applied sciences

Article Haptic Soft-Keyboard for Tablet-Sized

Sunjin Yu 1, Sung Hun Jin 2 and Kwangtaek Kim 3,*

1 School of Digital Media Engineering, Tongmyong University, Busan 48520, Korea 2 Department of Electronics Engineering, Incheon National University, Incheon 22012, Korea 3 Department of Information and Telecommunication Engineering, Incheon National University, Incheon 22012, Korea * Correspondence: [email protected]; Tel.: +82-835-8628

 Received: 4 June 2019; Accepted: 26 July 2019; Published: 31 July 2019 

Abstract: Most users are not satisfied with the typing vibration feedback from linear motors of mobile phones since it feels like buzzing rather than physical key pressing feedback. For larger touchscreens, such as 10.1 inch tablets, there is no commercial tactile soft keyboard developed yet. Therefore, there is the need to develop a haptic soft imitating the physical keyboard for the tablet-sized . In the present study, we present a new approach introducing an optimal design of a multilayered piezoelectric actuator that meets the need for a haptic soft keyboard for 10.1 inch tablets, implementing stronger tactile feedback that utilizes human sensitivity to time differences obtained by a psychophysical experiment and equalizes the force distribution of the touchscreen when using multiple piezoelectric actuators. The developed system was evaluated by a user study measuring typing performance.

Keywords: haptic feedback; soft keyboard; mobile input device; user interface; haptic soft keyboard; tablet touchscreen; tactile perception

1. Introduction These days, the soft keyboard is one of the most used input methods in devices from mobile phones to tablets. Researchers have been working to invent or develop advanced typing methods for mobile touchscreens for over a decade. Most of the efforts were put forth to propose new layouts of key characters by analyzing users’ typing behaviors or to improve typing productivity using add-on methods such as typing error correction and autocomplete. However, most of the frustration users feel is due to there being no real key click feedback when typing on the touchscreen. For this, phone users tend to use synchronized vibrations generated from a low-cost linear motor when they type. Nevertheless, most of the users are not satisfied with the built-in vibration feedback of mobile phones, since it feels like buzzing rather than physical key pressing feedback. For larger touchscreens, such as 10.1 inch tablets, there is no commercial tactile soft keyboard available yet. Therefore, there is a need to develop a haptic soft keyboard technology that imitates the physical keyboard for larger touchscreens of mobile devices, including tablets. To this end, piezoelectric actuators that have fast response time (10 to 50 µs) and strong deformation (even up to a few hundred millimeters when high voltage is applied) were introduced to be mounted inside touchscreens [1,2]. Despite the strength of piezoelectric actuators, there is little known about how to use them for implementing tactile key click effects on commercial tablet-sized touchscreens, because piezoelectric actuators generally require very high voltage to be deformed. For instance, single layered piezoelectric actuators are operated at 100 to 200 volts (peak to peak). The required voltage can be reduced by designing multilayered piezoelectric actuators; however, due to the capacitance, the input current should be increased proportional to the number of internal layers. Concerning the need for

Appl. Sci. 2019, 9, 3080; doi:10.3390/app9153080 www.mdpi.com/journal/applsci Appl. Sci. 2019, 9, 3080 2 of 14 stronger tactile feedback for larger mobile devices such as tablets, multilayer piezoelectric actuators are preferred to implement tactile feedback on touchscreens. To provide tactile key click effects on the entire touchscreen of a 10.1 inch tablet, multiple piezoelectric actuators should be embedded and controlled accurately using a mobile battery. However, driving multiple piezoelectric actuators at the same time on the touchscreen of a tablet is a challenging issue. The first prototype of a haptic soft keyboard using four piezoelectric actuators was developed with two commercial mobile devices (a and Surface Pro tablet) by Han and Kim [1]. They used a portable piezoelectric driver that can drive up to four multilayered piezoelectric actuators at the same time. For the haptic soft keyboard of the tablet, they developed the haptic soft keyboard and implemented an add-on interface for tactile feedback to the touchscreen. A user study measuring the performance of key typing on the touchscreen for three feedback conditions (visual only, visual-aural, and visual-haptic) was conducted, and the results showed improved key typing performance with haptic feedback in terms of accuracy, speed, and usability. They showed how to utilize multiple piezoelectric actuators to simulate a seemingly real key click effect on a larger touchscreen. However, there was a 10 ms time delay to get tactile feedback, the multilayer piezoelectric actuators were too thick, and the piezo driver circuit was too big to be embedded. In a following study, Kim developed a standalone haptic soft keyboard that can be simply added to existing tablet devices regardless of operating systems [2]. In the present study, we present a new approach introducing an optimal design for a multilayered piezoelectric actuator that meets the need of a haptic soft keyboard for 10.1 inch tablets, implementing stronger tactile feedback by utilizing human sensitivity to time differences obtained by a psychophysical experiment and equalizing the force distribution of the touchscreen. This paper is organized as follows. Section2 introduces related studies that deal with new designs of the soft keyboard by analyzing users’ typing behaviors or adding tactile feedback. Section3 presents how we found the optimal parameters and designed multilayer piezoelectric actuators needed for 10.1 inch touchscreens. Additionally, a perception study to estimate human sensitivity to time differences on touchscreens and an equalization algorithm for uniform force distributions are described. Experimental results of the quantitative evaluation of the developed piezoelectric actuators and force distribution method and the results of a user study measuring typing performance with the developed haptic soft keyboard are reported in Section4. Finally, discussions are provided in Section5.

2. Related Work Early studies focused on analyzing typing fingers and key typing patterns on touchscreens to improve existing soft keyboard designs. Findlater et al. designed a new QWERTY layout of a soft keyboard by adopting a typist’s typing patterns [3,4]. They conducted experiments to analyze typing positions and postures on touchscreens over a period of time and applied the results for a personalized layout. Their studies showed that a personalized layout based on typing patterns and behaviors significantly improved typing speed on the touchscreen of mobile devices. Similarly, Li et al. designed a QWERTY layout in a single line that consists of eight buttons and recognizes the user’s intention of the text input from typing finger postures [5]. The results demonstrated that users quickly learn the new text entry method. In line with this approach, Goel et al. proposed a new text entry method utilizing the user’s finger postures and motions to improve typing performance on touchscreens [6]. Most recently, Jo and Cho proposed a new method using random forest that automatically recommends a preferred layout for the user by adopting trained data of typing behaviors and patterns [7]. Due to all of these studies, the usability of the soft keyboard on touchscreens has been greatly improved. Unlike the physical keyboard, it is hard to confirm that a key was pressed or clicked on flat touchscreens. A simple way of confirming a pressed key is to change the color of the typed key character or to provide a key-click sound effect. However, users are not satisfied with these methods and prefer to feel a tactile key-click effect as they do with the physical keyboard. To add a tactile key effect, Poupyrev and Maruyama designed and implemented a tactile feedback interface for a Personal Appl. Sci. 2019, 9, 3080 3 of 14

Digital Assistant (PDA) touchscreen [8]. They conducted a user study with ten participants to prove the usability of the tactile feedback in comparison with audio feedback. Their results demonstrated the effectiveness of tactile feedback on the touchscreen of a PDA. Rabin and Gordon studied a similar study but for a medical application [9]. They showed that tactile feedback plays an important role in typing by letting users know the start position without looking at the touchscreen. From a practical perspective, Brewster et al. studied the effectiveness of vibro-tactile feedback using a commercial vibrator attached to the backside of a mobile device [10]. The results showed that typing speed and accuracy were improved through the tactile feedback generated from the backside. Hoggan et al. showed the same result for text typing on a touchscreen [11]. Koskinen et al. studied pleasant feelings on touchscreens with two different actuators, piezoelectric actuators and linear motors [12]. Their study showed that users feel more pleasant with piezoelectric actuators. Jansen et al. developed the MudPad, which is capable of localized haptic feedback on multi-touch surfaces [13]. Hoffman et al. demonstrated that tactile feedback can also be used for preventing typing errors [14], and McAdam et al. investigated tactile feedback on text entry with tabletop computers. In their study, tactile feedback improved typing speed, even though actuators were attached to users’ bodies [15]. Kaaresoja at al. showed that the latency of touch feedback did not significantly affect typing performance on the touchscreen [16]. In regards to developing a prototype with multiple piezoelectric actuators, Han and Kim showed the possibility of completely embedding the actuators into the commercial mobile devices (phones and tablets) [4,17]. Their study showed how to drive multiple piezoelectric actuators on a commercial tablet for the first time and the effectiveness of a tactile key click effect on touchscreens by conducting user studies. However, the prototype had a time delay (over 10 ms), which resulted in delayed tactile feedback to fast typing. Kim extended the approach by developing a standalone haptic soft keyboard system using a microcontroller [2]. Most recently, Jeon et al. studied the effectiveness of tactile feedback on the virtual keyboard [18].

