Can You Ear Me? — A Comparison of Different Private and Public Notification Channels for the Earlobe

The earlobe is a well-known location for wearing jewelry, but might also be promising for electronic output, such as presenting notifications. This work elaborates the pros and cons of different notification channels for the earlobe. Notifications on the earlobe can be private (only noticeable by the wearer) as well as public (noticeable in the immediate vicinity in a given social situation). A user study with 18 participants showed that the reaction times for the private channels ( Poke , Vibration , Private Sound , Electrotactile ) were on average less than 1s with an error rate (missed notifications) of less than 1%. Thermal Warm and Cold took significantly longer and Cold was least reliable (26% error rate). The participants preferred Electrotactile and Vibration . Among the public channels the recognition time did not differ significantly between Sound (738ms) and LED (828ms), but Display took much longer (3175 ms). At 22% the error rate of Display was highest. The participants generally felt comfortable wearing notification devices on their earlobe. The results show that the earlobe indeed is a suitable location for wearable technology, if properly miniaturized, which is possible for Electrotactile and LED . We present application scenarios ∗ These authors


INTRODUCTION
Notifications accompany us throughout the day on our smart devices, like the smartphone or the smartwatch, to inform us about current events.It is possible to reach a person almost immediately by calling or writing a message.By moving notifications from the smartphone in a user's pocket to the smartwatch on a user's wrist, notifications can be delivered and perceived faster and more reliably, as they are less likely to be missed [29].There are also other viable body locations for wearables but only a few are promising for social interactions.
Zeagler lists the following high potential locations: hand, wrist, forearm, upper arm, upper chest above the breast, forehead, ear, and mid thigh [47].
In this work, we focus on the earlobe as it is a common location to wear jewelry, like earrings or ear clips.The earlobe as a location for wearables not only offers the opportunity to inform the user about a notification, but, as an exposed location, is also suited to present information to people nearby.The earlobe and its potential for smart wearables has not yet been sufficiently explored.Only a few publications consider the earlobe as a location for smart wearables (e.g., [22,27]).The earlobe shares common benefits with other locations on or around the ear, like the proximity to the auditory canal.It also shares design constraints, like the need to minimize power consumption to keep the device small.In contrast to other ear locations the earlobe offers unique opportunities: It is visible to others in a face-to-face conversation as well as if the person is seen from the side, it offers a sufficiently large flat surface of about 2×2 cm for placing a wearable [1], and it is a socially accepted position to wear jewelry [44].Moreover, the traditional use of the earlobe as a place for jewelry demonstrates that the earlobe is suitable for convenient long-term wearability.Other ear locations, like in the auditory canal, can be used by in-ear devices to measure positional data, audio, and physiological parameters like the heart rate, blood pressure, respiration rate, or to present well-designed audio feedback.However, a major disadvantage of in-ear devices is that they interfere the auditory canal, preventing the wearer from fully hearing their surroundings.Moreover, wearing earbuds may be perceived by others as rude or as if the person is not interested in a conversation [42].A location inside the ear that does not occupy the auditory canal needs a customized design for every user as the shape of the ear is strongly individual.The ear has been explored for authentication, because of having distinctive structural features [16].Devices that are placed at the back of the ear interfere with objects like glasses.Again, creating a customized solution such as integrating the smart device into the temple of the glasses, would be necessary.The space behind the ear is not suitable for public notifications as the line of sight is blocked by the ear.As custom solutions are not suitable for everyone and require additional effort, the earlobe, as a natural location without major restrictions for wearing jewelry, has a high potential for wearing smart devices.
Based on these considerations, we see the earlobe as location with high potential for presenting information.Related work lacks a direct comparison of the suitability of different notification channels on the earlobe.We believe that in the future earrings can be digitally augmented to notify the wearer on the earlobe.To assess the suitability of different notification channels, we evaluated the noticeability and acceptability of vibration, electrotactile, poke, auditory, and thermal feedback on the earlobe.
As mentioned before, the earlobe is visible to people nearby and thus offers the opportunity of serving as a space for a wearable public display.As the earlobe is in the direct field of view during a conversation, it provides a suitable position for a public display to enhance the conversation with additional information.This led us to evaluate the noticeability of a single LED, a rectangular OLED display, and auditory feedback for people nearby.The opportunity to present information to the user and also to people nearby also enables various new interaction scenarios beyond exclusively private or exclusively public notifications.
In this paper, we contribute an evaluation of different private and public notification channels on the earlobe.We consider the use of smart ear clips for notifying either the wearer or people who are in the near vicinity of the wearer.Moreover, we present quantitative data like reaction times and error rates, qualitative feedback from a preliminary online survey, as well as feedback from the participants of our user study to draw conclusions for the future use of notification channels on the earlobe.In a small field study, we evaluated our prototype in a gym, in which a private and a public notification channel were combined to notify the wearer as well as a fitness instructor about the training intensity of the wearer.We present qualitative findings based on two interviews of the fitness instructor and of the person exercising.We show the suitability of the earlobe as a promising new location for miniaturized wearable notification devices.

