E-Fi: Evasive Wi-Fi Measures for Surviving LTE within 5 ...bocanegrac/papers/E-Fi_2018.pdf · posed...

E-Fi: Evasive Wi-Fi Measures for SurvivingLTE within 5 GHz Unlicensed Band

Carlos Bocanegra , Student Member, IEEE, Takai Eddine Kennouche,Member, IEEE,

Zhengnan Li , Student Member, IEEE, Lorenzo Favalli ,Member, IEEE,

Marco Di Felice, and Kaushik Chowdhury, Senior Member, IEEE

Abstract—The growing spectrum crunch has motivated exploratory efforts in the use of LTE in the 5 GHz bands for downlink traffic.

However, this paradigm raises concerns of fair sharing of the spectrum and the adverse impact of scheduled LTE frames on Wi-Fi

Packet Success Rates (PSR). To address this issue, we propose E-Fi, an interference-evasion mechanism that allows Wi-Fi devices to

survive LTE transmissions without any cooperation between these two different standards. Different from existing approaches, we

argue that the simple use of Almost Blank Subframes (ABS) within the LTE standard offering short channel access windows

overestimates opportunities for Wi-Fi. The pilots embedded in the ABS not only interfere with Wi-Fi but also adversely impact the

carrier sensing function. E-Fi mitigates this problem through a two-fold approach. It uses a combination of (i) Wi-Fi Direct with packet

relaying and (ii) classical distributed coordination function to reach distant nodes. Second, it ensures load balancing for both Wi-Fi

uplink and downlink traffic with high PSR by creating node-groups based with dedicated contention-based medium access intervals.

Our approach is validated by comprehensive simulation and experimental results that indicate significantly higher throughput in E-Fi

compared to classical Wi-Fi.

Index Terms—Coexistence, unlicensed band, LTE, Wi-Fi, optimal scheduling, matching, Hungarian algorithm, Almost Blank Subframes

(ABS), Packet Error Rate (PER), cell specific reference signals (CRS), further enhanced Inter-Cell Interference Coordination (feICIC)

Ç

1 INTRODUCTION

CELLULAR traffic has increased 4000 times over the lastten years, propelled by the growing adoption of smart-

phones nearing 50 percent of the electronic device market[1]. Furthermore, emerging areas like the Internet of Thingspredict billions of connected sensors worldwide within nextfew years, which will further stress existing communicationinfrastructures. One solution to this problem is the pro-posed LTE Unlicensed (LTE-U) paradigm that uses the 5GHz band for both enterprise-driven LTE and Wi-Fi net-works by assimilating spectrum from the unlicensed bands.However, the strict time-bound frame transmissions withinLTE and its extensive error recovery mechanisms raise con-cerns on the starvation of Wi-Fi in such shared spectrumuse. This paper attempts to address this issue by first dem-onstrating the limitations of existing standards-specifiedcoexistence techniques and then devising a new approachcalled Evasive WiFi (E-Fi) that combines Wi-Fi Direct andclassical 802.11 distributed coordination function (DCF).

When LTE and Wi-Fi coexist in the same spectrum, LTEis barely impacted, whereas the performance of Wi-Fi

drastically degrades to 70-100 percent packet error rate [2].LTE Release 10 includes eICIC (enhanced Inter-Cell Interfer-ence Coordination) that defines Almost Blank Subframes(ABS), which carry neither control nor data information. Pri-marily aimed for interference management between neigh-boring LTE small cells, the reuse of this technique for LTEand Wi-Fi coexistence was introduced in [3]. ABS allowsWi-Fi to gain access to the channel for a short time-fre-quency window, and by leveraging multiple ABS framesdevoid of cellular traffic this window can be extended.There is a fundamental assumption in the research involv-ing ABS scheduling so far: that Wi-Fi has truly undisturbedchannel access during the entirety of the ABS.

Our studies show that the simplistic assumption of acompletely interference-free ABS does not hold true in prac-tice, as the current LTE standards describe mandated andoptional reference signals (called pilots henceforth, shownby shaded time-frequency grid units in Fig. 1) within theABS that has significant impact on Wi-Fi. Release 11includes further eICIC (feICIC), with mechanisms for LTEusers to detect and cancel the signals from interfering cells.However, this capability is not present in Wi-Fi receivers.Release 13 describes Listen Before Talk (LBT), where LTE isexpected to perform carrier sensing and backoff before cap-turing the unlicensed channel. This proposal, however, hasnot been adopted in many key markets worldwide, includ-ing the U.S. Several prior works have relied on explicit feed-back from the Wi-Fi access points (APs) to the LTE basestation (BS) for sharing the medium. A differentiating aspectof our work is that the AP and the BS are unable to explicitlyexchange information; in fact there is no coordination mecha-nism defined up to the latest, still-evolving LTE Release 14.

� C. Bocanegra, Z. Li, and K. Chowdhury are with Northeastern University,Boston, MA 02215. E-mail: {bocanegrac, zhengnanlee, krc}@ece.neu.edu.

� T.E. Kennouche and L. Favalli are with the University of Pavia, Pavia27100, Italy. E-mail: {takaieddine.kennouche, lorenzo.favalli}@unipv.it.

� M. Di Felice is with the University of Bologna, Bologna 40126, Italy.E-mail: [email protected].

Manuscript received 15 June 2017; revised 12 June 2018; accepted 18 June2018. Date of publication 21 June 2018; date of current version 4 Mar. 2019.(Corresponding author: Carlos Bocanegra.)For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference the Digital Object Identifier below.Digital Object Identifier no. 10.1109/TMC.2018.2849409

830 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 18, NO. 4, APRIL 2019

1536-1233� 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See ht _tp://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Northeastern University. Downloaded on March 27,2020 at 17:14:55 UTC from IEEE Xplore. Restrictions apply.

https://orcid.org/0000-0003-2985-8521

https://orcid.org/0000-0003-2985-8521

https://orcid.org/0000-0003-2985-8521

https://orcid.org/0000-0003-2985-8521

https://orcid.org/0000-0003-2985-8521

https://orcid.org/0000-0002-3409-7942

https://orcid.org/0000-0002-3409-7942

https://orcid.org/0000-0002-3409-7942

https://orcid.org/0000-0002-3409-7942

https://orcid.org/0000-0002-3409-7942

https://orcid.org/0000-0002-4851-2393

https://orcid.org/0000-0002-4851-2393

https://orcid.org/0000-0002-4851-2393

https://orcid.org/0000-0002-4851-2393

https://orcid.org/0000-0002-4851-2393

mailto:

mailto:

mailto:

� E-Fi design goals and operational overview: E-Fi empowersthe Wi-Fi AP and its associated nodes to operate alongsideLTE transmissions, without dedicated feedback to/from theLTE BS. Instead, it reuses pre-set LTE ABS patterns toschedule its transmissions. Our solution partitions the Wi-Finetwork into self-organizing groups based on the observedPSR, using peer-to-peer communication without a Wi-Fi APvia Wi-Fi Direct [5]. Wi-Fi Direct designates a relay (akaGroup Owner), which emulates the Wi-Fi AP, and a Wi-FiDirect client, which interacts with the Group Owner. In E-Fi, the Group Owner (relaying node) forwards traffic to/from vulnerable nodes that are distant from the AP, butclose to the LTE BS. Thus, E-Fi purely looks at the coexis-tence problem from the Wi-Fi perspective. Recognizing thepractical difficulties in providing feedback to the LTE BS, aswell avoiding the complexities of solving optimizationproblems, E-Fi presents an algorithmic framework that ena-bles survival of the Wi-Fi network in uncooperative LTE-Udeployments and rogue small-cell installations.

The operating principle of E-Fi rests on three observa-tions: first, not all ABS offer equal transmission opportuni-ties for Wi-Fi. As shown in Fig. 1, depending upon wherethe ABS appears within the parent LTE downlink frame,i.e., its position in one or more of 0-9 subframes, it carriesdifferent types of pilot sub-carriers (so-called resource ele-ments in LTE). This in turn requires varying number ofcommitted resource units. For example, ABS 0 has signifi-cantly more presence of interfering pilots and offersreduced free channel access compared to others. Second, asshown in the right-hand side of Fig. 1, depending uponboth the specific ABS frame being considered and the sepa-ration between the LTE BS from the Wi-Fi nodes, there is anon-negligible impact on the carrier sensing and deferringprocess for the latter. Thus, Wi-Fi nodes close to the LTE BShave considerably less chances to transmit a packet [6]. Thismotivates our approach to schedule Wi-Fi groups into dif-ferent ABS, hence reducing the number of devices contend-ing for the channel. Third, SINR (and hence BER) can beconsiderably improved by reducing the transmission dis-tance, which motivates the approach of introducing relaynodes (e.g., m1 in Fig. 2) operating on Wi-Fi Direct. The E-Fimodule in the AP carefully assigns uplink and downlinkdurations (defined on the basis of whether traffic originatesat the AP or the nodes) and formsWi-Fi Direct groups basedon reported PSR. Though remote nodes (e.g., n1 and n2)now forward their traffic to the AP and vice-versa using

one-hop, this overhead is compensated with high PSR foreach link.

� Contributions: The main contributions of this work are:

1) We show that the assumptions in [2], [3] that Wi-Fihas undisturbed channel access during the entiretyof the ABS is a simplification that has practicalimpacts on coexistence.

2) Different from most approaches, we design E-Fiunder the assumption that the AP and the BS areunable to explicitly exchange information. We also donot introduce any changes within the LTE standard.

3) We undertake a methodical study on the impacts onWi-Fi PSR and carrier sensing mechanism caused byvarious pilot signals in different ABS configurationsthrough standards-compliant physical layer wave-form simulations.

4) We enable self-configuration of Wi-Fi nodes into Wi-Fi Direct groups with forwarding relays using PSRas a selection metric. We then propose a modifiedHungarian algorithm with well-defined complexityinstead of computationally expensive optimizationtechniques for the formation of such groups.

5) We formulate an ABS utilization strategy at the APthat partitions contention-based channel access timeinto distinct intervals, considering individual trafficloads and Wi-Fi Direct relay forwarding overheads.