3. Materials and Methods In this section, an optimal design of piezoelectric actuators, a perception study measuring discrimination thresholds on time differences of a tactile signal on touchscreen, and an algorithm that globally equalizes force distributions on touchscreens are presented.

3.1. Design of Haptic Actuators for Tablet Touchscreens

3.1.1. Piezoelectric Ceramic Actuators The piezoelectric effect is to generate an electric charge in response to mechanical stress, and its inverse effect refers to mechanical deformation when an electric field is applied in the polarization direction of piezoelectric materials. These effects can be used as both sensors and actuators. In this study, we focus on the inverse piezoelectric effect, which is the case for piezoelectric actuators. Fundamental equations for dielectrics can be written as:

A C =  (1) t Q = CV (2) where A, t, and  denote the area of the capacitor plate, the thickness of the plate, and the dielectric constant, and Q, C, and V are charge, capacitance, and voltage applied, respectively. Then, (2) can be rewritten by (1): AV Q = (3) t Appl. Sci. 2019, 9, 3080 4 of 14

Additionally, the mechanical displacement δ, which can be defined as charge density or the ratio of charge to the area of the capacitor is obtained as follows:

Q V δ = = (4) A t Further, δ can be derived as a simplified form for two layered and three layered actuators from [19]:

3d l2V δ = 31 (5) t2   m + 2 6S11d31 tm tp l V δ =   (6) m 2 + 2 + 3 + E 3 2S11 3tmtp 6tmtp 4tp S11tm m 2 E where S11 is the elastic coefficient of the supporting layer (m /N), S11 is the elastic coefficient of the 2 piezoelectric layer (m /N), tm is the thickness of the supporting layer (mm), tp is the thickness of the piezoelectric layer (mm), l is the length of the piezo, and d31 is the piezoelectric coefficient (mm/V).

3.1.2. Optimal Design of Piezoelectric Benders In this study, we focus on maximizing the mechanical displacement δ while meeting the size requirements (thickness less than 0.5 mm and width of 8 mm). To maximize the displacement δ assuming the material coefficients and the applied voltage to be constant, the layer thickness should be optimized by controlling the number of layers and the layer ratio. Thus, two factors (layers and the ratio of layers) were investigated by simulations, and then the structure of a piezoelectric bender to be mounted underneath the tablet touchscreen was designed and manufactured. There is limited space on the bezel of tablet touchscreens; thus, the width and the thickness of the actuator were fixed to 8 mm and 0.4 mm, respectively, due to the space limit [10 mm width on the bezel and a 0.5 mm gap between the touchscreen and the liquid crystal display (LCD) panel] of commonly used 10.1 inch tablets). A software tool, COMSOL Multiphysics, using a finite element method was used for simulating the bending effects of piezoelectric actuators from a single layer (called unimorph) to three layers. Our intention with the tool is to simulate the behavior of piezo bending in the z axis as a square wave (500 Hz, 20 Vpp) is applied, which will generate a force under the touchscreen that is eventually bent. A rectangle plate [(H) 40 (W) 8 (T) 0.4 mm] was designed and simulated with × × parameters [pole directions: down (unimorph); up and down (bimorph); up-up-down (three layers), layer thickness: 0.3 mm (unimorph); 0.15 mm each (bimorph); 0.1 mm and 0.14 mm (three layers) for Pe and Pi, respectively]. The pole directions were found by a pilot analysis investigating the bending strength versus pole directions (in series or parallel). The simulation was run on a Windows machine (CPU i5-6500 3.2 Hz, RAM 8 GB, Graphic GTX 1050). Figure1 shows the results of simulating piezoelectric benders driven by a square wave (500 Hz, 20 Vpp). The magnitude of displacement increased as the number of piezo’s layers increased. The displacement of the three-layer bender was three times greater than that of the two-layer bender in the series pole direction, which justified the need for three layers. Furthermore, the simulation indicated that the ratio of the three layers could affect the displacement performance. Thus, the other simulation was conducted to figure out the optimum ratio value. For the simulation, three layers were defined as Pe (external layer) and Pi (internal layer), and the ratio was computed by Pi/Pe. The right plot of Figure1 shows that the max displacement of the bender was obtained at 1.4 (i.e., Pi = 0.14 mm and Pe = 0.1 mm), which demonstrates that the thickness of the internal ceramic layer (Pi) significantly affected the bending performance of the piezo bender. Based on the simulations, a final design of piezoelectric actuators (three layers and 1.4 as its ratio) was determined, as seen in Figure2. The estimated max displacement was 600 µm at 100 Vpp. Appl. Sci. 2019, 7, x 5 of 14 Appl. Sci. 2019, 9, 3080 5 of 14 Appl. Sci. 2019, 7, x 5 of 14

Figure 1. Results of bending simulations for a piezoelectric plate [(H) 40 × (W) 8 × (T) 0.4 mm)] with Figure 1. Results of bending simulations for a piezoelectric plate [(H) 40 (W) 8 (T) 0.4 mm)] with COMSOLFigure 1. Results Multiphysics: of bending displacement simulations (μ m)for ina piezoelectthe z axis ricfrom plate a single [(H) 40layer ×× (W) to three8 × (T) layers 0.4 mm)] (left) with and COMSOL Multiphysics: displacement (µm) in the z axis from a single layer to three layers (left) and the theCOMSOL optimal Multiphysics: ratio (Pi/Pe) displacementfor a three-layer (μm) bender in the to z beaxis bent from most a single (right). layer A square to three wave layers (500 (left) Hz, and 20 optimal ratio (Pi/Pe) for a three-layer bender to be bent most (right). A square wave (500 Hz, 20 Vpp) theVpp) optimal was applied ratio (Pi/Pe) to the piezofor a three-layerplate for simulations. bender to be bent most (right). A square wave (500 Hz, 20 was applied to the piezo plate for simulations. Vpp) was applied to the piezo plate for simulations.

