Colorful Electrotactile Feedback on the Wrist

Providing rich feedback on small devices, like smartwatches, can be difficult. We propose colorful electrotactile feedback on the back of a smartwatch. Colorful electrotactile feedback provides private notifications, is energy efficient, and can express various sensations in different qualities. In a first study, 13 participants explored 49 different combinations of frequency and pulse width regarding the perceived “colorfulness” of electrotactile feedback. We investigated what sensations can be expressed with electrotactile feedback and which qualities of these sensations are conveyed. To describe the sensations, participants chose the best fitting terms from a list of 21 terms. The three most frequently selected terms were prickling (177), vibrating (163), and irritating (112). The three least frequently selected ones were twitching (31), tickling (29), and itching (28). In a second study with 17 participants we evaluated a reduced set of 9 sensations that we selected and refined based on the results of study 1. We evaluated these sensations regarding recognition rates and achieved recognition rates of up to 84% without prior learning. Furthermore, we investigated the acceptance of colorful electrotactile feedback and present a method for an easier and faster calibration of electrotactile feedback.


INTRODUCTION
The output space of smartwatches is strongly limited regarding the energy requirements of the feedback technology and the needed space.Common output technologies are visual displays and vibration motors.The display consumes significant amounts of energy but delivers dense information in a small space, while vibration feedback often just serves as a binary information channel, e.g. to notify the user.We suggest that establishing a feedback technology in between these extremes could enhance our interaction and experience with small wearables.Physical actuators are often only capable of creating a single tactile sensation.For example, a vibration motor produces the sensation of a vibration.It is not able to evoke a touch sensation or a poking sensation.This would require additional hardware.The use of more and more hardware to enrich the output space contradicts the trend of miniaturization, energy optimization, and cost reduction.To overcome this problem we propose the use of electrotactile feedback over the use of physical tactile feedback.Instead of stimulating with a single frequency or pulse width, we propose colorful electrotactile feedback, which is inspired by Kajimoto et al. [11].Electrotactile feedback is energy efficient and offers a rich design space of sensations.

Concept of Colorful Electrotactile Notifications
The qualification "colorful" means that the electrotactile feedback does not only provide a single sensation like vibration -like black or white -but a greater range of sensations.The term "colorful electrotactile feedback" is meant metaphorically.Electrotactile feedback can evoke a whole spectrum of perceptions, analogous to a spectrum of light colors.It is not about showing colors in a tactile way as investigated by Shin et al. [25].Our skin is the protective layer of our body, it lets us feel pain, heat and cold, or pleasant sensations like tickling or prodding, and uncomfortable sensations like pinching or itching.And there are subtle qualities of these sensations.They may be forceful or gentle, strong or faint, calming or energizing, diffuse or localized.We want to enrich our interaction with wearables by such sensations and qualities.We show that it is possible to create such sensations with a specific quality via electrotactile feedback.We call this enrichment of the tactile channel via electrical stimulation colorful electrotactile feedback as it brings the tactile channel to a higher level, with more opportunities, nuances, and potential emotions.
Our goal is to enable a more natural and expressive interaction by replacing the vibrotactile feedback of on-skin wearables with colorful electrotactile feedback.As we show, electrotactile feedback is also able to present the sensation of vibration, with the additional option to freely and independently adjust the parameters frequency and amplitude to create a smoother vibration or rougher, pulsating feedback.Regarding the physical integration, we focus on enhancing the output space without changing the design of a smartwatch.This can be achieved by simply placing the electrodes for electrotactile feedback on the back of a smartwatch, as the back provides sufficient contact area for the electrodes.This placement protects the electrodes from being touched inadvertently.Moreover, unlike the wristband the back of the smartwatch is not mechanically strained and rarely directly touched when the smartwatch is put on or off.Other on-skin sensors, like pulse oximeters, are also placed at this area.

Contribution
In this work we contribute the concept of colorful electrotactile feedback and application scenarios that benefit from colorful electrotactile feedback.In a study we explored the sensations that can be evoked by different combinations of frequency and pulse width.The results can be used to further investigate particular sensations, e.g. to find the optimal parameter set for the prickling sensation or to build up applications that use the explored sensations.In a second study we reduced the set of sensations to an easily distinguishable set of 9 sensations.Without knowing the mapping between a stimulation and descriptive term for a sensation, the participants were able to recognize 5 of the 9 sensations with an accuracy of 40% or more.This shows that a natural mapping is possible.We believe that further investigations of optimal parameters for specific sensations will improve the accuracy.Also we believe that a learned mapping will perform even better.Lastly we used the acquired data set to enhance the calibration speed via regression.

RELATED WORK
This section covers related work on electrotactile feedback as well as on work that tries to enhance the output space of smartwatches.We will also highlight wristbands and other approaches on the forearm that try to enhance the output space of smartwatches.