The rest of the paper is organized as follows: Section 2describes the related work. Section 3 introduces the conceptof safe zones in E-Fi. Section 4 presents E-Fi’s spatial relayselection and a group formation strategy. Section 5 explainsthe scheduling and channel access mechanisms for singleand group node nodes. Section 6 validates our approach bymeans of an integrated MATLAB and NS2 simulation envi-ronments. The experimental results are reported in Section 7.Finally, Section 8 concludes the paper.

2 RELATED WORK

LTE-Wi-Fi coexistence strategies can be broadly catego-rized into spatial multiplexing, frequency multiplexingbased on channel selection, time multiplexing using dutycycling, LBT and ABS. Interference management hasmainly focused on the downlink due to the asymmetric

Fig. 1. Channel availability experimental set-up using an Atheros NIC,emulating a Wi-Fi device, and srsLTE, running on a Zedboard equippedwith the AD-FMCOMMS3, acting as an LTE interferer. The control sig-nals within an ABS Subframe (Sfm) are generated accordingly to theLTE standard and are included in the parent LTE frame [4].

Fig. 2. Example scenario showing impact of different ABS on Wi-Fipacket reception due to the control signals embedded in such frames.Wi-Fi coverage radius is chosen for 90 percent PSR and 17 dBm trans-mitted power. The dashed and dotted lines define the coverage areaswhen subframe 1 and 0 are designated ABS, respectively, showing thatdifferent subframes allow varying spatial coverage for 90 percent PSR.

BOCANEGRA ETAL.: E-FI: EVASIVE WI-FI MEASURES FOR SURVIVING LTE WITHIN 5 GHZ UNLICENSED BAND 831


nature of traffic in LTE, in the order of 8:1 [16]. Dutycycling for coexistence is proposed in [2], [3], [7], [8], [10],[11], [17] whereas, [12], [13], [14], [15] explore other possi-ble coexistence schemes, such as LBT. A comparison of theaforementioned mechanisms is shown in Table 1. Addi-tionally, [18], [19], [20] use stochastic geometry in interfer-ence and coverage area modeling.

The mutual impact on the performance of LTE and Wi-FIhas been studied in [3], [11] taking the average throughputper user as a quality metric. Though [11] proposed a mecha-nism that is backward compatible with no mutual signaling,Wi-Fi must perform traffic sniffing to predict the ABSpattern. Moreover, the assumption that ABS periods arecompletely free of all interference limits its application. Twodifferent solutions of Wi-Fi coexistence using ABS and inter-ference avoidance are proposed in [7]. However, the coordi-nated interference avoidance here relies mostly on cellclustering, and assignment of priority among cells, whichrequires major modification of current standards. [8] pro-posed a relatively fair resource allocation method thatformulates a convex optimization problem of minimizingLTE-U/Wi-Fi collisions. [21] provides a comprehensive sur-vey of related works. It also presents theoretical models ofthroughput and overhead for coexisting methods.

A framework that models the channel access of bothtechnologies with LTE adopting the LBT approach is givenin [12]. A fixed contention window is set in [13] that limitsthe performance of LTE in terms of user throughput whencollisions occur with Wi-Fi devices. A dual band approachis proposed in [14], where LTE sends its control signalsthrough the licensed band, and offloads data traffic onto theunlicensed band. However, this reduces the spectral effi-ciency of the system.

Recently, the use of stochastic geometry for characterizingthe interference, andmodeling the coverage area and through-put in WLAN and LTE systems has gained traction [20]. In[18], a simplistic fluid networkmodel is used to study the idealcoexistence scenario when no multipath and backoff is pres-ent. Device-To-Device (D2D) communications for LTE-Wi-Ficoexistence have been proposed in [22], [23] to increase theLTE throughput by offloading some of the messages to Wi-FiDirect. A spatio-temporal estimation of interference and a loadbalancing mechanisms are proposed in [24] and [25]. Ourapproach to useWi-Fi Direct for relaying purposes is validatedin studies like [26], improvements in throughput and energyconsumption fromD2D are shown ([27] and [28]).

3 WI-FI GROUPS AND RELAYING

In this Section, we show the impact of the pilotswithin ABS onthe Wi-Fi receiver through: (i) physical layer BER and framedetection studies when the AP and the LTE BS transmit stand-ards-complaint waveforms concurrently, and (ii) on the reduc-tion in channel access opportunities for Wi-Fi at the MAClayer.AlthoughABSdonot contain anydata or control signals,they may still have multiple embedded pilots (Fig. 1): (i) Cell-Specific Reference Signals (CRS), used for power estimation;(ii) Physical Broadcast Channel (PBSCH), used to announcethe bandwidth/frequency used by the BS; (iii) Primary Syn-chronization Channel (PSCH), needed for subframe-level syn-chronization; and (iv) Secondary Synchronization Channel(SSCH), needed for frame-level synchronization.

3.1 Impact of ABS on 802.11 PHY - Simulation StudyIn order to measure the impact on PSR, we create standards-compliant LTE and 802.11n waveforms using MATLABLTE- and WLAN Systems toolboxes. Here, the LTE BS isdeployed as an indoor small cell occupying 20 MHz channelin the 5 GHz band served by a 802.11n AP. The LTE BSoperates in FDD mode with its own link adaptation usingthe Channel Quality Indicator (CQI) reported by the UEwith the default periodicity of 8ms. We select the industry-standard TGn model B with 100ns Delay Spread for theindoor propagation model [29]. The transmission power isset to 17 dBm, following the ITU recommendations in [29]with the noise floor set to �95 dBm. The LTE BS is separatedby 60m from the Wi-Fi AP, which is also the latter’s cover-age radius (Fig. 2). A transmission is considered successfulwhen the parity check of the signal field returns true andthe overall bit error rate (BER) per packet is exactly 0 per-cent. Hence, the PSR is the Packet Error Rate (PER) flipped.

Simulation results in Fig. 3 show the observed Wi-Fi PSRduring ABS 1, plotted as a function of the SINR (that includesLTE interference) and its own received power from the AP.The intersecting horizontal plane indicates the combination ofthese two measurements that is assured to provide at least 90percent PSR (PSRTh). Fig. 2 shows two spatial regions of 90percent PSR centered around the AP depending uponwhether the ABS 0 (inner dotted boundary) or ABS 1 (outerdashed boundary) is used. We call these as safe zones. AWi-Fidevicemj is within the safe zone if its PSR is greater than a pre-decided threshold (PSRj � PSRTh). In the default implemen-tation of E-Fi, we define the PSRTh to be 90 percent. Thus, anyWi-Fi node can determine whether it lies within the safe zone

TABLE 1Coexistence Scheme Comparison

Category Coexist scheme LTE Modification WiFi Modification Coordination Overhead CRS

DutyCycle

E-Fi No Require PSR

No

No YesE. Almeida [3] Blank subframe Dual coex. modes NoH. Zhang [7] Spectrum Sensing Neighbor Discov.C. Cano [8] BS monitors channel

NoNo

N/ACSAT [9] Adaptive DCN. Rupasinghe [10] Q-learning adaptive DC

Z. Guan [11] No LTE packet sniffing Yes LTE

LBT

T. Tao [12] Adaptive CW

No

Yes

No N/AS. Hajmohammad [13] Fixed CW

NoF. Liu [14] Dual bandsY. Li [15] Adaptive CW Adaptive CCA th.



by measuring the tuple of SINR and its own received power,even before packet transmissions begin. As we will cover inmore detail in further sections, this information is used by E-Fi as a decision criterion to define the three connectivitymodes for the Wi-Fi nodes (relay, Wi-Fi Direct node that con-nects to the AP via the relay, or a regular node that directlyconnects to the AP).

3.2 Impact of ABS on 802.11MAC - Experimental Study

The ABS pilots not only affect the Wi-Fi transmission in thedownlink, which we quantitatively analyze through thePSR, they also impact the uplink transmissions by reducingchannel access opportunities. To characterize this impact,we consider the scenario of a Wi-Fi device continuouslyattempting to transmit under a continuous presence of ABStype of sub-frames (Fig. 1). That is, the LTE received poweralways exceeds the carrier sensing threshold, leading theWi-Fi device to backoff.

We modified the srsLTE [30] implementation of theDownlink frames on a USRP B210 series as a way to saturatethe channel with LTE frames under different ABS configu-rations. On the Wi-Fi side, a regular laptop equipped withan Atheros NIC emulates the behavior of an AP attemptingto access the channel and operating in saturation mode. Weuse iperf [31] to generate Downlink traffic from the AP, aim-ing to fill the MAC queues with outgoing traffic and forcingthe driver to always look for transmission opportunities.

The Linux 802.11 configuration API (cfg80211) with theAtheros card and ath9k driver [32] helps in measuring thetime the radio is active, and the amount of time the primarychannel was sensed busy. There are 4 main functions thatare of interest:

� Ath_get_survey: This function is called by a userspace process and collects the statistics fromath_hw_cycle_counters_update.

� Ath_hw_cycle_counters_update: Collects the sta-tistics through some hardware registers (AR_CCCNT,AR_RCCNT, AR_RFCNT, AR_TFCNT) that act as aninterface between the MAC state machine running onthe System onChip (SoC) and the kernel driver.

� ath_update_survey_stats: Converts cycles intoseconds according to the clock-rate configured in thesystem.

� Ath_tx_complete: Calledwhenever a packet is sentout successfully and updates the queue in the Kernel.

The Ath_get_survey allows us to extract measure-ments on the time the channel was sensed busy directlyfrom the driver. The average time to transmit is difficult toobtain, since the drivers do not have access to the stages ofthe MAC state machine. Thus, there is no reliable way todetermine when the backoff counter goes to zero and thepacket is transmitted. As a workaround, we employed adebugging mechanism within Linux Kernel called Kprobes[33]. Whenever a message is transmitted to the channel, aflag is set that is detected by ath9k driver. The driver thencalls the ath_tx_complete function on the Linux Kernel,clears packets from the queue whenever they are transmit-ted into the air. Kprobes allows us to track the calls to thisfunction with ms precision [33], thus determining when apacket is transmitted and ultimately allowing for a highlyaccurate computation of the inter-frame departure time.

