Risk-Informed Interference Assessment for Shared Spectrum Bands: A Wi-Fi/LTE Coexistence Case Study
Andra M. Voicu, Ljiljana Simi?, J. Pierre de Vries, Marina Petrova, Petri Mähönen
11 Risk-Informed Interference Assessmentfor Shared Spectrum Bands:A Wi-Fi/LTE Coexistence Case Study
Andra M. Voicu, Ljiljana Simi´c, J. Pierre de Vries,Marina Petrova and Petri M¨ah¨onen
Abstract —Interference evaluation is crucial when decidingwhether and how wireless technologies should operate. In thispaper we demonstrate the benefit of risk-informed interferenceassessment to aid spectrum regulators in making decisions, andto readily convey engineering insight. Our contributions are:we apply, for the first time, risk assessment to a problem ofinter-technology spectrum sharing, i.e. Wi-Fi/LTE in the 5 GHzunlicensed band, and we demonstrate that this method compre-hensively quantifies the interference impact. We perform simu-lations with our newly publicly-available tool and we considerthroughput degradation and fairness metrics to assess the risk fordifferent network densities, numbers of channels, and deploymentscenarios. Our results show that no regulatory intervention isneeded to ensure harmonious technical Wi-Fi/LTE coexistence:for the typically large number of channels available in the 5 GHzband, the risk for Wi-Fi from LTE is negligible, rendering policyand engineering concerns largely moot. As an engineering insight,Wi-Fi coexists better with itself in dense, but better with LTE, insparse deployments. Also, both main LTE-in-unlicensed variantscoexist well with Wi-Fi in general. For LTE intra-technologyinter-operator coexistence, both variants typically coexist wellin the 5 GHz band, but for dense deployments, implementinglisten-before-talk causes less interference.
Index Terms —coexistence, interference, LTE, risk assessment,spectrum regulation, Wi-Fi.
I. I
NTRODUCTION
Inter-technology spectrum sharing may generate coexistenceproblems in bands where mutual interference among differentsystems occurs. Dynamic spectrum access (DSA) techniquesseek to solve such problems by allowing access to the spec-trum on a primary-secondary basis, where the primary haspriority over secondary systems [2]. This problem is managedby each technology individually in the unlicensed bands,where all systems have equal rights to access the spectrum.Regardless of the spectrum access rights, inter-technologyspectrum sharing raises a two-stage question: (i) which tech-nologies should/can coexist based on the expected harm of mu-tual interference, and (ii) how to manage interactions betweentechnologies on a moment-by-moment basis? In this paper we
This paper was presented in part at IEEE DySPAN, Baltimore, March2017 [1].A. M. Voicu, L. Simi´c, and P. M¨ah¨onen are with the Institute forNetworked Systems, RWTH Aachen University, Aachen, Germany (e-mail: [email protected]; [email protected]; [email protected]). J. P. de Vries is with the Silicon Flatirons Center, University ofColorado, Boulder, U.S. (e-mail: [email protected]). M. Petrova is withthe School of Information and Communication Technology, KTH Royal In-stitute of Technology, 100 44 Stockholm, Sweden (e-mail: [email protected]). present an extensive case study of applying risk assessmentfor Wi-Fi/LTE coexistence in the unlicensed bands, in orderto evaluate the harm caused by inter-technology interferencein shared spectrum bands.Evaluating coexistence problems due to co- and adjacentchannel interference is of interest both to spectrum regulatorsseeking to establish operational bounds and to engineersdesigning and managing systems for optimized performancewithin the regulatory restrictions. Assessing interference isnot a trivial task; consequently, most of the studies managethis complexity by considering worst-case scenarios as thebaseline. Nevertheless, it is not clear how often or underwhat conditions such worst-case scenarios would occur inpractice. Making regulatory decisions based on worst-caseanalysis may even lead to a complete exclusion of newentrant technologies, so that the second question of interfer-ence management becomes irrelevant. As such, comprehensiveinterference assessment methods are essential for creating aregulatory environment that would enable the deployment ofadvanced spectrum-sharing techniques, e.g. for DSA-like sce-narios. Effective interference assessment methods are equallyimportant for the engineers who design, deploy, and managenetworks of different technologies coexisting in shared bands,e.g. IEEE 802.11g and n, and Wi-Fi/LTE in the unlicensedbands. Coexistence performance optimization of such net-works cannot be conducted under worst-case conditions only.In this paper we demonstrate the benefit of risk assess-ment as a complement to worst-case interference analysis.Importantly, risk assessment is a very new method in thefields of communications engineering and spectrum regulation,although it has been used successfully in other fields [3]. Weapply risk assessment to a Wi-Fi/LTE coexistence study inthe 5 GHz unlicensed band for different network densities,number of channels, and scenarios, both from the point ofview of Wi-Fi incumbents and LTE-in-unlicensed entrants. Ourcontributions are: (i) we are the first to apply risk-informedinterference assessment to a real-life, topical problem (dealingwith inter-technology spectrum sharing with wide relevancefor regulatory DSA-like scenarios); and (ii) we demonstratethe benefit of risk assessment as a method that comprehen-sively and quantitatively characterizes the harm caused byinterference in an intuitive and illustrative manner, from bothpolicy and engineering perspectives. Furthermore, we providea publicly-available network simulation tool [4] for risk-informed interference assessment of Wi-Fi/LTE coexistence, a r X i v : . [ c s . N I] J un implementing our simulation model in Section IV.Our analysis shows that no regulatory intervention isneeded to ensure harmonious technical coexistence betweenWi-Fi/LTE in the unlicensed bands. From an engineeringperspective, we show that Wi-Fi coexists better with itselfand worse with LTE in locally dense deployments, but thatthe opposite holds in sparse deployments, due to the specificsof Wi-Fi’s MAC. Also, given the large number of availablechannels expected in practice in the 5 GHz band, there istypically no risk of interference caused by LTE-in-unlicensedentrants, which renders both policy and engineering coexis-tence issues largely irrelevant. In general, both main proposedLTE-in-unlicensed entrant variants coexist equally well withWi-Fi. For LTE intra-technology inter-operator coexistence,both variants typically coexist well in the 5 GHz band, butfor very dense deployments, the variant implementing listen-before-talk (LBT) causes less mutual interference betweenoperators.The remainder of this paper is organised as follows. Sec-tion II gives a brief overview of LTE-in-unlicensed and priorwork on its coexistence with Wi-Fi. Section III presentsthe risk-informed interference assessment method. Section IVpresents the simulation and throughput model. Section V illus-trates and discusses the benefit of applying the risk assessmentmethod for our Wi-Fi/LTE case study, from the point of viewof Wi-Fi incumbents. Section VI presents and discusses riskanalysis results from the perspective of LTE-in-unlicensed andSection VII concludes the paper.II. LTE- IN - UNLICENSED : T HE S TORY SO F AR LTE operation in the unlicensed 5 GHz band has recentlybeen proposed by industry [5], [6]. Initially, the unlicensedband is aggregated only for user data transmissions, while thecontrol traffic is sent over the licensed bands for reliabilityreasons [7]. Two main LTE-in-unlicensed variants with funda-mentally different MAC mechanisms have emerged: (i) LTE-Uproposed by the LTE-U Forum [5]; and (ii) Licensed AssistedAccess (LAA) first standardized by 3GPP in Release 13 [7].LTE-U is based on an adaptive duty cycle MAC mechanism,which adjusts the periodic transmission duration of the devicesaccording to the number of other devices operating in thesame channel, such that all devices have an equal share ofthe channel in time. However, LTE-U devices do not senseand defer to ongoing transmissions before starting their owntransmissions, so collisions are likely. LTE-U is a pre-standardversion intended for markets where LBT is not required byregulators (e.g. the U.S.).LAA is based on LBT, a MAC mechanism in which devicesstart transmitting only after detecting that the channel isunoccupied. LBT is required by spectrum regulators in someregions (e.g. Europe), so LAA was proposed as a globallyapplicable standard.As Wi-Fi is currently the dominant technology in the 5 GHzband, it has been claimed by some parties (e.g. [8]) that in-troducing LTE-in-unlicensed would harm Wi-Fi operation. On Considering economic and policy coexistence issues, e.g. deploying LTE-in-unlicensed for anti-competitive practices, is out of the scope of this paper. the other hand, proponents have argued that LTE-in-unlicensedwould actually improve Wi-Fi performance compared to Wi-Ficoexisting with itself [5], [6]. The debate between the twocamps led the FCC to issue a Public Notice requestingcomments on LTE coexistence in the unlicensed bands [9],implicitly raising the question of whether regulatory interven-tion is required