A Comparative Study to Evaluate theUsability of Context-based Wi-Fi AccessMechanismsMatthias Budde, Till Riedel, Marcel Köpke, Matthias Berning and Michael BeiglTECO, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany,email: [email protected], www: paper presents a comparative study of six di erent tag and context basedauthentication schemes for open Wi-Fi access. All of the implemented methodsrequire only a smartphone and an HTML5 capable webbrowser, making them interchangeable and easy to incorporate into existing infrastructure. We recruited22 participants for the study and used two standardized questionnaires as well asadditional metrics to assess whether further investment in a systematic usabilityanalysis seems prudent. The evaluation shows that suitable alternatives for Wi-Fiauthentication exist and points out their limitations and opportunities.Universal Access, Practical Security, Usability, User Experience, User Study,Interfaces, Device Association, Authentication, Wi-Fi, Smart Environments, ContextKeywords:1 IntroductionFuture computing environments as driven by the notions of ubiquitous computingand ambient intelligence are expected to give rise to information technology that isembedded in everyday life and is spontaneously formed from ubiquitous devices, objects,and services that we can easily access. While humans understand how to access physicalresources at their disposal, it is often harder in a digital world. Accessible digital resourcesare a key factor to assistance and inclusion, especially in public spaces.Open Wi-Fiaccess, e.g. through hotspots, creates many issues (access control, liability and legalissues) on the user and institutional side, as malicious parties cannot be kept out of thenetwork. This is why today, usually username and password, entered into a web interface(a.k.a.captive portal ), are required to access wireless infrastructure.This is the author's version of the work, posted for personal use, not for redistribution.lication rights licensed to Springer Publishing.Pub-Definitive version published and presented inProceedings of the 16th International Conference on Human-Computer Interaction (HCII 2014),Creta Maris,Heraklion, Crete, Greece, 22-27 June 2014. The final publication is available at 319-07446-7 44DOI: 10.1007/978-3-319-07446-7 44 101

Providing seamless wireless network service is not only about network quality but alsoabout user experience and ease of access has become an important factor for quality oflife.Adding an extra burden on users particularly technically non-literate users orones with special needs actually excludes many people from access, especially whenusing complicatedusername:passwordschemes with media breaks.As a step towardsthe proliferation of accessible networks, this work evaluates the usability of ways forassociating handheld devices to Wi-Fi networks, while also regarding implementationand practical feasibility.2 Related WorkA great variety of schemes have been proposed in the past to pair mobile devices forspontaneous interaction, many of which could also be applied for the use case of authenticating to a Wi-Fi hotspot.One solution proposed is the use of Out-of-Band (OOB)information to establish a shared secret. Holmquist et al. [6] as well as Mayrhofer et al.[12] proposed to couple devices using their accelerometers by shaking them simultaneously. This, however, is not easily applicable to systems involving static infrastructure.A method to generate a shared secret from ambient audio was investigated by Sigg etal. [16]. These technical publications do not take human factors into account, which wetargeted in this work. Several other authors conducted comparative user studies on theusability of device pairing or Wi-Fi setup. Kostiainen et. al. [9] used formative interviews to assess user needs in home network access control and proposed a conceptualsystem for setup and management.Both Uzun [18] and Kainda et al. [8] ana-lyzed device pairing by di erent textual interfaces. Kumar et. al. [10] included variantsof eight device pairing methods in a large study implemented on a common platform.The usability assessment is mostly based on automatically logged user actions with noqualitative feedback. Ion et al. [7] used mock-ups to investigate the usability of devicepairing methods with regard to perceived security needs in di erent real-life situation.Their ndings indicate that the usability as well as user preference depend on the contextof the pairing situation.3 MethodologyIn this work, we conducted a user study to explore six di erent methods suitable toreplace classicusername:passwordauthentication for public Wi-Fi. In addition to tokenbased methods we adapted we adapted two relevant OOB techniques and included arecent approach using surface-con ned Wi-Fi [3] as context based methods. The methodswere selected to ful ll the following constraints: Non-mediated : Intuitive :The user can perform the login at any time by himself.Wi-Fi hotspots are broadly used by ordinary non-expert users. Platform-independent :Mobile devices and OSs are diverse and volatile.2