Electrotactile Feedback
Electrotactile stimulation is a valuable technique in HCI, especially as a feedback technology for virtual reality experiences to bring physical properties to virtual objects [12,30].In earlier research electrotactile feedback is associated with electrotactile displays, which consist of 2D electrode arrays that can display fine-granular electrotactile feedback on the user's skin [29].Strong and Troxel show that an electrotactile display can produce a texture effect that is similar to moving the finger across a textured surface.Moreover, electrotactile displays can be used for sensory substitution [8] to present e.g.audio signals to the skin.Kajimoto et al. presented an approach to control the strength of the electrotactile stimulation by measuring the touch force to create a more realistic sensation [10].SmartTouch [9] solves the issue of hindering the perception of real objects when wearing an electrotactile displays by a 3-layer design.Another tackled problem by Kajimoto et al. is that the quality of the created sensations of electrotactile feedback does not feel natural [11].To enhance the quality they investigated the individual activation of three of the four mechanoreceptors -namely Merkel cells, Meissner corpuscles, Ruffini endings, and Pacinian corpuscles -to create primary tactile colors.Other tactile sensations could be then created by mixing the primary tactile colors in analogy to the RGB approach for presenting various colors to the visual system.The creation of different tactile sensations via electrical stimulation is a topic of interest in HCI.Pohl and Hornbaek investigated the "itch" sensation at the wristband of a smartwatch for notification purposes [21].They found out that a frequency of 65 Hz and a pulse width of 3800 ms is sufficient to evoke itch.Itch is produced by nociceptors and therefore not covered by the tactile primary color approach [11].Alotaibi et al. [1] try to create Electrotactons similiar to vibrotactile Tactons.They argue that there is not enough knowledge about what stimulation parameters should be used for electrotactile feedback.Therefore they first analyzed the effect of the transferred charge (pulse width × amplitude) on the sensations arousal, valence, urgency, and annoyance.Increasing intensity caused significantly higher ratings of urgency (similar to [4]), annoyance, and arousal, but reduced valence.They found that at the palm in the investigated interval of 10 Hz -110 Hz (steps of 20 Hz) there exist only two distinguishable levels of frequency at 10 Hz and 110 Hz.Four different intensities in terms of transferred charge could be distinguished.Stanke et al. directly compared electrotactile feedback with vibrotactile feedback [27] around the wrist and the finger, respectively.They found that electrotactile feedback could be a viable alternative to vibrotactile feedback, as it exhibits slightly better pattern recognition and is judged as more comfortable.Dependent on the small form factor of electrotactile feedback it is also an alternative for earclips to notify users [26].It was one of the fastest notification channels and favored by the study participants because of its soundless nature, above vibration and all other evaluated notification channels.Also in the area of electrical muscle stimulation (EMS) [5], electrotactile stimulation is a known side effect, because of the lower activation threshold of the sensory system compared to the threshold to activate the motor neurons to evoke muscle contraction.Knibbe et al. investigated the experience of EMS by conducting explicitation interviews with participants regarding 48 combinations of stimulation parameters and locations for EMS [13].The combinations were applied at 3 different locations on the forearm, 4 different frequencies (20 Hz, 55 Hz, 90 Hz, 120 Hz) and 4 different pulse widths (100 µs, 150 µs, 200 µs, 250 µs).
Our approach of colorful electrotactile feedback is motivated by the discussed works, as we see, similar to [1], the need to more investigate the evoked effects of electrotactile stimulation.Therefore, we analyzed a much broader range of parameters of frequencies (2 Hz -1 kHz) and pulse widths (10 -1000 µs) than the previous works to identify evoked sensations.Instead of trying to create these sensations by overlaying primary electrotactile colors, we investigate the frequency and pulse width range of a single electrode pair to evoke different sensations.We also do not focus on the localization of the feedback, as [27], or to fine granular feedback to the fingertips [10], as the tactile resolution on the wrist is lower than on the fingertips [15].We believe that medium grained electrotactile feedback could be an excellent feedback technology on smartwatches.It is capable of evoking several distinct sensations that can be created by the combination of different frequencies and pulse widths.

Feedback on the Back of Smartwatches
There is a large body of related work regarding the output space of smartwatches including traditional feedback, like vibration, as well as new approaches to enhance the output capabilities of smartwatches.Panëels et al. and Matscheko et al. investigated the use of multiple vibration motors [16,17] on the dorsal side of the wrist and found out that placing vibration motors around the wrist increases the recognition rates of vibration patterns with multiple actuators.Besides increasing the output channel by placing multiple vibration motors also other tactors are used, e.g.Shim et al. propose the use of air flow combined with vibrotactile feedback to present multimodal feedback on the back of a smartwatch [24].They also mention the term "colored" tactile feedback.Pasquero et al. used a haptic piezoelectric mechanism in their prototype to present natural haptic feedback [18].A tactile drawing on the skin under a watch was investigated by Ion et al. [7].Their SkinDrag prototype is able to draw tactile pictures to the skin.Different from actuators that cause a tactile sensation, Pohl et al. proposes Scatter-Watch [22], a prototype that emits indirect light via the back of the smartwatch through the skin.