to ensure harmonious technical coexistencebetween LTE-in-unlicensed and Wi-Fi.Most existing Wi-Fi/LTE coexistence analyses are not thor-ough enough to answer the public policy question of whetherLTE is friend or foe to Wi-Fi in the unlicensed band. Someexisting work lacks a detailed description of algorithms andmodels (e.g. [5]), so that it is difficult to draw generalizableconclusions from the presented results. Other work considersonly one main LTE-in-unlicensed variant ( cf . classificationof related work in [10]), so that the results only partiallycharacterize the Wi-Fi/LTE coexistence problem. In our pre-vious work [11] we presented the results of a transparent,systematic, and extensive coexistence study and we showedthat LTE-in-unlicensed is neither friend nor foe to Wi-Fi.In this paper we extend our previous work by conducting,for the first time, a risk assessment of the Wi-Fi/LTE coex-istence problem, in order to show the effectiveness of thismethod for deriving regulatory and engineering insight fromquantitative results in a comprehensive, illustrative, and intu-itive manner. Furthermore, we extend our throughput modelfrom [10] by incorporating adjacent channel interference andwe consider throughput fairness as an additional coexistenceperformance metric. Finally, we present more detailed resultsthan in [10], [11] by showing the full distributions of ourconsidered metrics.III. R ISK -I NFORMED I NTERFERENCE A SSESSMENT
A. Introduction to Risk Assessment
Risk-informed interference assessment was introduced asa comprehensive, quantitative tool for a spectrum regulatorseeking to balance the interests of incumbents, new entrantsand the public when deciding whether and how to allocate newradio services [12]. It facilitates a balanced assessment of theadverse technical impact of new entrants on incumbents.Engineering risk assessment, a well-established methodused in many industries (e.g. nuclear energy, environmentalprotection, food safety, etc. [3]), considers the likelihood-consequence combinations for multiple hazard scenarios, andcomplements a “worst case” analysis that considers the singlescenario with the most severe consequence, regardless of itslikelihood. Charts that plot the severity of hazards against theirlikelihoods are frequently used to visualize and compare therisk of different hazards; see Fig. 2(b).To date, quantitative risk assessment has not been usedin spectrum management. The author in [13] proposed afour-step method for performing risk-informed interferenceassessment: (1) make an inventory of all significant harmfulinterference hazard modes; (2) define a consequence metric tocharacterize the severity of hazards; (3) assess the likelihoodand consequence of each hazard mode; and (4) aggregatethem into a basis for decision making. In [13], [14] it was
TABLE IS
CENARIOS AND ENTRANT VARIANTS
PARAMETER SCENARIO
Indoor/indoor (indoor incumbent,indoor entrant)
Outdoor/outdoor (outdoor incumbent,outdoor entrant)Network size incumbent : 10 APs entrant : 1–30 APs incumbent : 10 APs entrant : 1–10 APs
Maximum number ofavailable channels (Europe)
19 11
Coexistencemechanism
Channelselection incumbent : random or single channel entrant : random or sense (select channel with fewest incumbent APs) or single channel MAC incumbent : Wi-Fi : LBT, CS threshold of -82 dBm for co-channel Wi-Fi devices,and -62 dBm for co-channel non-Wi-Fi and all adjacent channel devices entrant : LAA : LBT, CS threshold of -62 dBm
LTE-U : ON/OFF with adaptive duty cycle based on number of entrant& incumbent APs within CS range (CS threshold = -62 dBm)
Wi-Fi : LBT, CS threshold of -82 dBm for co-channel Wi-Fi devices,and -62 dBm for co-channel non-Wi-Fi and all adjacent channel devices
PHY incumbent : Wi-Fi : IEEE 802.11n spectral efficiency ρ W iFi , noise figure NF=15 dB entrant : LAA : LTE spectral efficiency ρ LT E , NF=9 dB
LTE-U : LTE spectral efficiency ρ LT E , NF=9 dB
Wi-Fi : IEEE 802.11n spectral efficiency ρ W iFi , NF=15 dB
LBT parameters& assumptions binary exponential random backoff with CW min =15, CW max =1023,time slot duration σ =9 µ s, SIFS=16 µ s, DIFS=SIFS+2 σ =34 µ s ( cf . IEEE 802.11) LBT frameduration T f Wi-Fi : T f = f n ( rate , MSDU , PHY header , M AC header ) , MSDU =1500 Bytes,
PHY header =40 µ s, M AC header =320 bits ( cf . IEEE 802.11) LAA : T f =1 ms (i.e. duration of LTE subframe) Duty cycle ON-time
LTE-U : 100 ms (i.e. maximum ON-time specified in [6])
User distribution
Traffic model downlink full-buffered
Channel bandwidth
20 MHz
Frequency band
AP transmit power
23 dBm shown how this method could be used to analyse the riskof cellular interference to weather satellite earth stations for ahypothetical general case. By contrast, we are the first to applyrisk-informed interference assessment to a real-life problemand to inter-technology coexistence in the same spectrumband.
B. Applying Risk Assessment to Wi-Fi/LTE Coexistence
In this paper we evaluate co- and adjacent channel interfer-ence among LTE-in-unlicensed entrants and Wi-Fi incumbentsby applying risk assessment. In Section IV we present theinterference hazards corresponding to Step (1). In Section III-Cwe define the throughput consequence metrics to characterizehazard severity for Step (2). In Section V we demonstrateSteps (3) and (4) by assessing the hazard modes and byshowing the effectiveness of risk assessment when makingdecisions of regulatory and engineering concern, from thepoint of view of the Wi-Fi incumbents. In Section VI weextend our demonstration for Steps (3) and (4) by presentingthe LTE-in-unlicensed perspective.
C. Consequence Metrics for Risk Assessment
In this section we define the consequence metrics to char-acterize the severity of the interference hazards. In the context of Wi-Fi/LTE coexistence we select two throughput metrics that represent the hazard consequence for the incumbents:(i) the throughput degradation, which we consider the mostrelevant metric to quantify whether Wi-Fi gets a fair share ofthe channel and whether it experiences excessive interferencewhen coexisting with LTE-in-unlicensed, and thus to answerthe technical public policy coexistence question; and (ii) thethroughput unfairness among incumbents, which gives insightinto engineering optimization of inter-technology coexistencewithin the given regulatory context.We define the throughput degradation of the incumbentaccess points (APs) when coexisting with entrant APs withrespect to two different baselines: (i) the standalone Wi-Fiincumbent network, in order to capture the general throughputdegradation due to network densification; and (ii) the Wi-Fiincumbent network coexisting with a Wi-Fi entrant network,in order to directly focus on the question of whether LTE isa better neighbour to Wi-Fi than Wi-Fi is to itself. For anincumbent AP x we estimate the throughput degradation as ∆ R x = R x , baseline − R x R x , baseline , (1) We note that throughput has been the baseline network performanceevaluation metric in general and is also considered the only or primaryperformance metric in important Wi-Fi/LTE coexistence studies, e.g. [5].Although delay can be considered a relevant evaluation metric in some cases,it is typically applied for VoIP traffic [15], which does not represent themajority of the traffic. where R x , baseline is the baseline throughput of x and R x is thethroughput of x when coexisting with a given entrant variant.In order to quantify the throughput fairness among incum-bent APs, we apply Jain’s fairness index [16] for a set ofincumbent throughput results corresponding to APs in a singlenetwork realization, given by J = | (cid:205) nx = R x | n (cid:205) nx = R x , (2)where n is the number of incumbents in the network.For consistency with data representation in a risk assessmentchart (explained in Section V-A), we define the incumbentthroughput unfairness as the consequence metric, given by U = − J . (3)When considering the LTE-in-unlicensed perspective inSection VI, the throughput consequence metrics are analogousto those defined for the incumbents in (1)–(3).IV. S IMULATION & T
HROUGHPUT M ODELS
A. Simulation Model
We assume a population of Wi-Fi incumbent APs coexistingwith Wi-Fi, LAA or LTE-U entrant APs in two main scenarios,for realistic network densities, as summarized in Table I.The incumbent APs and their associated users are alwaysWi-Fi devices implementing the IEEE 802.11n PHY layer andLBT at the MAC layer with a carrier sense (CS) thresholdof -82 dBm for deferring to co-channel Wi-Fi devices, and-62 dBm for adjacent channel Wi-Fi devices and co- andadjacent channel non-Wi-Fi devices. The entrants are either(i) LAA implementing the LTE PHY and the LBT MACmechanism with -62 dBm CS threshold for deferring toall other devices, or (ii) LTE-U that adapts its duty cycleaccording to the number of detected APs based on the -62 dBmCS threshold. As the baseline for answering the question ofwhether LTE-in-unlicensed is friend or foe to Wi-Fi, we alsoconsider (iii) Wi-Fi entrants.Two main scenarios are considered, where each AP hasone associated user, i.e. the indoor/indoor scenario whereall incumbent and entrant devices are located indoors andthe outdoor/outdoor scenario where all devices are locatedoutdoors, as in Fig. 1. For the indoor/indoor scenario weassume a single-floor building, according to the 3GPP dualstripe model [17]. Each incumbent AP and its associated userare located randomly within a single apartment. The entrantAPs and their associated users are first randomly located inunoccupied apartments and then randomly occupy apartmentswith only one other AP, until all apartments contain up to twoAPs. This results in network densities of 600–12000 APs/km ,as (and more) dense as that seen in contemporary 2.4 GHz de-ployments, but not yet in 5 GHz [18]. For the outdoor/outdoor scenario we assume 20 real outdoor base station locations fromcentral London [19] and we randomly overlay buildings over We note that CSMA/CA is a specific variant of the more general LBTmechanism, so we refer to it as LBT. In this paper we assume LBT withbinary exponential random backoff throughout. (a)
