model and derive the system throughputin uplink random access phase. As mentioned in Section 3,the channel is divided into r RUs. Assume that all RUs areavailable in uplink random access phase for STAs. Similarto the legacy 802.11 network, we consider that there is nohidden terminal in network. Meanwhile, we assume thatthe channel condition is ideal and the packet error occursonly when multiple STAs transmit BSR frame in the sameRU, simultaneously. Different from legacy DCF mechanism,we redefine the channel states as idle, successful transmis-sion, and collision, according to the states of r RUs in thesame time domain. In particular, the detail description isshown in Figure 6.

Let PI′denote the probability of channel idle, i.e., all RUsare in idle states, in the same time domain. Ps′ presents thesuccessful transmission probability of channel, i.e., at leastone RU is in successful transmission state and others arein idle states in the same time domain. Similarly, Pc′ denotes

Input: G = <V , E > , I = ½Ii,j�ðm×mÞ, ði, j ∈DÞOutput: IS = fIS1, IS2,⋯, ISkg1: k⟵ 02: when V is not empty do3: k⟵ k + 1, Im ⟵ 04: Select a D2D pair v in V as the initial node5: Add v into ISk, and remove it from V6: for i⟵ 0 to lenðV) do7: for j⟵ 0 to len(ISk) do8: if Ii, j == 1 then9: Im ⟵ Im + 110: Break11: end if12: end for13: if Im == 0 then14: Add D2D pair j into ISk and remove it from V15: Update ISk16: end if17: end for18: end when19: Sort (IS)20:Return IS

Algorithm 2: Generating maximum independent sets


the collision probability of channel, i.e., at least one RU hasoccurred collision in the same time domain. Specifically, Ps′and PI′ are expressed as

Ps′= 〠r

i=1Ps

iPIr−i, ð16Þ

PI′= PIr: ð17Þ

Consider that r available RUs can be used for differentSTAs to transmit BSR in the same slot in the proposedalgorithm. Correspondingly, we can calculate the systemthroughput ST in uplink random access phase by the for-mula derived in Section 3, which can be expressed as

ST = ∑rk=1PsE P½ �

Ps′Ts + Pi′Ti + 1 − Ps′− Pc′� �

Tc

: ð18Þ

Input: NS, IS, L, rOutput: DR1: AP allocates RU for the STA in NS2: i⟵ 0, T ⟵ 03: when NS ≠∅ or i ≠ r do:4: c⟵ 05: STA⟵NS½i�6: for j⟵ 0 to lenðS½T�Þ do:7: D ← IS½0�½j�8: if LSTA,D == 0 then:9: Add D into DR and remove it from S½T�.10: c⟵ c + 1;11: end if12: end for13: if c ≠ 0 then14: T ⟵ T + 115: else16: Update IS, T ⟵ 117: end if18: i⟵ i + 1;19: end when20: Return DR

Algorithm 3: Resource allocation decision algorithm

IS = [[D1, D2, D3, D5], [D4, D6, D7]]

For Example Si in NS, Si

DR [i] = [D3, D5] IS = [[D4 ,D6, D7], [D1, D2]]

RU allocation forD2D link pairs Update IS

NS = [S1, S2, S3, S4, S5, S6, S7, S8, S9]

D1 D2 D3 D4 D5 D6 D7L [Si] = [ 1, 1, 0, 0, 0, 1, 1]IS [0] = [D1, D2, D3, D5]

NS⫙

Figure 5: An example for allocating RU to D2D link pairs.


Consequently, by substituting the above correlation for-mulas in (16) and (17), the system throughput ST in (18)can be rewritten as the following:

ST = rPsE P½ �1 −Qr

i=1Pi

� �Ts +

Qri=1Pi

� �Ti

= rnτ 1 − τð Þ n−1ð ÞE P½ �1 − τ 1 − τð Þnrð ÞTs + 1 − τð ÞnrTi

:

