Assessment of Treatment Influence in Mobile Network

Received August 6, 2020, accepted August 26, 2020, date of publication September 3, 2020, date of current version September 17, 2020.

Digital Object Identifier 10.1109/ACCESS.2020.3021647

Assessment of Treatment Influence in MobileNetwork Coverage on Board High-Speed TrainsORLANDO S. TRINDADE1, TAULANT BERISHA2, PHILIPP SVOBODA 1, (Senior Member, IEEE),EFSTATHIA BURA1, AND CHRISTOPH F. MECKLENBRÄUKER 1, (Senior Member, IEEE)1TU Wien, 1040 Vienna, Austria2Dimetor GmbH, 1040 Vienna, Austria

Corresponding author: Philipp Svoboda ([email protected])

This work was supported by the TU Wien Bibliothek through its Open Access Funding Program.

ABSTRACT In this paper, we investigate the systematic change in the signal quality of mobile telecom-munication systems inside a high-speed train. Using commercial mobile phones, we perform measurementsinside a carriage of a high-speed train with an installed repeater and inside a carriage of a different high-speed train equipped with low radio frequency attenuation windows. We carry out additional referencemeasurements inside regular carriages of a high speed train. We use the adjusted Harrell–Davis methodand the nonparametric bootstrap to test multiple quantiles of samples from the key performance indicatorsproduced by the commercial mobile phones. In addition, we propose a common framework to standardizefuture comparisons on such studies. The results confirm considerable improvements in signal qualityresulting from both the repeater and low RF attenuation windows.

INDEX TERMS Vehicular channel characterization, inferential statistics, mobile network measurement,mobile network service quality, RF signal, high-speed trains, multiple hypotheses, multiple quantile testing.

I. INTRODUCTIONOver the last decades, mobile terminals have become anessential communication device. The current mobile networkinfrastructure supports large-scale economies in a variety ofways. Commuters and travelers depend on reliable wirelessconnections to perform a variety of tasks that used to be tiedto fixed access points. This is especially true for professionalsand commuters who use High-speed Trains (HSTs) to travelfor long periods. In this sense, it is critical for mobile serviceproviders to monitor their mobile network periodically toassure that consumers are providedwith a dependable, secure,and high-quality service network [1]. Enabling mobile ser-vice providers to accurately access the improvements of theirmobile services ultimately determines their ability to enhanceuser experience [2]. Key Performance Indicators (KPIs) arethe metrics monitored in network infrastructures and MobileStation (MS). The KPIs are measured through the physicaland logical channels of the mobile communication protocolto enable providers to deploy, operate, and improve mobilecommunication networks efficiently.

The associate editor coordinating the review of this manuscript and

approving it for publication was Mauro Fadda .

KPIs are metrics estimated by the network infrastructureand the MS based on the measurements derived from thephysical and logical channels of the mobile communicationprotocol stack. Large-scale Key Performance Indicator (KPI)measurements and advanced statistical analyses are necessaryto deploy, operate, and improve mobile telecommunicationnetworks efficiently.

The KPI statistics of a mobile communication networkvary over space and time; KPI statistics also change dueto network improvements, varying loads, and disruptions inmobile networks.

However, a statistically sufficient characterization of chan-nels under High-speed Train (HST) context is yet to be pro-posed [3]. It is crucial to detect even small changes in the sta-tistical properties of KPIs that are noticeable by commercialusers of mobile services in the network.

Channel characterization becomes challenging due to: 1)the variety of frequency bands of mobile communicationsystems; 2) the velocity of the HST under investigation peaksaround 250 km/h, and forecasts indicate that this velocitywould further increase in the future; and [4]; 3) the large num-ber ofmobile users on boardHSTs. Accordingly, network andtransceiver designs in the several layers of the protocol stackneed to be improved to meet these challenges [5].

VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 162945

https://orcid.org/0000-0002-2277-0378

https://orcid.org/0000-0001-9571-0379

https://orcid.org/0000-0002-0615-6546

O. S. Trindade et al.: Assessment of Treatment Influence in Mobile Network Coverage on Board High-Speed Trains

Some of the significant issues on the physical layer ofthe protocol stack include small-scale fading, large-scalefading, and the time and frequency selectivity [5]–[7]. Onthe higher layers of the protocol stack, the so-called signal-ing storm is caused by the large number of mobile usersthat simultaneously access the network, which overloads theavailable signaling resources [5]. The diverse nature of thosechallenges impede mobile network providers from deliveringhigh-quality services to consumers while on board HSTs. Asa result, this motivates researchers to conduct measurementcampaigns in an attempt to address those challenges. Eval-uations performed in [8] report degradations in the form ofpacket drops, network disconnections, and Round Trip Time(RTT) at velocity ranging 200–300 km/h. A similar study in[9] has found sufficient throughput up to 100 km/h; how-ever, service quality significantly degenerated at velocitiesof 300 km/h and beyond.

Commercially available solutions to improve network onboard HST vary in applicability, achievable gains, and com-plexity. We first distinguish between active and passive net-work improvements. Among the active improvements, thereare the amplify-and-forward repeaters [10], which are trans-parent to the protocols, and the decode-and-forward repeaters[11]. Passive improvements constitute structural changes thatreduce signal penetration loss [12], [13]. The work in [14]attempted to optimize handover. The studies in [15] and [16]propose an improved resource-scheduling scheme to reduceQuality of Service (QoS) problems. These studies evaluatethe proposed scheme under static-, low-, and high-mobilitycontexts. In terms of statistical evaluation, the majority of thepreviously mentioned analyses have relied solely on descrip-tive statistics of the KPIs. An exception is the study in [17],where the authors suggest using nonparametric hypothesistesting and then divide the traveled trajectory into kilometer-long segments in order to evaluate the collected KPIs.

A. CHALLENGESThe KPIs in mobile networks vary at different time-spacescale, and this poses the first challenge in KPI evaluation.The variations in a KPI due to a specific modification in amobile network do not occur in isolation, but results from acombination of many different factors. Some of these factorsare due to variations in space and time of the propagationchannel, in the traffic load in the mobile network, and in themeasurement sample [18], [19]. Secondly, the characteristicsof the implemented modification need to be first evaluatedbefore the KPIs of interest are selected. KPIs differ fromone another in terms of the underlying statistical distributionand data type. This becomes a challenge for researcherswhen the KPIs need to be compared among diverse studies.Thirdly, the length of the trajectory, velocity of the vehicle,and the KPI’s recording rate produces large data sets withlarge variations in KPI samples. Some KPIs have samplesizes in the order of tens of thousands, whereas others havejust a few hundreds, therefore making it difficult to performstatistical comparisons. Furthermore, the mobile networks

contain several Radio Access Technologies (RATs) that havediscontinuities along the sampled trajectories [20].

For operational reasons, we cannot control the factors men-tioned above, and any statistical inference concerning a KPIneeds to be robust against these challenges. For this reason,we do not assume any specific form of KPI distribution. Thismotivates the use of nonparametricmethods [21] to detect andinfer changes in the KPIs of mobile communication networksunder operational constraints and high mobility. In addition,the lack of standard terminologies among studies that focuson detecting change in mobile networks is another constraint.This undermines the comparability of results from the currentliterature and those from new studies. These conditions makeit difficult for researchers to develop a protocol-independentmethod that can detect changes in the QoS of the mobilenetwork as perceived by the user.