Explicit :The system can record consent of the user when connecting.The last constraint distinguishes the evaluated methods from access control based onGeofencing[14] or overprovisioning [4], which both allow con ning Wi-Fi networks tocertain physical boundaries, enabling location based access. Although it can be arguedthatimplicitaccess is generally preferable, additional to ful lling legal contstraints, theproposed explicit methods do not require overprovisioning in the infrastructure and aretherefore also deployable in an light-weight and interchangable fashion.For platform-independence, we selected methods that can solely be implemented usingweb technology, thus easily be adapted to user needs and incorporated intoFor this paper we implemented all on top ofGetDeviceMotion().HTML5features likecaptive portals.GetUserMedia()andThis has the aditional bene t that the schemes are interchangeableand a set of di erent methods can be o ered to respect user preferences or hardwareconstraints. The source code of the implementations will be released as part of thePublic Inclusive InfrastructureUsername:passwordGlobal1(GPII) component repository and are freely available .This authentication scheme is well-known and standard today.A user enters his credentials into a form and presses a login-button to gain access. Weincluded it as a base line and to back the obvious hypothesis that this scheme is not wellsuited for handheld devices.QR CodesBy encoding a URL containing the login credentials into a QR code it canbe scanned with any camera phone. We accessed the camera image throughdecoded it using aNFCJavaScriptjsqrcode ).HTML5andlibrary (Login information can also be stored as URL in aNear Field Communication(NFC) tag and accessed remotely with many modern phones. An alternative would havebeen NFC-basedWi-Fi Protected Setup,which we refrained from using due to knownvulnerabilities [19] and lacking personalizability.2DST Sheet A detailed description ofDimensional Signal Transmission (2DST)sheet was published in previous work [3].users connect to the open Wi-Fi and thetheTwo-waveguideTo log on,captive portalpage prompts them to place the device on the sheet (seeFigure 1) and acknowledge the coupling. The device cansubsequently be removed from the sheet and access isgranted.KinectFigure 1: Phone, 2DST sheet.We adapted thePoint&Controlsystem [2] that uses theMicrosoft Kinect ( rstThe captive portalgeneration) for user-device association based on the user prompts the user to press a button and raise an arm to connect. If the gesture ismatched by the Kinect, access is granted. For this usability test we used this very simple1

scheme, that does not match the accelerometer pattern of the phone with theKinectmodel for added contextual prove, which is possible on most modern devices also usingHTML5[5].Audio Context The last context-based method employs ambient audio as OOB chanPintext paring [16]. The entropy of generated ngerprints generally makesnel, as used inthem suitable to be used as a shared secret [15]. To log in, a user presses a button onthecaptive portalnizesand both phone and server record for eight seconds. A server synchro-2 both recordings, calculates ngerprints and compares them. If they are su cientlysimilar, access is granted.3.1 Task and Session StructureParticipants were asked to connect to an open Wi-Fi, authenticate their device usingthe given mechanism and open a browser to access a web page.given aSamsumg Galaxy SIIIFor this, they weresmartphone, which supported all of the six methods. Thetest sessions were conducted in an o ce at our lab in German or English, dependingon the subject's preference.Participants were welcomed and guided to the test roomone at a time (1-on-1 moderated sessions, see Figure 2). The test's setup and intentionwere introduced and subjects read and signed a privacy statement.Subsequently, themoderator collected demographic data (age, gender, etc.) and some information on thesubject's habits regarding technology use (frequency of handheld Wi-Fi access, usageof public Wi-Fi hotspots, etc.)by means of a pre-test questionnaire.After that, themain phase of the test commenced: Subjects, in turn, completed the task i.e. log on tothe Wi-Fi and open a website for each method and ll in a questionnaire. The orderof the six methods was shu ed to avoid biases from practice or fatigue. Each methodwas explained beforehand and written descriptions of how to proceed were availablethroughout the test.After repeating the three steps for each method, the participantwas asked to ll in the post-test questionnaire. Sessions took between 52 and 100 minutes,with an average of 69 minutes.StartTest conductionComplete for each method (randomized)Task: log on and open web pageIntroductionTest intention, privacy agreementPre-test questionnaireDemographic data (age, gender,etc.)UEQ questionnairePost-test questionnaireComparative rating (ease of use,attractiveness, perceivedsecurity, )SUS questionnaireQualitative feedbackEndFigure 2: Session structure each participant ran through during the user study.2In our tests synchronization signi cantly slowed down this scheme, as acurate time sync could not berealized in HTML.4