Figure 2. Final design (a–c) of a three-layer piezoelectric actuator based on the results of simulations Figure 2. Final design (a–c) of a three-layer piezoelectric actuator based on the results of simulations and (d) is a manufactured actuator. Pe and Pi are 0.1 mm and 0.14 mm, respectively. A and B are and (d) is a manufactured actuator. Pe and Pi are 0.1 mm and 0.14 mm, respectively. A and B are positiveFigure 2. and Final negative design electrodes,(a–c) of a three-layer and G is the piezoelectric ground, respectively. actuator based The final on the thickness results (0.39of simulations mm) was positive and negative electrodes, and G is the ground, respectively. The final thickness (0.39 mm) was estimatedand (d) is includinga manufactured bonding. actuator. Pe and Pi are 0.1 mm and 0.14 mm, respectively. A and B are estimatedpositive and including negative bonding. electrodes, and G is the ground, respectively. The final thickness (0.39 mm) was 3.2. Perceptionestimated including Study to Estimatebonding. the Sensitivity to Time Difference of Tactile Key Click 3.2. Perception Study to Estimate the Sensitivity to Time Difference of Tactile Key Click Understanding the human perception of key click tactile feedback on touchscreens by piezoelectric 3.2. Perception Study to Estimate the Sensitivity to Time Difference of Tactile Key Click actuatorsUnderstanding is a crucial the step human to designing perception the haptic of key soft keyboardclick tactile due feedback to the limited on touchscreens tablet power by to consistentlypiezoelectricUnderstanding driveactuators multiple the is ahuman crucial piezoelectric stepperception to designing actuators. of key the Therefore, hapticclick tactile soft our keyboard strategicfeedback due approach on to thetouchscreens limited was to tablet drive by piezoelectricpower to consistently actuators driveis more a crucial multiple efficiently step piezoelectric to (saving designing power) actuators.the haptic while Therefore,soft maximizing keyboard our human duestrategic to the perceptibility approach limited tablet was on touchscreenspowerto drive to consistentlypiezoelectric by utilizing drive temporalactuators multiple discriminationmore piezoelectric efficiently thresholds actuators. (saving of Therefore, thepower) piezo bending whileour strategic maximizing effect. approach It was learnedhuman was throughtoperceptibility drive a pilotpiezoelectric on experiment touchscreens actuators that by humans utilizing more perceive efficientlytempor aal stronger discrimination(saving key power) click thresholds eff ectwhile when ofmaximizing subsequentthe piezo bendinghuman double bendingperceptibilityeffect. It imperceptiblywas learnedon touchscreens through occurs. Thisbya pilot utilizing double experiment bendingtempor alth eff discriminationatect humans of piezoelectric perceive thresholds actuators a stronger of canthe key bepiezo achievedclick bending effect by drivingeffect.when subsequentIt with was alearned pulse double train through thatbending hasa pilot twoimperceptibly peaksexperiment with occurs. anthat imperceptible humans This double perceive interval, bending a st whichronger effect should ofkey piezoelectric click be smaller effect thanwhenactuators a subsequent temporal can be discriminationachieved double bending by driving threshold. imperceptibly with To a estimatepulse occurs. train the Thisthat temporal doublehas two touch bending peaks discrimination with effect an of imperceptible piezoelectric thresholds onactuatorsinterval, the touchscreen which can be should achieved of a be 10.1 smaller by inch driving tabletthan awith devicetemporal a pulse (Samsung di scriminationtrain Galaxythat has threshold. Tab), two apeaks psychophysical To with estimate an imperceptible the experiment temporal wasinterval,touch designed discrimination which and should conducted thresholds be smaller using on than the an atouchscreen adaptive temporal psychophysical di ofscrimination a 10.1 inch ta methodthreshold.blet device for To two(Samsung estimate conditions, theGalaxy temporal haptics Tab), onlytoucha psychophysical and discrimination haptics combinedexperiment thresholds with wason audio the designed touchscreen (key click and soundof conducted a 10.1 with inch the usingtablet same devicean 500 adaptive Hz(Samsung square psychophysical Galaxy wave). Tab), The combinedamethod psychophysical for condition two conditions, experiment was also haptics tested was toonly designed see and the haptics increased and conductedcombined sensitivity withusing by audio adding an adaptive(key audible click feedback, psychophysicalsound with which the ismethodsame commonly 500 for Hz two usedsquare conditions, for wave). typing hapticsThe on mobilecombined only touchscreens. and condition haptics combined was also tested with audio to see (key the increasedclick sound sensitivity with the bysame adding 500 Hz audible square feedback, wave). The which combined is commonl conditiony used was for typingalso tested on mobile to see touchscreens.the increased sensitivity by adding audible feedback, which is commonly used for typing on mobile touchscreens.

Appl. Sci. 2019, 7, x 6 of 14

3.2.1. Participants Appl. Sci. 2019, 7, x 6 of 14 Appl. Sci.A total2019, 9of, 3080 fifteen participants (eight males and seven females, age range 22–30, average age6 of26.4 14 years3.2.1. Participantsold) took part in the experiment. All participants were right-handed by self-report. Two participants had previous experience with haptic interfaces and perception experiments. None of the 3.2.1.A Participants total of fifteen participants (eight males and seven females, age range 22–30, average age 26.4 participants reported any deficiencies in hearing or touch. years old) took part in the experiment. All participants were right-handed by self-report. Two A total of fifteen participants (eight males and seven females, age range 22–30, average age participants had previous experience with haptic interfaces and perception experiments. None of the 26.43.2.2. years Stimuli old) took part in the experiment. All participants were right-handed by self-report. Two participants reported any deficiencies in hearing or touch. participantsTwo stimuli, had previous the test experience and the reference with haptic (see interfaces Figure and3a,b), perception were designed experiments. and used None for of thethe participantsperception3.2.2. Stimuli study. reported In anythe deficienciestest pulse, the in hearing time interval or touch. t (Figure 3a) was controlled to increase or decrease from trial to trial based on participants’ responses during the experiment. For the reference 3.2.2.Two Stimuli stimuli, the test and the reference (see Figure 3a,b), were designed and used for the pulse, the duration was fixed to 2 ms defined by the period of 500 Hz. To drive piezoelectric actuators perception study. In the test pulse, the time interval t (Figure 3a) was controlled to increase or efficiently,Two stimuli, a simple the piezo test and driver the reference circuit was (see devel Figureoped3a,b), using were an designed 8 bit microcontroller and used for the (AtMega perception 128, decrease from trial to trial based on participants’ responses during the experiment. For the reference study.16 MHz) In thewith test an pulse, acoustic the signal time interval amplifiert (Figure (TI DRV3a) 2667, was controlled gain 38.4 dB). to increase As seen or in decrease Figure 4, from test trial and pulse, the duration was fixed to 2 ms defined by the period of 500 Hz. To drive piezoelectric actuators toreference trial based pulses on participants’were generated responses in the duringmicrocontr theoller experiment. and amplified For the to reference 65 Vpp pulse,with the the amplifier duration efficiently, a simple piezo driver circuit was developed using an 8 bit microcontroller (AtMega 128, wasand fixedthen sent to 2 msto a defined piezoelectric by the actuator period of attached 500 Hz. to To the drive 10.1 piezoelectric inch touchscreen actuators to simulate efficiently, a tactile a simple key 16 MHz) with an acoustic signal amplifier (TI DRV 2667, gain 38.4 dB). As seen in Figure 4, test and piezoclick effect. driver circuit was developed using an 8 bit microcontroller (AtMega 128, 16 MHz) with an reference pulses were generated in the microcontroller and amplified to 65 Vpp with the amplifier acoustic signal amplifier (TI DRV 2667, gain 38.4 dB). As seen in Figure4, test and reference pulses and then sent to a piezoelectric actuator attached to the 10.1 inch touchscreen to simulate a tactile key were generated in the microcontroller and amplified to 65 Vpp with the amplifier and then sent to a click effect. piezoelectric actuator attached to the 10.1 inch touchscreen to simulate a tactile key click effect.

Figure 3. Two stimuli, (a) test and (b) reference, simulating a key click effect used for the psychophysical experiment. The time interval (t) between two peaks was controlled during the Figureexperiment. 3. Two stimuli, (a) test and (b) reference, simulating a key click effect used for the psychophysical Figure 3. Two stimuli, (a) test and (b) reference, simulating a key click effect used for the experiment. The time interval (t) between two peaks was controlled during the experiment. psychophysical experiment. The time interval (t) between two peaks was controlled during the experiment.