B. CURRENT WORKWe rely on inferential statistics as it enables comparingprobability distributions of the underlying data generatingprocesses.

In addition, we introduce terminology to standardize com-parisons among similar studies. Likewise, the present con-tribution proposes an empirical KPI assessment methodol-ogy based on distribution-free hypothesis testing combinedwith the bootstrap [22], [23]. More specifically, the proposedscheme uses the adjusted Harrell–Davis (HD) estimator [24]developed byWilcox [25] which can compare multiple quan-tiles of two distributions simultaneously. This method helpsus to determine more accurately whether the distributionsdiffer and how they differ from each other.

In the experiment, the unmodified HST carriage is calledthe placebo carriage. In an experimental setup, KPI samplesare collected from a modified carriage and from a placebocarriage. We performed measurement in two experimentalsetups: repeater and low Radio Frequency (RF) attenuationwindows. We compared the KPI distribution under the influ-ence of a repeater and the KPI distribution under the influenceof low RF attenuation windows with that of the placebo. Thestatistical analysis indicates that signal quality significantlyimproved.

This paper is organized as follows: Sec. II defines theproposed terminology, formulates the problem statement, anddescribes the collected data. Section III discusses our methodfor analyzing the data, whereas Sec. IV describes two casestudies where the proposed method is applied. The results arediscussed in Sec. IV-F, and we conclude with some remarks.

II. ASSESSING THE KEY PERFORMANCE INDICATORS OFMOBILE COMMUNICATION SCENARIOSConducting an effective empirical assessment of KPIsrequires understanding the phenomena that induce thechanges in these KPIs. In the following subsections, we intro-duce a terminology from inferential statistics, describe themobile communication scenarios, and characterize the col-lected data.

162946 VOLUME 8, 2020


A. DEFINITIONS AND TERMINOLOGYThe following definitions characterize various measurementcontexts in a mobile network. Figure 1 illustrates two differ-ent contexts. Fig.1 (a) shows the controlled condition wherethere is no mobility. Under the controlled condition, it is pos-sible to isolate variables to a large degree. The uncontrolledcondition is illustrated in Fig.1 (b). In this scenario the mobilenetwork infrastructure is static and the users are mobile. Incontext (b), the variable isolation is limited and the samplesdisplay non-stationary statistical properties.

We further define the velocities of the transceivers v(p1)and v(p2) and the relative velocity as vr = v(p1) − v(p2).When the transceivers are fixed, vr = 0, we name it asstatic conditionwhich is identical to the controlled conditionsin Fig.1(a).

The non-static condition depicts the case in which eitherp1 or p2 moves along the traveled trajectory and vr 6= 0as illustrated in Fig.1 (b). In terms of space, we define V asa volume equivalent to an indoor environment. When V islarge, it can be further divided into spatial sections to improvethe representation of the context under evaluation.

Figure 2 illustrates the modifications done on the mobilenetworks or on a vehicle similar to those shown in Fig. 1. Anysystematic modification is called a treatment and is denotedas Q. The context without systematic modification is calledplacebo q.The effect of a treatment is called treatment effect and

its magnitude is the effect size. Assessing a treatment effectrequires comparing it with at least one of the following:1) reference values, 2) absence of treatment or placebo q,3) an alternative treatment e.g. Q1. In addition, we use y toindicate the observations from a treatment and x to denoteplacebo-related observations. The treatments, samples, andthe comparison with reference values are shown in Fig. 2 (d).

A treatment without active elements is called passive treat-ment. The effect of such treatment is related to the changesin the characteristics of volume V, such as building materialsand internal and external geometries [26], [27]. The treatmenteffect, which in this case is the Vehicular Penetration Loss(VPL), is estimated by the signal power levels Pinside andPoutside [28]. In controlled conditions, we control the trans-mitted signal power PTX at the transmitting antenna. Thus,we get a good estimate of PTX , namely PTX , instead of usingthe estimated Poutside [29]. Since the volume’s walls representthe treatment Q, we can estimate the treatment effect by

LQ(dB) = −10 log10

(PinsidePTX

)(dB) (1)

where LQ represents the transmission loss LQ due to a passivetreatment.

A treatment composed of active systems that superimposespower gain on the signal levels is called active treatment. Inthe controlled condition, we assess the treatment effects via

the signal power levels and express it as a gain GQ,

GQ(dB) = 10 log10

(PQPTX

)(dB) . (2)

The measure LQ in (1) is similar to the signal transmissionlosses in buildings [30] or vehicles [13], [31]. Both LQ andGQare compared to referential values, e.g., measurements insidean anechoic chamber [13].

The variables L and P are empirically derived from mea-suring power levels via averaging or a computation of a givenquantile. In this study, the passive treatment is a FrequencySelective Surface (FSS), and the active treatment is a repeater.

Measurements performed via commercial mobile phonesallow mobile network operators to assess the KPIs of inter-est. The number of observations available for a given KPIdepends on the recording rate r and on the duration of themeasurement d. The recording rate is the number of KPIpoints recorded per second. The product, r×d , represents theexpected observations from a measurement. We refer to theobserved data as recorded observations. The reliable obser-vations result from data preprocessing. In measurements per-formed under uncontrolled conditions, given a KPI and itsrecording rate, we define spatiotemporal granularity as thenumber of KPI points spread along the traveled trajectory.A mobile phone is said to be in a connected state whenit performs a task that actively generates user data, e.g.,a download or a phone call. Further, a mobile phone is said tobe in an idle state when it is registered and connected to themobile network without any active user data transfer.

B. THE NEED FOR A NONPARAMETRIC APPROACHFig. 1 (a), shows the controlled condition in which the mobil-ity state can be controlled through constant distance p1 − p2and has no velocity. The signals can be controlled throughtransmitted power levels, angle of incidence, and polarizationof the RF signal. Meanwhile, the propagation conditions canbe predesigned and controlled through reflective and disper-sive behavior. As a result, we can fully control the distributionof the received signal by limiting the physical conditions inthe experiment. This will ultimately control the statisticalproperties of the observed KPIs. Then, we compare with theKPI probability distributions inferred for the no-treatmentscenario to assess the treatment effect.

Figure 1 (b) depicts a context that is characterized bynon static and uncontrolled conditions, which constitute areal-world operational situation of transceivers inside HSTs.In this operational context, the users on board HSTs moveat trajectories s(x, y, z) while being served by the mobilenetwork. We assume that the Base Station (BS) is at a fixedposition p1(x, y, z), and the user is at position p2(x, y, z) onboard the moving HST. The relative instantaneous velocityvr = vp2 (x, y, z) along the trajectory contains stops, acceler-ations, and intervals of steady velocity.

The velocity of HSTs causes fast and frequent transi-tions in propagation conditions, that is, hilly and flat ter-rains, combinations of high and low building, and different

VOLUME 8, 2020 162947


FIGURE 1. Contexts in mobile communication networks. (a) controlled conditions with twotransceivers and no mobility and (b) uncontrolled and non static conditions.

FIGURE 2. Treatment in mobile communication systems: (a) placebo treatment, (b) network infrastructure, (c)treatment Q, and (d) assessment of treatment effect. Blocks B1 and B2 represent the user side, whereas block Arepresents the mobile network infrastructure.

mobile network deployments. In railway tracks with dedi-cated mobile network BSs, the distance between HSTs iswithin few hundredsmeters; thus, large blockages cause rapidtransitions in the Line of Sight (LOS), strong multipath com-ponents, and Non Line of Sight (NLOS). The conditional dis-tribution of the signal received by MSs inside HSTs changesunder LOS and NLOS, respectively. The position of the BSsrelative to the HSTs is considered to be random as there is nodedicated deployment for railway coverage. Hence, multipathand LOS propagation conditions are dominant in urban andrural geographical regions, respectively.