Indoor/indoor scenario: the incumbents and entrants arelocated inside a single-floor building with 20 apartments (eachof 10 m ×
10 m × Outdoor/outdoor scenario: the incumbent and entrant APsare randomly allocated one real outdoor location and areplaced at the roof-top level. The outdoor users are locatedin the coverage area of and at a maximum distance of 50 mfrom the AP that they are associated with, at a height of 1.5 m.The length of the buildings is randomly selected between 3–10apartments and the height is randomly selected between 3–5floors. The size of the total study area is 346 m ×
389 m,corresponding to the area in London where the real locationsof the outdoor APs were observed.Fig. 1. Example network layout based on the 3GPP dual stripe modelfor indoor deployments and real outdoor picocell locations for outdoordeployments, for the (a) indoor/indoor , and (b) outdoor/outdoor scenarios,showing locations of incumbent APs ( (cid:78) ), incumbent users ( (cid:4) ), entrant APs( (cid:7) ), and entrant users ( • ). the area where the real outdoor locations were observed, re-sulting in network densities of 7–150 APs/km . The associatedusers are located within the coverage area of the respectiveAPs and at a maximum distance of 50 m. As a worst-caseinterference scenario of low signal attenuation through wallsresulting in high interference among APs, we also considerthe indoor/indoor scenario without internal walls .Each incumbent AP randomly selects one of the availablechannels. The entrants either randomly select a channel, i.e. random , or apply sense , i.e. they randomly select a channelunoccupied by incumbents. We assume the maximum numberof channels in the 5 GHz band in Europe to be typicallyavailable in practice (i.e. 19 indoor and 11 outdoor channels),or only the 4 non-DFS channels, corresponding to less likely TABLE IIP
ARAMETERS FOR THROUGHPUT AND INTERFERENCE MODEL
ParameterAP type Incumbent Entrant S x defined in [10] defined in [10], if W-Fi/LAA entrant1, if LTE-U entrant r deg , x
0, if Wi-Fi/LAA entrantdefined in [10], if LTE-U entrant 0
AirTime x + | A x | + | B x | , if Wi-Fi/LAA entrant (cid:214) y ∈ B x (cid:16) − + | C y | + | D y | (cid:17) × + | A x | ,if LTE-U entrant + | A x | + | B x | ρ x ρ W iFi [20] ρ W iFi , if Wi-Fi entrant ρ LT E [21], if LAA/LTE-U entrant I cou (cid:213) z ∈( A co \ A cox )∪( B co \ B cox ) P z × AirTime z L u , z (cid:213) z ∈( A co \ A cox )∪( B co \ B cox ) P z × AirTime z L u , z , if Wi-Fi/LAA entrant (cid:213) z ∈( A co \ A cox )∪( B co ) P z × AirTime z L u , z , if LTE-U entrant I ad ju (cid:213) z ∈( A ad j \ A ad jx )∪( B ad j \ B ad jx ) P z × AirTime z L u , z × ACI R u , z (cid:213) z ∈( A ad j \ A ad jx )∪( B ad j \ B ad jx ) P z × AirTime z L u , z × ACI R u , z , if Wi-Fi/LAA entrant (cid:213) z ∈( A ad j \ A ad jx )∪( B ad j ) P z × AirTime z L u , z × ACI R u , z , if LTE-U entrant cases of either legacy devices that do not implement DFS,or devices with faulty DFS implementation (e.g. erroneouslydetecting radar channels as occupied). As a worst-case of highlocal AP density corresponding to a high level of interference,we also consider the single channel case.For the indoor links we assume a multi-wall-and-floormodel (MWF) model [22] and for the outdoor links theITU-R model for line-of-sight (LOS) propagation within streetcanyons and for non-line-of-sight (NLOS) with over roof-top propagation [23]. The model also takes into account log-normal shadowing with a standard deviation of 4 dB for indoorlinks and 7 dB for outdoor links [24].We perform extensive Monte Carlo simulations in MATLABwith 3000 network realizations for the indoor/indoor scenarioand indoor/indoor scenario without internal walls , and 1500realizations for the outdoor/outdoor scenario. We assumedownlink saturated traffic (i.e. most challenging coexistencecase) and we evaluate the network performance based on thedownlink throughput per AP, estimated at the associated user. B. Throughput Model
Our throughput and interference model for co-channel in-terference is described in detail in [10] and in this paper weapply it to both co- and adjacent channel interference.For Wi-Fi and LAA, we assume the LBT mechanism doesnot allow co- and adjacent channel APs within CS range of each other to transmit simultaneously. Each of these APsis thus allowed to transmit for only an approximately equalfraction of time. The co- and adjacent channel APs located For multiple users associated to a single AP, the user throughput is obtainedby dividing the per-AP throughput to the number of associated users. The CS range within which co- and adjacent channel APs are located isdefined according to the respective CS thresholds given in Section IV-A. Wenote that the adjacent channel interference ratio (ACIR) is taken into accountfor adjacent channel power calculations. outside the CS range interfere by decreasing the signal-to-interference-and-noise-ratio (SINR) at the associated user.For LTE-U, the adaptive duty cycle MAC mechanism ad-justs the duty cycle of each AP based on the number of co- andadjacent channel APs detected within the CS range. However,the LTE-U APs within the same CS range may interfere witheach other, as they do not check if the channel is unoccupiedbefore transmitting. Instead, they transmit periodically, wherewe assume uncoordinated LTE-U APs that randomly selectthe starting moment of their duty cycle period, so that theirtransmissions may overlap in time. The Wi-Fi incumbentssense the medium unoccupied by coexisting LTE-U entrantsfor a duration determined by the entrants’ adaptive duty cycle,and the likelihood of their overlapping transmissions. Conse-quently, when coexisting with LTE-U entrants, the incumbentsdetect the medium unoccupied for a different fraction of timethan when coexisting with LAA entrants. The co- and adjacentchannel LTE-U APs located outside the CS range decrease theSINR at the associated incumbent or entrant user.In general we estimate the downlink throughput of an AP x according to our model in [10] as R x = S x × ( − r deg , x ) × AirTime x × ρ x ( SI N R u ) , (4)where S x is the LBT MAC protocol efficiency accountingfor sensing time and collisions between LBT frames basedon Bianchi’s model [25], r deg , x is the additional throughputdegradation due to collisions between LBT and duty cycleframes, AirTime x is the fraction of time that AP x is allowedto transmit according to its own and the other within-CS-rangeAPs’ MAC mechanisms, ρ x is the auto-rate function mappingthe SINR to the bit rate, and SI N R u is the SINR at theassociated user u of x . A mathematical description of theseparameters in given in Table II, where A x is the set of co-and adjacent channel incumbent APs within CS range of x , B x is the set of co- and adjacent channel entrant APs within CS range of x , | A x | is the number of co- and adjacent channelincumbent APs within the CS range of x , | B x | is the number ofco- and adjacent channel entrant APs within the CS range of x , | C y | is the number of co- and adjacent channel incumbentAPs within CS range of AP y , and | D y | is the number of co-and adjacent channel entrant APs within CS range of AP y .We estimate the SINR at the associated user u as SI N R u = P x ( L u , x ) − I cou + I adju + N , (5)where P x is the transmit power of AP x , L u , x is the propaga-tion loss between user u and AP x , I cou is the interference fromco-channel APs, I adju is the interference from adjacent channelAPs, and N is the background noise (assumed -174 dBm/Hz).A mathematical description of these terms is given in Table II,where A co is the set of all co-channel incumbent APs of x , A cox is the set of co-channel incumbent APs within CS rangeof x , B co is the set of all co-channel entrant APs of x , B cox is the set of co-channel entrant APs within CS range of x , A adj is the set of all adjacent channel incumbent APs of x , A adjx is the set of adjacent channel incumbent APs within CSrange of x , B adj is the set of all adjacent channel entrant APsof x , B adjx is the set of adjacent channel entrant APs withinCS range of x , P z is the transmit power of AP z , AirTime z is the fraction of time AP z may transmit (defined similarlyas AirTime x ), L u , z is the propagation loss between z and u ,and ACI R u , z is the adjacent channel interference ratio givenby z ’s transmitter at u ’s receiver when operating on adjacentchannels. We assume the model in [7] defining ACI R u , z as ACI R u , z = ACLR z + ACS u , (6)where ACLR z is the adjacent channel leakage ratio of trans-mitter z and ACS u is the adjacent channel selectivity ofreceiver u . For Wi-Fi APs and users we assume ACLR z =26 dBand ACS x = ACS u =22 dB, corresponding to the least efficientWi-Fi transmitter and receiver, whereas for the LTE-in-unlicensed variants we assume ACLR z =45 dB, ACS x =46 dB,and ACS u =22 dB [7], corresponding to the most efficient LTEAP transmitter and receiver, and the same LTE user receiveras for Wi-Fi.V. R ISK A SSESSMENT FROM THE W I -F I I NCUMBENT P ERSPECTIVE