45 4940 4435 3930 3425 2920 24neverrarelymonthlyweeklydailyAge14PhD90 2 4 6 8 10Number of participantsHandheld Wi-Fi usage230314510402468Number of participants15Number of ly4Admin.2Master6Field of workCreative8Bachelor10EducationHigh School1Public Wi-Fi usage1210542Technical15051015Number of participantsiOSAndroidWindowsothern/a0 2 4 6 8 10Number of participantsHandheld OS37112220 2 4 6 8 10Number of participantsFigure 3: (Top) Participants by age, completed level of education and eld of work orstudies. (Bottom) Frequency of Wi-Fi network access with a handheld devicerespectively using open Wi-Fi hotspots, as well OS's installed on subjects'2personal devices .3.2 ParticipantsWe recruited 22 participants aged between 20 and 48, eight of them female. All of themattended voluntarily without being o ered a reward. Figure 3 shows the data on demographics and participants' habits collected using the pre-test questionnaire. The subjectscomposed a well-educated group, accustomed to the use of mobile devices and workingin di erent elds, the majority pursuing technical professions.None were security ex-perts. All participants reported accessing the Internet once or more per day. Most (14)also daily used handheld mobile devices to connect to Wi-Fi networks, some weekly (3).Three participants said that they rarely accessed Wi-Fi with a handheld device and twonever at all. Most subjects often used public Wi-Fi hotspots, only one never did, andtwo rarely. All others accessed public Wi-Fis at least monthly, four even daily. Overall,the subject group is suitable for an initial assessment of the selected methods, as theyare digitally literate and familiar with the presented task.3.3 Questionnaire DesignThe study we conducted is mostly summative, with additional formative aspects.Wewere looking to nd out whether the proposed solutions could in principle satisfy userneeds. To quickly assess both usability and user experience of the tested schemes, weconsidered di erent standardized questionnaires. TheSystem Usability Scale (SUS)[1]has been applied to a wide range of systems in the past 20 years, from printers overphones to desktop and web applications.Subjects express their level of agreement toten simple statements using a ve-point Likert. It is slim, short term viable and yields2Three participants chose multiple options.5

a single score as result, which is already generalizable at relatively small sample sizes[17].As alternative, we looked at theUser Experience Questionnaire (UEQ) [11]. Itannoying and enjoyable ). Users expressconsists of 26 pairs of opposing attributes (e.g.their agreement with them on a seven-point Likert. The UEQ yields six di erent scoresfor the categoriesnovelty.attractiveness, perspicuity, dependability, e ciency, stimulation,andWe decided to use both UEQ and SUS, as they can be lled in quickly and wewere interested to see if they yielded consistent results, since the SUS focuses on usabilitywhile the UEQ aims at assessing the whole user experience.Aside from summative data from the two questionnaires, we also collected three to vequalitative statements about what the subjects liked and disliked about each system. Allof this was done separately for each of the methods. In addition, we constructed a shortpost-test questionnaire which prompted the subjects to directly compare the systemswith each other, by ranking them regarding their ease of use, perceived security andattractiveness.Participants were also asked which of the systems (if any) they wouldrecommend to friends or acquaintances. Finally, they were given the possibility to specifyadditional free text comments. The moderator also recorded any unprompted statementsmade throughout the test.Aside from the questionnaires, we recorded the number ofattempts needed to complete the task and the time to do so.3.4 Data CleansingOverall, our implementations proved to run stably and the conduction of the study wentsmoothly. In two cases however, we experienced software problems that led to di cultiesin completing the task: A software crash interrupted the task completion in six cases oftheAudio Contextlogin and an error dialog was shown.In these cases, the task wasrepeated and participants were instructed to disregard the rst failed attempt. Failedtries were not included in the results for task times or number of attempts.In twoinstances involving the 2DST sheet, a software bug prevented the correct recognition ofthe device by the sheet, leading to an unusually high number of tries (12 respectively 8attempts). To avoid skewed results, the data from these two runs was removed from theset.Regarding the nal three ranking questions, subjects explicitly were given the optionto rank two systems equally by assigning the same ordinal number. However this leadcompetition ranking (i.e. leaving a gap in the ranking whendense ranking (no gaps). To reach a realistic rankall participants, we transformed the data to fractional rankingto some participants usingseveral systems tied) and others usinging when averaging overscores, as those have the property that the ranking numbers' sum is the same as understrict ordinal ranking.4 ResultsThis section presents the results of our analyses. First, we show the quantitative metrics (attempts, task time), followed by the SUS and UEQ scores. Finally, comparativestatements and qualitative feedback ratings are presented.6