The velocity of the HSTs cause fast transitions in thepropagation condition and it can be reasonably assumed thatthe received signal is characterized by a mixed distribution.We assume that a shorter travelled trajectory would decreasethe number of such combinations. The treatment effect is thensuperimposed on effects from the propagation conditions. Insuch experimental setting, the resulting data do not have aneasily identifiable or tractable distribution. Hence, we rely on

a nonparametric approach and consider the distributions asunknown.

Mobile network resources are not uniformly distributed inspace. That is, the available resources vary along the travelledtrajectory. This variation is primarily due to the differences inthe characteristics of the network deployment between denseurban, suburban, and rural regions. In this case, the spatialvariation in radio resources results in variations in the QoS.The user perceives this variation through conditions such asNo Service (NoS), call drops, insufficient data throughput[19], or frequent Radio Access Technology (RAT) handover[20]. Network providers have little or no control over thevariables being measured when the KPIs are being assessedunder operational conditions. In addition, accessing someKPIs requires a special setup such as that in [32], which isnot feasible in HSTs.

Mobile operators have full control of their mobile networksand can collect KPI samples from both the infrastructureand the user side. The combined analyses of both sides is of

162948 VOLUME 8, 2020


strategic interest for a mobile operator. Therefore, the mobileoperator wishes not to disclose KPI observations at the infras-tructure side. Since we aim to assess the treatments regardlessof any specific mobile operator, we are then forced to basethe assessment solely on KPI samples at the user’s side. Inthis way, Fig. 2 (a) and (c) represent the user’s side. Themobile network side – e.g. BSs – is represented by Fig. 2(b). In this study, we compare the KPI distribution underthe treatment context with that under the no treatment, giventhat both distributions are unknown. Also, the distribution of1xQ is unknown. In the following paragraphs, we discusssome possible changes in the KPI distributions resulting fromtreatment Q.

a) Treatment effect 1xQ has no statistical signifi-cance. At a given level of significance, we cannotdistinguish between the two KPIs that are with andwithout treatments.

b) Treatment effect 1xQ changes the mean of xq,i.e., My = Mx + 1Mx , where 1Mx is a constant.Here, we infer the KPI distribution of yQ is relatedto the KPI distribution of xq by a deterministic shift.

c) Treatment effect1xQ impacts the variance of xq,i.e., σ 2

yQ 6= σ2xq . Consequently, the yQ and xq differ.

C. DATA DESCRIPTIONFigure 3 illustrates specific attributes of the KPI samplesconsidered in this work: namely, the mobile network oper-ator, radio access technology, frequency band, and physicalchannel.

We analyse KPI distributions for specific choices of theattributes. The required filtering according to Fig. 3 isparametrized in the measurement setup and performed duringdata preprocessing. In this manner, we obtain a data set forone chosen set of attributes, which is then used to infer thetreatment effect. The data set used in this work is attributedby one network operator, the RAT 3G, frequency band 1, andthe downlink direction., cf. Fig. 3.The mobile phone, MS, is a system that recovers informa-

tion from the application layer contained in g(t, ω, s), e.g.,voice, text files, pictures. In order to accomplish this task, theBS andMS exchange a large amount of information andmanyKPIs are computed from both sides. The data array D in (3)is a metadata file that contains records of the computed KPIsand mobile network information, e.g., system information,user information.

D =

A11, a12, a13, . . . , a1nC21, c22, c23, c24D31, d32, d33A41, a42, a43, . . . , a4nC51, c52, c53, . . . , c5nF61, f62, f63, . . . , fmn

(3)

We represent the data array from the contexts with andwithout treatment by DQ and Dq, respectively. A represen-

tation of the treatment effect through the data array is

DQ = Dq +1DQ, (4)

where 1DQ is the treatment effect in the data array.The rows in (3) list A,B, . . . , E as measurement events.

Both the measurement events and their content depend onvarious factors, such as the type of task performed on theapplication layer, e.g., file transfer, voice call, and position.Other factors are events from the mobile network duringthe measurement, e.g., absence of signal and RAT cover-age. Under optimum conditions, the measurement events arerecorded at a fixed interval. The outcome of a KPI is anelement of a measurement event and appears in the samecolumn of (3). For instance, a sample from a KPI (ai2) in themeasurement event A is

Ai2 = (a12, a42, . . . , am2)T . (5)

Here, the samples of a selected KPI from the contexts withand without treatment are xq and yQ, respectively,

xq = (x1, . . . , xn)T , (6)

yQ = (y1, . . . , ym)T . (7)

Using (4), we write the expression of the KPI under thetreatment as

yQ = xq +1xQ, (8)

where 1xQ is the treatment effect observed via a selectedKPI x.As the duration of the measurement increases, so does the

length of the KPI vector in (6). A mobile phone connects onlyto one RAT at a time, and the distribution of the RAT is notnecessarily continuous along a travelled trajectory. Given aRAT dependent KPI, the sample size of this KPI decreasesby the proportion of the RAT available along the traveledtrajectory, as depicted in Fig. 8. Computing a KPI dependson the parameters measured by the MS and on the valuesreported by the BS. However, mobile signal coverage gapsalong the traveled trajectory are not uncommon and result inmissing reports from the BS at many positions, which reducesthe KPI sample size. The samples from the vector in (6) and(7) contain information on the temporal and spatial dynamicsof the stochastic process.

In addition, train stations have dedicated cellular networkthat have better QoS than that along the railway. The trainsthat approach or stop at a train station are at least partiallyserved by this dedicated cellular network. In this case, the spa-tial variation of the signal g(t, ω, s) is considerably smallerthan the variation inside an HST that is moving steadily awayfrom a train station. Both signal levels and service qualityare better under these conditions. Consequently, a portion ofthe KPI samples in x and y contain only temporal variations,which results in better signal and service levels.

VOLUME 8, 2020 162949


FIGURE 3. Representation of KPIs and its filtering in mobile communication networks.

III. DATA ANALYSISFor the statistical evaluation, we use a nonparametric statisti-cal approach based on multiple hypotheses testing combinedwith bootstrap to test for treatment effect and to estimate theeffect size. This approach assesses whether the treatment Qresults in different KPI distribution based on the collectedsamples.

In order to design the data evaluation approach, we assumethe following:

A1 The variations in the observed KPIs are consideredrandom independent identically distributed within agiven geographical region and a single travel direc-tion.

A2 The KPI probability distribution only depends ontreatment Q. Other factors can be neglected.

A3 The KPI distribution is unknown.To satisfy assumption A1, the travel trajectory is divided sys-tematically into geographical regions. This results in urbanregions Su1, . . . , Suku and rural regions Sr1, . . . , Srkr . Thenumber of urban regions ku and rural regions kr depend onthe length of the trajectory s(x, y), among other features.Selecting a geographical region results in a pair of data setsDQ and Dq which contain samples from yQ under treatmentQ and samples xq under treatment q (placebo).