ð19Þ

Hence, the system throughput of the uplink randomaccess phase in the MISD algorithm is determined by r, n,Ps, and Ts. Thereby, the maximum system throughput inuplink random access phase can be obtained by employingthe optimal CW for a given n and Ts, which has beenderived in Section 3.2.2. Moreover, by integrating D2D into802.11ax network, less nodes n need to contend for channelresource, and hence, Ps in MISD can be further optimizedespecially in dense scenarios. Meanwhile, it also reducesthe invalid channel occupancy time by canceling ACK forBSR, i.e., the transmission time for data Ts is increased.Obviously, the MAC access efficiency can be improved byenabling the enhanced back-off mechanism with the opti-mal CW in MISD. Besides, the higher MAC access effi-ciency implies that the more channel time can be usedfor data transmission, which indirectly optimizes the spec-trum efficiency.

5. Performance Simulation

5.1. Evaluation Settings. In this section, we conduct extensivesimulations to evaluate and analyze the performance of theproposed algorithm in terms of system throughput, comple-tion time, collision rate, and channel utilization using a cus-tom simulator in Matlab. The proposed algorithm iscompared with the legacy 802.11ax, G-OFDMA, and InGRA[31]. InGRA is an efficient resource allocation scheme forD2D link pairs in cellular network, and the related simula-tion results indicate that it also obtains the good perfor-mance in WLANs. Table 1 summarizes the relatedsimulation parameters. In the simulations, we focus onuplink transmission condition in a single BSS scenario.Wherein, AP is deployed at the center of a circle with aradius of 100m and acts as a centralized controller to allo-

cate RU resource and schedule uplink transmission. Allnodes (10-100) including STAs and D2D pairs are saturatedstates and randomly distributed in the simulation field. Inparticular, the maximum distance between the receiverand transmitter for each D2D pair is less than 30m. Weconsider the channel bandwidth is 40MHz, which isdivided into 8 or 18 RUs. Moreover, each simulation runlasts 100 s and the corresponding result is averaged over10 independent experiments.

5.2. Simulation Results. In what follows, we consider twosimulation scenarios: (1) different number of node (10-100) scenario, where the number of D2D pairs varies withthe network topology; (2) different number of D2D pairs(4-28) for a given number of node (100) scenario. Besides,the impact of different RUs (8 and 18) on network perfor-mance is considered in both cases. Besides, we investigatethe performance of system throughput, completion time,channel utilization, and collision rate, respectively.

5.2.1. Different Number of Nodes. Figure 7 illustrates theimpact of the different number of nodes on system through-put when the MISD, 802.11ax, G-OFDMA, and InGRAalgorithms are used in first considered simulation scenario.Referring to Figure 7, both the proposed MISD and InGRAalgorithms have achieved the significant performanceimprovements in terms of the system throughput, especiallyin dense scenarios. This result is expected because D2Dcommunication and the channel resource reuse technologiesare exploited in both algorithms. Meanwhile, as the increas-ing of nodes in the network, nodes are offloaded to D2Dcommunication mode with higher probability. Correspond-ingly, less STAs need to contend channel in uplink randomaccess phase and more D2D pairs can transmit packets inparallel, which improves spectrum access efficiency and fur-ther increases spectrum reuse probability. In MISD, AP canschedule more D2D pairs to share the same RU, since italways preferentially allocates RU for the D2D pairs in thelargest MIS, which achieves higher resource reuse thanInGRA. Particularly, while the number of nodes is less than30, the G-OFDMA outperforms the MISD in 18 RU sce-nario. However, the throughput performance of G-OFDMA is almost unchanged as the number of nodesincreases. The main reason is that the G-OFDMA adapts

Ni: number of i in one time slotNs: number of s in one time slotNc: number of c in one time slot

RU1RU2RU3RU4…

RUr

iiiiii

isiiii

isciii

Idle Success Collision

Idle on the RU, marked as i

Collision on the RU, marked as cSuccess on the RU, marked as s

Ni = r, Ns = 0, Nc = 0PI

Ps

Pc´

Ns > = 1, Ni + Ns = r, Nc = 0

Nc > = 1, Ni + Ns + Nc = r

Figure 6: Illustration of channel status on time domain.


the fixed uplink random access stages and the maximumsystem throughput is limited by the available RUs.