Figure 4. Experimental setup of a perception study measuring temporal discrimination thresholds on aFigure 10.1 inch 4. Experimental touchscreen. setup The participantof a perception passively study measuring felt tactile feedbacktemporal ondiscrimination the touchscreen thresholds and then on respondeda 10.1 inch totouchscreen. the provided The stimulus. participant passively felt tactile feedback on the touchscreen and then responded to the provided stimulus. 3.2.3.Figure Procedures 4. Experimental setup of a perception study measuring temporal discrimination thresholds on 3.2.3.a Procedures 10.1 inch touchscreen. The participant passively felt tactile feedback on the touchscreen and then Aresponded two-interval to the forced provided choice stimulus. (2IFC) one-up one-down adaptive procedure [20] was used for the experimentA two-interval to estimate forced the choice discrimination (2IFC) one-up thresholds one-down of the adaptive tactile procedure key click e [20]ffect was simulated used for by the a piezoelectricexperiment3.2.3. Procedures to actuator estimate designed the discrimination in this study. threshol Participantsds of the were tactile tested key with click two effect conditions, simulated haptics by a alonepiezoelectric (H) or haptics actuator combined designed with in audiblethis study. sound Participants (HA). On were each trial,tested for with the Htwo (or conditions, the HA) condition, haptics A two-interval forced choice (2IFC) one-up one-down adaptive procedure [20] was used for the thealone participants (H) or haptics put theircombined index with finger audible on the sound tablet touchscreen(HA). On each (Samsung trial, for Galaxythe H (or Tab) the and HA) felt condition, the key experiment to estimate the discrimination thresholds of the tactile key click effect simulated by a clickthe participants effect simulated put their by the index piezoelectric finger on actuator the tablet driven touchscreen by the stimulus (Samsung signal Galaxy (a 500 Tab) Hz square and felt wave the piezoelectric actuator designed in this study. Participants were tested with two conditions, haptics signal).key click For effect the simulated HA condition, by the the piezoelectric participants actuator perceived driven the by key the click stimulus effect signal with the(a 500 same Hz squaresquare alone (H) or haptics combined with audible sound (HA). On each trial, for the H (or the HA) condition, wavewave (500signal). Hz) For through the HA both condition, the touchscreen the participants and the headphone.perceived the The key sound click was effect generated with the with same a the participants put their index finger on the tablet touchscreen (Samsung Galaxy Tab) and felt the soundsquare play wave function (500 Hz) implemented through both in thethe microprocessor.touchscreen and The the reference headphone. stimulus The sound was presented was generated with a key click effect simulated by the piezoelectric actuator driven by the stimulus signal (a 500 Hz square 2with ms squarea sound wave, play while function the test implemented stimulus was in controlledthe microprocessor. by a time interval The reference (t), as seen stimulus in Figure was3. wave signal). For the HA condition, the participants perceived the key click effect with the same Thepresented reference with and a 2 the ms test square stimuli wave, were while randomly the test displayed stimulus to was avoid controlled subjects’ by expectation a time interval errors. (t), The as square wave (500 Hz) through both the touchscreen and the headphone. The sound was generated participant’sseen in Figure task 3. was The to reference indicate whichand the one test was st feltimuli as twowere pulses. randomly According displayed to the to one-up avoid one-down subjects’ with a sound play function implemented in the microprocessor. The reference stimulus was adaptiveexpectation rule, errors. the value The participant’s of t in the test task pulse was was to indicate increased which after one an incorrectwas felt as response two pulses. and According decreased presented with a 2 ms square wave, while the test stimulus was controlled by a time interval (t), as after a correct response. The initial t value was chosen to be large enough (15 ms) so that the two peaks seen in Figure 3. The reference and the test stimuli were randomly displayed to avoid subjects’ of a test pulse train were clearly felt to the participant. The value of t then decreased or increased by a expectation errors. The participant’s task was to indicate which one was felt as two pulses. According

Appl. Sci. 2019, 7, x 7 of 14 to the one-up one-down adaptive rule, the value of t in the test pulse was increased after an incorrect Appl.response Sci. 2019 and, 9 ,decreased 3080 after a correct response. The initial t value was chosen to be large enough7 of (15 14 ms) so that the two peaks of a test pulse train were clearly felt to the participant. The value of t then decreased or increased by a step size of 6 dB depending on the participant’s responses. After the step size of 6 dB depending on the participant’s responses. After the initial three reversals, the value of initial three reversals, the value of t changed by 2 dB. The initial larger change in t was necessary for t changed by 2 dB. The initial larger change in t was necessary for fast convergence, whereas the later fast convergence, whereas the later smaller change improved the accuracy of threshold estimates. smaller change improved the accuracy of threshold estimates. Each adaptive series was terminated Each adaptive series was terminated after eight reversals at the smaller step size. The participants after eight reversals at the smaller step size. The participants were comfortably seated in front of were comfortably seated in front of the touchscreen with a keypad, as shown in Figure 4. They wore the touchscreen with a keypad, as shown in Figure4. They wore headphones to block any sound headphones to block any sound from the equipment for the H condition. Initial training was provided, from the equipment for the H condition. Initial training was provided, where a series of stimuli were where a series of stimuli were presented to familiarize the participants with the tactile key click effect presented to familiarize the participants with the tactile key click effect on the touchscreen and the two on the touchscreen and the two experimental conditions (H and HA). The training time varied from experimental conditions (H and HA). The training time varied from participant to participant and participant to participant and averaged 10 min. Each participant was tested six times per condition, averaged 10 min. Each participant was tested six times per condition, resulting in a total of twelve resulting in a total of twelve adaptive series per participant. It took between forty minutes to one adaptive series per participant. It took between forty minutes to one hour for each participant to hour for each participant to complete all the 12 series. complete all the 12 series.

3.2.4. Data Data Analysis Analysis For each adaptive series, thresholds were calculat calculateded from the values over the last eight reversals at the 2 dB step size. Specifically, Specifically, two thresholds were estimated by averagin averagingg the last eight reversals for thethe two two conditions conditions (H, (H, HA). HA). The discriminationThe discriminati thresholdson thresholds were equally were consideredequally considered “just noticeable “just noticeablediscrimination” discrimination” (JND) on temporally (JND) on temporally displayed tactiledisplayed pulses tactile on the pulses touchscreen. on the touchscreen.

3.2.5. Results Results of of the the Psychophysical Experiment Figure5 shows5 shows the resultsthe results of the psychophysicalof the psychophys experimentical experiment that estimated that temporal estimated discrimination temporal discriminationthresholds (3.96 thresholds ms for H and(3.96 2.27 ms msfor forH and HA) 2.27 of key ms clicksfor HA) on of a tabletkey clicks touchscreen. on a tablet The touchscreen. graph bars Therepresent graph the bars thresholds represent for the all thresholds participants for for all the pa tworticipants conditions, for the H andtwo HA,conditions, and the H error and bars HA, show and thestandard error bars deviations. show standard In the left deviations. plot, it was In the observed left plot, that it was the thresholdsobserved that for the Hthresholds condition for were the Hall condition higher than were those all ofhigher the HA than condition those of over the allHA the condition participants. over Itall was the alsoparticipants. apparent inIt was the rightalso apparentthat the average in the right threshold that the for average H was threshold larger than fo thatr H was of HA. larger A one-waythan that ANOVAof HA. A withone-way the condition ANOVA (H,with HA) the condition showed that (H, there HA) showed was a significant that there di wasfference a significant between difference the two conditions between the [F (1,two 118) conditions= 160.1, [pF(<<1, 1180.0001) = ].160.1, The resultsp << 0.0001]. demonstrated The results that demonstrated (1) a tactile display that (1 of) sequentiala tactile display two peaks of sequential with a smaller two peaksinterval with than a 3.96smaller ms couldinterval be than perceived 3.96 ms as could one peak, be perceived and (2) the as perception one peak, and of tactile (2) the key perception clicks could of tactilebe more key sensitive clicks could by adding be more a synchronized sensitive by audioadding cue, a synchronized which can serve audio as acue, guideline which can in both serve sound as a guidelineand touch in feedback both sound on touchscreens. and touch feedback The touch on sensitivitytouchscreens. (about The 4 touch ms) to sensitivity time difference (about measured 4 ms) to timein this difference study was measured slightly in smaller this study than was the 5slightly ms reported smaller in than Goldstein’s the 5 ms book reported [21]. Thein Goldstein’s increased booktouch [21. sensitivity The increased when combined touch sensitivity with aural when feedback combined was with expected aural due feedback to hearing was dominance—theexpected due to hearingsensitivity dominance—the to time difference sensitivity was about to 0.01time ms difference reported was in [21 about]. In this 0.01 study, ms reported the threshold in [21]. estimated In this study,by the the haptics threshold only condition estimated was by the adopted haptics for only the condition haptic soft was keyboard. adopted for the haptic soft keyboard.