Feedback from Wristbands and Forearm Wearables
Further opportunities for the integration of additional feedback on the lower arm is the integration of actuators in armbands, e.g. in the wristband of a smartwatch.The integration of up to 8 vibrotactile motors around the wrist was evaluated by Matscheko et al. [16] and Hong et al. [6], showing that presenting vibrotactile feedback around the wrist is more accurate than presenting vibrotactile feedback on the back of a smartwatch and that a setup of 4 vibration motors provides better results than a setup of 8 motors.Vibro-band [23] evaluates a setup up of 8 vibration motors worn in a wristband around the dominant arm to support needlebased interventions.The results show that the vibro-band performs equivalent compared to visual guidance in terms of completion time, accuracy, task load and usability.Yet not only vibration feedback is evaluated for around-the-wrist output.Besides vibrotactile feedback thermal feedback integrated in a smartwatch wristband in ThermalBracelet [19], brushing was evaluated in BrushTouch [28] and compression feedback for notifications on the wrist in Squeezeback [20] [2].Based on our goal to extend the output capability of smartwatches, all presented wristband approaches are viable alternatives.But the integration of electronics and actuators in the watch strap poses several technical and design challenges.First, the function of wrist straps is to fix the watch to the wrist.Different wrist sizes require different wrist strap sizes to achieve suitable layouts of the actuators.Second, the watch straps need an electrical connection in addition to the physical connection to the smartwatch to be powered and controlled.This requires paying attention to the fact that wrist straps are also flexible and experience forces when fastening, which could reduce the durability of electrical connections in the straps.Third, watch straps are often a customizable design feature of a watch that could be realized with a wide spectrum of materials like leather, silicone, different metals, etc.Each material presents different challenges in terms of integrating the components.Based on these considerations we favor the back of the smartwatch for the integration of technical elements rather than the wrist straps.

PROTOTYPE
Figure 1 shows our prototype that consists of four components: a smartwatch stub, a RehaMove 3 device, an Adafruit Feather, and electrodes on the underside of the smartwatch case (carbon rubber and gold plated electrodes).The RehaMove 3 is a powerful stimulator for functional electrical stimulation (FES) or EMS that is intended for research and development.It is mobile and can be controlled via USB by the open ScienceMode1 protocol and offers four separately controllable channels.The frequency can be adjusted in the range of 1 -1000 Hz in steps of 1 Hz.The pulse width can be adjusted in the range of 10 -4095 µs in steps of 1 µs.The current can be adjusted in the range of 0 -130 mA in steps of 0.5 mA.The device also offers the opportunity to create different pulse shapes, but we only used standard biphasic rectangular pulses.For mobile operation of the system we turned an Adafruit Feather M0 Bluefruit LE2 into a USB host controller to communicate with the RehaMove 3. As there was no library of the ScienceMode protocol for microcontrollers, we implemented the communication protocol on our own.Through this protocol it is possible to control the stimulator via Bluetooth LE from a smartphone for mobile use cases.The smartwatch prototype consists of a 3D-printed casing with a diameter of 46 mm.The smartwatch casing mainly serves as a stub for connecting different electrodes that can be exchanged with a screw thread for a fast evaluation of different electrode designs.Initially, we evaluated two different layouts for the electrodes, which we cut of carbon rubber electrodes as presented in [3].The concentric layout failed, because the conductivity in the carbon rubber material was too inhomogeneous, so that the stimulation was not felt under the whole prototype, but only at a single point.For that reason we continued with the half-circle layout shown in Figure 1.Further investigations in producing custom-shaped electrodes that have a low resistance and are not corrosive led us to use gold-plated electrodes in study 2. The gold plating was applied to the 3D printed casing in our workshop.Another advantage of the gold electrodes is that they are extremely thin (< 0.1 mm) and therefore are easy to integrate.

STUDY 1: EXPLORATION OF ELECTROTACTILE FEEDBACK STIMULATION PARAMETERS
To explore the set of possible sensations that could be produced with electrical on-skin stimulation we conducted this study with a within-subject design.The study complied with our ethical review process.The independent variables are the frequency (IV1) and pulse width (IV2).The dependent variable was the set of terms that each participant selected for a given combination of frequency and pulse width at the amplitude that was calibrated for that participant.As the possible parameter space given by the stimulation device is huge -with about 4 million possible different combinationswe had to reduce the set of frequencies and pulse widths.For IV1 we chose seven different levels: 2 Hz, 10 Hz, 50 Hz, 100 Hz, 250 Hz, 500 Hz, and 1000 Hz.For IV2 we chose six different levels: 10 µs, 50 µs, 100 µs, 250 µs, 500 µs and 1000 µs.This results in 39 combinations, because three combinations are not possible (1000 Hz and 1000 µs, 1000 Hz and 500 µs, and 500 Hz and 1000 µs).The parameter set was chosen by the following reasoning.We chose the frequencies out of different ranges.The first range is from 1 to 20 -30 Hz, in which every pulse is felt separately.The next range from 20 -30 Hz upwards to 200 Hz is typically used by EMS/FES and electrotactile applications [1,13,14].The range above 200 Hz and upwards is not covered by related work.We tried to cover at least 2 frequencies from each range.Similarly, we tried to cover different pulse widths: Pulse widths from 50 to 300 µs are used for EMS/FES applications [13] or 50 to 200 µs in [1], [21] evaluated a pulse width of 3800 µs.Moreover, we tried to cover the range of pulse widths below 50 µs and the range above 300 µs.
The playback time of a stimulus with a particular stimulation parameter combination was set to 500 ms.The participants had to choose zero or more terms from a list of 21 terms [21] that they found appropriate to describe the experienced sensation.The terms were presented to the participants in their native language.To overcome order effects we shuffled for each participant the order of the combinations of stimulation parameters as well as the order of the describing terms that were presented to the participants in the application (cf. Figure 2 right).