Figure 8 compares the variation of the average comple-tion time for transmitting 1 Mega bit file when OFDMA-based algorithms and D2D offloading in OFDMA algo-rithms are used, respectively. Note that more transmissionopportunities and higher successful transmission probabilitycan reduce the completion time of transmitting the givendata. Irrespective of the number of nodes, both MISD andInGRA algorithms yield the better performance in terms ofcompletion time than that of the legacy 802.11ax and G-OFDMA. This is due to the fact that both MISD and InGRAalgorithms offload nodes to D2D communication mode,where D2D pairs can share the spectrum resource with STAsfor data transmission, i.e., more nodes can access channel totransmit data, simultaneously, which gives more transmis-sion opportunities for nodes. It also can be observed thatthe completion time of four algorithms markedly increaseswith the increases of nodes. Compared with 802.11ax andG-OFDMA, the performance of MISD is significantlyimproved and outperforms InGRA in most cases. More spe-cifically, the completion time of MISD is 76%, 62%, and 5%less than that of the 802.11ax, G-OFDMA, and InGRA,

respectively (8 RUs and 100 nodes). This can be attributedto two factors: (1) the MISD algorithm relieves the data col-lision in uplink random access by offloading nodes to D2Dcommunication mode and providing higher successfultransmission probability for STAs with the enhanced back-off mechanism; (2) in MISD, AP schedules multiple D2Dpairs and STA to share the same RU with low time complex-ity and offers more transmission opportunity for D2D pairsbased on the MISs.

Figure 9 compares the collision rate of the four algo-rithms, where the number of nodes ranges from 10 to 100,where both 8 and 18 RU cases are considered. Note thatthe collision rate results are directly related to the uplinkrandom access mechanism and the number of contentionnodes. The results depicted in Figure 9 are diverse. The col-lision rate increases with the increasing number of nodes inboth 802.11ax and InGRA. However, the results are almostunchanged for G-OFDMA. More interestingly, the collisionrate of MISD decreases as the number of nodes increases. Inparticular, the collision rate gain of MISD is 64%, 44%, and64% compared with 802.11ax, G-OFDMA, and InGRA inthe case of 8 RUs and 100 nodes, respectively. This can beexplained by the fact that the proposed MISD algorithm

Table 1: Simulation parameters.

Parameters Value Parameters Value

Subchannel number 8 or 18 Channel bandwidth 40MHz

SIFS 20 us Data rate 12 or 6Mbps

Slot time 10 us CWmin 15

DIFS 50 us CWmax 64

B-ACK length 144 bits MAC header length 240 bits

Packet length 1500 bytes Physical header length 120 bits

100

25

50

Syste

m th

roug

hput

(Mbp

s)

75

100

125

150

175

200

20 30 40 50Number of nodes

60 70 80 90 100

802.11ax+8802.11ax+18MISD+8MISD+18

G_OFDMA+8G_OFDMA+18InGRA+8InGRA+18

Figure 7: System throughput vs. different nodes.


reduces the number of contention nodes by enabling D2Doffloading technique and the proposed enhanced back-offmechanism further relieves data collision by allowing STAaccess channel only once. Hence, the number of STAs con-tending for channel has been greatly reduced. Besides, APderives the optimal CW according to the network condition,which reduces the unnecessary data collisions so as toenhance MAC access efficiency.

Figure 10 compares the channel utilization of the fouralgorithms as a function of the number of nodes. It can beobserved that the trends of channel utilization results of allalgorithms are almost similar with the number of nodechanges, since the OFDMA technology has been enabledin all algorithms and each RU will be used for at least onenode (STA or D2D pair) to transmit data during uplinktransmission phase. As a result, the channel utilization

0.0

0.2

04

0.6

0.8

1.0

Com

plet

ion

time (

sec)

10 20 30 40 50Number of nodes

60 70 80 90 100

802.11ax+8802.11ax+18MISD+8MISD+18


Figure 8: Completion time vs. different nodes.