In this section we present a selection of our simulationresults that illustrate the effectiveness of risk assessment forWi-Fi/LTE coexistence. Specifically, we evaluate the risk ofco- and adjacent channel interference for the Wi-Fi incum-bents and we show its relevance for spectrum regulators, i.e.in deciding whether regulatory action is required to ensureharmonious inter-technology coexistence, and for engineersdesigning and optimizing such networks.We apply the consequence metrics defined in Section III-Cas follows. The throughput degradation is estimated for each We note
ACS x is needed as the power received from adjacent channelsis also estimated at AP x , since its level may be high enough, such that x may detect the channel busy and may share its channel in time with (or adaptits duty cycle according to) transmissions in the adjacent channels. incumbent AP in each Monte Carlo network realization, re-sulting in a distribution of throughput degradation over allincumbents in all network realizations. Jain’s unfairness is es-timated for each network realization, over the set of incumbentthroughput values within a single network realization, resultingin a distribution of unfairness over all network realizations.In this section we first discuss how to read risk assessmentcharts in general and for our case study. Then we focus onindividually assessing the risk of interference for different net-work densities, channel availability, and deployment scenarios,from the Wi-Fi incumbent perspective. A. Reading Risk Assessment Charts
Risk assessment representations in general are likelihood-consequence charts where the curves show an increasing riskof harm from the lower left corner to the upper right corner,as indicated by the red arrow in the example of Fig. 2(b). Forthe Wi-Fi/LTE coexistence case, the likelihood-consequencecharts illustrate the risk of interference that the incumbentssuffer when coexisting with different entrants, by following thegeneral rule of increased risk towards the upper right corner.We represent the likelihood as the CCDF of the throughputconsequence metrics, for consistency with this rule.In our figures showing throughput degradation, e.g.Figs. 2(b) and 2(c), a positive throughput degradation isequivalent to an actual decrease in throughput comparedwith the considered baseline, whereas a negative throughputdegradation shows an increase in throughput.
B. Effect of Network Density
In this section we demonstrate the advantage of risk overconventional representation of our coexistence study resultswhen assessing interference for various network densities.Fig. 2 shows an example of conventional and risk represen-tations of incumbent throughput performance results, for the indoor/indoor scenario with 10 incumbents and 0–30 entrants,for single channel (i.e. co-channel interference only). Specif-ically, Fig. 2(a) shows an example of a conventional repre-sentation as the CDF of the incumbent AP throughput R x .When the number of entrants increases from 0 to 30, theincumbent throughput decreases from e.g. 10 to 2.5 Mbpsfor the median value. Also, Fig. 2(a) shows that for a fixednumber of entrants, the throughput of incumbents coexistingwith Wi-Fi entrants is sometimes higher and sometimes lowerthan when coexisting with LAA or LTE-U entrants. Thissuggests that LTE-in-unlicensed entrants are sometimes friendand sometimes foe to Wi-Fi, but does not readily providefurther insight. Although such a representation of the absolutethroughput is important for coexistence cases since it providesthe baseline for calculating the throughput degradation as arelative metric, the performance degradation caused by variousentrants cannot be quantified in a straightforward way.Fig. 2(b) shows the results in Fig. 2(a) in the form ofa likelihood-consequence chart, i.e. the CCDF vs. incum-bent throughput degradation with the standalone incumbentthroughput (i.e. no entrant) as baseline. Fig. 2(b) shows ingeneral that the risk increases significantly when the number Throughput per incumbent AP (Mbps) C D F NO entrant1 entrant: Wi-Fi5 entrants: Wi-Fi10 entrants: Wi-Fi30 entrants: Wi-Fi1 entrant: LAA5 entrants: LAA10 entrants: LAA30 entrants: LAA1 entrant: LTE-U adaptive duty cycle5 entrants: LTE-U adaptive duty cycle10 entrants: LTE-U adaptive duty cycle30 entrants: LTE-U adaptive duty cycle (a) Conventional representation -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100
Throughput degradation per incumbent AP (%) CC D F friendfoe (b) Risk representation -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 Throughput degradation per incumbent AP (%) CC D F (c) Risk representationFig. 2. Example of conventional (a) and risk (b) and (c) representationsof incumbent AP performance results for the indoor/indoor scenario, for single channel , for 10 incumbent and 0–30 entrant APs, as (a) distribution ofthroughput per incumbent AP; (b) distribution of throughput degradation perincumbent AP with the standalone incumbents as baseline; and (c) distributionof throughput degradation per incumbent AP with the incumbents coexistingwith Wi-Fi entrants as baseline. of entrants increases, irrespective of the entrant technology.The median incumbent throughput degradation increases from0% to 75% for 0 to 30 entrants. Also, for each numberof entrant APs there is a switching point where the orderof the curves corresponding to Wi-Fi and LAA or LTE-Uis reversed. Consequently, the risk of incumbent throughput degradation when coexisting with Wi-Fi entrants is sometimeshigher and sometimes lower than when coexisting with LTE-in-unlicensed. LTE-in-unlicensed is thus neither consistentlyfriend nor foe to Wi-Fi, suggesting the engineering policyquestion is moot.From a more detailed engineering perspective, it is evidentfrom Fig. 2(b) that the Wi-Fi entrants pose greater risk incase of lower negative impact, whereas the LAA and LTE-Uentrants pose greater risk in case of higher negative impact.Let us consider the example case of 30 entrants, where theswitching point occurs at a throughput degradation of 72%.For a throughput degradation lower than 72%, the risk posedby Wi-Fi entrants is higher than for LAA or LTE-U entrants,whereas for a throughput degradation higher than 72% theopposite holds. This effect occurs due to the value of the CSthreshold according to which the incumbent APs defer to theentrants, i.e. the incumbents apply a -82 dBm and -62 dBmthreshold to defer to Wi-Fi and LTE-in-unlicensed entrants,respectively ( cf . IEEE 802.11). For the lower CS thresholdthe incumbents are more conservative and avoid strong in-terference, by deferring to more entrants and transmitting lessoften. A lower CS threshold is thus suitable for (locally) densedeployments with strong interference, whereas it causes theincumbents to defer unnecessarily in sparse deployments withlow interference. The opposite holds for a higher CS threshold.Fig. 2(b) also shows that for a fixed number of entrants, thethroughput degradation from LTE-U and LAA is similar, soLTE-U and LAA are almost equally good neighbours to Wi-Fi.The risk is somewhat higher from LTE-U than LAA, due to theadditional collisions in term r x , deg and the adjustment of theentrant duty cycle based on the number of devices detected bythe entrants only. Consequently, some incumbents are allowedto transmit for a lower fraction of time than their equal sharewhen considering the number of APs in their own CS range. Finally, Fig. 2(b) shows that some of the incumbents have anegative throughput degradation when coexisting with entrantscompared with the standalone (i.e. no entrant) network. Thesecases are due to hidden nodes that are continuous sourcesof interference in the standalone incumbent network, butthat interfere only for a fraction of time when they deferto entrants deployed in the coexistence cases. However, thenegative throughput degradation is in some cases an artefactof our throughput model, where the MAC efficiency term S x is averaged over the entire CS range, sometimes resultingin higher average values for the incumbents when LTE-in-unlicensed entrants with higher S x are located within the CSrange.In order to focus directly on the question of whether LTE-in-unlicensed is friend or foe to Wi-Fi, Fig. 2(c) shows an alter-native risk representation of Fig. 2(b), where the baseline forincumbent throughput degradation is the incumbent throughputwhen coexisting with Wi-Fi entrants. A positive throughputdegradation thus corresponds to LTE being foe, whereas anegative throughput degradation corresponds to LTE being The opposite effect was shown in [10] for low incumbent and high entrantdensities, where the likelihood of short duty cycles and overlapping entranttransmissions is higher, such that the incumbents find the medium unoccupiedby entrants for a longer fraction of time.