A. ESTIMATION OF TREATMENT EFFECTThe adjusted HD method is a nonparametric approach forestimating the difference between sample quantiles of inde-pendent groups [24], [25]. This method uses the HD estimate[24] to compute the pth sample quantile F−1(p) = θp. Then,the difference between quantiles is computed via multiplecomparisons.

Let a pair of quantiles from the samples with and withouttreatment be θpq and θpQ, respectively. We estimate the dif-ference between those quantiles as

1θp = θpq − θpQ. (9)

The null hypothesis of no treatment effect is

H0 : θpq = θpQ, (10)

where θpq and θpQ are the true quantiles of the with andwithout treatment distributions, respectively. The functionqcomhd in R package WRS2 [33] tests whether the corre-sponding single or multiple quantiles of two distributions arethe same. That is, it computes (9), tests (10) and provides the1− α confidence interval.In order to specify the sample size for this estimation,

we define a time window of 150 seconds in the measurementdata. The samples from the desired KPI inside this timewindow form the sample size. The time window correspondsto the duration of the longest preprogrammed task and main-tains partial agreement with recommendations in [34]. Thisresults in 300 and 450 scores for the recording rates of 2 and3 samples/sec, respectively. When the available number ofscores in a KPI vector is smaller than 300, the entire sampleis used. In order to perform the test, the function qcomhdproduces bootstrap replicates with size equal to the size ofthe input data. For example, given input data of 100 scores,the function produces 2000 bootstrap replicates with size 100.

Our study also includes cases where the size of the inputdata is as large as 5000, and we exploit this large sampleby extracting subsamples with smaller size. For this reason,we adjust the package to produce bootstrap replicates withprespecified sample size as the input in the function qcomhd.

B. STATISTICAL SIGNIFICANCEThe testing design in [25] is based on the null hypothe-sis significance testing [35]. A Type I error occurs whenthe null hypothesis is true, but is wrongfully rejected. Theprespecified statistical significance, α = 0.05, representsthe acceptable probability to commit Type I error. In [25],the empirical p-value, that is the empirical probability of type

162950 VOLUME 8, 2020


I error or α, is defined as

p∗ =A+ 0.5C

B, (11)

where A is the number of times that H0 is rejected, C is thenumber of times that H0 is not rejected, and B is the numberof bootstrap replicates.

Using (11), we further compute the p-value as 2min(p∗, 1−p∗). The HD method applies multiple comparisons, andthe overall significance level is controlled via the classicalBonferroni method [36], in which the prespecified statisticalsignificance is adjusted as

αn =α

n, ∀n = 1, . . . , k, (12)

where k is the number of tested quantiles. The null hypothesisis rejected if, and only if, at least one p-value satisfies

p-valuen < αn. (13)

IV. CASE STUDYIn this case study, we apply the aforementioned nonparamet-ric statistical approach based on multiple hypotheses testingto investigate the advantages of two commercial solutions toimprove mobile network services inside HSTs, see Fig. 4.

B1 A passive treatment based on modified windowsequipped with a FSS. Fig. 5 depicts a FSS [30]that reduces the VPL of HSTs and hence improvesmobile network services inside HSTs.

B2 An active treatment using rooftop antennas andAmplify-and-Forward (AF) repeaters. In Fig. 6 theAF repeater creates an path that bypass the VPL inHSTs. Due to its external antenna, the repeater alsoallows users inside HSTs to access mobile networkinfrastructures near and distant from the railwaytrack.

The Device under Test (DUT) is the entire 210 -meter-longHST which consists of individual carriages of 26.5 -meterseach. Each treatment is prototypically applied to carriages oftwo separate railjet trains [37]. TheMeasurement Entity (ME)shown in Fig. 7 is deployed on board railjet trains. Selectingand applying treatments Q and q to two different carriagesyields VQ and Vq. In the following sub-sections we furtherexplore those treatments and their respective effects.

A. PASSIVE TREATMENT: FREQUENCY SELECTIVEWINDOWSCurrent HSTs are equipped with metal-coated windows toprovide thermal and visual comfort to passengers. As a sideeffect, the metal coating attenuates the wireless signal trans-missions of cellular phones and Global Positioning System(GPS) up to 40 dB. The untreated HST is fully equipped withlightly coated windows, which are referred to as RW [12]. Ingeneral, a FSS consists of a two-dimensional periodic geo-metric structure composed of metallic and dielectric mate-rials [30], [38]. The geometric structure defines the effective

electromagnetic properties, i.e. permittivity, conductivity, andmagnetic permeability.

The transmission of electromagnetic waves through a FSSis highly linear over a large dynamic range of wave ampli-tudes. Its transfer function depends on the frequency, polar-ization, and angle of incidence of the waves [12], [39],[40]. An engineered FSS functions as a frequency selectiveabsorber, passive repeater [41], filter, or reflector [42]. Thepassive treatment consists in equipping a section of a HSTwith a FSSwindow, named SW, that yields awide-band band-pass filter within the frequency range 0.7—3.5GHz [43].This SW features substantially smaller RF attenuation than aRW does. The Austrian Federal Railways partially upgradeda single carriage of a HST with SWs to evaluate how thismodification improves the QoS of the mobile network insidethe HST.

Using (1), we estimate the transmission loss of a HSTwithout treatment by

Lq = −10 log10

(PqPTX

)(dB) (14)

and the loss of a HST with treatment by,

LQ = −10 log10

(PQPTX

)(dB). (15)

Here Pq and PQ are the received signal power estimates insidethe HST.

To arrive at an equivalent representation for (8), we subtract(14) and (15) to express the estimated difference in transmis-sion loss 1L,

1L = Lq − LQ (dB). (16)

Taking the difference ensures that the value of PTX cancels,as well as the antenna characteristics, cable losses, and manyother setup specific properties. In the context with treatment,the reduced transmission loss increases the received signalpower levels. The condition that attributes improvement tothe current treatment is expressed as

1L > 0 ⇐⇒ Lq > LQ. (17)

The improved signal power level can be represented as

PQ = Pq +1P, (18)

where Pq, PQ are quantile estimates of the KPI samples andrepresent the received signal power levels.

B. ACTIVE TREATMENT: AF REPEATERThe active treatment consists of deploying Moving RelayNodes (MRNs) on board the HST. A Moving Relay Node(MRN) is a system composed of rooftop mounted donorantennas, AF repeaters, and coaxial leaky cables [10]. Asingle omnidirectional antenna is used as donor antenna fora single AF repeater.

The AF repeater amplifies signal and noise at selectedsignal bands in both the up- and the downlink of the Fre-quency Division Duplex (FDD) mobile network. The bands

VOLUME 8, 2020 162951


FIGURE 4. Case study with two different treatments and a reference: (a) Regular Windows (RW), (b) Structured Windows (SW), (c)Amplify-and-Forward (AF) repeater.

FIGURE 5. Structured window (passive treatment). Protocol stack (A1) represents the mobile network side and (B1)represents smart phones deployed inside the train.

for amplification are selected by digitally programmablefilters. The filter transfer functions have steep transitionsbetween the pass- and stopbands which are associated withincreased group delay [44].

The AF repeater is a device with non-linear behaviorbecause there is a maximum output power level due to physi-cal limits of the RF circuit and its power transmission limits.Furthermore, the AF repeater is a Single-Input Single-Output(SISO) device that uses a single antenna. This fundamentallyinfluences the achievable data rates for high-speed mobilenetwork services that employ Multiple Input Multiple Output(MIMO) techniques using multiple antennas at the BS andMS.