Participants
We recruited 13 participants (1 female & 12 male) with a mean age of 24.2 years ( = 2 years).Five participants were left-handed, eight were right-handed.From eight participants that regularly wear a watch, three wear a smartwatch, a fitness tracker, or a similar device.Three participants had prior experience with electrotactile feedback or EMS from previous studies.Before the experiment we asked the participant about their expectations how electrotactile feedback could feel.The participants answered free text and stated their own terms (multiple mentions possible).Eight participants answered that it would be prickling, four mentioned hurting and three itching.The following terms were mentioned once: vibrating, tingling, pinching, and twitching.

Procedure
First the participants filled out a consent form and a demographic questionnaire.The consent form informed the participants about the study and especially about their right to end the study at any time without giving any reason.Also the form presented a list of health-related points to them under which they were not allowed to participate.The participants could indicate whether one of the points applied to them without explicitly naming it.If that occurred the participant was excluded from the experiment and received a bar of chocolate.Then the participants put on the smartwatch prototype on the arm, on which they would wear a wristwatch (cf. Figure 2 left).They could also control how tight the device was fastened to their wrist.The smartphone application (cf. Figure 2 middle) allowed the participants to calibrate the amplitude for each combination of frequency and pulse width on their own.We told the participants to increase the amplitude to that point that they clearly felt the stimulation, the feeling is not hurting and no muscle contracts.The evaluation started, after the last combination had been calibrated.The participants could play back the electrotactile signal as often as they liked and choose terms that matched to their sensation (multiple choices possible) (cf. Figure 2 right).This procedure was repeated for all 39 combinations.After finishing the evaluation, the devices were turned off and the participants took off the smartwatch prototype.A closing questionnaire followed and at last each participant received each a bar of chocolate for participation.The duration of the study took about 1 hour per participant.

Results
One male participant was excluded from the further evaluation, because he mentioned that he felt the current even at 0.0 mA.This might have occurred since the stimulation device always sends a probe pulse for safety reasons to ensure that the electrodes are connected.But this pulse is so weak that none of the other 12 participants reported about that.Further the excluded participant selected the sensation "hurting" for 0.0 mA for a low frequency of 10 Hz, which suggests that the participant has a very sensitive skin.The calibrated amplitude for the other 12 participants ranged from 0.5 -150 mA.So the parameter range of the stimulation device was fully used.Some participants also reported that they would have increased the amplitude for some stimulations above the maximum of 150 mA if that had been possible.
The participants selected 1594 terms in total.On average 3.14 terms were selected per stimulation.The term prickling was selected most often with 177 selections, followed by vibration with 163 selections.The least frequently selected term was itching with only 28 selections.
For the analysis, we split the list of terms in two groups.The terms in the first group describes a felt sensation, e.g.vibrating and the terms in the second group describe the felt sensation, e.g.forceful.The results of the sensation selection are shown in Figure 3 and Figure 4.The ratio in each square is calculated as the count of mentions of the term for the pair of frequency and pulse width divided by the maximum count of that term over all pairs.For example, 11 participants found the term vibrating matching for the pair (50 Hz, 50 µs).As 11 is the maximum count for the sensation of vibration over all pairs, the resulting value for the square is 1.The heatmaps show a very strong diffusion of the felt sensations, but also highlight some tendencies.For example, the term pulsating (Figure 3), depends mainly on the frequency of 10 Hz and not on the pulse width, whereas the participants felt it mostly clearly at 50 µs.Other sensations like vibrating and prickling occur mostly only on frequencies higher than 10 Hz, independently of the pulse width.The maximum for vibration is found at (50 Hz, 50 µs), but prickling has three maxima: (250 Hz, 500 µs), (50 Hz, 250 µs) and (50 Hz, 1000 µs).The sensation jabbing is selected by frequencies below 50 Hz and most clearly at (2 Hz, 100 µs), even greater pulse widths than 100 µs are also described as jabbing.The sensations itching and stinging are selected more often at high pulse widths.They were mentioned most frequently at a pulse width of 1000 µs.twitching has its maximum number of mentions at (10 Hz, 100 µs).squeezing was felt the most at high frequencies of 500 Hz and 1000 Hz with pulse widths of 50 µs and 100 µs, respectively.tickling was mentioned most frequently at frequencies between 50 Hz and 250 Hz: (50 Hz, 250 µs), (100 Hz, 500 µs) and (250 Hz, 10 µs).Pulling was felt mostly by the participants at higher frequencies of 100 Hz and more and pulse width above 50 µs.It was mentioned most often at (1000 Hz, 50 µs) and (250 Hz, 100 µs).Irritating is mostly felt with medium pulse width and frequencies and has its maxima at (500 Hz, 50µs) and (100Hz, 50µs).
Figure 4 shows the distribution of the mentions for the more describing terms.The sensations soothing and calming behave quite similarly as they were mentioned mentioned the most at (50Hz, 10µs).The number of mentions for sensations at the opposite end of the spectrum, like strong and hurting, is very low at 10 µs.hurting was mentioned the most at (100Hz, 1000µs), while the maximum for strong is at (2Hz, 500µs).Other describing terms that intend a high intensity, like energizing (2Hz, 250µs) and forceful [(1000Hz, 100µs), (2Hz, 500µs)], have a similar distribution that is either related to a high frequency or high pulse width.Both conditions are related to a high intensity of the stimulation signal.The sensations faint (2Hz, 10µs) and gentle (100Hz, 10µs), which received a high number of mentions at low pulse widths, also received more nominations for rather low frequencies.A stimulation is felt mostly as localized if the stimulation frequency is low, both maxima occur at 2 Hz with a pulse width of 10 µs and 250 µs, respectively.In contrast, a sensation is felt as diffuse for the pair (500Hz, 250µs).Both distributions are not without overlaps, but they seem to be complementary to each other.
In total, it seems that the pulse width primarily has an impact on strength-related aspects of the sensations as described above.0.17 0 0 0.25 0.33 0.17 0 0 0.08 0.08 0 0.08 0.17 0.08 0.17 0.08 0.08 0.08 0 0 0.08 0.25 0.25 0.42 0.25 0.17 0.08 0.67 1 0.83 0.67 0.75 0.75 0 0.08 0 0.17 0.17 Frequencies, on the other hand, seem to be strongly correlated with tactile aspects of the sensation like vibrating or jabbing.Very low frequencies evoke rhythmical sensations like jabbing.Increasing the frequency leads to more continuous sensations like prickling or vibrating.