The results are shown in Fig. 4. The busy rate numbersprove the intuition that the number of control signals has anon-negligible impact on the availability of the channel.Same conclusion can be applied to the time to transmit,where the configuration ABS 0 certainly detriments WiFi’sperformance. In addition, a high number of devices contend-ing for the channel require greater need of ABS resources.This last observation serves further justifies the groupingprocedure in E-Fi, where we reduce the number of contend-ing devices to ensure faster channel access.

Algorithm 1. E-Fi: Group Formation & Resource Dist

1: LTE presence awareness and ABS detection (Section 4.1).2: PSR Calculation based on PRX and SINR (Section 3.1).3: Pre-categorization into GOc andWCc (Section 3.1).4: GOc discoverWCc and exchange PSR (Section 4.2).5: GOc follows Eq. (3) to generate candidate set (Section 4.3).6: AP gathers GOc data to form final groups (Section 4.4).7: AP gathers traffic info and allocates resources (Section 5.2).

3.3 Relays for Resilient Transmission duringABS Subframes

In this section, we motivate the key strategy of elevatingselected nodes to a relay position to counter the impact ofthe ABS pilots. These relay nodes forward traffic to andfrom the AP, connecting remote nodes affected by low PSR.E-Fi requires such relays to become Wi-Fi Direct group

Fig. 4. The CRS embedded within the ABS increase the channel occu-pancy as shown in the ECDF busy rate (percent). As a result, the averagetime to access the channel and transmit aWiFi packet increases, revealingthat the number of CRS impact negatively on the channel accessibility.

Fig. 3. PSR at the Wi-Fi receiver as a function of the received power andthe SINR during concurrent LTE ABS 1 (subframe 1 in LTE frame asABS) and Wi-Fi transmissions using modulation and coding scheme(MCS) 0 (BPSK and coding rate 1/2).



owners, and by reducing the link distance in the presence ofLTE pilots it improves the collective PSR of the network.

Consider a Wi-Fi AP connected to V nodes in its coveragearea and represented by the set VWi�Fi ¼ ðv1; v2; . . . ; vV Þ. AnLTE BS serves U user equipments (UEs). The LTE BS canschedule a number of ABS independently of the Wi-Fi, andany such ABS pattern is valid for 40 ms. Consider a subsetVSZ of M nodes that happen to be inside the safe zone(PSRj � PSRTh, 8j 2 f1;Mg) and a subset (VNSZ) ofN nodesthat happen to be outside it (PSRi < PSRTh, 8i 2 f1;Ng).Therefore, V ¼ M þN . Any device (mj, 8j 2 f1;Mg) withinthe safe zone, defined on the basis of PSR, is a potential relaycandidate. Those nodes that are outside this range are non-safe zone nodes that attempt to associate with a distinct relaynode. All data communication between the relays and suchnon-safe zone nodes occurs via Wi-Fi Direct within a givenABS. The traffic exchange between the relay and the APoccurs via regular 802.11 in a different ABS.

As we describe in Section 4, E-Fi distributes the availableABS for the two sets of nodes: Set I containing Wi-Fi Directgroups composed of both relays and non-safe zone nodes.Set II containing (i) non-safe zone nodes who are unableto connect to intermediate relays, and (ii) safe zone nodeswho do not serve as relays. E-Fi further introduces differen-tial backoff duration to ensure that the remote non-safezone nodes in Set II (that suffer from lower PSR) getincreased transmission opportunities to recover from likelyhigher errors in Section 5.2. From Fig. 5, Set I =fm1;m2; n2; n3; n4gwhile Set II = fn1; c1; c2g.

4 RELAY SELECTION AND DEVICE GROUPING

We formulate the Wi-Fi Direct group formation as a General-izedAssignment Problem (GAP), whose aim is tomaximize thenumber of non-safe zone nodes connected to relays under theobjective function of minimizing the average number of trans-mission in thedownlink.Hence, this is theminimizationversionof GAP orMINGAP [34]).We choose this approach for two rea-sons- (i) Wi-Fi Direct standard allows a maximum number of 8connections per group [5] and (ii) high PSR is desired per nodein the downlink given the asymmetric flow of traffic. Hence,each group is owned by a relay (mj; j 2 f1;Mg) that serves aset of associated non-safe zone nodes (Cj). The cardinality ofCj

is denoted by jCjj. All nodes who are not in any relay-ownedgroup are consolidated intoSetII and representedby the vari-ableCC that directly connect to theAP (see Fig. 5).

The formal description of the problem is given in Eq. (1),where the objective function is to minimize the number ofoverall expected transmissions, subject to a maximum num-ber of Wi-Fi device connections K per group and improvedPSR for every node in the network compared to direct con-nection with the AP (Eq. (2)).

min Ntx ¼ minXj�1

1þ jCjjPSRAP

mj

þXi2Cj

1

PSRmjni

0@

1A

þXi2CC

1

PSRAPni

(1)

subject to:

PSRvi� � PSRvi ; 8vi 2 VWi�Fi

jCjj � K ; 8j 2 f1;Mg: (2)

Here, PSRAPni

, PSRmjni and PSRAP

mjrepresent the estimated

PSR for the direct transmissions by the AP and received at thenon-safe zone node, the PSR for transmissions by the relayand received at the node and direct transmissions by the APand received at the relay, respectively. Other terms aredefined earlier in Section 3.3. To ensure that the group forma-tion has bounded complexity, this organization into groups isundertaken centrally at the AP using a modified version ofthe Hungarian Algorithm [35]. Section 4.4 explains in detailhow the constraints of the MINGAP problem are relaxed toreduce it to the Linear Assignment problem (LAP). The pro-posed algorithm solves the LAP in polynomial time by usingPSR collected at the individual nodes and considering all pos-sible non-safe zone node to relay associations from the devicediscovery phase fromWi-Fi Direct.

4.1 Interference Awareness and ABSPattern Detection

A sudden drop in the performance of the Wi-Fi networkcaused by in-band LTE triggers the initialization of the E-Fiprocedure. Consequently, the AP notifies the devices andforces them to defer their transmission and detect the LTEABSPattern configured at the BS. Existing methods such as theones proposed in [37], [38] and [39] use an RSSI sampler, avail-able at everyWi-Fi device, to detect and characterize the inter-ference. As for determining the start of the frame, the Wi-Fidevices may employ pattern recognition techniques such assymbol folding, which detects periodic signals (i.e., the BCCHinABS0 and the PSCH/SSCH in theABS0/5) in noisy environ-ments [40]. Finally, the devices report themeasurements to theAP,who announces the start of the discovery phase.

To validate whether these mechanisms can detect ABSover time, we set-up a basic Pattern detector based on the oneused in [36] (Fig. 6). The detector has two stages. First, a pri-mary stage composed by a diode and a capacitor allows forenvelop detection. Second, a combination of resistors andcapacitors allows us to compute the threshold by measuringthe fluctuations in the signal. For simplicity, the system wasconfigured to provide a threshold equal to 0:5 � ðVmin þ VmaxÞ,where Vmin and Vmax representing the minimum and maxi-mum voltage, respectively. The LTE frame generation follows

Fig. 5. The safe zone is enclosed by the dashed lines. The nodes aregrouped into: Cc ¼ fc1; c2; n1g; C1 ¼ fm1; n2; n3g and C2 ¼ fm2; n4g.Group C1 and C2 are in Set I while Cc a collection of Set II devices:non-safe zone nodes without relays (n1) and safe zone nodes who arenot relays (c1 and c2). At the lower end, sample ABS resource allocationfor each group is shown with separate uplink/downlink duration.



the procedure described in detail in Section 3.2, where aUSRPB210 sends LTE signals over the air according to the ABS Pat-tern defined and configured by the srsLTE software.

Fig. 7 shows the behavior of theABSdetector. TheABSPat-tern configured for this experiments is [0101001000], wheresubframes 1, 3 and 6 are configured asABSwhile the rest carrydummydata. The top figure shows the LTE signal transmittedover the air. In spite of the presence of control signals duringABS, the power level measured at the detector during those isway below the threshold. Thus, allowing the system to clearlydifferentiate between ABS and non-ABS subframes by com-paring itwith a predefined and static threshold.

4.2 Device Discovery and PSR ExchangeOnce devices are notified, all nodes measure their expectedindividual PSR (using SINR and received power from AP,see Fig. 2) that allows them to self-determine whether theylie in the safe zone. Any node in this zone is a potential relayand assumes the role of a Wi-Fi Direct group leader. It thenbegins the device discovery process by issuing discoverybeacons and logs all non-safe zone nodes that initiate con-nection requests. Similarly, a given node i that identifiesitself to be in the non-safe zone, will send reply beacons con-taining its ID, the estimated PSR for direct transmissions bythe AP, i.e., PSRAP

ni, and the estimated PSR for the short-

range link between itself and the relay candidate PSRmjni . All

potential relays also compute their estimated PSR for theAP’s transmissions, i.e., PSRAP

mj, as this is the metric used to

identify which nodes are in safe/non-safe zones.

4.3 Forming Initial Relay GroupsOn receiving a set of replies from non-safe zone nodes, thecandidate relaymj determines which neighbor node iwouldexperience improved PSR through a one-hop Wi-Fi Direct-based relaying versus direct AP communication. First, usingthe measurement of the received power from the non-safezone nodes, it calculates the new SINR and computes thePSR of the link between itself and the non-safe zone nodes(PSR

mjni ). Second, it checks if the expected number of trans-

missions through the one-hop communication (w�ij) is lesser

than the direct one to the AP (wi) for that node, as

1

PSRAPmj

þ 1

PSRmjni

zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{wij�

<1

PSRAPni

zfflfflfflffl}|fflfflfflffl{wi

: (3)

The subset Fj contains non-safe zone nodes i (i 2 f1; Ng)that meet this condition for relay mj j 2 f1;Mg, and hence,

benefit from association with it. Each candidate relay node jcreates a vector of PSR estimates yyj that includes its ownPSRAP

mjas well as the effective PSR wij

� estimated for thenon-safe zone nodes i 2 Fj that associate with it as follows

yyj ¼ ½PSRAPmj

waj�wbj

� . . .wgj�

|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}xxj

� 8a;b; g 2 Fj: (4)

This vector is sent to theAP by all the candidate relays. Thefinal group formation is completed at theAPusing amodifiedHungarian Algorithm (Section 4.4). The individual groupmembership is then relayed back to the network by the AP.