The active treatment creates an additional propagation pathbetween the outside and the inside of the HST in parallelto the passive propagation paths through the windows. Theperformance of this active treatment strongly depends on the

optimization of the AF repeater’s configuration as well as thescenario under test, e.g. base station deployment along thetrack [45].

Similarly to Sec. IV-A, the gain under active treatment iswritten as

1G = 10 log(PQPq

)(dB). (19)

The condition that attributes improvement to the currenttreatment is

1G > 0 ⇐⇒ Pq < PQ. (20)

PQ = Pq +1G, (21)

where Pq, PQ are quantile estimates of KPIs samples andrepresent the received signal power levels.

In the equations above, the notation P is a selected KPI thatrepresents an estimate of the received signal power level.

162952 VOLUME 8, 2020


FIGURE 6. AF-repeater (active treatment). Protocol stack (A1) represents the mobile network side and (B1) representssmart phones deployed inside the train.

C. MEASUREMENT SETUPWe carried out measurement campaigns in two railjet trains[37]. The train with a carriage equipped with SW traveledfrom Vienna to Graz (198 km, 3 hour duration), while theother with a carriage equipped with an AF repeater trav-eled from Vienna to Salzburg (305 km, 3 hour duration),as detailed in Fig. 8.

The ME in Fig. 7, consists of a group of six commercialoff-the shelf smart phones. For our measurement, we dividedthe smart phones in three groups of two. The first group waslocked to 2G RAT only. The second group was locked to 2Gand 3G RATs. The third group had access to 2G, 3G, and 4G.Each mobile was equipped with dedicated software (AniteNemoWalker Air) [46] which was configured to repeat a fixedsequence of tasks. The tasks consisted of one voice call withtotal duration of 120s, one download and one upload bothwith 10 MB data volume. The tasks emulate user activities inquasi-real usage conditions. In both measurements, one MEwas deployed inside section VQ, where treatmentQ predomi-nates, and another ME in section Vq under the predominanceof treatment q. The ME is placed on a passenger’s seating,as shown in Fig. 7.

D. VARIABLES OF INTERESTWe expect to detect a statistically significant treatment effectin any selected KPI under the direct influence of the treat-ment. Different KPIs show different sensitivities to the treat-ments. Table 1 displays the group of KPIs used to assess thetreatment effect.

Both the AF repeater and the SW are expected to improvelink quality at the physical layer. For this reason, the treatmenteffect is expected to be associated with the metrics fromthis layer. Based on this association, three KPIs report signallevels in the downlink: Received Signal Strength Indicator(RSSI), Received Signal Code Power (RSCP) and receivedSignal-to-Interference Ratio (SIR).

TABLE 1. Description of the selected KPIs.

In Universal Mobile Terrestrial Services (UMTS), the UserEquipment (UE) reports the RSSI and RSCP according to theThird Generation Partnership (3GPP) standards. The RSSIreport measures the total received signal plus interferenceand noise power within the signal band of interest. It alsoserves as a rudimentary indicator of the power level of thesignal of interest [47] and forms the basis for estimating morespecific link quality-related KPIs [48]. The selected RSCPfrom the primary cell is such a more specific measure of linkquality. The SIR indicates the signal quality and is estimatedvia the Dedicated Physical Control Channel (DPCCH) [47].It is a metric that contains interference from co-channels andneighbor cells. Therefore, the variation in this KPI dependson the radio resource control and network topology (amongothers). InMRN contexts, the SIR is affected by themultipathcomponents passing through the vehicle frame [49].

E. DATA PREPROCESSINGThe size of the data sets depends on the length of the regions,the average velocity of the train, and the recording rate.Before the statistical evaluation of the KPIs discussed in Sec.IV-C, we filter the data sets with respect to mobile networkoperator and RAT of interest, cf. Sec. II-C. In addition,we select the geographical regions from the travelled trajec-tory. The 3G (UMTS) network offers continuous coveragealong the travel trajectories and sufficient spatial granularityof the data sample.

VOLUME 8, 2020 162953


FIGURE 7. Measurement entity deployed in railjet trains. (a) measurement entity, (b) placement onpassenger’s sitting, and (c) GPS antenna and modem.

We select the group of mobile phones that was configuredto access 3G preferably and 2G alternatively (see Sec. IV-C)for a single operator. The mobiles phones repeatedly executethe tasks described in Sec. IV-C, and hence they remainpredominantly in a connected state (only 2.8% in idle statealong the route Vienna to Graz). The route Vienna to Salzburgtraverses a well-covered flat terrain, thus contribution of thesamples from an idle state is neglected in the statistical eval-uation from both tracks. The stops at train stations are partof the regular train commute, and we have not imposed anyspecial treatment of those.

With support from the Eurostat [50], we classify seg-ments of the travelled trajectories Vienna—Graz s1(x, y)and Vienna—Salzburg s2(x, y) into regions as described inSec. III. Figure 8 shows this classification, where eachregion corresponds to data sets with treatment Q and withouttreatment q.

F. KPI HISTOGRAMSFigures 9 and 10 present histograms of KPIs comparingdata with and without active treatment in urban and ruralregions. In these histograms, we observe that the treatmenteffect imposes a considerable shift on the lower quantilesand imposes truncation due the limited output power of theAF repeater on the upper quantiles. The KPI histograms withtreatment indicate left-skewed distributions, upper-bounded,and show a large overlap with the histograms without treat-ment.

Figures 11 and 12 show histograms of KPIs comparingdata with and without passive treatment in urban and ruralregions. Here the histograms show that the treatment effectimposes a moderate shift on the lower quantiles and a smallshift on the upper quantiles. The histograms indicate that thedistributions with and without passive treatment share a largeoverlapped region.

G. EVALUATION FOR TREATMENT EFFECTTables 2 and 3 summarize the changes in the distributionsfrom both active and passive treatments and attribute statisti-cal significance to these.

Traditionally, the median is used, for assessing changesin KPIs from the physical layer of mobile communicationsystems from measurements.

The treatment effect of the active treatment assessed viathe median is summarized in Table 2. Its magnitude for theRSCP varies from 12.38 dB to 15.12 dB and from 16.08 dBto 20.02 dB in the urban and rural regions, respectively. Thetreatment effect of the passive treatment assessed via themedian is summarized in Table 2. Its magnitude for the RSCPvaries 1.80 dB to 7.97 dB and from 3.53 dB to 10.67 dB inthe urban and rural regions, respectively.

Due to the characteristics of the active and passive treat-ments, the effect size that we evaluated solely via the medianprevents the treatment effects from being detected in lowerand upper quantiles. Indeed, the treatment effects from theactive and passive treatments are not equal among the sam-ple quantiles, as seen in Figs. 9–12. This behavior of thetreatment effect is presented in Table 3 displaying results ofmultiple hypotheses testing carried out by the HD method.

In the following, we discuss the effect size for the 0.25 and0.75 quantiles as summarized in Table 3. The column ‘‘Effectsize’’ gives the 95% confidence interval for the shift in thedata due to the treatment.

In the active treatment, the magnitude of the effect size forRSCP varies from 9.93 dB to 14.91 dB and from 15.56 dB to27.93 dB in the urban and rural regions, respectively.