Discussion
The results show that it is possible to create different kinds of sensations with electrotactile feedback and that the created sensation depends on both frequency and pulse width.Table 1 shows for each term at which combination of frequency and pulse with it was mentioned most often by our participants.The table could be used to further build on this work.Moreover, each heatmap is a rough map of how a sensation is perceived when the stimulation parameter changes and could be used for further explorations of individual sensations.The results also show that electrotactile feedback should not be limited to a frequency range of (10 -110 Hz) as well as the pulse width range should be analyzed up to the mentioned 2000 µs as reported by [21].
As the calibration was a very time consuming process in study 1, we used the gathered data visualized in Figure 5 (left) to reduce the required time.We observed that the amplitude is specific to each parameter set, but if someone prefers a high amplitude in one parameter set that person typically also prefers a high amplitude in another parameter set.We found that the desired maximum amplitude can be estimated using linear interpolation between known amplitudes for given pulse widths and frequencies.Equation 1 shows the regression result of the log of the pulse width (pw) and the log of the frequency (f) to the log of the amplitude.The prediction was found to be significant (F(2, 465) = 1084.7,p < 0.001), with  2 = 0.823.() = −0.8192* () − 0.1786 * ( ) + 6.7757 (1)

STUDY 2: RECOGNITION OF SPECIFIC ELECTROTACTILE FEEDBACK WITHOUT PRIOR LEARNING
To use the by the stimuli evoked sensations for HCI purposes, e.g. as an output modality for smartwatches, the parameter space has to be reduced and meaningful sensations have to be found that users can easily describe and recognize.We therefore evaluate in this study if users map the sensations created by electrotactile feedback without prior learning, only by the perceived feeling of the electrical stimulation.The study complied with our ethical review process.
For this reason we conducted a second study as a within-subject design.Our independent variable is our set of 9 different sensations, which is shown with the related stimulation parameters in Table 2.   1) plotted against the  of the chosen amplitude.Each marker represents one amplitude of one participant for a given combination of frequency and pulse width.
The parameters are based on the cluster centers that were identified in study 1.
We chose jabbing, pulsating, and vibrating as the distributions of these three sensations should be distinguishable by their frequency (compare Figure 3).We used slightly different pulse widths for jabbing (80 µs) and pulsating (10 µs) than in Table 1.For the sensations irritating, pulling, and squeezing we chose the frequency with the highest number of mentions in study 1 and tried to find an optimal pulse width within the range of the highest number of mentions (cf. Figure 3).The sensation itching, was rarely selected in study 1.Therefore we used the mean values for frequency and pulse width that were identified by Pohl and Hornbaek in [21].As our sensation term list from study 1 does not cover all possible sensations, we looked at the distributions of the more descriptive terms in Figure 4. We found that the sensations gentle and faint have similar distributions.We tried to come up with a term for the sensation that is created by a low frequency and a low pulse width.We found the term prodding similar to a slight touch to fit well and added it to our list of sensations.The same was done for stroking, which was created by considering the distributions of calming and soothing.Both terms created similar distributions in Figure 4 and describe a pleasant sensation.Thus we tried to find a term that fits these characteristics and decided to use the term stroking.We excluded the sensations prickling, as the feeling is wide spread over all parameter combinations, tickling and twitching, because of the small number of mentions in study 1, and stinging as it is an unpleasant sensation.
In the study the participants had to assign one of the 9 sensation terms of Table 2 to the played stimulation without knowing beforehand what term belongs to which parameter set.As in study 1, the terms were presented to the participants in their native language.We repeated this mapping task 5 times for each participant and each parameter set.The participants got no feedback if there choice was the correct term.