Fairness (Jain's index) for incumbent APs C D F NO entrant1 entrant: Wi-Fi5 entrants: Wi-Fi10 entrants: Wi-Fi30 entrants: Wi-Fi1 entrant: LAA5 entrants: LAA10 entrants: LAA30 entrants: LAA1 entrant: LTE-U adaptive duty cycle5 entrants: LTE-U adaptive duty cycle10 entrants: LTE-U adaptive duty cycle30 entrants: LTE-U adaptive duty cycle (a) Conventional representation
Unfairness for incumbent APs CC D F (b) Risk representationFig. 3. Example of conventional (a) and risk (b) representations of incumbentAP performance results for the indoor/indoor scenario, for single channel , for10 incumbent and 0–30 entrant APs, as (a) distribution of Jain’s fairness indexfor incumbent APs in each network realization; and (b) distribution of Jain’sunfairness index for incumbent APs in each network realization. friend to Wi-Fi in unlicensed bands. For a given number ofentrants, the percentage of incumbents for which the entrantsare friends or foes is similar, with up to 50% being friends and50% foes for 30 entrants. This clearly shows that for the typical indoor/indoor scenario no regulatory intervention is required.In the rest of this paper we focus on the throughput degradationgiven the standalone incumbent network as baseline (as inFig. 2(b)), as this case provides better insight into the moregeneral network densification problem.Let us now consider the second consequence metric, i.e.Jain’s unfairness. Fig. 3 shows an example of conventional andrisk representations of Jain’s fairness/unfairness among incum-bents, for the indoor/indoor scenario with 10 incumbents and0–30 entrants, corresponding to the throughput degradation inFig. 2. Specifically, Fig. 3(a) shows a conventional represen-tation as the CDF of the fairness index J . For consistencywith the likelihood-consequence charts, Fig. 3(b) shows thesame results as Fig. 3(a) in the form of CCDF of throughputunfairness U , where the risk increases towards the upper rightcorner. We will thus comment only on Fig. 3(b). For a fixednumber of entrants, the risk of incumbent unfairness is higherfor LAA or LTE-U entrants than for Wi-Fi entrants, consistent with our results in Fig. 2(b), which show that the risk ofhigh throughput degradation is higher for LTE-in-unlicensed,resulting in larger variation of the throughput degradation.Also, the risk of unfairness increases with the number ofentrants for LAA or LTE-U, whereas it decreases for Wi-Fi,given the different CS thresholds that the incumbents apply.Moreover, the risk of unfairness decreases for Wi-Fi below therisk for the standalone incumbent network. Also, LTE-U has ahigher risk of unfairness compared with LAA, consistent withits higher throughput degradation for only some incumbents.Importantly, our results show that for single channel the riskis qualitatively different for the two considered consequencemetrics. The risk of throughput degradation (relevant for theengineering policy question) in Fig. 2(b) is sometimes higherand sometimes lower for coexistence with LAA or LTE-Uthan with Wi-Fi (i.e. LTE-in-unlicensed is sometimes friendand sometimes foe). By contrast, the risk of Jain’s unfair-ness among incumbents (relevant for engineering performanceoptimization) in Fig. 3(b) is always higher with LAA orLTE-U than with Wi-Fi (i.e. with Wi-Fi, all incumbents areaffected in a similar way). This illustrates the importanceof choosing a metric that effectively quantifies policy goals,as different metrics, encoding different values, may lead todifferent conclusions. C. Effect of Channel Availability
In this section we assess the risk of interference for theWi-Fi/LTE coexistence case, for different numbers of channels(i.e. with co- and adjacent channel interference) and channelselection schemes. Fig. 4 shows the incumbent throughputdegradation and Jain’s unfairness, for the indoor/indoor sce-nario, for 10 incumbents and 10 entrants (i.e. an example witha single AP in each apartment), and 1, 4 and 19 channelswith sense and random . The risk of throughput degradationin Fig. 4(a) increases when the number of channels decreases,from 0% median throughput degradation for 19 channels to40-50% median throughput degradation for single channel .For sense with the maximum number of 19 channels(typically available in practice), near-perfect coexistence isensured between incumbents and entrants (i.e. 0% incumbentthroughput degradation), due to the large number of unoccu-pied channels that the entrants can select from. Also, random with 19 channels has similar performance, with only a smallpercentage of incumbent APs suffering a rather low throughputdegradation. This shows that no regulatory or engineeringaction is needed to ensure harmonious coexistence. As anengineering insight, Fig. 4 reveals that sense does not bringsignificant benefit for such a high number of channels.For non-DFS devices operating on 4 channels with theentrants implementing sense , the throughput degradation issimilar to the one for 19 channels, whereas for 4 channelswith random the throughput degradation increases signifi-cantly, showing that engineers should implement sense forthe rare cases of such a low number of channels. Also, theswitching point delimiting the friend/foe entrants (explained inSection V-B) is visible for 4 channels random and for singlechannel ; for the other cases LTE-in-unlicensed is an equallygood or better neighbour to Wi-Fi than Wi-Fi is to itself. -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100
Throughput degradation per incumbent AP (%) CC D F (a) Unfairness for incumbent APs CC D F
10 entrants: Wi-Fi; 1 ch10 entrants: LAA; 1 ch10 entrants: LTE-U; 1 ch10 entrants: Wi-Fi; 4 ch; random
10 entrants: LAA; 4 ch; random
10 entrants: LTE-U; 4 ch; random
10 entrants: Wi-Fi; 4 ch; sense
10 entrants: LAA; 4 ch; sense
10 entrants: LTE-U; 4 ch; sense
10 entrants: Wi-Fi; 19 ch; random
10 entrants: LAA; 19 ch; random
10 entrants: LTE-U; 19 ch; random
10 entrants: Wi-Fi; 19 ch; sense
10 entrants: LAA; 19 ch; sense
10 entrants: LTE-U; 19 ch; sense (b)Fig. 4. Risk representation of incumbent AP performance results for the indoor/indoor scenario, for different number of channels , for 10 incumbentand 10 entrant APs, as (a) distribution of throughput degradation per incum-bent AP with the standalone incumbents as baseline; and (b) distribution ofJain’s unfairness index for incumbent APs in each network realization.
Fig. 4(b) shows the CCDF of Jain’s incumbent unfairnessfor 1 to 19 channels, where the unfairness increases when thenumber of channels decreases, with the exception of Wi-Fi, for single channel . The highest unfairness is caused by the LAA orLTE-U entrants for single channel , but for 4 and 19 channelsthe unfairness is similar to the one caused by Wi-Fi entrants,consistent with the similar throughput degradation results forall entrant technologies for these number of channels. Impor-tantly, for the typical 19 and also for 4 non-DFS availablechannels, both consequence metrics consistently show thatthere is no coexistence problem relevant for engineering policyor engineering optimization.