Figure 5. Discrimination thresholds (temporal sensitivity) estimated on tactile key click effects on tablet touchscreens: thresholds estimated per participant (left) and the average thresholds (right) for the haptics alone (H) and haptics with audible sound (HA) conditions. Appl. Sci. 2019, 7, x 8 of 14

Figure 5. Discrimination thresholds (temporal sensitivity) estimated on tactile key click effects on Appl. Sci.tablet2019 touchscreens:, 9, 3080 thresholds estimated per participant (left) and the average thresholds (right) for8 of 14 the haptics alone (H) and haptics with audible sound (HA) conditions.

3.3. Global Equalization of Tactile KeyKey ClickClick onon TabletTablet SizedSized TouchscreensTouchscreens Until now, thethe best designs of piezoelectric actuatorsactuators and an imperceptible two-peak pulse were found to provide a strong bending eeffectffect within the required thickness (0.4 mm) and to eefficientlyfficiently maximize perceptual key click feedback on touchscreens.touchscreens. However, these findings findings are focused on increasing thethe strength strength of of local local tactile tactile feedback. feedback. Equalizing Equalizing force force feedback feedback on a touchscreen on a touchscreen is a diffi cultis a problemdifficult problem because, because, in general, in thegeneral, strength the strength of tactile of feedback tactile feedback is too strong is too at strong the bezel at the but bezel too weak but too at theweak center at the of thecenter touchscreen. of the touchscreen. To globally To equalize globally the equalize strength the of keystrength click feedbackof key click on touchscreens,feedback on wetouchscreens, propose a newwe propose algorithm a new that algorithm adaptively that changes adap thetively strength changes of tactilethe strength feedback, of tactile as summarized feedback, inas Figuresummarized6. The strategyin Figure is 6. to The reduce strategy strength is to onreduce or near strength the bezel on whileor near keeping the bezel the while maximum keeping piezo the bendingmaximum at piezo the weak bending area. at Asthe seen weak in area. Figure As6 ,seen tactile in feedbackFigure 6, (acceleration)tactile feedback on (acceleration) a touchscreen on is a proportionaltouchscreen tois theproportional input voltage to the driven input to thevoltage piezoelectric driven to actuator. the piezoelectric The linear relationshipactuator. The between linear accelerationrelationship andbetween input acceleration voltage can and be representedinput voltage as can a linear be represented regression as model a linear found regression by line model fitting (seefound the by left line of fitting Figure 6(see). From the left the of linear Figure regression 6). From model, the linear an inverse regression function model, is derivedan inverse to estimatefunction anis derived input voltage to estimate value: an input voltage value: G + 0.21 V = 𝐺 + 0.21 (7) 𝑉i = 1.17 (7) 1.17 where G is a desired acceleration value, and Vi denotes an input voltage value to piezoelectric actuators where G is a desired acceleration value, and Vi denotes an input voltage value to piezoelectric to be controlled. In this study, G was set to 1.5 G, which was above the detection threshold (1.17 G) actuators to be controlled. In this study, G was set to 1.5 G, which was above the detection threshold found by Kim [5]. The overall algorithm can be summarized as follows: (1.17 G) found by Kim [5]. The overall algorithm can be summarized as follows: 1. Get the position (x,y) of finger touch on the touchscreen from the digitizer of the tablet device. 1. Get the position (x,y) of finger touch on the touchscreen from the digitizer of the tablet device. 2. Compute the distance (vertical and horizontal) to the bezel. 2. Compute the distance (vertical and horizontal) to the bezel. 3. IfIf the the distance distance is is smaller smaller than than the the distance threshold δ, then estimate an input voltage from the equationequation above. above. 4. IfIf the the distance distance is is larger larger than than the distance threshold δ, then put a maximum input voltage. 5. SendSend the the amplified amplified input signal (a two-peak pulse) to piezoelectric actuators. 6. RepeatRepeat Steps Steps 1 1 through through 5. A pseudo code of the algorithm isis describeddescribed inin FigureFigure6 6..

Figure 6. A proposed algorithm to equalize tactile key c clicklick effects effects on the table touchscreen; a linear regression functionfunction that that has has the the best best fit fit to to the the data da pointsta points (embedded (embedded actuators) actuators) and and a pseudo a pseudo code code that implementsthat implements the key the clickkey click equalization equalization algorithm algorithm (right). (right). The implemented algorithm equalizing the strength of tactile feedback on the touchscreen was The implemented algorithm equalizing the strength of tactile feedback on the touchscreen was tested and compared with the measured force distributions generated from four actuators. Two tested and compared with the measured force distributions generated from four actuators. Two actuators were attached next to the home button, and the other two actuators were located at the middle of two vertical sides, respectively. The same accelerometer was used for measuring tactile Appl. Sci. 2019, 7, x 9 of 14 Appl. Sci. 2019, 9, 3080 9 of 14 actuators were attached next to the home button, and the other two actuators were located at the middle of two vertical sides, respectively. The same accelerometer was used for measuring tactile forceforce feedbackfeedback onon thethe touchscreen.touchscreen. AccelerationsAccelerations atat sixteensixteen pointspoints equallyequally spacedspaced onon thethe touchscreentouchscreen werewere measuredmeasured whenwhen allall ofof thethe piezoelectricpiezoelectric actuatorsactuators werewere bentbent atat thethe samesame time.time. Then, the measured valuesvalues werewere graphicallygraphically visualizedvisualized inin FigureFigure7 7.. InIn thethe plots,plots, thethe redderredder isis thethe strongerstronger tactiletactile feedback.feedback. ItIt waswas obviousobvious fromfrom thethe resultsresults thatthat thethe forceforce distributiondistribution waswas greatlygreatly improvedimproved afterafter applyingapplying ourour proposedproposed globalglobal equalizationequalization method.method.

(a)

(b)

FigureFigure 7.7. Accelerations distributed on a 10.1 inch commercial touchscreen (Samsung Galaxy Tab); non-uniformnon-uniform tactiletactile feedbackfeedback distributiondistribution ((aa)) andand thethe globallyglobally equalizedequalized distributiondistribution byby ourour proposedproposed algorithmalgorithm (b(b).). Note Note that that the the redder redder area area means means the strongerthe stronger tactile tactile feedback, feedback, and black and circlesblack circles are points are wherepoints accelerationswhere accelerations were measured. were measured.