60 70 80 90 1000.0

0.2

0.4

0.6

0.8

1.0

Col

lisio

n ra

te

802.11ax+8802.11ax+18MISD+8MISD+18


Figure 9: Collision rate vs. different nodes.


performance is limited by the available RUs and total datatransmission time. For both D2D offloading algorithms,i.e., MISD and InGRA can obtain the higher channel utili-zation compared to the 802.11ax and G-OFDMA under dif-ferent network densities. This is due to the fact that theMAC access efficiency in uplink random access phase canbe improved by reducing the number of contention nodesand hence all nodes have more opportunities to transmitdata instead of contending channel resource for the givensimulation time.

5.2.2. Different Number of D2D Link Pairs. In this subsec-tion, we deploy 100 nodes as a dense scenario and furtherevaluate the performance of MISD with different numbersof D2D pairs. Wherein, we set the number of D2D pairsfrom 4 to 28 by limiting the total number of D2D pairsunder the same network topology. Besides, different num-bers of RUs (8 and 18) are considered in this scenario.

Figure 11 plots the system throughput versus the num-ber of D2D pairs when the MISD and InGRA are used,respectively. It is evident that the system throughput results


60 70 80 90 1000.5

0.6

0.7

0.8

0.9

1.0

Chan

nel u

tiliz

atio

n

802.11ax+8802.11ax+18MISD+8MISD+18


Figure 10: Channel utilization vs. different nodes.

40

25

50

75

Syste

m th

roug

hput

(Mbb

ps)

100

125

150

175

200

8 12 16Number of D2D link pairs

20 24 28

MISD+8MISD+18

InGRA+8InGRA+18

Figure 11: System throughput vs. different D2D pairs.

0.0

0.2

0.4

0.6

0.8

1.0

Col

lisio

n ra

te

4 8 12 16Number of D2D link pairs

20 24 28

MISD+8MISD+18

InGRA+8InGRA+18

Figure 12: Collision rate vs. different D2D pairs.


of MISD and InGRA increase as the number of D2D pairsincreases. Moreover, both two algorithms have similar sys-tem throughput when the number of D2D pairs is less than8. The result is reasonable because the system throughputgain is limited by the number of D2D pairs while it is lessthan the available RU. However, as the number of D2D pairsincreases, the throughput performance of MISD is betterthan InGRA. In particular, when the number of D2D pairsis 28, the system throughput of MISD is higher than thatof InGRA by about 21% and 11% under 8 and 18 RUs,respectively. The results confirm that the proposed MISDalgorithm effectively improves the spectrum reuse efficiencyby scheduling more D2D pairs to share the same RU withthe higher probability based on the MISs compared to theInGRA especially in the dense network. Moreover, MISDexploits an enhanced back-off mechanism including adap-tive CW adjustment strategy, which reduces the controloverheads and further improves the MAC access efficiencyin uplink random access phase.

Figure 12 compares the collision rate of the two algo-rithms as a function of the number of D2D pairs. The colli-sion rate decreases with the increasing number of D2D pairsin both algorithms, where the collision only occurs in uplinkrandom access phase in both algorithms. Reasonably, thenumber of STAs will decrease with the number of the D2Dpairs that increases for the given number of node scenario.Moreover, it is obvious that the collision rate of using theproposed MISD algorithm is lower than that of the InGRAalgorithm. More specifically, the collision rate of MISD canbe reduced by 64% over InGRA under 28 D2D pairs and 8RU case. The result is intuitive because that the proposedMISD algorithm exploits an enhanced back-off mechanism,in which each STA employs the optimal CW to choose itsback-off value and only contends channel once. Thus, thecollision can be effectively relieved by reducing the number

of the contention STAs so as to avoid secondary collisionin considered random access phase in MISD.

Figure 13 illustrates the variation of the average comple-tion time for transmitting 1 Mega bit file, when the MISDand InGRA algorithms are used. The completion timeresults for the two algorithms decrease with the increasingof D2D pairs. This result is expected because, for the givennumber of nodes, more nodes adopt the D2D communica-tion mode that is beneficial not only in reducing the timefor uplink random access but also in providing more datatransmission opportunities for nodes by enabling spectrumreuse technology. Hence, the node can access the specificRU and transmit data quickly even in highly dense D2D pairscenario. Besides, it can be observed that the completiontime of the proposed MISD is slightly lower than that ofInGRA. Specifically, when the number of D2D pairs is 28,the MISD achieves about 4% and 2% performance improve-ment over InGRA with 8 and 18 RUs, respectively. Theslight improvement in performance can be attributed tothe enhanced back-off mechanism in MISD, where AP willnot respond to ACK for each BSR which reduces the channeloverheads. Moreover, the optimal CW is enabled to enhancethe MAC access efficiency. Consequently, AP can schedulenodes (D2D pairs and STAs) to transmit data timely in datatransmission period with the proposed MISD algorithm.