D. Effect of Deployment Scenario
This section shows the benefit of risk assessment whenquantifying the harm of interference in different scenarios,i.e. indoor/indoor , indoor/indoor without internal walls , and outdoor/outdoor . Fig. 5 shows how different scenarios affectthe incumbent throughput degradation and Jain’s unfairness for single channel , for 10 incumbents and 10 entrants. Importantly,Fig. 5(a) shows a consistent switching point between Wi-Fi and LAA or LTE-U curves across different scenarios at -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 Throughput degradation per incumbent AP (%) CC D F
10 entrants: Wi-Fi; in/in10 entrants: LAA; in/in10 entrants: LTE-U; in/in10 entrants: Wi-Fi; in/in w/o walls10 entrants: LAA; in/in w/o walls10 entrants: LTE-U; in/in w/o walls10 entrants: Wi-Fi; out/out10 entrants: LAA; out/out10 entrants: LTE-U; out/out (a)
Unfairness for incumbent APs (%) CC D F (b)Fig. 5. Risk representation of incumbent AP performance results for the indoor/indoor , indoor/indoor without internal walls , and outdoor/outdoor scenarios, for single channel , for 10 incumbent and 10 entrant APs, as (a) dis-tribution of throughput degradation per incumbent AP with the standaloneincumbents as baseline; and (b) distribution of Jain’s unfairness index forincumbent APs in each network realization. outdoor/outdoor scenario and thehighest risk of low degradation for the indoor/indoor sce-nario without internal walls ; (ii) the highest risk of highthroughput degradation is achieved for the outdoor/outdoor scenario and the lowest risk of high degradation for the indoor/indoor scenario without internal walls ; and (iii) forthe indoor/indoor scenario there is a moderate risk of highand low throughput degradation. This shows that the variationof incumbent throughput is highest in the outdoor/outdoor scenario, moderate for the indoor/indoor scenario, and low forthe indoor/indoor scenario without internal walls . This effectis consistent with the interference conditions in each scenario.For the indoor/indoor scenario without internal walls wherethe interference is high and the APs are located close to eachother, the incumbents detect more entrants and are able tobetter avoid strong interference by deferring to them, at the expense of sharing the channel in time. Specifically, almostall incumbents suffer a degradation of at least 20%, and forcoexistence with Wi-Fi entrants the incumbent degradationis constant and equal to 52%, as every incumbent detectsall incumbents and entrants within CS range and the MACefficiency also changes accordingly. In the outdoor/outdoor scenario the AP network deployment is more sparse and theusers are located at a wider range of distances from the APsthat they are associated with. Consequently, users close totheir corresponding APs experience low risk of degradation,but users far from their corresponding APs may face hiddennode problems (i.e. at least 15% of the APs have a throughputdegradation of 100%). The interference in the indoor/indoor scenario where the APs are separated by walls is moderatecompared with the other scenarios.We note that in our previous work [10], [11] we observedthat in the indoor/outdoor scenario, i.e. where the incumbentsare located indoors and the entrants are located outdoors,the incumbents and entrants are isolated from each other,due to the high attenuation through the external walls. Thecorresponding risk of interference from the entrants to theincumbents would therefore be zero, so we do not presentresults for this scenario in this paper.Fig. 5(b) shows Jain’s throughput unfairness among in-cumbents for different scenarios. Consistent with our resultsin Fig. 5(a) and the corresponding discussion, the lowestunfairness is achieved for the indoor/indoor scenario w/ointernal walls with down to zero unfairness for incumbentscoexisting with Wi-Fi entrants. A moderate risk of unfairnessis shown for the indoor/indoor scenario, whereas for the outdoor/outdoor scenario the unfairness is large. Also, for eachspecific scenario, the unfairness when coexisting with Wi-Fientrants is lower than when coexisting with LAA or LTE-Uentrants, consistent with the values of the CS threshold thatthe incumbents implement.VI. R ISK A SSESSMENT FROM THE
LTE- IN - UNLICENSED P ERSPECTIVE
In this section we apply risk analysis for Wi-Fi/LTE co-existence from the point of view of the LTE-in-unlicensedtechnology. In Section VI-A we show a selection of thethroughput results for LAA or LTE-U entrants coexistingwith Wi-Fi incumbents, as complementary results to thosefor the Wi-Fi incumbents in Section V. We note that theLTE-in-unlicensed entrant results are outside the scope of thetechnical policy question of whether LTE is friend or foeto Wi-Fi. However, they provide further engineering insightinto Wi-Fi/LTE coexistence, which can be used by LTE-in-unlicensed operators to decide which variant to deploy for bestnetwork performance. In Section VI-B we consider a furtherseparate case of LTE-in-unlicensed inter-operator coexistence,where there are no Wi-Fi incumbents. We thus explore thehypothetical future case where LTE is the dominant technologyin the 5 GHz unlicensed band and we analyse which choiceof technology is better suited from the operator perspectivewhen coexisting with other operators. -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100
Throughput degradation per entrant AP (%) CC D F (a) Unfairness for entrant APs CC D F
10 entrants: LAA; 1 ch10 entrants: LTE-U; 1 ch10 entrants: LAA; 4 ch; random
10 entrants: LTE-U; 4 ch; random
10 entrants: LAA; 4 ch; sense
10 entrants: LTE-U; 4 ch; sense
10 entrants: LAA; 19 ch; random
10 entrants: LTE-U; 19 ch; random
10 entrants: LAA; 19 ch; sense
10 entrants: LTE-U; 19 ch; sense (b)Fig. 6. Risk representation of entrant
AP performance results for the indoor/indoor scenario, for different number of channels , for 10 entrantand 10 Wi-Fi incumbent APs, as (a) distribution of throughput degradationper entrant AP with the standalone incumbents as baseline; and (b) distributionof Jain’s unfairness index for entrant APs in each network realization.
A. Wi-Fi/LTE Coexistence from the Entrant Perspective
In this section we present a selection of the throughput re-sults for the LTE-in-unlicensed entrants when coexisting withWi-Fi incumbents, for the scenarios in Section IV-A. For theentrants we apply the same throughput consequence metrics asthose defined for the incumbents in Section III-C. Specifically,we consider the throughput degradation per entrant, where thebaseline is the throughput of the standalone entrants (i.e. whenthere is no incumbent), and the unfairness among entrants.Fig. 6 shows the risk representation of entrant throughputresults, for the indoor/indoor scenario with 10 entrant APscoexisting with 10 incumbent APs, for different numbersof channels and channel selection schemes. Fig. 4 showsthe incumbent results for the same scenario and channelconfigurations. The throughput degradation for LAA entrantsin Fig. 6(a) is consistently but marginally higher than for LTE-U entrants, for single channel (by up to 15 percentage pointsfor the 70 th percentile). This shows that the LAA entrants areaffected more by the coexistence with Wi-Fi incumbents, asthe LAA entrants have to defer to the neighbouring incum-bents and thus transmit for a shorter time, according to theimplemented LBT mechanism. Although the LTE-U entrants also reduce their duty cycle transmission time according tothe number of detected incumbents, the likelihood of randomoverlapping transmissions from the neighbouring LTE-U en-trants causing strong interference is also reduced comparedto the standalone LTE-U entrants. The resulting throughputdegradation is thus lower for LTE-U than for LAA. Althoughthis metric shows that LAA performs worse than LTE-U,we note that this is a relative metric, which depends on theabsolute throughput value for LAA and LTE-U, respectively.The entrant throughput degradation can thus be consideredin practical cases for e.g. determining the risk for an LTEvariant when coexisting with Wi-Fi in shared spectrum bandsvs. operating in dedicated spectrum bands. For cases wherethe absolute throughput is more relevant, we provide furtherentrant results in [10], [11].For a larger number of available channels (i.e. 4 or 19),LAA and LTE-U have a similar throughput degradation dueto Wi-Fi incumbents, showing that both LTE-in-unlicensedvariants coexist equally well with the incumbents for typicalcoexistence cases in the 5 GHz band. We note that a marginaldifference in throughput degradation can be observed by com-paring the results for LAA or LTE-U for 4 channels random against sense . A fraction of 13% of the entrants experience anegative throughput degradation (i.e. throughput increase) and37% a positive throughput degradation with sense , whereasonly 2% of the entrants experience a negative throughputdegradation and 30% a positive throughput degradation with random . The throughput degradation thus varies less with random than with sense . This effect occurs due to the higherdynamics of the sense channel selection mechanism, comparedto random . Specifically, with random each of the entrantAPs always transmits on the same randomly selected channel,whereas with sense each of the entrant APs selects a differentchannel when coexisting with incumbent APs. Consequently,when estimating the throughput degradation for entrants with sense , the throughput obtained for coexistence is based ona different channel allocation than the baseline throughputfor standalone entrants. Moreover, with sense the entrantsmay cause more mutual interference among themselves, byavoiding the same channels occupied by the incumbents, butthe mutual interference between incumbents and entrants isthus reduced. As such, the overall interference at the entrantsdepends on the specific deployment and the channels selectedby the incumbents. For 19 available channels we observethe same trend as for 4 channels. However, the throughputdegradation varies even less, due to the low number of co-channel APs.The unfairness for the LTE-U entrants in Fig. 6(b) ismarginally but consistently higher than for LAA, for singlechannel (up to 0.5 difference in unfairness). This effect iscaused by the random duty cycle transmissions of neighbour-ing LTE-U entrants, which can cause strong mutual inter- We note that with sense the entrant APs first check which channels arealready occupied by Wi-Fi incumbents and then select an unoccupied channel( cf.