4. Results

Three evaluation experiments were conducted to verify the piezoelectric actuators designed with optimum parameters (three layers and the layer ratio) and the haptic discrimination threshold and to measure typing performance with the tactile feedback system. A piezoelectric actuator manufactured

Appl. Sci. 2019, 7, x 10 of 14

4. Results Three evaluation experiments were conducted to verify the piezoelectric actuators designed Appl.with Sci. optimum2019, 9, 3080 parameters (three layers and the layer ratio) and the haptic discrimination threshold10 of 14 and to measure typing performance with the tactile feedback system. A piezoelectric actuator manufactured based on the final design was tested on a 10.1 inch touchscreen under a laboratory basedenvironment. on the finalThe strength design was of vi testedbro-tactile on a feedback 10.1 inch (acceleration touchscreen) underon a commercial a laboratory touchscreen environment. was Themeasured strength as ofincreasing vibro-tactile input feedback voltage (acceleration)to a piezoelectric on aactuator. commercial Accelerations touchscreen were was measured measured with as increasingtwo conditions, input free voltage stroke to aon piezoelectric a detached actuator.touchscreen Accelerations and after embedding were measured the actuator with two in conditions, a 10.1 inch freetablet. stroke For onthe a free detached stroke touchscreen condition, the and piezo after embeddingactuator was the glued actuator on a in separated a 10.1 inch touch tablet. glass For (see the freeFigure stroke 8), and condition, for the theembedding piezo actuator condition, was the glued actu onator a separated was attached touch to glassthe bezel (see space Figure between8), and forthe theLCD embedding panel and condition, the touchscreen the actuator of a tablet. was attached Then, an to theaccelerometer bezel space (KISTLER between the 8688A50) LCD panel was and placed the touchscreenon the touchscreen of a tablet. to measure Then, an accelerations accelerometer as (KISTLERthe input voltage 8688A50) (a wassquare placed wave on of the 500 touchscreen Hz) of the topiezo measure actuator accelerations was incrementally as the input altered. voltage The (ainput square signal wave to the of 500piezo Hz) was of controlled the piezo actuatorby a function was incrementallygenerator (Agilent altered. 33120). The Figure input signal8 shows to the the accele piezorations was controlled measured by with a function the two generator conditions (Agilent (before 33120).and after Figure embedding8 shows theactuators). accelerations The results measured showed with that, the two for conditionsthe embedding (before condition, and after the embedding strength actuators).of tactile feedback The results was reduced showed that,by 27.3% forthe due embedding to piezoelectric condition, actuators’ the strengthcompression of tactile by the feedback process wasof reassembly reduced by of 27.3% the tablet due todevice. piezoelectric actuators’ compression by the process of reassembly of the tablet device.

Figure 8. Acceleration values (left) measured on a 10.1 inch touchscreen with a three-layer piezoelectric actuatorFigure 8. as aAcceleration function of inputvalues voltage (left) (Vpp)measured for two on conditions, a 10.1 inch free stroketouchscreen on a detached with a touchscreenthree-layer (right)piezoelectric and after actuator embedding as a thefunction actuators of input in a tablet. voltage (Vpp) for two conditions, free stroke on a detached touchscreen (right) and after embedding the actuators in a tablet. The second experiment was conducted to verify whether the two-peak tactile pulse made users feel betterThe second about theexperiment touchscreen. was conduc For thisted experiment, to verify whether ten participants the two-peak (five tactile males pulse and five made females, users agefeel rangebetter 21–29,about the average touchscreen. age 25.5 For years this experiment old) who had, ten never participants experienced (five males haptic and interfaces five females, took partage range in the 21–29, experiment. average The age experiment25.5 years old) setup who was had the never same experienced as shown inhaptic Figure interfaces4, but the took value part ofin ttheof experiment. the test stimulus The experiment in Figure3 asetup was was set to the 3 ms,same which as shown was in much Figure smaller 4, but than the value the estimated of t of the discriminationtest stimulus in threshold Figure 3a(3.919 was set to0.393 3 ms,) to which ensure was imperceptibility. much smaller Thethan test the andestimated the reference discrimination stimuli ± werethreshold randomly (3.919 displayed ± 0.393 ) toto avoid ensure subjects’ imperceptibility. expectation The errors. test No and sound the feedbackreference wasstimuli provided were duringrandomly the displayed experiment. to avoid The participant’s subjects’ expectation task was errors. to indicate No sound which feedback one felt was better provided as a key during click. Eachthe experiment. participant wasThe testedparticipant’s ten times task for was both to stimuli, indicate resulting which inone a totalfelt ofbetter ten trialsas a perkey participant.click. Each Itparticipant took about was five tested minutes ten for times each for participant both stimuli, to complete resulting all in of a the total trials. of ten The trials result per was participant. that 81% ofIt participanttook about responsesfive minutes were for that each the participant two-peak pulseto comp waslete better, all of but the there trials. was The a result variation was from that person81% of toparticipant person due responses to different were sensitivities. that the two-peak pulse was better, but there was a variation from person to personThe last due experiment to different wassensitivities. conducted to evaluate typing performance (efficiency, speed, and accuracy)The last with experiment the developed was tactileconducted touchscreen to evaluate for a 10.1typing inch performance tablet device. (efficiency, For the experiment,speed, and aaccuracy) commercial with QWERTY the developed keyboard tactile was touchscreen implemented for on a the 10.1 Android inch tablet platform device. of aFor tablet the deviceexperiment, (uPad), a ascommercial shown in QWERTY Figure9. The keyboard developed was implemented system had a on latency the Android of 1.87 msplatform from sensingof a tablet touch device using (uPad), the Androidas shown touch in Figure event 9. listenerThe developed to tactile system feedback. had Aa latency two-peak of 1.87 tactile ms pulse from (sensingt = 3 ms) touch was using used forthe tactileAndroid key touch click feedback,event listener and theto tactile signal wasfeedback. amplified A two-peak to 65 Vpp tactile by the pulse amplifier (t = 3 (TI ms) DRV was 2667, used gain for 38.4 dB). The tactile feedback was synchronized with the key-down event provided from the Android platform. Well-known metrics, key strokes per character (KSPC), words per minute (WPM), and uncorrected error rate (UER) [1,17,22], were used for quantitatively measuring typing performance Appl. Sci. 2019, 7, x 11 of 14

tactile key click feedback, and the signal was amplified to 65 Vpp by the amplifier (TI DRV 2667, gain 38.4 dB). The tactile feedback was synchronized with the key-down event provided from the Android

Appl.platform. Sci. 2019 Well-known, 9, 3080 metrics, key strokes per character (KSPC), words per minute (WPM),11 ofand 14 uncorrected error rate (UER) [1,17,22], were used for quantitatively measuring typing performance across two conditions, visual (V) only and visual and haptic (VH), to see the effectiveness of tactile acrossfeedback two when conditions, addingvisual to an existing (V) only soft and keyboard. visual and The haptic two (VH), conditions to see were the e ffcounterbalancedectiveness of tactile and feedbackrandomized. when For adding the V tocondition, an existing the soft pressed keyboard. key was The highlighted two conditions in white, were while counterbalanced tactile feedback and randomized.was additionally For theprovided V condition, for the the VH pressed condition. key The was experiment highlighted consisted in white, of while ten blocks tactile feedbackper condition, was additionallyand each block provided had fifteen for the sentences VH condition. that were The experimentrandomly selected consisted from of ten a phrase blocks dataset per condition, created and by each[23]. blockEach sentence had fifteen was sentences displayed that in werered at randomly the top of selected the touchscreen, from a phrase as seen dataset in Figure created 9, until by [ 23the]. Eachuser finished sentence transcribing was displayed the indisplayed red at the sentence. top of the The touchscreen, white text box as seenshowed in Figure what the9, until user thetyped user in finishedreal time. transcribing Ten participants the displayed (six males sentence. and four The females, white text average box showed 24.2 years what theold) user took typed part inin realthe time.experiment. Ten participants All participants (six males were and skilled four females,typists an averaged had at 24.2 least years two old) years took of part experiences in the experiment. using the Allsoft participants keyboard of were a 10.1 skilled inch tablet typists device. and had Before at least starting two years the experiment, of experiences a training using thesession soft keyboard(about 5– of10 amin) 10.1 was inch provided tablet device. to participants Before starting to get the familiar experiment, with the a trainingtactile soft session keyboard. (about Each 5–10 participant min) was providedwas then toinstructed participants to transcribe to get familiar displayed with sentences the tactile as soft quickly keyboard. and accurately Each participant as possible. was thenThe instructedparticipants to wore transcribe earplugs displayed to block sentences noise during as quickly the experiment. and accurately It took as about possible. fifty Theminutes participants for each woreparticipant earplugs to complete to block noise the experiment. during the experiment. It took about fifty minutes for each participant to complete the experiment.