RELATED WORK
In current research the earlobe is given relatively little attention as a location for wearables [36,47].With a length of about 2 cm, which increases during life, the earlobe is either attached to the cheek (nonpendulous) or not (pendulous) [1].Although the earlobe is not in one's field of view, it is visible for others.Ni and Baudisch [27] see the earlobe as potential location for future devices.They suggest that such devices could either be implanted or mounted and could offer gesture input or audio input and output.Further, the earlobe can be considered as a culturally accepted location for wearing jewelry [41].An integration of technology in jewelry offers a wide range of possibilities for personalization and individualization and for presenting user-specific feedback and content [25].Several works investigated the other parts of the ear as a location for providing feedback or using it as an additional input surface [36].The explored prototypes target different locations like the auricular [12,22,26], the area behind the ear [2,4,23,46], the earlobe [9,12,22], and the auditory canal [2,6,19,20].Below, we first discuss related work on the notification channels that we propose for earlobe output.Then we present related work on the different ear locations in terms of what has been evaluated on the input and output side of the ear.

Selection of Notification Channels
We propose the earlobe as a location for various types of notification output.Comparisons of different notification channels have already been conducted for other body locations, with a focus on smartwatches and rings.Roumen et al. [37] presented NotiRing, a set of five interactive rings for the notification channels light, sound, vibration, poke, and warmth.They evaluated the reaction times and error rates for different levels of physical activity (laying, sitting, standing, walking, and running).The results show that the participants reacted most quickly to vibration, which was also the most reliable notification channel.Physical activity did not affect the reaction times for vibration, sound, and poke.In contrast, the noticeability of light and thermal feedback was affected by the physical activity.
An evaluation of ten different channels (press, poke, pinch, heat, cool, blow, suck, vibrate, moisture, and brush) at six different body locations (collar bone, shoulder, stomach, side of the torso, upper arm, lower back) under clothes for delivering notifications was conducted by Bhatia et al. [3].The results show that the location affects the error rate, but the reaction time is only dependent of the channel.Further, Warnock et al. [45] evaluated different modalities (visual, auditory, tactile, and olfactory) for presenting notifications in a desktop environment.Participants had to react to these during a card game as the primary task.The modality did not affect the error rate of the primary task, but the olfactory and tactile modalities slowed down the performance of the primary task.Roumen et al. [37], Bhatia et al. [3] and Warnock et al. [45] explore various sets of notification channels, while we focus on a smaller and more commonly used set of notifications channels (poke, vibration, thermal warm and cold, sound, visual for public scenarios) for our evaluation.For some more exotic notification channels evaluated by Bhatia et al. [3] and Warnock et al. [45], we see problems in miniaturizing them for daily use: blow and suck need an additional air pump, brush needs an additional servo motor, olfactory, and moisture need some refilling of liquids as well as some volume to store it.The notification channels press, pinch, and poke of Bhatia et al. [3] could not be evaluated easily, as the ear clip always adds some pressure on the earlobe to fasten the prototype.Further, when the solenoid is activated the pin compresses the earlobe between itself and the ear clip, which could be perceived as increased pressure, poke, or a pinch on the earlobe.
Electrotactile feedback for wearables was evaluated in TactileWear [43].Different locations around the wrist and ring finger were compared for electrotactile and vibration feedback.The results suggest that electrotactile feedback is more localized than vibration feedback.The evaluation of different patterns showed that the participants recognized the patterns more accurately with electrotactile feedback than with vibration feedback.However, the participants preferred vibration feedback.We included electrotactile feedback in our set of notification channels for ear clips, as it produces soundless tactile feedback and is fairly energy efficient [18].

The Ear as Location for User Input
On the input side, well-known earables, like the eSense platform from Nokia Bell Labs [19], enable playing and measuring sound, sensing motion, and communicating via Bluetooth.Röddiger et al. [35] evaluated the design space and usability of an earable prototyping platform that is based on the eSense platform with a focus on problem-oriented learning.Unlike the eSense platform, Eartouch [20] adds photo-reflective sensors to an in-ear headphone to measure shape deformations of the ear.The Eartouch prototype recognizes five ear-pulling gestures with an average accuracy of 77.43 %.EarBit [2] compares different sensor settings for recognizing chewing.Bedri et al. places a proximity sensor near the auditory canal to measure the deformation of the auditory canal, an IMU behind the ear, and a microphone around the neck.A comparison of the recognition rates of the different sensors and their combinations showed that the IMU behind the ear as the sole sensor, achieved the best recognition rates and was rated as the most comfortable variant.Auracle [4] is a wearable earpiece with a contact microphone that is worn behind the ear to detect chewing.Auracle achieves an accuracy of up to 92.8%.
Several works propose wearables that are not worn in the auditory canal.Earput [23] is an arc-shaped accessory worn behind the ear, consisting of 12 distinct electrodes that can recognize single-touch or multi-touch events as well as grasp interactions and mid-air gestures via capacitive sensing.Lissermann et al. provide an analysis of the design space behind the ear for input purposes and report the results of an experiment that shows that different regions behind the ear can be targeted reliably.iSkin [46] enables the production of custom touch sensitive and flexible skin stickers.For example, a sticker that is attached to the back of the ear offers touch buttons to control a music application.The recognition of face gestures is also possible with ear-worn devices.Futami et al. [12] present three different prototypes, located either on the earlobe, the auricular, or the tragus.Optical sensors measure the deformations of the ear when performing a face gesture.Nine face gestures could be recognized with high F-scores of above 94 %.While we are investigating the output capabilities of ear clips, the mentioned works in this subsection did not influence our work strongly.However, they show the high relevance of research on wearables on or at the ear, as it is an excellent location to sense input and present information.