Section IV-A), whereas for entrants implementing random , the channeloccupation is irrelevant when selecting a channel. In practice, random and sense would correspond to quasi-static channel allocation and frequent channel(re-)allocation, respectively. -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100
Throughput degradation per entrant AP (%) CC D F
10 entrants: LAA; in/in10 entrants: LTE-U; in/in10 entrants: LAA; in/in w/o walls10 entrants: LTE-U; in/in w/o walls10 entrants: LAA; out/out10 entrants: LTE-U; out/out (a)
Unfairness for entrant APs CC D F (b)Fig. 7. Risk representation of entrant AP performance results for the indoor/indoor , indoor/indoor without internal walls , and outdoor/outdoor scenarios, for single channel , for 10 entrant and 10 Wi-Fi incumbent APs, as(a) distribution of throughput degradation per entrant AP with the standaloneentrants as baseline; and (b) distribution of Jain’s unfairness index for entrantAPs in each network realization. ference, if they overlap in time. By contrast, LAA entrantsimplementing LBT always defer to other neighbouring trans-missions and avoid such strong interference, resulting in alower variation of the entrant throughput. A similar marginaldifference between LAA and LTE-U unfairness occurs for4 available channels. For the typical case of 19 availablechannels in the 5 GHz band, the unfairness is similar forLAA and LTE-U entrants, as the number of co-channel APsis considerably reduced.Fig. 7 shows the risk representation of entrant throughputresults, for 10 entrant and 10 incumbent APs, for differ-ent scenarios and single channel ( cf. incumbent results inFig. 5). Fig. 7(a) shows that for the indoor/indoor scenario without internal walls , the throughput degradation for theLAA entrants is significantly higher than for the LTE-Uentrants (e.g. 35 percentage points difference for the medianthroughput degradation). This shows that, as discussed for the indoor/indoor scenario in Fig. 6(a), the LTE-U entrants are lessaffected by coexistence with incumbents than LAA entrants.This difference in LAA vs. LTE-U throughput degradationis more pronounced for the indoor/indoor scenario withoutinternal walls than for the indoor/indoor scenario, due to the stronger interference among APs in the open-plan indoorscenario. The LAA entrants thus defer to more APs, whereasthe LTE-U entrants already suffer from strong interferencefrom other LTE-U entrants, such that the coexistence with Wi-Fi incumbents does not affect them to the same extent thatLAA is affected. Moreover, we note that 6% of the LTE-Uentrants in the indoor/indoor scenario without internal walls have 0 Mbps throughput (i.e. the distribution curve does notreach 1 on the y axis). For the outdoor/outdoor scenario,the LAA throughput degradation is larger than for LTE-U,consistent with the trend observed for the other scenarios.However, the difference in throughput degradation for LAAvs. LTE-U is only marginal, given the sparser deployment withfewer neighbouring APs.Fig. 7(b) shows that the unfairness for the indoor/indoor scenario without internal walls is significantly higher for LTE-U than for LAA (i.e. unfairness of up to 0.18 for LAA andup to 0.65 for LTE-U), due to the high likelihood of stronginterference among neighbouring LTE-U entrants. By contrast,the difference between the LTE-U and LAA unfairness forthe outdoor/outdoor scenario is much lower (i.e. at most 0.1difference), consistent with the throughput degradation resultsin Fig. 7(a). However, the unfairness for LTE-in-unlicensedentrants is overall larger for the outdoor/outdoor scenario thanthe others, due to the wider range of distances between the APsand the respective associated users, which results in a largervariation of the received power. Furthermore, hidden terminalproblems are more likely for this scenario.Overall, our LTE-in-unlicensed entrant results show thatfor the typically large number of available channels in the5 GHz band, LAA and LTE-U coexist equally well with Wi-Fi incumbents. However, for dense deployments with a largenumber of co-channel APs, the two considered consequencemetrics show different results: from the perspective of thethroughput degradation, LTE-U coexists better with Wi-Fi thanLAA; from the perspective of the unfairness, LTE-U coexistsworse with Wi-Fi than LAA. We note, however, that theseresults are also due to the interactions of the entrants amongthemselves, according to the implemented MAC mechanism. B. LTE-in-unlicensed Inter-Operator Coexistence
In this section we use risk analysis to also explore LTE-in-unlicensed inter-operator coexistence. We consider the hy-pothetical future case where two operators, i.e.
Operator A and
Operator B , deploy APs of the same LTE-in-unlicensedtechnology, i.e. either LAA or LTE-U. We thus focus ontwo major inter-operator coexistence cases, as determined bydifferent regional regulatory requirements for the MAC inthe 5 GHz unlicensed band. For regions where LBT is notrequired, it is expected that the operators will initially deployLTE-U, as a less complex variant of LTE-in-unlicensed, forwhich compliant devices are already available [26], [27]. Forregions where LBT is required, the operators have to imple-ment LAA, so only LAA/LAA inter-operator coexistence ispossible. Importantly, such coexistence cases are of interest tooperators, since inter-operator coordination cannot be achievedin a straight-forward manner and thus LTE-in-unlicensed inter- operator coexistence may pose similar problems to Wi-Fi/LTEcoexistence.We note that LAA/LTE-U inter-operator coexistence mayalso occur, in regions where LBT is not required. Althoughwe do not specifically consider this case in our analysis,LAA/LTE-U coexistence is similar to the Wi-Fi/LTE-U co-existence case analysed in Sections V and VI-A, as both LAAand Wi-Fi implement LBT with a CS threshold of -62 dBmfor deferring to other LTE-in-unlicensed APs. For evaluatingcoexistence between other variants of LBT (e.g. with differentCS thresholds) when coexisting with duty cycle devices, weencourage the reader to use our publicly-available simulationtool [4].We consider the scenarios in Section IV-A with the samelocations as previously. For the APs of Operator A we considersimilar throughput consequence metrics as defined for theincumbents in Section III-C. Specifically, we consider thethroughput degradation per Operator A AP, where the baselineis the throughput of the standalone Operator A (i.e. OperatorB is not active), and the unfairness among Operator A APs.We present a selection of our results for the Operator A APs,similar to the Wi-Fi incumbent results in Section V and theLTE entrant results in Section VI-A.Fig. 8 shows the risk representation of Operator A through-put results, for the indoor/indoor scenario with 10 OperatorA APs and 10 Operator B APs, for different numbers ofchannels and different channel selection schemes. Fig. 8(a)shows that the throughput degradation per Operator A APis similar for LAA and LTE-U, regardless of the numberof available channels. However, for single channel and 4available channels, the risk of high degradation (i.e. morethan 50%) is somewhat larger for LTE-U than for LAA,due to random, overlapping LTE-U duty cycle transmissionsfrom neighbouring APs, which are avoided by LAA. Asdiscussed for Fig. 7(a), there is a larger variation in throughputdegradation for 4 channels with sense vs. random , due to themore dynamic sense channel selection scheme.The unfairness among Operator A APs in Fig. 8(b) isconsistently larger for LTE-U than for LAA, regardless of thenumber of available channels, as discussed for the unfairnessin Fig. 6(b). However, the difference between the LAA andLTE-U unfairness reduces when increasing the number ofavailable channels (i.e. up to 0.12 difference for single channel and up to 0.05 for 19 channels), due to the low number of co-channel APs. Moreover, we note that the unfairness does notchange for random vs. sense for LAA and LTE-U, respectively,regardless of the number of available channels. This showsthat the range of long-term average throughput per AP over agiven network realization is not sensitive to the dynamics ofthe channel selection scheme.In general, we observe that the choice of LBT or adap-tive duty cycle MAC mechanism only marginally affects thethroughput degradation of the entrants in the indoor/indoor scenario, for the realistic cases of 4 or 19 available channels.Consequently, network operators can deploy LTE-U or LAAdevices with equivalent network throughput performance. Wenote that the dynamics of the channel selection scheme has aslightly stronger impact on the throughput degradation than the -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 Throughput degradation per Operator A AP (%) CC D F (a) Unfairness for Operator A APs CC D F LAA: Operator A & Operator B; 1 chLTE-U: Operator A & Operator B; 1 chLAA: Operator A & Operator B; 4 ch; random
LTE-U: Operator A & Operator B; 4 ch; random
LAA: Operator A & Operator B; 4 ch; sense
LTE-U: Operator A & Operator B; 4 ch; sense
LAA: Operator A & Operator B; 19 ch; random
LTE-U: Operator A & Operator B; 19 ch; random
LAA: Operator A & Operator B; 19 ch; sense
LTE-U: Operator A & Operator B; 19 ch; sense (b)Fig. 8. Risk representation of LTE-in-unlicensed