In the passive treatment, the magnitude of the treatmenteffect for RSCP varies between 3.53 dB and 14.08 dB forurban and from 5.22 dB to 22.79 dB for rural regions, respec-tively. The treatment effect in rural regions is more varied,i.e., the samples exhibit a wider range. This is due to the poorsignal coverage in those areas. As a result, both treatmentslead to larger improvements in those regions.

The values presented in Table 3 show that the treatmenteffect among the quantiles is unequal. This implies that bothtreatments change the KPI distribution. The null hypothesisof equal distribution is rejected for all evaluated quantiles.

The active treatment effect is more considerable in the ruralregion (Fig. 10) then in urban (Fig. 9). This is explained bythe low signal levels in rural regions at the input of the AF

162954 VOLUME 8, 2020


FIGURE 8. Segmentation of track with one example of 3G service availability each for rural and urban regions. Top:Passive treatment is indicated by SW and no treatment by RW. Bottom: Active treatment is indicated by ON and notreatment by OFF.

FIGURE 9. Histograms of KPIs without (a, b, c) and with (d, e, f) active treatment (‘‘Repeater’’) in urban region.

repeater. The received signals are amplified with the desiredgain because the RF amplifiers of the AF repeater operate farbelow its output power limit.

The passive treatment shows no output power limit.However, the treatment effect depends on distribution ofAngle of Arrival (AoA) and polarization. The distributionof AoA and polarization differs between the urban and rural

regions, as a result of the differences in deployment ofmobile network and characteristics of the terrain betweenthose regions. Multipath propagation conditions are domi-nant in urban regions, whereas LOS is dominant in ruralregions.

The FSS used in the passive treatment imposes higherattenuation on the weaker multipath components with highly

VOLUME 8, 2020 162955


FIGURE 10. Histograms of KPIs without (a, b, c) and with (d, e, f) active treatment (‘‘Repeater’’) in rural region.

FIGURE 11. Histograms of KPIs without (a, b, c) and with (d, e, f) passive treatment (‘‘Structured Windows’’) in urbanregion.

162956 VOLUME 8, 2020


FIGURE 12. Histograms of KPIs without (a, b, c) and with (d, e, f) passive treatment (‘‘Structured Windows’’) in rural region.

TABLE 2. Estimated medians of treatment effect - Active and passive treatments.

scattered AoA. Accordingly, the passive treatment showsbenefits in urban regions.

The analysis shows that the effect size depends on thecharacteristics of the treatment (active or passive), as well asthe network deployment (rural or urban). The mobile servicecoverage in the urban regions is better than in the rural regionswhere signal levels are much lower. Therefore, the overallbenefits of a treatment depend on themix of networks deploy-ments along the travelled trajectory.

The chosen segmentation of the travelled trajectory interms of length (km) and type (urban vs. rural) influences thevalidity of assumption A2 in Sec. III, see Fig. 8. Short seg-ments, e.g., 5–10 km, result in small spatial variations in theKPIs with treatment effect. Long segments, e.g., hundreds ofkilometers, hide significant variations in the treatment effecton the KPI. This effect can be observed when comparing theRSCP statistics of the whole data set with the RSCP statisticsin rural region for the active treatment.

VOLUME 8, 2020 162957


TABLE 3. Estimated quantiles of treatment effect - active and passive treatments.

The results presented in Table 2 show this bias: the effectsize along the 64-km track in the rural regions is largerat approximately 18 dB, compared to and an effect size ofapproximately 12 dB along the entire 305-km trajectory.

The treatment effect differs among the set of inves-tigated KPIs. Some KPIs may consistently display noeffect, in contrast to others. In our case study, the goalof the treatment is an enhanced signal level on boardthe HST.

Specifically, we emphasize that the results in Table 2show that the effect sizes in the Rx SIR under both activeand passive treatments are much smaller than in the RSCPand RSSI. This is partially explainable by the closed-looptransmit power control in the network for voice calls. Thiscontrol loop varies the transmit power to achieve an operator-specified target value for Rx SIR.

In regions, where the interference level is lower than thenoise level plus VPL, the Rx SIR on board the HST is closeto the Rx Signal-to-Noise Ratio (SNR). This is true for bothtreatments: active and passive. Here, any treatment reducingthe VPL has the potential of improving the measured Rx SIRon board the HST.

Active treatments can completely compensate the VPLin rural areas where the received signal levels are low, butcome at the expense of adding some noise. Passive treatmentscan only partially compensate the VPL and they do not addnoise. Therefore, the active treatment shows more consid-erable KPI improvements in rural regions than the passivetreatment. In urban regions, neither treatment shows muchimprovement.

V. CONCLUSIONIn this paper, we inferred significant benefits of treatments inmobile network coverage on board high-speed trains (HSTs)in urban and rural regions through measurements of KPIsin Austria. Two treatments are assessed: an active treatmentbased on AF repeaters and a passive one using structuredwindows with low RF attenuation.

We use the modified HD method and nonparametric boot-strap to test multiple sample quantiles from collected key per-formance indicators (KPIs) for significance. This approachhandles well the differences in mobility, regional conditions,and from the varying sample sizes along the railway track.

The results support the hypothesis that signal qualityimproves with both treatments. The improvement in themobile network coverage on board the HST with the AFrepeater is higher than that due to the installed low RF atten-uation windows. The enhancements from both treatments arelarger when the HST travels in rural regions.

Regarding the effect of the difference in signal qual-ity, prior research has shown that QoS enhancement withboth treatments is similar along the trajectories travelled inAustria. The experiments in this study cannot consider theeffect of the network deployment along the railroad.

ACKNOWLEDGMENTThe authors thank J. Resch and S. Ojak at the Austrian Fed-eral Railways (OBB) for providing access to prototypical rail-jets equipped with Structured Windows (SW) and Amplify-and-Forward (AF) repeaters. The measurement campaignwas funded by OBB.

162958 VOLUME 8, 2020


REFERENCES[1] B. Kurt, E. Zeydan, U. Yabas, I. A. Karatepe, G. K. Kurt, and A. T. Cemgil,

‘‘A network monitoring system for high speed network traffic,’’ in Proc.13th Annu. IEEE Int. Conf. Sens., Commun., Netw. (SECON), Jun. 2016,pp. 1–3.

[2] A. Botha, K. Calteaux, M. Herselman, A. S. Grover, and E. Barnard,‘‘Mobile user experience for voice services: A theoretical framework,’’ inProc. Int. Conf. Mobile Commun. Develop., vol. 18, p. 10, Feb. 2012.

[3] C.-X. Wang, A. Ghazal, B. Ai, Y. Liu, and P. Fan, ‘‘Channel measurementsand models for high-speed train communication systems: A survey,’’ IEEECommun. Surveys Tuts., vol. 18, no. 2, pp. 974–987, 2nd Quart., 2016.

[4] A. Schafer andD.G.Victor, ‘‘The futuremobility of theworld population,’’Transp. Res. A, Policy Pract., vol. 34, no. 3, pp. 171–205, Apr. 2000.

[5] J. Wu and P. Fan, ‘‘A survey on high mobility wireless communica-tions: Challenges, opportunities and solutions,’’ IEEE Access, vol. 4,pp. 450–476, 2016.

[6] Y. Wen, Y. Ma, X. Zhang, X. Jin, and F. Wang, ‘‘Channel fading statisticsin high-speed mobile environment,’’ in Proc. IEEE-APS Topical Conf.Antennas Propag. Wireless Commun. (APWC), Sep. 2012, pp. 1209–1212.