In this simulation, we further evaluate the impact of thenumber of D2D pairs on the performance of channel utiliza-tion. Referring to Figure 14, both InGRA and MISD canachieve high channel utilization especially in highly densenetwork scenarios by offloading nodes to D2D communica-tion mode. The channel utilization results are near-optimalbecause the RU resources are fully used by STAs and D2Dpairs according to the direct scheduling of AP during theuplink transmission phase. Besides, simulation results ofchannel utilization also confirm that enabling D2D


20 24 280.0

0.1

0.2

0.3

0.4

0.5

0.6

Com

plet

ion

time (

Sec)

MISD+8MISD+18

InGRA+8InGRA+18

Figure 13: Completion time vs. different D2D pairs.


offloading schemes in 802.11ax is beneficial not only inimproving the spectrum efficiency by enabling resourcereuse technology in scheduling mode but also in shorteningthe period of the random-access process by reducing thenumber of contention nodes. As a result, high channel utili-zation can be obtained in both uplink random access anddata transmission phases.

6. Conclusions

In this paper, we have investigated the uplink resource allo-cation problem for dense WLAN to improve the spectrumefficiency by introducing the D2D offloading into IEEE802.11ax. We first developed an adaptive CW optimizationmodel for the enhanced back-off mechanism in uplink ran-dom phase to optimize the MAC access efficiency. The con-ception of MIS has been introduced to manage theinterference among D2D pairs. Based on this, we proposedan efficient resource allocation algorithm to maximize thespectrum efficiency, i.e., AP effectively schedules multipleD2D pairs and specific STA to share the same RU. Theperformance analysis results have demonstrated that theproposed algorithm minimizes the completion time andimproves system throughput, channel utilization. Besides,it has shown that the use of the enhanced back-off mecha-nism with optimal CW renders the random-access processmore efficient and also reduces the control overhead. Theresults confirm the feasibility of the proposed algorithm,which can be flexibly integrated into the next generationIEEE 802.11ax without extra hardware modification require-ments. However, the proposed MISD algorithm is only astarting point to explore innovation resource allocation pro-tocol by combining D2D offloading and 802.11ax. In futureresearch, it can examine the feasibility of extending the pro-posed MISD algorithm to the cases of dynamic network sce-nario and heterogeneous node types. Moreover, the deep

reinforcement learning-based approach can be designedand trained to intelligently optimize the resource allocationfor D2D pairs and STAs.

Data Availability

The key source code and data used to support the findings ofthis paper are available upon request to the correspondingauthor.

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper.

Acknowledgments

This work was supported in part by the Scientific andTechnological Research Program of Chongqing MunicipalEducation Commission under Grant KJQN201900626 andthe Funding for Doctoral Program of Chongqing Universityof Posts and Telecommunications (No. BYJS202116).

References

[1] E. Khorov, A. Kiryanov, A. Lyakhov, and G. Bianchi, “A tuto-rial on IEEE 802.11ax high efficiency WLANs,” IEEE Commu-nications Surveys and Tutorials, vol. 21, no. 1, pp. 197–216,2019.

[2] Y. Li, B. Li, M. Yang, and Z. Yan, “A spatial clustering groupdivision-based OFDMA access protocol for the next genera-tion WLAN,” Wireless Networks, vol. 25, no. 8, pp. 5083–5097, 2019.

[3] K. Wang and K. Psounis, “Scheduling and resource allocationin 802.11ax,” in Proceedings of IEEE INFOCOM - IEEE Confer-ence on Computer Communications, pp. 279–287, Honolulu,HI, USA, 2018.