4.4 Forming Final Groups at the APConsider the set of vectors representing PSR measurementsreported by all the candidate relay nodes to the AP, i.e., (yyj,8j 2 f1;Mg). It is possible that the same non-safe zone nodesoccur in multiple tentative groups formed by the candidaterelays. Wi-Fi Direct requires each node to be linked to onlyone group owner. Hence, the goal of this stage is to (i) finalizethe groups such that the nodes connect to one relay only, and(ii) distribute nodes uniformly throughout the groups withinthe network to maximize the overall PSR. For this purpose,we use a modifiedHungarian Algorithm that matches nodesto relays using the PSR vector described above.

4.4.1 Algorithm Description

The AP builds an N ðM þ 1Þ matrix WW containing themeasurements vectors xxj, 8j 2 f1;Mg (Eq. (4)), forwarded byN candidate relays (Eq. (5)) as well as the initial PSR values(wi, 8i 2 VWi�Fi). Given that the cardinal of the set Fj mayvary for differentmj, the AP assigns 0 for the situations where(i) there exists no connection between the relay and the non-safe zone node, or (ii) the one-hop forwarding is not beneficialfor that node (i.e., wij

� ¼ 0 8i =2 Fj). Along the same lines, amatrix PP is defined as PP ¼ limn!1½11� ð11�WWÞn]. ThematrixPP contains 1’s if the relay communication betweenmj and i isbeneficial (wij

� > 0) and 0’s otherwise (wij� ¼ 0).

m1 ::: mM AP

WW ¼n1

..

.

nN

w11� ::: w1M

� w1

..

. . .. ..

. ...

wN1� ::: wNM

� wN

0BB@

1CCA:

(5)

Fig. 6. ABS detection enabled by an envelope detector, which smoothensout the fluctuations, and a comparator, which detects whether the powerlevel exceeds a pre-defined threshold [36]. The LTE interference is gen-erated using a USRP B210 empowered with srsLTE (for further details,see Section 3.2).

Fig. 7. At the top, the received LTE signal at the entry of the ABS PatternDetector. The ABS Pattern configured is [0101001000], meaning thatthe frame embodies 3 ABS (subframes 1, 3, and 6). On the bottom, thesignal power detected evolution and the decision whether or not a sub-frame is ABS or non-ABS based on the threshold.



We define the vectors zz ¼ PP � 11 and tt ¼ 11t � PP . The formerone shows the number of favorable connections for a givennode, whereas the latter one shows the number of favorableconnections that each relay mj can offer. In other words, ttshows the cardinal of F (tj ¼ jFjj 8j 2 f1;Mg).

We model the network as a Bipartite Graph G ¼ ðC;S;WÞ, where the set C contains contains the non-safe zonenodes (VNSZ) and the set S contains the candidate relays(VSZ) and AP. Recall that the weight of the edge from nodeni (i 2 C) to relay mj (j 2 S) is wij

� and that to the AP is wi.The group formation problem is solved using the Hungar-ian Algorithm [35], which finds the optimal matching toreturn the maximum PSR for the entire network. Thus,every node in set C is linked to one node in set S, and afterthe match, it is removed from the set C. Given that relayscan forward traffic to/from more that one node, a modifiedBipartite Graph G0 is formed by adding dummy relays andAPs so that a perfect match can be found. The algorithm ter-minates when C is an empty set or when no further matchesare possible in a given iteration. Table 3 summarizes themain variables in the E-Fi system.

Note that all non-safe zone nodes that could not asso-ciate with a relay are automatically included in CC , i.e.,the set of all nodes who are not in any relay-ownedgroup. All candidate relay nodes that were not matchedwith at least one non-safe zone node are also included inCC . The multiple groups formed through the matchingalgorithm compose Set I. We show next how the APdistributes the ABS for both these category of nodesbased on network loads.

4.4.2 Algorithm Complexity

The Hungarian Algorithm has polynomial complexity givenby Oðn3Þ. Although a simplistic solution to form G0 wouldinvolve adding ðK � 1Þ � jSj dummy relays and ðK � 1Þ � jSjdummy APs to the set S to force the matching, this drasti-cally raises the complexity to Oðð2 �K � jSjÞ3Þ. Instead, E-Fiintelligently adds dummy nodes when needed. Table 2shows the conditions under which it is necessary to adddummy relays or AP nodes in G0 based on the 3-tuple(M;K;N) and the candidate sets Fwith the purpose of min-imizing the complexity. Vector tt shows the number of nodesthat need to be inserted in the modified graph for each nodein S. The maximum cardinality of set S is 11t � PP � 11. A simpli-fied scenario is shown in Fig. 8 where groups are formedusing the modified Bipartite Graph G0. Table 2 shows theneed to add two dummy relays (m1

0 and m20) and dummy

APs (m00 and m0

00). The number of dummy nodes is givenby tt (t1 � 1 form1, t2 � 1 form2 and tMþ1 � 1 form0).

5 ABS RESOURCE DISTRIBUTION FOR GROUPS

E-Fi adopts the strategy of fair resource sharing, wherein theAP assigns ABS to individual groups. Further, this resourceallocation is done per group and also split between downlinkand uplink. The short window of transmission opportunitywithin theABS framesworks only under conditions of limitedcontention between nodes. We define the downlink durationfor transfer of data traffic from the AP to relays or to theirassociated nodes (i.e., AP ! mj or AP ! mj ! ni) and alsocorresponding ACKs that traverse the links in the reversedirection. On the other hand, the uplink duration is for datapackets originating from the relays or associated nodes, withthe AP as the destination. The ACKs in this case also arrive inthe opposite direction within the same window of transmis-sion. It is possible that a relay may not have sufficient time toforward data packets that it receives within the same ABSwindow. In such cases, the relay packet queue grows andtransmission resumes the next time the group is assigned theABS. The contention mechanism within a group is explainedin detail later in Section 5.2.

E-Fi defines a load factor for every group j, denoted byhj, as a weighted expression of wij

� (Eq. (3)) of (i) number ofnodes in the group |Cj| and the application load �i and mi,representing the throughput desired in the downlink(Eq. (6)) and uplink (Eq. (7)), respectively. For Set II,which contains the individual nodes (CC), we set wj ¼ 1 inthe above equations (we assume the AP itself is sufficientlyspaced from the LTE BS and not affected by the LTE inter-ference) and �mj

¼ 0;mmj¼ 0 (no relay is present).

hjDL ¼ �mj

� wj þX

i s:t:ni2Cj

�ni � wij� (6)

hjUL ¼ mmj

þX

i s:t:ni2Cj

mni� wji

�: (7)

5.1 Inter-ABS Resource AllocationEach LTE frame has a number of included ABS subframes.Though the Wi-Fi AP cannot influence this number via feed-back to the LTE BS, in E-Fi, it can recognize the ABS patternand knows when such ABS are scheduled. Let the corre-sponding vector that indicates the presence of the ABS loca-tions within the LTE frame be given byAA, with the number of

TABLE 2Conditions under which Dummy Nodes Need to be

Added to the Bipartite Graph G0

M �N M �K ¼ N M �K > N

mj0 AP mj

0 AP mj0 AP

8i; j; i 6¼ j; s:t:@ @ @

Fi

TFj ¼ ;

9i; j; i 6¼ j; s:t:@ @ @ @

Fi

TFj 6¼ ;

K = maximum number of Wi-Fi Direct nodes, M relays, and N non-safe zonenodes.

Fig. 8. Modified Bipartite Graph G0 given three non-safe zone nodes(N ¼ 3), two candidate relays (M ¼ 2) whose number of Wi-Fi Directconnections are limited by the standard (K ¼ 7). The final matching isshown in bold, where all the clients (top in G0) are connected to exactlyone Relay/AP (bottom in G0).



such ABS represented by jAAj. For instance, AA ¼ ½1100001100�,where 4 subframes are designated ABS in the LTE frame. TheAP assigns resources to the groups proportional to the loadfactor in theDownlink (Eq. (9)) andUplink (Eq. (8)).

jAAjjUL ¼ hjUL

Pj�1 hj

UL þ hCUL

� jAAjUL (8)

jAAjjDL ¼ hjDL

Pj�1 hj

DL þ hCDL

� jAAjDL: (9)

5.2 Intra-ABS Resource AllocationIn this Section, we explain how E-Fi handles collisions andmedium contentions within the ABS frame. Such a framecan be allotted for either uplink or downlink, and we sepa-rately consider both these situations. The key idea here isthat each device uses a slightly shifted carrier sensing starttime while accessing the channel depending upon the num-ber of packets in its MAC layer queue and the reliability ofthe links (PSR). This time shift results in preferential accessto the channel for certain stressed nodes who experiencegrowing queues (such as relays) and distant nodes with lowPSR (such as non-safe zone nodes without relays).

5.2.1 Downlink

Both the AP and the relay of a group contend for the channel.If the former wins the contention, then the destination is therelay. If the relay wins, then it begins to forward the queuedpackets to its associated (and downstream) Wi-Fi Directgroup members. Through a control parameter a, E-Fi ensuresthat (i) the AP has enough opportunities to successfully trans-mit the packets to the relays (link l1) according to the applica-tion load demanded by the Wi-Fi Direct nodes (Eq. (10)), and(ii) relays have priority to forward the packets from the AP tothe respective Wi-Fi Direct nodes (link l2) as soon as theyreceive them (Eq. (11)). These conditions provide an upperand lower bound for a (Eq. (12)) as follows:

�mjþ

Xi2Cj

�ni

zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{�j

� Tj � aTtx � wj

(10)

Tj � aTtx � wj

� Tj � ð1� aÞTtx � wij

� (11)

�j � Ttx � wj

Tj� a � wj

wj þ wij� ; (12)

where, Tj ¼ jAAjjDL � 1 ms=40 ms, and 40 ms is the durationfor which a given ABS pattern is active. wij

� is the averagePSR in a cluster. Moreover, Ttx is the expected time for asuccessful packet transmission with no collisions. Thisparameter depends on the exponential backoff time, andPKT, which is the time to transmit a packet. The probabilitythat the AP or the relay gets to transmit is given by a and(1� a), respectively, (Note that a ¼ 1 for Set II) andallows for modifying both the duration of DIFS and the con-tention window as following Eq. (13). Note that Eq. (12)returns a feasible range if a PSRth is selected following thesteps in Section 4 and the requirement in Eq. (3) is met.