Table 1: Automatically collected metrics (sorted by median time per attempt).Meth.NFCKinect2DSTQRPwdAudio# Attemptsmin max med. mean conf.31111.00 n/a1411.64 0.381311.45 0.331311.55 0.331211.09 0.421211.18 0.17Task time (overall)min max med. mean conf.2s 15s6s6.5s 1.367s 57s 11s 19.3s 5.667s 56s 14s 19.3s 5.8110s 91s 33s 44.9s 11.9034s 153s 51s 60.3s 1.6777s 381s 126s 157.3s 33.43Task time (per attempt)min max med. mean conf.2s 15s6s6.5s 1.367s 17s 10s 10.5s 1.167s 34s 11s 13.2s 2.656s 81s 26s 30.4s 7.5134s 111s 51s 54.4s 1.2577s 215s 119s 130.3s 16.254.1 Quantitative Performance MetricsTable 1 shows the number of attempts and the time needed to perform the login task(both overall and averaged per attempt).NFC was the only scheme which took allsubjects only one attempt, all others had to be repeated at least once by at least oneparticipant.andWhile this was seldomly necessary forAudio Contextusername:password(2, mistyping)(4, ngerprints too di erent), it happened more often with the 2DSTsheet (6, device removed from sheet to early). More than a third of the subjects had torepeat their attempt using QR codes (8, recognition failed) or thefailed). Regarding task times, all methods exceptAudio ContextKinect(9, trackingperformed signi cantlyfaster than entering passwords. Again, NFC stood out, closely followed byKinectand2DST sheet. The use of QR codes still took only half the time of entering text credentials,while the audio login took much longer due to the long processing times.4.2 SUS and UEQ scoresFigure 4 and Table 2 show the UEQ category results, the overall UEQ score as well asthe SUS score over all participants.When looking at SUS scores, a percentile rankof below 60.0 is considered poor and an indicator for severe usability problems, whilevalues over 80.0 are generally good [17], 100.0 being the maximum possible. Regardingthe UEQ score, values below -0.8 indicate a negative rating, over 0.8 a positive one, andin between neutral. However, Schrepp et al. [13] point out that the actual interpretationof the ratings depends on the weight of the categories in the concrete application andattractiveness as most important,followed by perspicuity and dependability, and thirdly e ciency. The authors also provideintended user group. For normal end users, they regarda benchmark which is based on 163 studies with a total of 4818 participants and setsthe score boundary between above-average and below-average di erent for the individualattractiveness : 1.09, perspicuity :stimulation : 1.0, and novelty : 0.63).categories (0.9,dependability :1.06,e ciency :0.84 ,Basing our interpretations on these preliminary considerations, the ratings show aslightly di erent picture than the performance metrics before: Judging from the SUSscores, forKinectand QR, no compelling conclusion can be drawn from the SUS score.Both NFC and the 2DST sheet login can be considered good systems with no apparentusability problems, while3Audio Contextseems to have severe issues.Username:passwordAll con dence intervals ( ) in this work are constructed at a 95% con dence level.7