Figure 9. Soft keyboard design and a user interface implemented on the Android platform with a Figure 9. Soft keyboard design and a user interface implemented on the Android platform with a 10.1 10.1 inch commercial touchscreen (uPad) for a user study. The white text field was an area for a user inch commercial touchscreen (uPad) for a user study. The white text field was an area for a user to to transcribe a red sentence randomly selected from a test set of sentences created by [23]. For the transcribe a red sentence randomly selected from a test set of sentences created by [23]. For the visual visual (V) condition, the pressed key was highlighted in white, while tactile key click feedback was (V) condition, the pressed key was highlighted in white, while tactile key click feedback was additionally provided for the visual and haptic (VH) condition. additionally provided for the visual and haptic (VH) condition. Figure 10 shows the results of typing performance (KSPC, WPM, UER) tested with the haptic soft keyboard.Figure 10 shows The bars the represent results of the typing average performa valuesnce of each(KSPC, metric WPM, obtained UER) tested from thewith experiment, the haptic andsoft thekeyboard. error bars The show bars therepresent standard the deviations. average values Note of that each a lower metric value obtained meant from better the performance, experiment, exceptand the with error WPM, bars whichshow the was standard interpreted deviations. as typing Note speed. that As a lower an effi valueciency meant index, better KSPC performance, ranged from 1.17except to 1.38,with withWPM, VH which appearing was interpreted to be more as e ffitypingcient, speed. which As showed an efficiency better productivityindex, KSPC whenranged tactile from feedback1.17 to 1.38, was with added. VH Theappearing average to UERs,be more typing efficient, errors, which were showed 0.54 and better 0.8 for productivity VH and V, respectively,when tactile whichfeedback demonstrated was added. improved The average accuracy UERs, withtyping tactile errors, feedback. were 0.54 In contrast,and 0.8 for there VH wasand noV, direspectively,fference in typingwhich demonstrated speed observed improved from the accuracy average WPMwith tactile values feedback. from 28.4 In and contrast, 28.9 for there V and was VH, no difference respectively. in Atyping one-way speed ANOVA observed showed from thatthe average there was WPM no significantvalues from di ff28.4erence and (28.9p >> for0.05) V and across VH, conditions respectively. for theA one-way three metrics. ANOVA showed that there was no significant difference (p >> 0.05) across conditions for the three metrics.

Appl. Sci. 2019, 9, 3080 12 of 14 Appl. Sci. 2019, 7, x 12 of 14

Figure 10. Typing performance results measured by a user study across two conditions V and VH. Figure 10. Typing performance results measured by a user study across two conditions V and VH. Key strokes per character (KSPC), words per minute (WPM), and uncorrected error rate (UER) were Key strokes per character (KSPC), words per minute (WPM), and uncorrected error rate (UER) were interpreted as typing efficiency, speed, and accuracy on the 10.1 inch touchscreen. The bars represent interpreted as typing efficiency, speed, and accuracy on the 10.1 inch touchscreen. The bars represent means and standard deviations of each metric for the two conditions. A lower value meant better means and standard deviations of each metric for the two conditions. A lower value meant better performance except for WPM. performance except for WPM. 5. Discussion 5. Discussion In the present study, we investigated an effective way of providing a stronger tactile feeling of key clickIn the on present tabletsized study, touchscreens. we investigated It is imperativean effective becauseway of providing deformation a stronger of the larger tactile touchscreen feeling of keycan beclick achieved on tablet by sized the strongertouchscreens. piezoelectric It is impera bender.tive because It is known deformation by Equation of the (4) larger that the touchscreen thickness canof a be ceramic achieved layer by is the inversely stronger proportional piezoelectric to bender mechanical. It is displacement,known by Equation which (4) justifies that the the thickness need for ofmulti-layer a ceramic piezoelectric layer is inversely actuators proportional that require to more mechanical expensive displacement, manufacturing which cost justifies and current the need because for multi-layerthe capacitance piezoelectric increases withactuators the number that require of layers. more For thisexpensive reason, findingmanufacturing optimized cost parameters and current (the becausenumber ofthe layers capacitance and the increases layer ratio) with is crucial, the number especially of layers. for mobile For devicesthis reason, that havefinding limited optimized power parametersand space to (the add number actuators. of layers The best and parameters the layer accordingratio) is crucial, to the especially simulations for were mobile three devices layers that and 1.4have as limited the ratio power of the and internal space ceramic to add layer actuators. to the externalThe best ceramic parameters layer. according It was also to foundthe simulations from pilot wereexperiments three layers that theand bending 1.4 as the performance ratio of the was internal better ceramic when the layer poles to ofthe the external three layers ceramic were layer. connected It was alsoin a series,found asfrom seen pilot in Figure experiments2. The bending that the performancebending performance of the manufactured was better actuatorwhen the was poles tested of the by threemeasuring layers accelerations. were connected Theresults in a series, from Figure as seen8 demonstrate in Figure the2. The su ffi bendingcient strength performance to simulate of keythe manufacturedclicks on a 10.1 actuator inch touchscreen was tested with by measuring low voltage accelerations. by referring toThe the results detection from threshold Figure 8 demonstrate (1.14 G) on a thecommercial sufficient touchscreen strength to simulate [1]. However, key clicks it was on also a 10.1 found inch thattouchscreen there was with a significant low voltage drop by referring (average to27.3%) the detection of tactile forcethreshold display (1.14 on G) the on touchscreen a commercial due touchscreen to the compression [1]. However, force caused it wasby also reassembling found that therethe tablet was devicea significant after bonding drop (average the actuators. 27.3%) This of ta issuectile couldforce bedisplay resolved on bythe designing touchscreen a new due internal to the compressionstructure that force preserves caused the by mechanical reassembling bending the whenevertablet device voltage after is bonding applied tothe piezoelectric actuators. This actuators. issue couldIn be addition, resolved the by discriminationdesigning a new thresholds internal (JNDs)structure measured that preserves in this studythe mechanical to understand bending the wheneversensitivity voltage to time is di appliedfferences to of piezoelectric a tactile pulse actuators. or a tactile pulse combined with sound were the first resultsIn everaddition, reported the discrimination for 10.1 inch commercial thresholds touchscreens.(JNDs) measured The discriminationin this study to threshold understand for the sensitivityhaptics only to condition time differences was used of as a antactile upper pulse bound or a perception tactile pulse value combined to increase with the sound perceptual were strengththe first resultsof a tactile ever key reported click eff forect. 10.1 The inch HA thresholdcommercial can touchscreens. be a guideline The for designingdiscrimination a multimodal threshold feedback for the hapticssignal commonly only condition used inwas mobile used devices.as an upper Through boun thisd perception approach, usersvalue canto increase surely feel the the perceptual stronger strengthtactile feedback of a tactile on 10.1 key inchclick touchscreens, effect. The HA and threshold the use of can mobile be a guideline battery driving for designing multiple a actuators multimodal can feedbackalso be reduced, signal commonly since an imperceptible used in mobile two-peak devices. pulse Through consumes this approach, less energy users than can a one-long-peaksurely feel the pulse.stronger The tactile desired feedback acceleration on 10.1 value inch on touchscreen the touchscreens, and wasthe use also of determined mobile battery by taking driving into multiple account actuatorsthe perception can also data be (detection reduced, thresholdsince an imperceptible 1.17 G) estimated two-peak from pulse [2] for consumes an algorithm less equalizingenergy than the a one-long-peakdistribution of tactilepulse. forces,The desired as shown acceleration in Figure value6. This on study the touchscreen shows a practical was also example determined of how toby takingemploy into human account perception the perception data, but furtherdata (detection studies are threshold needed to 1.17 apply G) various estimated form from factors [2] in for terms an algorithmof the size andequalizing the material the distribution of touchscreens. of tactile Another forces, contribution as shown of in this Figure study is6. theThis proposal study shows of a new a practicalalgorithm example that can of equalize how to theemploy distribution human percepti of forceon feedback data, but on further a 10.1 inch studies touchscreen. are needed Due to apply to the various form factors in terms of the size and the material of touchscreens. Another contribution of this study is the proposal of a new algorithm that can equalize the distribution of force feedback on