Participants
We recruited 17 participants (1 female, 16 male) with a mean age of 24.7 years ( = 2.94 years).One participant was left-handed, 14 were right-handed and 2 participants were ambidextrous.From 13 participants that wear a watch regularly, 5 wear a smartwatch, a fitness tracker, or a similar device.Eight participants had prior experience with electrotactile feedback or electrical muscle stimulation from previous studies, from that 3 had participated in study 1.As in study 1 we asked how electrotactile feedback could feel (multiple mentions possible), eight participants imagined it would be prickling, four mentioned vibrating and three pulsating.The following terms were mentions twice: hurting, pinching, stinging.Lastly the following terms were mentions once: irritating, pushing, itching, like pressure, tickling, tingling and pulling.

Procedure
Similar to Study 1, the participants first filled out a consent form and a demographic questionnaire.The consent form informed the participants about the study and especially about their right to end the study at any time without giving any reason.Also the form presented a list of health-related points to them under which they were not allowed to participate.The participants could indicate whether one of the points applied to them without explicitly naming it.If that occurred the participant was excluded from the experiment and received a bar of chocolate.After that they put on the smartwatch prototype and got the smartphone that guides them through the study (cf. Figure 6).To speed up the calibration of the participants, we used the regression Equation 1 as follows.We calibrated the sensations for each participant starting with the sensation vibration to determine the individual offset from the regression model in Equation 1.For that we started with a current of 6 mA instead of 0 mA, because about 95 % of the amplitudes of study 1 were above this value.By shifting the regression plane with that determined offset, we obtained the approximated amplitudes for all other sensations.The currents were then rounded down to the next 0.5 mA step.For each sensation we checked the calibration and let the participant increase or decrease the amplitude to find their personal optimum.Then the main part of the study started.A parameter combination was played and the participant selected zero or more of the 9 terms as shown in Figure 6.The intended mapping between the stimulated sensation and term was not presented to the participants.The main part proceeded in 5 blocks, such that each sensation was evaluated 5 times.Each participant gave a total number of 45 responses.After that we revealed the intended mapping between the terms and sensations, as shown  2, to the participants and asked them to judge how well the term fits the experienced sensation and if they would use the sensation as a feedback modality, e.g. for notifications in their daily life.At the end the participants received a bar of chocolate for their participation.The study took about 1 hour per participant.

Results
As we allowed the selection of multiple terms for each played sensation, we provide two measures of the quality of the recognition of the sensations as reported in Table 3.The first measure is the recall (percentage of correctly recognized sensations).The second measure is the precision and shows how accurately the term was used.Therefore the number of correctly recognized sensations for a given sensations was divided by the number of all selections of that term, e.g. the number of correctly recognized jabbing sensations is divided by all selections of the term jabbing.
Table 3: The first column shows the recall for each term.For example, jabbing was played and the term jabbing was in the selection set of the participant.The precision shows the relation of all correct selections of a term divided by all selections of that term.The third column shows the  1 score, i.e., the harmonic mean of precision and recall.The recall of all sensations is higher than the chance of randomly picking the right term (11.1 %).The four best recognized sensation are prodding, jabbing, vibrating, and pulsating.The precision is quite poor and reaches a maximum of 53 % for jabbing.The sensations with the best precision are jabbing, prodding, stroking, and pulsating.The sensation vibration has a low precision of 22 %.
Before the final part of the questionnaire, we presented our mapping of parameter pairs to sensations as defined in Table 2 to the participants the first time.After knowing our mapping, the participants rated on a 5-point Likert scale, how appropriate they judged the term for the given stimulation parameter pair.Figure 7 left shows that the majority of the participants agreed with our mapping of the terms.Only stroking was judged as inappropriate by the majority of the participants.We also asked the participants to state on a 5-point Likert scale if they would like to use the 9 sensations in an everyday life scenario, e.g. for notifications on a smartwatch.As shown in Figure 7 right, considering the usage of such colorful electrotactile feedback for the exemplary use case the majority of the participants could only imagine using vibrating, jabbing, prodding, and pulsating.Squeezing and pulling got an equal amount of positive and negative mentions with a tendency towards the negative.Itching was rated as worst for use in a notification context, which confirms the results of [21].