DIFSmj�>ni ¼ ð1� aÞ �DIFS

CWmj¼ ð1� aÞ � CW

DIFSAP�>mj¼ a �DIFS

CWAP ¼ a � CW:

(13)

5.2.2 Uplink

The uplink consists of two different situations that arise fornodes in Set I and CC . Consider nodes in Set I, where arelay j and its associated devices jCjj contend for the chan-nel. E-Fi gives priority to the relay to forward the packetsfrom the Wi-Fi Direct nodes to the AP (Eq. (11)). Thus, theequation 10 is valid with wj ¼ 1 (as the PSR from the relaysto the AP is assumed to be 100 percent) and wji

� remainsthe reverse-path PSR in the UL.

For the nodes that belong inCC and form the Set II, theindividual devices outside of the safe zone also contendwithothers within the safe zone. From Fig. 1, we see that the firstcolumn of the ABS always carries pilots that may also causethe Wi-Fi clients closer to the LTE BS to persistently backoff,while the ones within the safe zone (and hence farther fromthe BS) may discover the channel to be free. To address thisinequality at both PHY and MAC layers, E-Fi defines a timeshifted window for the safe-zone nodes. All nodes that are inthe safe zone andmember of Set IImust wait for 1 resourceunit time from the start of the ABS before starting the DIFS.There is no such wait period imposed on the non-safe zonenodes. The intuition here is to give the nodes with low PSRadditional opportunities to transmit within the ABS andbring about some measure of fairness in the link throughputfor each node. We evaluate this design decision using Jain’sfairness index in Section 6.

6 PERFORMANCE EVALUATION

0We evaluate E-Fi using an integrated MATLAB and NS-2simulation environment. MATLAB is used to model the sig-nal waveforms at the PHY layer that are 100 percent stand-ards-compliant usingWLANand LTE System toolboxes. Thisallows studying interference caused by LTE on a per-resourceunit basis for various separation distances of Wi-Fi nodes.The spatiotemporal interference map is then imported intotheNS-2 simulator, wherewe simulate theWi-Fi Direct groupformation and E-Fi’s enhanced channel accessmechanism.

We first characterize the optimum range of PSR thresholdto perceive the maximum improvement from E-Fi for several

TABLE 3Main Variables and Their Definitions

Variable Definition

M Total number of Relay CandidatesN Total number of Nodes outside the S.Z.mj Relay or Relay Candidate j, 8j 2 f1;Mgni P2P or P2P Candidate j, 8i 2 f1; NgFj P2P Candidate set ofmj

Cj P2P Nodes attached tomj, jCjj � 7

a Modified Backoff for DL and ULwi , wj PSR of Client i and Relay jwij

� , wji� PSR of ni throughmj in the DL and UL

�mj, �ni Application Load ofmj and ni

hjDL , hj

UL Load Factor of group j in the DL and UL

jAAjjDL , jAAjjUL Number of ABS assigned to group j Tx



AP-BS distances. Further, we evaluate the improvement intro-duced by E-Fi as a function of the distance between the AP-BSaiming to avoid topology restrictions. Knowing that E-Fi per-formance is tightly related with the number of potentialrelays, we evaluate E-Fi’s performance for several networkdensities and selected a commercial range. Finally, we per-form a broad study on the throughput and Packet DeliveryRatio (PDR) for different network configurations as well as acomprehensive overview of the combined impact of the LTEcontrol signals and channel contention on the Clear ChannelAssessments (CCA) embedded inWi-Fi’s DCF.

The simulations performed in this section consider17 dBm to be the transmit power for both Wi-Fi-AP and BS-LTE, which is the maximum transmit power output perantenna in the 802.11n standards for indoor communica-tions in this band. Furthermore, the initial coverage area isselected as for to provide 90 percent PSR when no LTE inter-ference is present. Commercial deployments require PERlevels within the range of 10-30 percent to provide reliableand uninterrupted communication without affecting theuser experience. The initial Wi-Fi transmit rate was config-ured to be MCS-3 (16-QAM with coding rate 1/2) withenabled automatic rate fallback (ARF).

For space reasons, we report only the results referred toas downlink traffic scenario, i.e., the AP generates equalConstant Bit Rate (CBR) traffic for all Wi-Fi nodes given by� pkts=s. Let Tshare be the temporal share of the channelusage between the LTE and the Wi-Fi network, dependingon how many ABS frames are included (it is impossible forWi-Fi to operate in LTE data frames). Since E-Fi works inuncooperative LTE deployments, the value of Tshare istunable.

6.1 PSR Threshold Selection in E-FiThe Safe Zone, directly defined by the PSR threshold, deter-mines the number of nodes that could become Group Owners.As the PSR threshold decreases, more devices could meet thecriteria and be elected as a Group Owner by the AP. Fig. 9shows how the PSR threshold determines the relay roles, anddefines the regions outside the Safe Zonewhere nodes aremostlikely to be assigned a relay to increase their PSR. As the PSRthreshold decreases, the Safe Zone area widens, more devicesare categorized as GroupOwners and, in turn, more nodes out-side the Safe Zone can relayed. Also, Fig. 9 shows that the devi-ces elected asW-Fi direct clients are the ones closest to the LTE-BS, and thus suffering from severe interference. However, thereis a small set of nodes, the closest one to the BS, for which E-Ficant find an improvement given the high levels of interference.

Fig. 10 shows the optimum value for the PSR thresholdso that E-Fi provides the maximum PSR. The optimumrange is defined as the one that provides the highest averagePSR for the considered AP-BS distance. The modified Hun-garian Algorithm finds the maximum average PSR whenthe threshold is selected within the range 40-70 percent,where the AP has enough Group Owners, and nodes withlow PSR that lie outside the Safe Zone become Wi-Fi DirectClients. The node distribution is shown in Fig. 16.

6.2 Impact of the Distance between the AP and BSSince we consider the transmit power to be fixed to the max-imum, we evaluate E-Fis performance for different AP-BS

Fig. 9. Node categorization and distribution in the coverage area whendifferent PSRth are selected. As the PSRth decreases, more devices canbe elected as relays and hence, serve more devices in the outside the S.Z. The Wi-Fi AP is located at (0,0) and the LTE BS at (60,0).

Fig. 10. Average PSR of the Wi-Fi devices when several PSR thresholdsare selected defining the Safe Zone. The optimum value lies between 70and 40 percent.

Fig. 11. Average PSR of Wi-Fi devices for different AP-BS distances.The Wi-Fi devices are deployed within the initial Coverage Area, definedto be 90 percent PSR with no LTE interference. The PSR threshold ischosen to be 70 percent. E-Fi proves to provide the highest improvementwhen the BS is located 20-80 meters far from the AP.

Fig. 12. Average PSR of Wi-Fi devices when different number of Wi-Fidevices are deployed in the network. The PSR threshold is chosen to be70 percent. E-Fi always improves the PSR and achieves a maximum of10 percent improvement.



distances. Fig. 11 depicts the range where the maximumimprovement is reached. The node distribution is shown inFig. 16. As expected, when no LTE interference is present,the average PSR converges to the initial PSR thresholddefining the coverage area.

6.3 Impact of the Number of Wi-Fi DevicesE-Fi relies on a certain number of nodes perceiving highPSR so that by enabling relaying capabilities, the PSR ofother nodes can be increased. Fig. 12 shows the PSR changeas the number of nodes increases. The node distribution isshown in Fig. 16. Note that the increasing trend is becausethe PSR metric does not consider the impact of the channelavailability or collisions, but it just accounts for the reliabil-ity of the link against LTE interference.

6.4 PHY throughput ImprovementIntroduced By E-Fi

Next, we evaluate the improvement on the average PHYthroughput that Wi-Fi Direct clients perceive by employingrelaying capabilities. Fig. 13 shows the throughput distribu-tion for a network where 20 Wi-Fi devices operating withinthe coverage area being interfered by an LTE-BS located 60mfar from theAP. The PSR threshold is chosen to be 70 percent.

We note few key findings: First, E-Fi introduces 50-70 percentimprovement on the throughput of the nodes affected by theLTE interference. Second, the Group Owners tend to havelesser PSR than the Safe Zone Clients, meaning they arelocated closer to the Safe Zone boundary. Third, E-Fi helpsthe devices that aremost affected by the LTE interference.

6.5 MAC-Based Analysis of E-Fi - Impacton the CCA

While other proposals that use duty cycling, such as LTE-U,require a wider time window for the Wi-Fi devices to trans-mit, E-Fi shortens the available transmission time by cluster-ing the devices into groups and allocating fewABS to each ofthem. Fig. 14 shows a statistical distribution of nodes per E-Fi group, proving that groups contain between one and threenodes roughly 95 percent of the times. Fig. 15 shows the aver-age time to successfully transmit a packet for different groupsizes, accounting for the impact of the LTE control signals onthe CCA carried out in the DCF. In short, the groupingmech-anism allows for the reduction of the time given to Wi-Fi totransmit down to a fewmilliseconds (ABS subframes).

Fig. 13. CDFof the Throughput of the Wi-Fi devices when 20 devices areoperating in the network, the PSR threshold is chosen to be 70 percentand the BS is located 60m far from the AP. E-Fi improves the PHYthroughput of the Wi-Fi Direct Clients 50-70 percent. Section 6 describesthe communication set-up, i.e., deployment scenario, transmit power andPSRth.