4Table 2: Mean UEQ category and SUS scores, including con dence , and median SUS.Meth.PwdQRNFC2DSTKinectAudioUEQ category scoresSUS scoreattrac. perspic. depend. e c.stimul. novel.mean conf. mean conf. mean conf. mean conf. mean conf. mean conf. median mean conf.-0.85 0.45 1.56 0.50 1.22 0.31 0.00 0.51 -1.30 0.41 -2.50 0.32 63.8 66.3 6.020.94 0.48 1.76 0.33 1.11 0.41 0.88 0.46 0.59 0.41 0.76 0.4473.877.7 5.562.11 0.36 2.72 0.21 2.02 0.40 2.24 0.39 1.60 0.37 1.81 0.40 95.0 91.6 4.451.67 0.38 2.31 0.34 1.30 0.45 1.78 0.35 1.39 0.45 1.90 0.45 86.3 83.5 4.720.75 0.60 1.83 0.43 0.22 0.50 1.13 0.38 1.48 0.40 2.16 0.2972.569.9 7.67-0.12 0.60 1.27 0.48 -0.15 0.32 -0.42 0.39 0.24 0.42 2.10 0.4155.058.5 gure 4: Mean scores for each access method: UEQ categories (left) and SUS (right).barely scores better than ambient audio, only just exceeding 60.0 points.When looking at the UEQ, we see a di erence between the overall UEQ and the SUSscore: The overall UEQ rating for thecodes, andAudio ContextKinect based system is betterusername:password.than that of QRscores higher thanRegarding the most relevant UEQ categories, the use of passwords scores high in theperspicuity and dependability, while receiving low ratings for attractivenesse ciency. This illustrates the additional information that can be drawn from thecategoriesandUEQ scores.Similar observations can be made for the other systems: While the QRmethod scores more or less average in all remaining categories, it performs well regardingperspicuity.Both NFC and the 2DST sheet login clearly score high in all categories.According to the UEQ scores, the main shortcoming of theKinect scheme is dependability,number ofwhich is in line with the observations made from the performance metricattemptswhile the e ciency score adequately re ects the short completion times of theAudio context scores badly across the board, with the exception of the categoriesperspicuity and novelty. An aspect clearly di ering from the performance metrics are thenon-task related hedonic quality aspects, that express how novel and stimulating usersscheme.percieve a system to be [11]. Not surprising, the password scheme has a very low ratinghere, while the context-based methods score high.8

NFC2DSTQRKinectAudioPasswordMost UsableMost 5.144.4352Most Secure1.6144.615.4362462.14Mean Fractional Rank (J better 122530510 15 20Number of participantsFigure 5: Subjective ranking of the methods regarding usability, perceived security andattractiveness (left) and which schemes participants would recommend (right).4.3 Comparison & PreferencesThe last section of the post-test questionnaire speci cally prompted the users to rank thesix methods compared to each other. The resulting ranking (see Figure 5) is consistentusability and attractiveness : NFC clearly spearheads, followed by theKinect and then Audio Context. The classic username:passwordscheme came out last. The rankings so far are in agreement with the UEQ attractivenessfor the two aspects2DST sheet, QR codes,rating.Interestingly, theperceived securityranking paints a di erent picture: Generally, par-ticipants felt that the token based methods were more secure than the context-basedschemes (see Figure 5).Altogether, almost all (20) participants stated that they would recommend NFC asWi-Fi access method to friends or acquaintances.More than half would recommendusing QR codes (12) or the 2DST sheet for contextual access (13). Notably, despite theotherwise rather poor ratings of the audio based scheme, still ve participants wouldrecommend the system, more than the classic password (3) or theKinectbased scheme(2).4.4 Qualitative Statements & User FeedbackAs expected, the study con rmed thatusername:passwordcredentials are not suitableintuitive(8) and secure (10), almost all users said that smartphones are inadequate for enteringrandom strings (18). As unique property, subjects saw that text is memorizable (P 06),for Wi-Fi authentication on mobile devices.While users rate the method asproviding an abstract token. Regarding QR codes, we observed that opinions were divided: Participants both describe the method to beas well asfastandslowintuitive(15) andcomplicated(9)(9 each). Some (6) also stated that they either experienced oranticipated problems in bad lighting. (P 07:the shadows of hand and phone interfered ).Regarding the necessary permission for the browser to access the camera,P 22reported:giving deep system access without exactly knowing what is going on makes me uneasy.Concerning NFC, participants predominantly gave positive comments (see Table 3):Most subjects characterized it asintuitve(18) andfast(17). However, a few users (5)also voiced technical concerns, ranging from fear of high energy consumption (P 14:always on? )NFCor losing the tag to security concerns, as the phone did not ask permission9