Appl. Sci. 2019, 9, 3080 13 of 14 material characteristics of 10.1 inch commercial touchscreens, vibro-tactile signals are not transparently delivered to the center from the bezel, where multiple piezoelectric actuators are bonded. People feel uncomfortable near the bezel if the tactile force is not uniformly distributed. Our proposed algorithm improves the force distribution by adaptively reducing the force values to the distance of the bezel, but further study is needed to improve the upper area that still appeared relatively blue, as visualized Figure7. Finally, the developed tactile feedback system was verified by measuring typing performance using well-known metrics, KSPC, WPM, and UER, which represented typing efficiency (key strokes per character), typing speed, and accuracy (uncorrected errors), respectively. The results demonstrated that typing efficiency and accuracy were improved with tactile feedback, whereas there was no difference in typing speed; however, there were no statistically significant differences between the two conditions due to different typing skills from naïve to skilled typists. Note that our previous study showed that tactile feedback was more effective with skilled typists []. Several factors may have contributed to this result. First, typing confirmation with tactile feedback played an important role in reducing typing errors or correcting errors, because skilled typists tended to press keys without looking at each key character, which resulted in higher KSPCs and UERs with the V condition. Second, a time delay in the tactile feedback for confirming key clicks caused lower typing speed with the VH condition. The time delay often resulted in unsynchronized tactile feedback, especially with fast typists. Third, typing speed may have been affected by learning during the experiment. It is possible that participants did not show good performance in typing speed since they were still getting used to the tactile feedback condition by the end of the experiment. This could be proved by a further study investigating the effect of tactile feedback with a longer time of training.

Author Contributions: The first author (S.Y.) and the corresponding author (K.K.) developed algorithms and the piezo driving circuit system, and the second author (S.H.J.) verified and analyzed the research results. The corresponding author (K.K.) guided the research direction. All authors made substantial contributions in the writing and revision of the paper. Funding: This work was supported by Incheon National University Research Grant in 2016 (Grant No. 2016). Conflicts of Interest: The authors declare no conflict of interest.

References

1. Han, B.; Kim, K. Typing performance evaluation with multimodal soft keyboard completely integrated in commercial mobile devices. J. Multimodal User Interfaces 2015, 9, 173–181. [CrossRef] 2. Kim, K. Perception-based tactile soft keyboard for the touchscreen of tablets. Mob. Inf. Syst. 2018, 2018, 9. [CrossRef] 3. Findlater, L.; Wobbrock, J.O.; Wigdor, D. Typing on flat glass: Examining ten-finger expert typing patterns on touch surfaces. In Proceedings of the SIGCHI Conference on HUMAN Factors in Computing Systems, Vancouver, BC, Canada, 7–12 May 2011; pp. 2453–2462. 4. Findlater, L.; Wobbrock, J. Personalized input: Improving ten-finger touchscreen typing through automatic adaptation. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Austin, TX, USA, 5–10 May 2012; pp. 815–824. 5. Li, F.C.Y.; Guy, R.T.; Yatani, K.; Truong, K.N. The 1line keyboard: A QWERTY layout in a single line. In Proceedings of the 24th annual ACM symposium on User interface software and technology, Santa Barbara, CA, USA, 16–19 October 2011; pp. 461–470. 6. Goel, M.; Jansen, A.; Mandel, T.; Patel, S.N.; Wobbrock, J.O. ContextType: Using hand posture information to improve mobile touch screen text entry. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Paris, France, 27 April–2 May 2013; pp. 2795–2798. 7. Jo, S.M.; Cho, S.B. A Personalized Context-aware Soft Keyboard adapted by Random Forest trained with Additional Data of Same Cluster. Neurocomputing 2019, 353, 17–27. [CrossRef] Appl. Sci. 2019, 9, 3080 14 of 14

8. Poupyrev, I.; Mauyama, S. Tactile interfaces for small touch screens. In Proceedings of the 16th Annual ACM Symposium on User Interface Software and Technology, Vancouver, BC, Canada, 2–5 November 2003; pp. 217–220. 9. Rabin, E.; Gordon, A.M. Tactile feedback contributes to consistency of finger movements during typing. Exp. Brain Res. 2004, 155, 362–369. [CrossRef][PubMed] 10. Brewster, S.; Brewster, S.; Chohan, F.; Brown, L. Tactile feedback for mobile interactions. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, San Jose, CA, USA, 28 April–3 May 2007; pp. 159–162. 11. Hoggan, E.; Brewster, S.A.; Johnston, J. Investigating the effectiveness of tactile feedback for mobile touchscreens. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Florence, Italy, 5–10 April 2008; pp. 1573–1582. 12. Koskinen, E.; Kaaresoja, T.; Laitinen, P. Feel-Good Touch: Finding the Most Pleasant Tactile Feedback for a Mobile Touch Screen Button; ACM: New York, NY, USA, 2008. 13. Jansen, Y.; Karrer, T.; Borchers, J. MudPad: Localized Tactile Feedback on Touch Surfaces. In Proceedings of the 23nd annual ACM symposium on User interface software and technology, New York, NY, USA, 3–6 October 2010; ACM: New York, NY, USA, 2010; pp. 385–386. 14. Hoffmann, A.; Spelmezan, D.; Borchers, J. TypeRight: a keyboard with tactile error prevention. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Boston, MA, USA, 4–9 April 2009; pp. 2265–2268. 15. McAdam, C.; Brewster, S. Distal tactile feedback for text entry on tabletop computers. In Proceedings of the 23rd British HCI Group Annual Conference on People and Computers: Celebrating People and Technology, Cambridge, UK, 1–5 September 2009; pp. 504–511. 16. Kaaresoja, T.; Anttila, E.; Hoggan, E. The effect of tactile feedback latency in touchscreen interaction. In Proceedings of the 2011 IEEE World Haptics Conference, Istanbul, Turkey, 21–24 June 2011; pp. 65–70. 17. Han, B.K.; Kim, K.; Yatani, K.; Tan, H.Z. Text entry performance evaluation of haptic soft QWERTY keyboard on a tablet device. In Proceedings of the International Conference on Human Haptic Sensing and Touch Enabled Computer Applications, Versailles, France, 24–26 June 2014; Springer: Berlin/Heidelberg, Germany, 2014; pp. 325–332. 18. Jeon, S.; Lee, H.; Jung, J.; Kim, J.R. User-Adaptive Key Click Vibration on Virtual Keyboard. Mob. Inf. Syst. 2018, 2018, 1–9. [CrossRef] 19. Wang, Q.M.; Cross, L.E. Constitutive equations of symmetrical triple layer piezoelectric benders. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 1999, 46, 1343–1351. [CrossRef][PubMed] 20. Levitt, H. Transformed up-down methods in psychoacoustics. J. Acoust. Soc. Am. 1971, 49, 467–477. [CrossRef] 21. Goldstein, E.B. Sensation and Perception, 9th ed.; Cengage Learning: Boston, MA, USA, 2014. 22. Soukoreff, R.W.; Mackenzie, I.S. Metrics for text entry research: An evaluation of MSD and KSPC, and a new unified error metric. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Lauderdale, FL, USA, 5–10 April 2003; pp. 113–120. 23. Mackenzie, I.S.; Soukoreff, R.W. Phrase sets for evaluating text entry techniques. In Proceedings of the CHI’03 Extended Abstracts on Human Factors in Computing Systems, Ft. Lauderdale, FL, USA, 5–10 April 2003; ACM: New York, NY, USA, 2013; pp. 754–755.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).