Fig. 14. Dimensionality of the E-Fi groups for ABS 1. Roughly 95 percentof the groups end up having between 1 and 3 devices.

Fig. 15. Transmission time accounting for the LTE control signals and thecontention between the Wi-Fi nodes.

Fig. 16. Analysis of the Statistical Role distribution of Wi-Fi nodes whensubframe 1 is designated ABS when we vary the PSR threshold definingthe Safe Zone, the distance between the AP and the BS in meters, andthe number of Wi-Fi nodes operating in the network. Default values arePSRth ¼ 0:6, AP-BS distance ¼ 50 (m) and 20 Wi-Fi nodes.



6.6 Benefits of Traffic Forwarding via RelaysWe next evaluate how relays improve the application layerPacket Delivery Ratio (PDR) by comparing five differentschemes with modular enhancements:

� LTE OFF, Legacy Wi-Fi. No interference is present inthis configuration (Blank Subframes). Used as a base-line comparison.

� LTE ON, Legacy Wi-Fi. LTE interference is character-ized by the Matlab results in Section 3.1. Realisticand current scenario.

� LTE ON, Random WiFi relay selection. The systemgroups the nodes based on a random pattern.

� LTE ON, Random WiFi relay selection over the RelayCandidate (RC) set. The pairing follows the Eq. (3),where the candidates must lay within the Safe Zone.

� LTE ON, Hungarian-based WiFi relay selection. The sys-tem groups the nodes according to the modifiedHungarian algorithm (Section 4.4).

The parameters under study for this section are: Nnodes,representing the number of Wi-Fi nodes; Tshare, denotingthe time ratio assigned to Wi-Fi (i.e., the LTE and WI-FI net-works have equal share when Tshare ¼ 50%); and �, repre-senting the Wi-Fi traffic rate in packets per second. Fig. 17ashows the impact of the Nnodes, from which we can drawsome conclusions. First, the LTE OFF case overestimatesdata delivery without capturing packet losses. Second, thelegacy Wi-Fi incurs up to 60 percent of packet losses forNnodes = 20, primarily due to: (i) channel errors caused bythe BS interference or (ii) buffer overflow at the AP. Third,The pure random selection scheme worsens the situation

due to the inefficient relay selection and may end up lower-ing the quality of the link. Finally and despite of the factthat the random grouping over the RC scheme equally dis-tributes the nodes amongst Wi-Fi Direct groups, the Hun-garian-based algorithm maximizes the probability ofsuccessful data delivery (20 percent or more compared tothe legacy Wi-Fi) by also taking into account the quality ofeach wireless link. The same improvement was observedfor the network throughput, which was not included duethe shortage of space.

Fig. 17b shows the impact of the Tshare on the averagePDR. Regardless of the configuration, the performanceincreases as Tshare increases due the higher chances toaccess the channel. For the case LTE OFF, the system expe-rience some packet dropping at Tshare � 33 percent due tobuffer overflow at the AP. These results confirm that theHungarian Algorithm approach maximizes the perfor-mance, with an improvement of 31% PDR compared to thelegacy Wi-Fi for Tshare = 75%. The same conclusions can bederived from Fig. 17c, where we show the average PDR asa function of �.

6.7 ABS Resource Allocation Network AnalysisA comprehensive analysis of the resource allocation mecha-nism introduced in E-Fi is presented in this section. Theschemes being compared are:

� Legacy Wi-Fi, i.e., the current practical systemsdeployed today.

� The Hungarian-based relay selection, using the legacy802.11-DCF at the MAC layer.

Fig. 17. Impact of Nnodes (a), Tshare (b), and � (c) on the PDR. The default values areNnodes ¼ 10, Tshare ¼ 50%, and � ¼ 25.

Fig. 18. Impact of the configuration schemes (Section 6.7) on the PDR (a), throughput (b), and Jain’s fairness index (c). The default values areTshare ¼ 50% and � ¼ 25.



� E-Fi framework, where both the Hungarian-relayselection (Sections 4.3 and 4.4) and the ABS alloca-tion algorithm (Section 5) are employed with differ-entiated contention access.

Fig. 18a shows the average PDR for the three schemes.We see that (i) the ABS allocation algorithm provides a sig-nificant improvement to the performance of E-Fi, whichnow outperforms the legacy Wi-Fi standard by almost 100percent percent in highly dense scenarios (i.e., N = 20); (ii)even in these extreme situations, the PDR of E-Fi gets con-siderably close to the baseline reference (LTE OFF inFig. 18a), with only 20 percent difference in terms of PDR.Hence we argue that the Wi-Fi network can really surviveLTE-U using E-Fi, mitigating most of the interference com-ing from the BS. The same inference can be derived in termsof throughput shown in Fig. 18b.

Finally, Fig. 18c shows the throughput fairness amongthe N data flows, computed using the well-known Jain’sFairness index. Also in this case, E-Fi provides the best per-formance because: (i) the Hungarian-relay selection mecha-nism guarantees average higher link quality for vulnerableclients; (ii) the inter-ABS scheduler allocates channel

opportunities in a fair way based on the load factor of eachWi-Fi Direct group (Eq. (6)); (iii) the intra-ABS scheduleradjusts the MAC back-off parameters so that the AP and therelaywill have proportional channel access during eachABS.

6.8 Case in Point: E-Fi versus LTE-UEfforts to standardize LTE in the unlicensed spectrum haveresulted in two main implementation proposals, Licensed-Assisted Access (LAA) that is supported by 3GPP anddefined in Release 13 as part of its work plan for 5G [41],and LTE-U from LTE-U Forum that is based on 3GPPReleases 10/11/12 [42]. Both proposals envision consider-able changes to currently deployed LTE specifications. LAAmanages access to the unlicensed spectrum using a Listen-Before-Talk scheme that resembles CSMA/CA and exploitscarrier-aggregation to anchor unlicensed band access to it toa licensed band, while LTE-U adopts a Carrier Sense Adap-tive Transmission (CSAT) as mechanism deployed on Sec-ondary Cell (SCell) in the downlink, and employs an on/offduty cycle as a mechanism to share the medium with exist-ing Wi-Fi networks.

As opposed to LTE-U and LAA, E-Fi groups WiFi devi-ces as to minimize the expected number of transmissions(1=PSR) and further allocates transmission time to thembased on the result and their application load (pkts/s). E-Fiinherently relies on the ABS distribution that LTE previ-ously configured to fulfill its own interference requirements,binding its performance to the available number of ABS.

Comparison has been made also against LTE-U withduty cycling considering the default 80 ms value[43]. It hasbeen shown in [44] that LBT(LAA) and CSAT(LTE-U) con-verge for sufficiently long LTE transmissions, and a choiceof either is mainly driven by LTE operators interests. There-fore, we analyzed the average required time (in ms) for adetermined number of Wi-Fi nodes to access the channeland successfully complete a transmission. Results areshown in Fig. 19 for different network sizes and all the pos-sible ABS configurations (1 to 10 ABS available). Load isevenly distributed among all the Wi-Fi devices.

7 TESTBED AND EXPERIMENTAL RESULTS

In this section,we evaluate the performance of E-Fi on a small-case testbed.More specifically, our goal is to demonstrate thatE-Fi is able to improve the performance of Wi-Fi nodes under

Fig. 19. Latency comparison between LTE-U, with Ton/off equal to 80ms,and E-Fi with equal load factor across the nodes. E-Fi performancedepends on the number of available ABS to schedule its transmissions.E-Fi outperforms LTE-U when the number of Wi-Fi nodes is greater orequal than 10. As the number of nodes increases, E-Fi requires lessABS to outperform LTE-U. Thus, proving the effectivity of the groupingand resource allocation mechanisms proposed in E-Fi.

Fig. 20. Experimental set-up in Section 7. The WC is constituted by aNexus 6 smartphone running Android 6.0 Operating System (OS), andthe GO is a Nexus 5 smartphone running Android 5.0 OS (above). AnAndroid application is installed on the WC and GO and implements theprocedures described in Sections 4 and 5.

Fig. 21. Relay candidacy and final relay selection in a deployment with20 devices, BS-AP distance of 50m and PSRth equal to 0.4. ThePSRAP�WC represents the link quality of the regular Wi-Fi communica-tion whereas PSRAP�GO and PSRGO�WC represent the link quality in E-Fi. The inner area delimited by the line represents the feasible set, meet-ing the criteria in Eq. (3). The Hungarian always selects the combinationwith highest PSR.



several different network topologies andLTE interference lev-els in spite the overhead introduced by Wi-Fi Direct and theone-hop communication. To this aim, we built the testbedcomposed of four nodes: one Wi-Fi Direct Client (WC); onelegacy IEEE 802.11 AP; one Corresponding node (CN) con-nected to the AP via Ethernet; and a Wi-Fi Direct GO (GO),which forwards the traffic from the WC to the CN via AP(Fig. 20). The WC node generates a constant number of UDPpackets per second (k); the packet size is equal to 1000 bytes.The LTE interference was modeled with the PSR and isdenoted as PSRAP

WC , on the WC-AP link; PSRGOWC on the WC-

GO link; and PSRAPGO on the GO-AP link. The PSR valueswere

extracted from a feasible set of combinations (i.e., Fig. 21shows the feasible set when PSRAP

WC is 20 and 30 percent)Therefore,we devised two communication scenarios:

� C1 - No-Relay: The WC sends the packets directly tothe CN (state-of-the-art Wi-Fi).

� C2 - Relay: The WC first transmits to the GO, whichforwards it to the CN afterwards.

Performance metrics, such as the network throughput, thePDR and the delay were extracted upon completion and dis-played on the Application GUI. Notice that processes such asnetwork set-up, i.e., the PSR exchange, device discovery androle selection are handled by the E-FI mobile application(Fig. 20). Fig. 22a shows the achievable throughput account-ing for the overhead incurred in C2 by the GO node for3 baseline scenarios (PSRAP

WC ¼ 60%; 40%; 20%). Since C1 isindependent from the relay, the results are constant across thex plane. We selected reasonable set of values for PSRGO

WC

(Fig. 21) while keeping k constant at 300. The regions whenconfiguration C2 outperforms C1 correspond to those in thegraphwhere the curves lay above the baseline.