QUALITATIVE RESULTS
As smartwatches are often used to present notification we asked especially for this use case in the questionnaires.In the first study 9 of 13 participants and in the second study 12 of 17 participants agreed to use different types of notifications for different situations, e.g. to not disturb others, to distinguish different sources of notifications, or to react to different contexts, like a loud environment.So there is an interest of the users to encode more details in a notification.We also asked the participants of study 2 to rank the notification methods in their preferred order after they finished the study.Vibrotactile feedback was rated best with a mean rank of 1.8, followed by electrotactile feedback (1.9) and status led (2.2).Sound was rated worst (3.8).All three best ranked notifications methods have in common that they can be unobtrusive.Based on the ranking, participants could imagine using colorful electrotactile feedback as well as vibration for notifications.
Before the participants got in contact with colorful electrotactile feedback, the questionnaires of both studies show that the imagination of the participants of how electrotactile feedback feels, is somehow limited to the sensations of prickling, hurting, and vibrating.Regarding the results of study 2 (cf. Figure 7 right) the opinions changed.The majority could imagine to use most of the sensations for notifications in everyday life.The electrotactile vibrating sensation leads the preference ranking, followed by jabbing, prodding, and pulsating.15 of 17 participants rated the electrotactile feedback as not tiring.
Regarding the design of our prototype, we asked the participants for their opinion.The design of the prototype with the gold electrodes was judged positively.All 17 participants of study 2 agreed that wearing the prototype was pleasant.The size of the prototype was rated neutrally: 8 participants found it too big and another 8 participants as not too big (1 neutral).The armband was rated only by one participant as too wide and by three other participants as too tight.Only 6 participants could imagine using the prototype in everyday life (4 neutral).On the other hand 13 of 17 participants could imagine using the system integrated in a smartwatch.

DISCUSSION
The results of both studies show a high degree of variability.But the results also show that it is possible by varying the frequency and the pulse width to evoke sensations of different qualities.This offers a big playground for presenting diverse colorful feedback that could enrich our interaction with wearables.Furthermore, study 2 shows that it is possible to create natural sensations based on electrotactile feedback that can be recognized with no prior learning.
The results of study 1 might be better if we had forced the participants to only select the most fitting term among a given set of terms.We also believe that practice will increase the recognition rates.Also the results indicate over all participants tendencies whether a particular parameter combination of frequency and pulse width creates a specific sensation.The results are limited, as we only asked if a sensation was felt, e.g.vibrating or not.Alotaibi et al. [1] used a 7-point Likert scale for the four sensations of urgency, annoyance, valence, and arousal, which results in a more fine-granular resolution for the sensations.But this would increase the needed time for the study.To reduce the participants' effort, it might be reasonable to build semantic differentials for fitting terms, on which participants locate their sensation, e.g.localized -diffuse, energizing -calming, forceful -gentle, or strong -faint.For the other terms that do not fit into a semantic differential, at least a 5-point Likert scale should be used to get insights on how a parameter pair was perceived.This would even help to answer the question of why participants reported 2.4 % of the 1594 selected terms as hurting, because the calibration should have avoided that.Similarly, Pohl and Hornbaek [21] reported that 2 % of the mentioned terms in their study were rated as hurting.Other negatively connotated terms, like itching and stinging, also have their maxima for long pulse width of 500 µs and 1000 µs.This may indicate to use the areas of high pulse width only carefully.Nevertheless, we deem it important to also ask participants about negatively connotated terms to explore the full parameter space for creating sensations and to get a better understanding of both suitable parameter combinations for usable sensations as well as parameter combinations that should 20  Would you use the sensation in everyday life for notifcations?
strongly disagree disagree neutral agree strongly agree be avoided.The sensation vibration is felt over a wide frequency range and over a wide pulse width range.So the term is naturally connected to this kind of stimulation, this is also reflected by the low precision of vibration in study 2. With smartwatches that use electrotactile feedback it is possible to omit the vibration motor and produce a similar sensation on the skin with electrotactile feedback.The advantage is also that the electrotactile feedback can be controlled in a more fine-granular way.The amplitude and the frequency of the electrotactile vibration can be controlled freely and independently of each other, something that is not possible with eccentric rotating mass (ERM) vibration motors or linear resonant actuators (LRA).Unlike these actuators, which are housed in the smartwatch, the electrodes of the proposed method needs direct skin contact under the smartwatch casing.This area is also occupied by sensors in modern smartwatch generations.For example, the Samsung Galaxy Watch 6 3 features optical and electrical heart sensors, a bioelectrical impedance analysis sensor, and an infrared temperature sensors, that are located on the back of the smartwatch.We believe that this does not conflict with electrotactile feedback.First, the electrodes for electrotactile feedback can be scaled down, as shown by their usage in electrotactile displays [9].Further the electrotactile feedback does not directly interfere with optical sensors.On the other hand, of course, it will interfere the electrical measurement of the heart rate, as well as of the skin impedance.
When the electrotactile feedback is presented, a simultaneous and continuous measurement of both sensors would not be possible.This effect could be mitigated by synchronizing the sensor's measurements with the playback of the electrotactile feedback.A similar technique was implemented for the time-based interleaving of EMS and EMG on the same electrodes in MuscleIO [4].So it could also be possible to use the existing electrodes of the impedance sensors to present electrotactile feedback.The parameter range is so large that the presented results are only a first step in the direction of colorful electrotactile feedback.The heatmaps show some interesting spots for some sensations, e.g. for pulsating at 10 Hz, but maybe the sensation is felt much more strongly at 15 Hz.Either the parameter range has to be covered more 3 https://www.samsung.com/us/watches/galaxy-watch6/densely or future research could start by focusing on a particular sensation and investigate parameter combinations based on the presented heatmaps.The latter approach was taken by Pohl and Hornbaek for the sensation itching in [21].As mentioned before Knibbe et al. investigated the experience of EMS [13].They covered a frequency range of 20 Hz to 120 Hz and a pulse width range of 100 to 250 µs more deeply.However, the results are not fully applicable to electrotactile stimulation as the current amplitude is much higher in EMS than in electrotactile feedback.Therefore the reported sensations are valid for EMS, but not necessarily for electrotactile stimulation.
The medically certificated stimulator that we used in this work could be a good starting point for such explorations, as it provides a reasonable frequency, pulse width, and amplitude range.
A major problem that we could solve was the reduction of the time needed for calibration.The calibration in study 1 took longer than the actual term selection process.The whole study 1 took about 1 hour for each participant.This only allowed one presentation of each parameter combination and term selection.With the proposed regression approach the calibration process could be completed in study 2 in about one third of the time that was required in study 1.The exploration of a larger set of frequencies and pulse widths was primarily enabled by our regression model.Even if some parameters of the setting change, e.g., because electrodes of different sizes are used, the approach should still be beneficial.