[4] B. Bellalta and K. Kosek-Szott, “AP-initiated multi-user trans-missions in IEEE 802.11ax WLANs,” Ad Hoc Networks,vol. 85, pp. 145–159, 2019.

[5] B. Bellalta, “IEEE 802.11ax: high-efficiency WLANs,” IEEEWireless Communications, vol. 23, no. 1, pp. 38–46, 2016.

[6] V. T. Q. Duong, H. D. Tuan, and A. Kortun, “Optimal videostreaming in dense 5G networks with D2D communications,”IEEE Access, vol. 6, pp. 209–223, 2018.

[7] G. Naik, S. Bhattarai, and J. Park, “Performance analysis ofuplink multi-user OFDMA in IEEE 802.11ax,” in Proceedingsof IEEE International Conference on Communications (ICC),pp. 1–6, Kansas City, MO, USA, 2018.

[8] R. Ali, S. W. Kim, B. S. Kim, and Y. Park, “Design of MAClayer resource allocation schemes for IEEE 802.11ax: futuredirections,” IETE Technical Review, vol. 35, no. 1, pp. 28–52,2018.

[9] Q. Qu, B. Li, M. Yang, and Z. Yan, “AnOFDMA based concur-rent multiuser MAC for upcoming IEEE 802.11ax,” in Pro-ceedings of IEEE Wireless Communications and NetworkingConference Workshops (WCNCW), p. 134, New Orleans, LA,USA, 2015.

[10] T. Uwai, T. Miyamoto, Y. Nagao, L. Lanante, M. Kurosaki, andH. Ochi, “Adaptive backoff mechanism for OFDMA randomaccess with finite service period in IEEE802.11ax,” in


20 24 280.0

0.2

0.4

0.6

0.8

1.0

Chan

nel u

tiliz

atio

n

MISD+8MISD+18

InGRA+8InGRA+18

Figure 14: Channel utilization vs. different D2D pairs.


Proceedings of IEEE Conference on Standards for Communica-tions and Networking (CSCN), pp. 1–6, Berlin, Germany, 2016.

[11] N. Shahin, R. Ali, S. W. Kim, and Y. T. Kim, “Cognitive backoffmechanism for IEEE802.11ax high-efficiency WLANs,” Jour-nal of Communications and Networks, vol. 21, no. 2,pp. 158–167, 2019.

[12] J. Wang, M.Wu, Q. Chen, Y. Zheng, and Y. H. Zhu, “Probabil-ity complementary transmission scheme for uplink OFDMA-based random access in 802.11ax WLAN,” in Proceedings ofIEEE Wireless Communications and Networking Conference,pp. 1–7, Marrakesh, Morocco, 2019.

[13] S. Lin, H. Qi, X. Wen, Z. Lu, and Z. Hu, “An efficient group-based OFDMA MAC protocol for multiuser access in denseWLAN systems,” in Proceedings of IEEE International Confer-ence on Communications (ICC), pp. 1–6, Kansas City, MO,USA, 2018.

[14] D. Bankov, A. Didenko, E. Khorov, and A. Lyakhov, “OFDMAuplink scheduling in IEEE 802.11ax networks,” in Proceedingsof IEEE International Conference on Communications (ICC),pp. 1–6, Kansas City, MO, USA, 2018.

[15] D. Bankov, A. Didenko, E. Khorov, V. Loginov, andA. Lyakhov, “IEEE 802.11ax uplink scheduler to minimizedelay: a classic problem with new constraints,” in Proceedingsof IEEE 28th Annual International Symposium on Personal,Indoor, and Mobile Radio Communications (PIMRC), pp. 1–5, Montreal, QC, Canada, 2017.

[16] K. Dovelos and B. Bellalta, “Optimal resource allocation inIEEE 802.11ax uplink OFDMA with scheduled access,” 2018,https://arxiv.org/abs/1811.00957.

[17] J. Lee and D. J. Deng, “OFDMA-based hybrid channel accessfor IEEE 802.11ax WLAN,” in Proceedings of the 14th Interna-tional Wireless Communications and Mobile Computing Con-ference (IWCMC), pp. 188–193, Limassol, Cyprus, 2018.