The PDR (Fig. 22b) gives us a direct metric of the perfor-mance accounting for retransmissions. Similar to the through-put analysis, E-Fi perceives a better PDR thanWi-Fi in certaincases. For instance, given the combination PSRAP

WC = 20%,PSRGO

WC = 80%, PSRAPGO = 80% (feasibility shown in Fig. 21),

E-Fi raises the PDR from 20 percent to almost 60 percent. As aconclusion, we notice that the one-hop communication is suit-able for moderate or severe interference conditions (i.e.,PSRAP

WC = 20% and PSRAPWC = 40%), verifying the simulation

analysis in Section 6.1.In Fig. 22c, we show the throughput gain of the E-Fi

scheme, considering k = 300 and severe LTE interference con-ditions (PSRAP

WC = 20%), and by varying the PSRGOWC parameter

(on the x-axis), and the PSRAPGO parameter (on the y-axis). The

gain is computed as the throughput increase (in percentage)compared to the no-relay scheme. In our implementation, E-Fi employs the utilization of the GO relay only when the PSRcondition in Eq. (3) is met; otherwise, the WC will transmitdata directly to the CN as in the no-relay scheme (henceachieving a zero gain in these cases). From Fig. 22c, we noticethat the throughput gain can exceed the +500 percent undersome configurations, with amean value of +85 percent.

Aiming to provide some metrics on the incurred delay bythe multi-hop communication, we configure C1 and C2 withdifferent system loads and evaluate the PDR and delay jointly(Fig. 23). We consider a configuration with PSRGO

WC = 80%,PSRAP

GO = 80% and PSRWCGO = 20%. The PDR results reveal that

E-Fi always outperforms Wi-Fi for any system load. Viceversa, E-Fi introduces a higher delay thanWi-Fi, although thevalues are quite close for any system load. Several factorsneed to be consideredwhen interpreting this result. Fist, delaymeasurements only consider packets that were received suc-cessfully, and hence are clearly affected by the congestionissues which might originate at the GO device. E-Fi is able tosuccessfully deliver a considerably higher ratio of packets,and this also implies that each packet experience -on average-a longer buffer delay at the queue of the intermediate GOrelay. Second, the experiment set-up serves for an under-standing of the delayed incurred by the LTE control signalsand in the relay node. However, it does not consider the delayincurred due to the contention amongst Wi-Fi devices (actu-ally only oneWCdevice is considered).

Fig. 22. Testbed results. The network throughput and PDR for different configurations of PSRAPWC and PSRGO

WC are depicted in (a) and (b), respectively.The E-Fi throughput gain is depicted in (c).

Fig. 23. Delay incurred by the direct and one-hop communication for dif-ferent system loads. Although E-Fi may incur in a higher communicationdelay, it lowers the probability of dropping and, in turn, increases thePDR.



In accordance with the analysis covered in Section 6.8,the E-Fi produces an effective performance gain in terms ofdelay when the number of WC is higher than a thresholdwhich depends on the number of ABS (see Fig. 19). Finally,it is worth remarking that the current testbed implementsthe E-Fi functionalities at the application layer; hence, eachmessage forwarding operation incurs in an additional proc-essing delay, which might be nullified when deploying theE-Fi scheme at the MAC layer.

8 CONCLUSION

We proposed E-Fi, which allows Wi-Fi devices to surviveuncoordinated LTE transmissions through grouping ofnodes into relays, separating Uplink/Downlink traffic dura-tions and modifications to the backoff based on PSR. Ourdesign is motivated through studies conducted with realdevices, network simulators and standards-compliant LTEand WLAN physical layer waveforms. We show for the firsttime that Wi-Fi can intelligently adapt its operation to handlehigh PER and lack of channel access opportunities throughthe use of ABS frames. Through a mix of packet-level simula-tion using a standards-compliant physical layer, as well astestbed experiments, we show that E-Fi’s improvementsover classical Wi-Fi range from 25-50 percent for throughputand 15 percent for PER under severe interference from LTE.We also provide a comprehensive comparison between E-Fiand LTE-U, the strongest proposal to be included in the nextrelease, as well as conditions under which E-Fi outperformsLTE-U. Given the involvement of P2P device querying andthe required measurement exchange with the AP, E-Fi ismore applicable in low-mobility scenarios. Our work demon-strates that it is indeed possible to coexist without assump-tions of direct feedback between these two very differentaccess technologies in the ISM band. The multiple AP sce-nario where the APs operate in the same channel is beinginvestigated and will be incorporated in future work. Weare evaluating a game theory-based approach that will allowE-Fi APs to select the channel in multi-AP scenario. We arealso considering cases where more than one LTE stationshares the spectrum.

ACKNOWLEDGMENTS

This work is supported in part by MathWorks under theDevelopment-Collaboration Research Grant and by the U.S.Office of Naval Research under grant number N00014-16-1-2651. The authors would like to thank Mike McLernon,Ethem Sozer, Rameez Ahmed, and Kunal Sankhe for theircontinued support and guidance on this project.

REFERENCES

[1] Cisco, Global Mobile Data Traffic Forecast Update, 2015–2020White Paper. Accessed on: 18 Apr. 2016.

[2] A. M. Cavalcante, E. Almeida, R. D. Vieira, S. Choudhury,E. Tuomaala, K. Doppler, F. Chaves, R. C. Paiva, and F. Abinader,“Performance evaluation of LTE andWi-Fi coexistence in unlicensedbands,” inProc. IEEE 77thVeh. Technol. Conf., 2013, pp. 1–6.

[3] E. Almeida, A. M. Cavalcante, R. C. Paiva, F. S. Chaves,F. M. Abinader, R. D. Vieira, S. Choudhury, E. Tuomaala, andK. Doppler, “Enabling LTE/WiFi coexistence by LTE blanksubframe allocation,” in Proc. IEEE Int. Conf. Commun., 2013,pp. 5083–5088.

[4] S. Sesia, I. Toufik, andM. Baker, LTE, TheUMTSLongTermEvolution:FromTheory to Practice. Hoboken, NJ, USA:Wiley Publishing, 2009.

[5] Wi-Fi Alliance, “Wi-Fi peer-to-peer (P2P) technical specification,”2010. [Online]. Available: www.wi-fi.org/Wi-Fi_Direct.php

[6] A. Babaei, J. Andreoli-Fang, Y. Pang, and B. Hamzeh, “On theimpact of LTE-U on Wi-Fi performance,” Int. J. Wireless Inf. Netw.,vol. 22, no. 4, pp. 336–344, 2015.

[7] H. Zhang, X. Chu, W. Guo, and S. Wang, “Coexistence of Wi-Fi andheterogeneous small cell networks sharing unlicensed spectrum,”IEEECommun.Mag., vol. 53, no. 3, pp. 158–164,Mar. 2015.

[8] C. Cano and D. J. Leith, “Coexistence of WiFi and LTE in unli-censed bands: A proportional fair allocation scheme,” in Proc.IEEE Int. Conf. Commun. Workshop, 2015, pp. 2288–2293.

[9] A. K. Sadek, “Carrier sense adaptive transmission (csat) in unli-censed spectrum,” U.S. Patent 9,491,632, Nov. 8, 2016.

[10] N. Rupasinghe and _I. G€uvenc, “Reinforcement learning forlicensed-assisted access of lte in the unlicensed spectrum,” inProc. IEEE Wireless Commun. Netw. Conf., 2015, pp. 1279–1284.

[11] Z. Guan and T. Melodia, “CU-LTE: Spectrally-efficient and faircoexistence between LTE and Wi-Fi in unlicensed bands,” in Proc.35th Annu. IEEE Int. Conf. Comput. Commun., 2016, pp. 1–9.

[12] T. Tao, F. Han, and Y. Liu, “Enhanced lbt algorithm for lte-laa inunlicensed band,” in Proc. IEEE 26th Annu. Int. Symp. PersonalIndoor Mobile Radio Commun., 2015, pp. 1907–1911.

[13] S. Hajmohammad, H. Elbiaze, and W. Ajib, “Fine-tuning theFemtocell performance in unlicensed bands: Case of WiFi Co-existence,” in IEEE Proc. Int. Wireless Commun. Mobile Comput.Conf., 2014, pp. 250–255.

[14] F. Liu, E. Bala, E. Erkip, and R. Yang, “A framework for femtocellsto access both licensed and unlicensed bands,” in Proc. IEEEWiOpt, 2011, pp. 407–411.

[15] Y. Li, J. Zheng, and Q. Li, “Enhanced listen-before-talk scheme forfrequency reuse of licensed-assisted access using lte,” in Proc.IEEE 26th Annu. Int. Symp. Personal Indoor Mobile Radio Commun.,2015, pp. 1918–1923.

[16] S. Nielsen and A. Toskala, “LTE in unlicensed spectrum: Europeanregulation and co-existence considerations,” presented at the 3GPPWorkshop LTE Unlicensed Spectrum, Sophia Antipolis, France, 2014.

[17] S. Deb, P. Monogioudis, J. Miernik, and J. P. Seymour,“Algorithms for enhanced inter-cell interference coordination(eICIC) in LTE HetNets,” IEEE/ACM Trans. Netw., vol. 22, no. 1,pp. 137–150, Feb. 2014.

[18] J. Jeon, Q. C. Li, H. Niu, A. Papathanassiou, and G. Wu, “LTE inthe unlicensed spectrum: A novel coexistence analysis withWLAN systems,” in Proc. IEEE GLOBECOM, 2014, pp. 3459–3464.

[19] S. Sagari, S. Baysting, D. Saha, I. Seskar, W. Trappe, and D. Ray-chaudhuri, “Coordinated dynamic spectrum management ofLTE-U and Wi-Fi networks,” in Proc. IEEE DySPAN, 2015,pp. 209–220.

[20] A. Bhorkar, C. Ibars, and P. Zong, “On the throughput analysis ofLTE and WiFi in unlicensed band,” in Proc. IEEE 48th AsilomarConf. Signals Syst. Comput., 2014, pp. 1309–1313.