[7] F. J. Martin-Vega, I. M. Delgado-Luque, F. Blanquez-Casado, G. Gomez,M. C. Aguayo-Torres, and J. T. Entrambasaguas, ‘‘LTE performance overhigh speed railway channel,’’ in Proc. IEEE 78th Veh. Technol. Conf.,Sep. 2013, pp. 1–5.

[8] L. Li, K. Xu, D.Wang, C. Peng, Q. Xiao, and R.Mijumbi, ‘‘Ameasurementstudy on TCP behaviors in HSPA+ networks on high-speed rails,’’ in Proc.IEEE Conf. Comput. Commun. (INFOCOM), Apr. 2015, pp. 2731–2739.

[9] Q. Xiao, K. Xu, D. Wang, L. Li, and Y. Zhong, ‘‘TCP performance overmobile networks in high-speed mobility scenarios,’’ in Proc. IEEE 22ndInt. Conf. Netw. Protocols, Oct. 2014, pp. 281–286.

[10] T. Berisha, P. Svoboda, S. Ojak, and C. F. Mecklenbrauker, ‘‘Cellularnetwork quality improvements for high speed train passengers by on-boardamplify-and-forward relays,’’ in Proc. Int. Symp. Wireless Commun. Syst.(ISWCS), Sep. 2016, pp. 325–329.

[11] Y. Sui, J. Vihriala, A. Papadogiannis, M. Sternad, W. Yang, andT. Svensson, ‘‘Moving cells: A promising solution to boost performancefor vehicular users,’’ IEEE Commun. Mag., vol. 51, no. 6, pp. 62–68,Jun. 2013.

[12] M. Madjdi, L. W. Mayer, and A. Demmer, ‘‘Hochfrequenz-durchlassigefensterscheiben für Regional-und Fernverkehrsfahrzeuge,’’ ZEVrail,vol. 140, no. 9, pp. 376–383, 2016.

[13] M. Lerch, P. Svoboda, S. Ojak, M. Rupp, and C. Mecklenbraeuker, ‘‘Dis-tributed measurements of the penetration loss of railroad cars,’’ in Proc.IEEE 86th Veh. Technol. Conf., Sep. 2017, pp. 1–5.

[14] C.-W. Lee, M.-C. Chuang, M. C. Chen, and Y. S. Sun, ‘‘Seamless handoverfor high-speed trains using femtocell-based multiple egress network inter-faces,’’ IEEE Trans. Wireless Commun., vol. 13, no. 12, pp. 6619–6628,Dec. 2014.

[15] P. Dong, X. Du, T. Zheng, and H. Zhang, ‘‘Improving QoS on high-speedvehicle by multipath transmission based on practical experiment,’’ in Proc.IEEE Veh. Netw. Conf. (VNC), Dec. 2015, pp. 32–35.

[16] H. Gao, Y. Ouyang, H. Hu, and Y. Koucheryavy, ‘‘A QoS-guaranteedresource scheduling algorithm in high-speed mobile convergence net-work,’’ in Proc. IEEE Wireless Commun. Netw. Conf. Workshops(WCNCW), Apr. 2013, pp. 45–50.

[17] T. Berisha, P. Svoboda, S. Ojak, and C. F. Mecklenbrauker, ‘‘SegHy-Per: Segmentation- and hypothesis based network performance evaluationfor high speed train users,’’ in Proc. IEEE Int. Conf. Commun. (ICC),May 2017, pp. 1–6.

[18] P. Dong, B. Song, H. Zhang, and X. Du, ‘‘Improving onboard Internetservices for high-speed vehicles by multipath transmission in heteroge-neous wireless networks,’’ IEEE Trans. Veh. Technol., vol. 65, no. 12,pp. 9493–9507, Dec. 2016.

[19] F. P. Tso, J. Teng, W. Jia, and D. Xuan, ‘‘Mobility: A double-edged swordfor HSPA networks: A large-scale test on hong kong mobile HSPA net-works,’’ IEEE Trans. Parallel Distrib. Syst., vol. 23, no. 10, pp. 1895–1907,Oct. 2012.

[20] A. Lutu, O. Alay, D. Baltrunas, and A. Elmokashfi, ‘‘Profiling mobilebroadband coverage,’’ in Proc. Int. Workshop Traffic Monit. Anal. (TMA),2016, pp. 1–5.

[21] V. Bagdonavicius, J. Kruopis, and M. Nikulin, Nonparametric Tests forComplete Data. Hoboken, NJ, USA: Wiley, 2013.

[22] B. Efron, The Jackknife, Bootstrap, Other Resampling Plans. Philadelphia,PA, USA: SIAM, 1982.

[23] A.M. Zoubir and D. R. Iskander, Bootstrap Techniques for Signal Process-ing. Cambridge, U.K.: Cambridge Univ. Press, 2004.

[24] F. E. Harrell and C. E. Davis, ‘‘A new distribution-free quantile estimator,’’Biometrika, vol. 69, no. 3, pp. 635–640, 1982.

[25] R. R. Wilcox, D. M. Erceg-Hurn, F. Clark, and M. Carlson, ‘‘Comparingtwo independent groups via the lower and upper quantiles,’’ J. Stat. Com-put. Simul., vol. 84, no. 7, pp. 1543–1551, Jul. 2014.

[26] A. Valcarce and J. Zhang, ‘‘Empirical Indoor-to-Outdoor propagationmodel for residential areas at 0.9–3.5 GHz,’’ IEEE Antennas WirelessPropag. Lett., vol. 9, pp. 682–685, 2010.

[27] L. Subrt and P. Pechac, ‘‘The influence of dynamic changes of indoorscenarios on electromagnetic wave propagation,’’ in Proc. 4th Eur. Conf.Antennas Propag., pp. 1–5, Apr. 2010.

[28] E. Tanghe, W. Joseph, L. Verloock, and L. Martens, ‘‘Evaluation of vehiclepenetration loss at wireless communication frequencies,’’ IEEE Trans. Veh.Technol., vol. 57, no. 4, pp. 2036–2041, Jul. 2008.

[29] P. Karlsson, C. Bergljung, E. Thomsen, and H. Borjeson, ‘‘Widebandmeasurement and analysis of penetration loss in the 5 GHz band,’’ in Proc.21st Century Commun. Village., Sep. 1999, pp. 2323–2328.

[30] I. Ullah, D. Habibi, X. Zhao, and G. Kiani, ‘‘Design of RF/Microwave effi-cient buildings using frequency selective surface,’’ in Proc. IEEE 22nd Int.Symp. Pers., Indoor Mobile Radio Commun., Sep. 2011, pp. 2070–2074.

[31] U. T. Virk, K. Haneda, V.-M. Kolmonen, P. Vainikainen, and Y. Kaipainen,‘‘Characterization of vehicle penetration loss at wireless communica-tion frequencies,’’ in Proc. 8th Eur. Conf. Antennas Propag. (EuCAP),Apr. 2014, pp. 234–238.

[32] Ribbenfjard, Lindmark, Karlsson, and Eklund, ‘‘Omnidirectional vehicleantenna for measurement of radio coverage at 2 GHz v.2.0,’’ IEEE Anten-nas Wireless Propag. Lett., vol. 3, pp. 269–272, 2004.

[33] P. Mair and R. Wilcox, ‘‘Robust statistical methods in R using the WRS2package,’’ Behav. Res. Methods, vol. 52, pp. 464–488, 2020.