before opening the webpage.ParticipantP 01one has to use both hands.also speci Some usersTable 3: Amount of posi-(4) also pointed out that not every device features NFC (e.g.tive and negativecally disapproved thatP 15: my iPhone doesn't have an RFID reader ).Regarding the 2DST sheet, users reported that it wasvative /cool(6) as well asfast(13) andeasyinno-(21) empha-sizing that they needed to do very little (P 06:that's it? ).Some users perceived the system as limited in terms of beingxed to a location (3) and one subject did not grasp the concept of context-based access (P 01:not t in my bag ).big and bulky and wouldAnother user was concerned regardingpossible radiation from the sheet.comments(bothpromptedandunprompted).Method Pos. Neg. 247783.281.971.191.130.980.62Kinect, positive and negative comments nearly balanced each other: On onefast (13), intuitive (14) and cool (13). On the other, somepeople were embarrassed (8, e.g. P 08: don't want to jump about and attract attention inpublic ) and some voiced their concerns on being recorded and possible privacy implications (2, e.g. P 05: sense of being under surveillance ). Others again liked the aspect ofperforming an activity in order to log on (8). Regarding the Ambient Audio login, theusers' main issue was the long wait (21), followed by technical concerns, such as interference from handling noises or unreliability in silent ambiences (2, e.g. P 11: especiallyproblematic for mutes ). Four users expressed disbelief that the method would work atall (P 13: I have the feeling that this will often fail ). Regarding the security, the methodfelt both insecure (5, e.g. P 18: I guess that many false positives occur ) and very secure(4) to the users. As with the Kinect, privacy concerns were also expressed (2). On theother hand, participants characterized the system as innovative (8) and intuitive /magic(12, e.g. P 07: great, I don't have to do anything ).As for theside, users saw the system asGeneral comments mostly concerned lack of understanding regarding context-basedaccess applicability (e.g.P 18: I don't see use cases, except maybe in trains ).5 DiscussionWe implemented sixHTML5 -basedtechniques for associating handheld devices to Wi-Fi networks and evaluated them regarding their usability. As expected, it backed thatusername:passwordcredentials are unsuitable for handheld Wi-Fi access, and that theother ve schemes may to a varying degree present viable alternatives.Our study yielded some interesting results: A general observation is that purely summative studies may not re ect the full range of relevant aspects.Although both SUSand UEQ seem suitable to determine if severe issues exist, caution should be exercisedwhen using them to rank systems. While we can see that they generally show similartendencies, the UEQ addresses the whole user experience and its categories can providehelpful additional insights regarding the area possible problems may reside in. This isespecially true for systems whose SUS score lies in the gray area between 60.0 and80.0 percent. Placing too much emphasis on mere speed or completion rates as a factor10

may be misleading as well, especially in the context of usable security. In this area, itis important to augment standard usability testing with some metric that speci callyaddresses aspects like perceived security, trust, etc.The multitude of di erent user statements revealing interesting issues both actual andperceived underlines the importance of also collecting qualitative feedback. While forthe two best performing schemes (NFC and 2DST) most metrics are in agreement witheach other, singular issues are revealed by other metrics, such as qualitative feedbackfor NFC or the number of attempts with the sheet.For some methods, only the fullrange of metrics paints an adequate picture for the assessment of the system, especiallythose on which opinions are divided: TheKinectachieved average to high summativeratings and the second to best completion times, but subjects perceived it as insecureand least recommended it. As for theAudio Contextscheme, despite mostly bad ratingsand performance, more than half of the subjects regarded it as intuitive and almost aquarter would recommend it. This illustrates that multiple metrics also allow discerningbetween fundamental issues and speci c problems that can e.g. be attributed to theimplementation and may be remedied in the future.An important realization regarding methods involving cameras or microphones wasthat many participants voiced their concerns on being recorded and possible privacyimplications, as well as a sense of being under surveillance. We conclude that systemsinvolving video or audio recording should probably be avoided, and if such methods areconsidered, it is important to convey to the users that their privacy is protected. Furthermore, deep system access by web apps (e.g. camera or sensor access or automaticallyopening scanned URLs) should be transparent and only occur with user consent. Whenconsidering methods involving visible activity, embarrassment is an important factor,even if theKinectbased scheme caused both positive and negative feedback regardingthe activity. We conclude that the applicability of such a system depends strongly onthe situation, user group and maybe also cultural aspects. As a design guideline for sys-Ambient Audio ) and/or losing focus of the screen (as2DST ), non-visual feedback such as an auditory signal istems that involve little interaction (sometimes seen withKinectoradvisable to indicate success of the association process.6 Conclusion & Future WorkWe believe that good integration of classic usability studies and metric based analyses,as well as an analysis of other requirements on the

TECO, Karlsruhe Institute of ecThnology (KIT), Karlsruhe, German,y . al. [16]. These technical publications do not take human factors into account, which we targeted in this work. Several other authors conducted comparative user studies on the . High School Bachelor Master PhD 8 2 8 4