[21] A. M. Voicu, L. Simi�c, and M. Petrova, “Inter-technology coexis-tence in a spectrum commons: A case study of wi-fi and lte in the5-ghz unlicensed band,” IEEE J. Select. Areas Commun., vol. 34,no. 11, pp. 3062–3077, Nov. 2016.

[22] A. Pyattaev, K. Johnsson, S. Andreev, and Y. Koucheryavy, “3GPPLTE traffic offloading onto WiFi Direct,” in Proc. IEEE WirelessCommun. Netw. Conf. Workshops, 2014, pp. 135–140.

[23] A. Asadi, P. Jacko, and V. Mancuso, “Modeling D2D communica-tions with LTE and WiFi,” ACM SIGMETRICS Perform. Eval. Rev.,vol. 42, no. 2, pp. 55–57, 2014.

[24] D. M. Gutierrez-Estevez, B. Canberk, and I. F. Akyildiz, “Spatio-temporal estimation for interference management in femtocellnetworks,” in Proc. IEEE 23rd Int. Symp. Personal Indoor MobileRadio Commun., 2012, pp. 1137–1142.

[25] J. Xie and I. Howitt, “Multi-domain wlan load balancing in wlan/wpan interference environments,” IEEE Trans. Wireless Commun.,vol. 8, no. 9, pp. 4884–4894, Sep. 2009.

[26] A. Asadi and V. Mancuso, “WiFi Direct and LTE D2D in action,”in Proc. Wireless Days, 2013, pp. 1–8.

[27] G. A. Shah, F. Alagoz, E. A. Fadel, and O. B. Akan, “A Spectrum-aware clustering for efficient multimedia routing in cognitiveradio sensor networks,” IEEE Trans. Veh. Technol., vol. 63, no. 7,pp. 3369–3380, Sep. 2014.

[28] B. Gulbahar and O. B. Akan, “Information theoretical optimiza-tion gains in energy adaptive data gathering and relaying in cog-nitive radio sensor networks,” IEEE Trans. Wireless Commun.,vol. 11, no. 5, pp. 1788–1796, May 2012.



www.wi-fi.org/Wi-Fi_Direct.php

[29] P. Series, “Propagation data and prediction methods for the plan-ning of indoor radiocommunication systems and radio local areanetworks in the frequency range 900 MHz to 100 GHz,” Recom-mendation ITU-R, pp. 1238–7, 2012.

[30] J. Dugan, S. Elliott, B. A. Mah, J. Poskanzer, and K. Prabhu, “Iperf:The TCP/UDP bandwidth measurement tool,” https://iperf.fr/,2005.

[31] A. Tirumala, F. Qin, J. Dugan, J. Ferguson, and K. Gibbs, “Iperf:The TCP/UDP bandwidth measurement tool,” 2005. [Online].Available: http://dast.nlanr.net/Projects

[32] Atheros Communications Inc., “Drivers for Atheros IEEE 802.11nPCI/PCI-Express and AHB WLAN based chipsets,” 2005.[Online]. Available: https://wireless.wiki.kernel.org/en/users/drivers/ath9k

[33] R. Krishnakumar, “Kernel korner: kprobes-a kernel debugger,”Linux J., vol. 2005, no. 133, 2005, Art. no. 11.

[34] C. Chekuri and S. Khanna, “A polynomial time approximationscheme for the multiple knapsack problem,” SIAM J. Comput.,vol. 35, no. 3, pp. 713–728, 2005.

[35] H. W. Kuhn, “The Hungarian method for the assignment prob-lem,”Naval Res. Logistics Quart., vol. 2, no. 1/2, pp. 83–97, 1955.

[36] V. Liu, A. Parks, V. Talla, S. Gollakota, D. Wetherall, andJ. R. Smith, “Ambient backscatter: Wireless communication out ofthin air,” in Proc. ACM SIGCOMM 2013 Conf. SIGCOMM, 2013,pp. 39–50.

[37] S. M. Kim and T. He, “Freebee: Cross-technology communicationvia free side-channel,” in Proc. 21st Annu. Int. Conf. Mobile Comput.Netw., 2015, pp. 317–330.

[38] Z. Chi, Z. Huang, Y. Yao, T. Xie, H. Sun, and T. Zhu, “Emf:Embedding multiple flows of information in existing traffic forconcurrent communication among heterogeneous iot devices,” inProc. IEEE Conf. Comput. Commun., 2017, pp. 1494–1502.

[39] X. Guo, X. Zheng, and Y. He, “Wizig: Cross-technology energycommunication over a noisy channel,” in Proc. IEEE Conf. Comput.Commun., 2017, pp. 1485–1493.

[40] D. H. Staelin, “Fast folding algorithm for detection of periodicpulse trains,” Proc. IEEE, vol. 57, no. 4, pp. 724–725, Apr. 1969.

[41] LTE-U Forum, “LTE-U SDLCoexistence Specifications v1.3,” 2015.[42] Alliance, Wi-Fi, “Coexistence guidelines for LTE in unlicensed

spectrum studies V2.0,” 2016.[43] C. Cano and D. J. Leith, “Unlicensed LTE/WiFi coexistence: Is LBT

inherently fairer thanCSAT?” in Proc. 2016 IEEE Int. Conf. Commun.(ICC), May 2016, pp. 1–6, doi: 10.1109/ICC.2016.7510910.

[44] C. Cano and D. J. Leith, “Unlicensed lte/wifi coexistence: Is lbtinherently fairer than csat?” in Proc. IEEE Int. Conf. Commun.,2016, pp. 1–6.

Carlos Bocanegra received the bachelor’s degreefrom the Polytechnic University of Catalonia (UPC),Barcelona, Spain, in 2015. He is working toward thePhD degree in the Electrical and Computer Engi-neering Department, Northeastern University. Hewas a visiting scholar with Northeastern Universityfrom October 2014 to July 2015. His researchinterests includemobile communications, millimeterwaves communications, software-defined radios(SDR), optimum scheduling mechanisms, and net-work slicing. He is a studentmember of the IEEE.

Takai Eddine Kennouche received the graduatedegree from The National Preparatory School forEngineering Studies of Algiers, Algeria, in 2010and the master’s degree in telecommunicationsengineering from the National Institute of Telecom-munications and ICTof Oran, Algeria, in 2013. Heis currently working toward the PhD degree inthe Department of Industrial and Information Engi-neering, University of Pavia, Italy. His researchinterests include coexistence of heterogeneouswireless networks, cross layer optimizations, cogni-tive radio, andSDR. He is amember of the IEEE.

Zhengnan Li received the bachelor’s degree incommunication engineering from the ShandongUniversity of Technology, Zibo, China. He is cur-rently working toward the MS degree in electricaland computer engineering at Northeastern Uni-versity, Boston, Massachusetts. His researchinterests include wireless communications, 5G,and SDR. He is a student member of the IEEE.

Lorenzo Favalli received the graduate degree inelectronic engineering and the PhD degree fromthe Polytechnic University of Milan, in 1987 and1991, respectively. He joined theUniversity of Paviain 1991 as an assistant professor and became anassociate professor, in 2000. His teaching dutiesinclude courses in digital communications, wirelesscommunications systems, andmultimedia commu-nications. His research activity covers variousaspects of signal and video analysis and transmis-sion in both wireless and wired networks. His work

also encompasses the exploitation of adaptive techniques to improve flexi-bility and reliability of the communications chain, source and networkmodeling, and improvements in signal detection techniques in heteroge-neouswireless environments. He is amember of the IEEE.

Marco Di Felice received the Laurea (summa cumlaude) and PhD degrees in computer science fromthe University of Bologna, Italy, in 2004 and 2008,respectively. In 2007, he was a visiting researcherwith the Broadband Wireless Networking Labora-tory, Georgia Institute of Technology, Atlanta, Geor-gia. In 2009, he was a visiting researcherwith Northeastern University, Boston Massachu-setts. Currently, he is an associate professor incomputer science with the University of Bologna.His research interests include self-organizing wire-

less networks, cognitive radio and vehicular systems, mobile applications,and services. He currently serves on the editorial board of theElseviers AdHoc Networks Journal. He authored more than 80 papers on wireless andmobile systems. He joint several national and international research proj-ects. He received the Best Paper Award at the Association for ComputingMachinery International Symposium on Mobility Management and Wire-less Access (MOBIWAC) in 2012 and at the IEEE Annual MediterraneanAdHocNetworkingWorkshop (MED-HOC- NET) in 2013.

Kaushik Chowdhury (M’09-SM’15) received thePhD degree from the Georgia Institute of Tech-nology, Atlanta, in 2009. He is currently an asso-ciate professor and a faculty fellow with theElectrical and Computer Engineering Depart-ment, Northeastern University, Boston, Massa-chusetts. He was awarded the Presidential EarlyCareer Award for Scientists and Engineers(PECASE), in Jan. 2017, the DARPA Young Fac-ulty Award in 2017, the Office of Naval ResearchDirector of Research Early Career Award in

2016, and the US National Science Foundation (NSF) CAREER awardin 2015. He received multiple best paper awards, including three in theIEEE ICC conference, in 2009, 2012, and 2013, and the ICNC confer-ence in 2013. He serves on the editorial board of the IEEE Transactionson Wireless Communications. His research has been supported bythe NSF, Office of Naval Research, DARPA, and MathWorks, amongothers. His current research interests are in dynamic spectrum accessnetworks and systems, wearables and implant communication, andenergy harvesting sensors. He is a senior member of the IEEE.

" For more information on this or any other computing topic,please visit our Digital Library at www.computer.org/publications/dlib.



https://iperf.fr/

http://dast.nlanr.net/Projects

https://wireless.wiki.kernel.org/en/users/drivers/ath9k

https://wireless.wiki.kernel.org/en/users/drivers/ath9k

http://dx.doi.org/10.1109/ICC.2016.7510910

Date post:	06-Aug-2020
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

E-Fi: Evasive Wi-Fi Measures for Surviving LTE within 5 ...bocanegrac/papers/E-Fi_2018.pdf · posed...

Documents