Operator A
AP performanceresults for the indoor/indoor scenario, for different number of channels ,for 10 Operator A and 10 Operator B APs, as (a) distribution of throughputdegradation per Operator A AP; and (b) distribution of Jain’s unfairness indexfor the Operator A APs in each network realization. choice of MAC mechanism. However, this impact is reflectedby our throughput degradation consequence metric, due toits sensitivity to per-AP variations of the selected channel intime, whereas the unfairness capturing the long-term averagethroughput results over the entire network remains unchanged.This highlights the importance of selecting the consequencemetric that reflects the engineering design goal of a particulardeployment.Fig. 9 shows the risk representation of Operator A through-put results, for 10 Operator A and 10 Operator B APs, fordifferent scenarios and single channel . Fig. 9(a) shows thatfor any of the considered scenarios, there is a switchingpoint between the throughput degradation curves for LAAand LTE-U. There are thus two regimes: of higher risk oflow degradation for LAA vs. LTE-U; and of higher risk ofhigh degradation for LTE-U vs. LAA. These results show thatLAA implementing LBT protects the APs better against stronginterference, unlike LTE-U with adaptive duty cycle, for whichthe APs are more likely to suffer from strong interference.Fig. 9(b) shows that the unfairness for LTE-U is significantlylarger than for LAA, regardless of the scenario, consistent withthe results in Fig. 9(a). The largest difference in unfairness -100-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100
Throughput degradation per Operator A AP (%) CC D F LAA: Operator A & Operator B; in/inLTE-U: Operator A & Operator B; in/inLAA: Operator A & Operator B; in/in w/o wallsLTE-U: Operator A & Operator B; in/in w/o wallsLAA: Operator A & Operator B; out/outLTE-U: Operator A & Operator B; out/out (a)
Unfairness for Operator A APs CC D F (b)Fig. 9. Risk representation of Operator A
AP performance results for the indoor/indoor , indoor/indoor without internal walls , and outdoor/outdoor scenarios, for single channel , for 10 Operator A and 10 Operator B APs, as(a) distribution of throughput degradation per Operator A AP; and (b) distribu-tion of Jain’s unfairness index for Operator A APs in each network realization. between LAA and LTE-U is observed for the indoor/indoor scenario without internal walls (i.e. up to 0.2 unfairness forLAA and up to 0.8 unfairness for LTE-U).Our results show in general that for very dense deploymentswith a large number of LTE-in-unlicensed APs deployed bydifferent operators (i.e. uncoordinated co-channel APs), it ismore beneficial to deploy LAA than LTE-U, regardless ofthe scenario, since LAA is more robust against interferenceand achieves a more uniform throughput among APs. Wenote that this is, of course, consistent with the use of LBTas the preferred random access MAC mechanism for hightraffic load and high density scenarios (e.g. in Wi-Fi). Bycontrast, uncoordinated devices implementing adaptive dutycycle are likely to cause strong interference to each other fordense deployments and high traffic load. We emphasize thatthis result is relevant for recent regulatory discussions [9]. Forfrequency bands like the 5 GHz unlicensed band, where manychannels are available, deploying LAA or LTE-U (i.e. LBT oradaptive duty cycle MAC in general) in practice results in asimilar throughput performance. VII. C
ONCLUSIONS
In this paper we presented a case study of Wi-Fi/LTE coex-istence in the 5 GHz band, in order to demonstrate the valueof risk-informed interference assessment in making regulatorydecisions and for providing engineering insight. We appliedrisk assessment methods to this coexistence problem from bothincumbent and entrant perspectives by (i) identifying co- andadjacent channel interference as hazard modes, (ii) definingthe throughput degradation and Jain’s throughput unfairnessas consequence metrics, and (iii) assessing the likelihood andconsequence for different network densities, numbers of avail-able channels, and scenarios (i.e. indoor/indoor , indoor/indoorwithout internal walls , and outdoor/outdoor ). We performedextensive Monte Carlo simulations for Wi-Fi incumbents co-existing with LTE-in-unlicensed entrants and we estimatedthe downlink throughput by considering co- and adjacentchannel interference. Furthermore, we highlighted our newlypublicly-available network simulation tool for risk assessmentof Wi-Fi/LTE coexistence [4].We demonstrated that risk assessment is an effective methodfor evaluating the harm caused by interference in a compre-hensive and intuitive manner. Our analysis clearly showedthat LTE-in-unlicensed is neither friend nor foe to Wi-Fi ingeneral, and thus that no regulatory intervention is needed toensure harmonious technical coexistence. From an engineeringperspective, our results showed that Wi-Fi incumbents suffer alower risk of interference when coexisting with Wi-Fi entrantscompared with LTE-in-unlicensed entrants in locally densedeployments, but the opposite holds for sparse deployments,due to the Wi-Fi MAC design. Also, for the high num-ber of available channels expected in practice, there is anegligible risk of interference for Wi-Fi incumbents fromLTE-in-unlicensed entrants, which renders both policy andengineering coexistence issues largely irrelevant. From theLTE-in-unlicensed entrant perspective, both LAA and LTE-Uvariants coexist equally well with Wi-Fi incumbents. For LTEintra-technology inter-operator coexistence, both variants typi-cally coexist well in the 5 GHz band. However, for very densedeployments, LAA causes less mutual interference betweenoperators than LTE-U, due to implementing LBT.R EFERENCES[1] A. M. Voicu, L. Simi´c, J. P. de Vries, M. Petrova, and P. M¨ah¨onen,“Analysing Wi-Fi/LTE coexistence to demonstrate de value of risk-informed interference assessment,” in
Proc. IEEE DySPAN , Baltimore,2017.[2] Y.-C. Liang, K.-C. Chen, G. Y. Li, and P. M¨ah¨onen, “Cognitive radionetworking and communications: An overview,”
IEEE Trans. on Vehic-ular Technology , vol. 60, pp. 3386 – 3407, Sept. 2011.[3] J. P. de Vries, “Risk-informed interference assessment: A quantitativebasis for spectrum allocation decisions,”
Telecommunications Policy ,2017. [Online]. Available: http://dx.doi.org/10.1016/j.telpol.2016.12.007[4] iNETS inter-technology wireless coexistence risk assessment tool
IEEE J. Sel. Areas Commun. , vol. 34, no. 11, pp.3062–3077, Nov. 2016.[11] L. Simi´c, A. M. Voicu, P. M¨ah¨onen, M. Petrova, and J. P. de Vries, “LTEin unlicensed bands is neither friend nor foe to Wi-Fi,”
IEEE Access ,vol. 4, pp. 6416–6426, Sept. 2016.[12] FCC Technological Advisory Council, “A quick introduction torisk-informed interference assessment,” V 1.00, April 2015. [Online].Available: http://transition.fcc.gov/bureaus/oet/tac/tacdocs/meeting4115/Intro-to-RIA-v100.pdf[13] J. P. de Vries, “Risk-informed interference assessment: A quantitativebasis for spectrum allocation decisions,” in
Proc. TPRC , Arlington,2015. [Online]. Available: http://ssrn.com/abstract=2574459[14] J. P. de Vries, U. Livnat, and S. Tonkin, “A risk-informed interferenceassessment of MetSat/LTE coexistence,”
IEEE Access , Mar. 2017.[Online]. Available: https://doi.org/10.1109/ACCESS.2017.2685592[15] Wi-Fi Alliance, “Coexistence test plan,” V 1.0, 2016.[16] R. K. Jain, D.-M. W. Chiu, and W. R. Hawe, “A quantitativemeasure of fairness and discrimination for resource allocation in sharedcomputer system,” submitted to
ACM Transaction on Computer Systems ,September 1984. [Online]. Available: https://arxiv.org/abs/cs/9809099[17] Alcatel-Lucent, picoChip Designs, and Vodafone, “Simulation assump-tions and parameters for FDD HeNB RF requirements,” May 2009,3GPP TSG RAN WG4 Meeting 51, R4-092042.[18] A. Achtzehn, L. Simi´c, P. Gronerth, and P. M¨ah¨onen, “Survey ofIEEE 802.11 Wi-Fi deployments for deriving the spatial structure ofopportunistic networks,” in
Proc. IEEE PIMRC,
London, 2013.[19] Mozilla Location Service, August 2015. [Online]. Available:https://location.services.mozilla.com/downloads.[20]
IEEE Standard for Information technology - Telecommunications andinformation exchange between systems; Local and metropolitan areanetworks - Specific requirements; Part 11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) Specifications , IEEE Std.802.11, Mar. 2012.[21] 3GPP, “E-UTRA; Radio Frequency (RF) system scenarios,” TR 36.942V8.2.0, July 2009.[22] M. Lott and I. Forkel, “A multi-wall-and-floor model for indoor radiopropagation,” in
Proc. IEEE VTC , Rhode, 2001.[23] ITU-R, “Propagation data and prediction methods for the planning ofshort-range outdoor radiocommunication systems and radio local areanetworks in the frequency range 300 MHz to 100 GHz,” Recommenda-tion P.1411-7, Sept. 2013.[24] 3GPP, “E-UTRA; Further advancements for E-UTRA physical layeraspects (Release 9),” TR 36.814 V9.0.0, Mar. 2010.[25] G. Bianchi, “Performance analysis of the IEEE 802.11 distributedcoordination function,”
IEEE J. Sel. Areas Commun. , vol. 18, no. 3,pp. 535–547, Mar. 2000.[26]
Ericsson AB RBS 6402 LTE Base Station , “FCC IDTA8AKRD90106083,” FCC ID Database. [Online]. Available:https://fccid.io/TA8AKRD90106083[27]