[18] A. Yang, B. Li, M. Yang, and Z. Yan, “Concept and analysis ofcapacity entropy for uplink multi-user media access controlfor the next-generation WLANs,”Mobile Networks and Appli-cations, vol. 24, no. 5, pp. 1572–1586, 2018.

[19] M. Karaca, S. Bastani, B. E. Priyanto, M. Safavi, andB. Landfeldt, “Resource management for OFDMA based nextgeneration 802.11 WLANs,” in Proceedings of IFIP Wirelessand Mobile Networking Conference (WMNC), pp. 57–64, Col-mar, France, 2016.

[20] P. Mach, Z. Becvar, and T. Vanek, “In-band device-to-devicecommunication in OFDMA cellular networks: a survey andchallenges,” IEEE Communications Surveys and Tutorials,vol. 7, no. 4, pp. 1885–1922, 2015.

[21] S. T. Shah, S. F. Hasan, B. C. Seet, P. H. Chong, and M. Y.Chung, “Device-to-device communications: a contemporarysurvey,” Wireless Personal Communications, vol. 98, no. 1,pp. 1247–1284, 2018.

[22] Y. Cai, F. R. Yu, C. Liang, B. Sun, and Q. Yan, “Softwaredefined device-to-device (D2D) communications in virtualwireless networks with imperfect network state information(NSI),” IEEE Transactions on Vehicular Technology, vol. 65,no. 9, pp. 7349–7360, 2015.

[23] R. Cheng, C. Huang, and S. Pan, “WiFi offloading using thedevice-to-device (D2D) communication paradigm based onthe software defined network (SDN) architecture,” Journal ofNetwork and Computer Applications, vol. 112, pp. 18–28, 2018.

[24] A. Pyattaev, K. Johnsson, A. Surak, R. Florea, and S. Andreev,“Network-assisted D2D communications: implementing a

technology prototype for cellular traffic offloading,” in Pro-ceedings of Wireless Communications and Networking Confer-ence, pp. 3266–3271, Istanbul, Turkey, 2014.

[25] L. Zhao, H. Wang, and X. Zhong, “Interference graph basedchannel assignment algorithm for D2D cellular networks,”IEEE Access, vol. 6, pp. 3270–3279, 2018.

[26] S. Wen, X. Zhu, Z. Lin, X. Zhang, and D. Yang, “Energy effi-cient power allocation schemes for device-to-device(D2D)communication,” in Proceedings of Vehicular Technology Con-ference (VTC Fall), pp. 3266–3271, Las Vegas, NV, USA, 2013.

[27] M. Seyedebrahimi, A. Raschellà, F. Bouhafs, M. Mackay,Q. Shi, and M. H. Eiza, “A centralized Wi-Fi managementframework for D2D communications in dense Wi-Fi net-works,” in Proceedings of IEEE Conference on Standards forCommunications and Networking (CSCN), pp. 1–6, Berlin,Germany, 2016.

[28] P. Thulasiraman and X. Shen, “Interference aware resourceallocation for hybrid hierarchical wireless networks,” Com-puter Networks, vol. 54, no. 13, pp. 2271–2280, 2010.

[29] T. Nabetani, N. Madhavan, H. Mori, and T. Aoki, “A novellow-overhead channel sounding protocol for downlinkmulti-user MIMO in IEEE 802.11ax WLAN,” IEICE Transac-tions on Communications, vol. E101.B, no. 3, pp. 924–932,2018.

[30] J. Lei and H. Yun, “Decoupling channel contention and datatransmission in dense wireless infrastructure network,” in Pro-ceedings of the 8th Computer Science on-Line Conference,Cham, 2019.

[31] R. Zhang, X. Cheng, L. Yang, and B. Jiao, “Interference graph-based resource allocation (InGRA) for D2D communicationsunderlaying cellular networks,” IEEE Transactions on Vehicu-lar Technology, vol. 64, no. 8, pp. 3844–3850, 2015.


https://arxiv.org/abs/1811.00957

Documents

Enabling Device-to-Device (D2D) Communication for the Next