[34] Technical Specification ETSI 102 250-5, document V2.4.2, 2015.[35] S. Nakagawa and I. C. Cuthill, ‘‘Effect size, confidence interval and

statistical significance: A practical guide for biologists,’’Biol. Rev., vol. 82,no. 4, pp. 591–605, Nov. 2007.

[36] Y. Benjamini and Y. Hochberg, ‘‘Controlling the false discovery rate:A practical and powerful approach to multiple testing,’’ J. Roy. Stat. Soc.,Ser. B Methodol., vol. 57, no. 1, pp. 289–300, Jan. 1995.

[37] Wikipedia, OBB, Vienna, Austria, 2020.[38] M. W. B. Silva and L. C. Kretly, ‘‘An efficient method based on

equivalent-circuit modeling for analysis of frequency selective surfaces,’’in Proc. SBMO/IEEE MTT-S Int. Microw. Optoelectronics Conf. (IMOC),Aug. 2013, pp. 1–4.

[39] B. Peswani, S. Yadav, and M. M. Sharma, ‘‘A novel band pass double-layered frequency selective superstrate for WLAN applications,’’ in Proc.5th Int. Conf. - Confluence Next Gener. Inf. Technol. Summit (Confluence),Sep. 2014, pp. 447–451.

[40] A. Ben Munk, Frequency Selective Surfaces: Theory and Design, chapter[1] and [7]. Hoboken, NJ, USA: Wiley, 2000.

[41] M. W. B. Silva, L. C. Kretly, and S. E. Barbin, ‘‘Practical guidelinesfor the design and implementation of microwave absorber using FSS-frequency selective surfaces,’’ in Proc. 20th Int. Conf. Microw., RadarWireless Commun. (MIKON), Jun. 2014, pp. 1–4.

[42] S. Hamid, B. Karnbach, H. Shakhtour, and D. Heberling, ‘‘Thin multi-layer frequency selective surface absorber with wide absorption response,’’in Proc. Loughborough Antennas Propag. Conf. (LAPC), Nov. 2015,pp. 1–5.

[43] L. W. Mayer, A. Demmer, A. Hofmann, and M. Schiefer, ‘‘Metal-coatedwindowpane, particularly for rail vehicles,’’ U.S. Patent 20 140 718 024,Feb. 17 2016.

[44] M. Lerch, P. Svoboda, O. Trindade, J. Resch, V. Raida, and M. Rupp,‘‘Identifying multipath propagation in vehicular repeater deploymentsby LTE measurements,’’ in Proc. IEEE 89th Veh. Technol. Conf. (VTC-Spring), Apr. 2019, pp. 1–5.

[45] M. Lerch, P. Svoboda, D. Maierhofer, J. Resch, A. Brantner, V. Raida,and M. Rupp, ‘‘Measurement based modelling of in-train repeater deploy-ments,’’ in Proc. IEEE 89th Veh. Technol. Conf. (VTC-Spring), Apr. 2019,pp. 1–6.

[46] Keysight Technologies. Nemo Handy Handheld Measurement Solution |Keysight, 2018.

[47] Technical Specification TS25.215, document TR V9.0.0, 3GPP, 2011.[48] R. Kreher, UMTS Performance Measurement: A Practical Guide to KPIs

for the UTRAN Environment. Hoboken, NJ, USA: Wiley, 2006.

VOLUME 8, 2020 162959


[49] S. Scott, J. Leinonen, P. Pirinen, J. Vihriala, V. Van Phan, andM. Latva-aho,‘‘A cooperative moving relay node system deployment in a high speedtrain,’’ in Proc. IEEE 77th Veh. Technol. Conf. (VTC Spring), Jun. 2013,pp. 1–5.

[50] Eurostat Regional Yearbook, Eur. Commission, Brussels, Belgium 2016.

ORLANDO S. TRINDADE received the bache-lor’s degree in electrical engineering and the mas-ter’s degree in telematics from the State Universityof Campinas (Unicamp), São Paulo, Brazil. Heis currently undergoing research with the Insti-tute of Telecommunications, Technische Univer-sität Wien, Vienna, Austria. Some of his researchinterests are stochastic modeling, industrial net-works, Macro Raman spectroscopy, mobile net-works quality assessment, vehicular communica-

tions, the IoT, and IIoT.

TAULANT BERISHA received the B.S. and M.S.degrees in electrical engineering from the Uni-versity of Prishtina, Kosovo, in 2011 and 2014,respectively, and the Ph.D. degree in telecommu-nications from TUWien, Vienna, Austria, in 2019.From 2015 to 2019, he was a Project Assistantwith TU Wien, and since 2019, he has beena Researcher with Dimetor GmbH, Vienna. Hisresearch interests are in vehicular communica-tions, BVLoSUAVoperations, benchmarking, sta-

tistical modeling, and service quality measurements.

PHILIPP SVOBODA (Senior Member, IEEE)received the Dr.Ing. degree in electrical engineer-ing from TU Wien. He is currently a Senior Sci-entist with TU Wien, with a research focus onthe performance aspects of mobile cellular tech-nologies. He is currently examining the feasibilityof using crowdsourcing to conduct performancemeasurements on 4G and 5G mobile networks.His research aims to establish a common frame-work for evaluating the performance of mobile

networks, guaranteeing reliable, and fair connectivity for end-users.

EFSTATHIA BURA received the B.S. degreein mathematics from the University of Athens,Greece, the M.S. degree in mathematics from theUniversity of Illinois at Chicago, and the M.S. andPh.D. degrees in statistics from the University ofMinnesota, USA, in 1996. She served as an Assis-tant, Associate (with tenure), and Full Professorwith the Statistics Department, George Washing-ton University, and currently holds the Chair ofApplied Statistics at the Institute of Statistics and

Mathematical Methods in Economics, Faculty of Mathematics, TU Wien.She is an elected member of the International Statistical Institute and a Ful-bright scholar, has published widely in methodological and applied statisticsand has been serving in several editorial boards.

CHRISTOPH F. MECKLENBRÄUKER (SeniorMember, IEEE) received the Dipl.-Ing. degree(Hons.) in electrical engineering from TechnischeUniversität Wien, Vienna, Austria, in 1992, andthe Dr.-Ing. degree (Hons.) from Ruhr-UniversitätBochum, Bochum, Germany, in 1998.

From 1997 to 2000, he worked at SiemensAG Austria and engaged in the standardizationof UMTS. From 2000 to 2006, he held a seniorresearch position with the Telecommunications

Research Center Vienna (FTW), Vienna. In 2006, he joined TU Wien as aFull Professor. From 2009 to 2016, he lead the Christian Doppler Labora-tory for Wireless Technologies for Sustainable Mobility. He has authoredapproximately 250 papers in international journals and conferences, forwhich he has also served as a reviewer, and was granted several patents inthe field of mobile cellular networks. His current research interests includeradio interfaces for peer-to-peer networks (vehicular connectivity, sensornetworks), ultra-wideband radio, and MIMO transceivers.

Dr.Mecklenbräuker is a member of the Antennas and Propagation Society,the Intelligent Transportation Society, the Vehicular Technology society,the Signal Processing society, as well as VDE and EURASIP. His doctoraldissertation received the Gert-Massenberg Prize, in 1998. He is the councilorof the IEEE Student Branch Wien.

162960 VOLUME 8, 2020

Documents

Assessment of Treatment Influence in Mobile Network