LIMITATIONS
The results are limited by the fact, that we could not explore the full parameter space.So there might be a global optimum for a sensation that we have not found yet.Also the number of participants and their young age limit the expressive power of our work.Our results are also limited by the device that we used.Even if the stimulator offers a wide range of supported pulse widths and frequencies, the parameter set is limited.The creation of pulse widths longer than about 4 ms or shorter than 10 µs and frequencies of more than 1 kHz is not possible.Also the granularity of the amplitude could be finer for sensations that depend on a high frequency or pulse width.For example, Pohl and Hornbaek [21] mentioned a current of 0.2 mA for creating itching.Furthermore the device always sends a probe pulse preceding any biphasic pulse, which may affect the results.The particular coefficients of our regression model are specific to our prototype and setup, as we have used a fixed electrode size and did not systematically evaluate different locations or electrode sizes with our approach.The results of study 1 are limited by the used list of terms.Terms that are not on the list could not be selected by the participants.So the expressiveness of colorful electrotactile feedback could be broader than what we present here.Terms like stroking could be added, other terms could be removed.

CONCLUSION & FUTURE WORK
We showed that colorful electrotactile feedback can present a wide range of different sensations.The frequency and the pulse width determine which sensation the electrotactile feedback evokes.By controlling these two parameters colorful sensations can be used for various scenarios to enrich our interaction with wearables like smartwatches.Notifications can be encoded with different qualities, like the source or the urgency.Table 1 provides other researchers an overview on which parameters could be used to create a specific sensation.The presented heatmaps in Figure 3 and Figure 4 give a rough overview over the distributions of the sensations in the huge parameter space and could be used for fine tuning of the parameters.The sensations that we evoked were mostly not solely perceived.For example, when we played the stimulation for pulsating the participants selected not only the term pulsating but also other terms.Further, we presented a calibration approach that substantially reduces the needed time for the calibration.The regression equation and the parameters are reported and can serve as a starting point for future work in this area.It considerably speeds up the calibration process when evaluating a massive parameter set.
In study 2 we analyzed if the electrotactile feedback can produce sensations that can be recognized and understood without prior learning.The results show that the participants achieved, without any practice, recognition rates well above the rate of randomly choosing one term, which is at 11.1 %.The four best recognized sensations without prior learning in that order are (with parameters) prodding (f = 2 Hz, pw = 10 µs), jabbing (f = 2 Hz, pw = 80 µs), vibrating (f = 50 Hz, pw = 50 µs), and pulsating (f = 10 Hz, pw = 10 µs).Training specific sensations with the users will likely increase these recognition rates.
Our work could be used as starting point for further work investigating the parameter space.Our concept of colorful electrotactile feedback could also be used for patterns, like double playback of jabbing, and could enhance concepts as Electrotactons [1].The intensity of the sensation could be adjusted over time or sensations could be presented on multi-electrode setups on the back of the smartwatch.

Figure 1 :
Figure 1: Left: The prototype, consisting of the smartwatch stub, the RehaMove 3 device, and a microcontroller with a battery.Right: Carbon rubber electrodes used in study 1 and gold plated electrodes used in study 2.

Figure 2 :
Figure 2: Left to right: Study setup, calibration view, term selection view.

Figure 3 :
Figure3: The heatmaps show the relative amount of selected terms depending on the frequency and pulse width for the terms that are related to a sensation on the skin.

FaintFigure 4 :
Figure4: The heatmaps show the relative amount of selected terms depending on the frequency and pulse width for the terms that are describing a felt sensations.

Figure 5 :
Figure5: Left: The calibrated amplitudes for all participants, the larger rhombus shows the mean value.Right: The results of the log-log-regression (Equation1) plotted against the  of the chosen amplitude.Each marker represents one amplitude of one participant for a given combination of frequency and pulse width.

Figure 6 :
Figure 6: Left to right: study setup, term selection view.

Figure 7 :
Figure 7: On a 5-point Likert scale participants rated how well the term fits to the sensation created by the parameters and which sensation they could imagine to use in everyday life for notifications.

Table 1 :
This diagram shows the pairs of frequency and pulse width at which our participants selected the mentioned terms most often.Cells filled by an x indicate am impossible parameter pair for biphasic stimulation signals.

Table 2 :
Evoked sensations with their frequency and pulse width based on the cluster centers that were identified in study 1.