Presenting Information on the Ear
On the output side, different modalities have been evaluated.For example, Nasser et al. [26] investigated thermal feedback at the auricular area of the ear by presenting cold and hot stimuli with an earhook prototype with multiple Peltier elements.The mean single-point accuracy was above 85 % and the average accuracy for a multipoint pattern stimulation was reported with 85.3 %.ActivEarring [22] is a vibration ring on the ear with three actuators, two on the auricle and one on the earlobe.The authors analyzed single actuations as well as spatial and temporal patterns at the mentioned locations.The results show a mean accuracy of 58.4 % for 25 patterns using both ears with three actuators each.A smaller and more lightweight implementation of the actuators can be found in FLECTILE [9].It is a flexible electromagnetic actuator consisting of a coil in a flexible material and a magnet.A current flow through the coil generates a magnetic field that causes attractive or repulsive forces moving the actuator.Fang et al. [9] propose wearing it on the earlobe as an exemplary application scenario.Toning [6] suggests adding a display or LED to earbuds to show a color related to the played music.The color is used to share the playlist with others in the vicinity and indicates the music genre.The colors were determined in a survey.As this concept is mainly bound to the context of hearing music, it is also highly related to our idea of presenting notifications to people in the immediate vicinity.
Apart from mounting actuators on the ear, the ears themselves can become the output device.Pull-Navi [21] is a 3D navigation mechanism that is implemented by pulling the ears in the desired direction.Wiggleears [32] used the ears for public output by flapping them with servo motors based on biodata to express emotions.Other work focused on public output by mounting displays on regular over-ear headphones [42].To the best of our knowledge there has not yet been a systematic investigation of different notification channels on the earlobe.
For this work, we included the notification channels that were evaluated in NotiRing [37].Additionally, we chose thermal cold feedback as well as electrotactile feedback, as they were common candidates for comparison [26,43].Because we are also interested in presenting public notifications on the earlobe, we chose LED (as used in NotiRing [37]), display [42], and sound [37] as public notification channels.

IMPLEMENTATION OF THE EAR CLIPS
Our implementation consists of a custom control PCB (see Figure 2) with a microcontroller (an Adafruit Feather M0 Bluefruit1 ) and an Android smartphone application.The PCB allows the connection of eight different ear clip prototypes as depicted in Figure 1 -one for each notification channel, except for Thermal Warm and Thermal  1) that cannot be drawn from the microcontroller pins (maximum 46 mA).For that purpose MOSFETs were used for the Peltier element and solenoid to provide a higher current via the power source.A motordriver DRV2605L2 that is stacked on the PCB was used for the vibration feedback.The control of the electrotactile feedback needs galvanic isolation from the control logic to protect the circuit of high voltages from the electrotactile feedback.This is similar to the circuit used by Pfeiffer et al. [33] in the LetYourBodyMove toolkit.The schematics as well as the layout of the PCB can be found in the supplemental material.The microcontroller communicates with the Android smartphone via Bluetooth LE.The Android application controls which notification channel to use.
The electronic parts that we used for the notification channels on the ear clip prototypes are shown in Table 1.We distinguish between private and public output of notifications.This is necessary to address different usage scenarios in which (a) only the wearer gets a notification and (b) one or more persons in the vicinity of the wearer get a notification.The parts are mounted on so-called screw-hinge clip-on earrings, which have a screw to optimally fasten the clip for each participant.The solenoid needed an additional mount due to its weight of 15 g (for comparison: the vibration ear clip has a weight of 3 g).We also attached a small heat sink to the Peltier element to minimize the heat reflow for the Thermal Cold channel.For better perceptibility, we created a small 3D-printed translucent sphere that we attached on top of the RGB LED to scatter the light (see Figure 1).The electrotactile signals were biphasic pulses that were generated with a medically approved EMS/TENS device (cf.Table 1).

PRELIMINARY ONLINE SURVEY
To gain first insights into the acceptance of providing public notifications on the earlobe and showing information to people nearby, we conducted a preliminary online survey.In this survey, we focused on the possibility of presenting public notifications, as it is unconventional to notify people nearby with your own wearable.Therefore, we explored each public channel in an individual scenario.We recruited 57 participants (16 female, 41 male, age 19-56, M = 30.2,SD = 11.3).20 participants were experienced in wearing earrings.Of these 20 participants, 9 wear earrings only, 1 wears ear clips only, and 10 wear both earring types.
We presented three scenarios (see Figure 3) to the participants and asked the participants for their opinion.Scenario 1: We showed our participants Figure 3 (left) and told them that they should imagine a scenario in which a passenger is sleeping on an airplane, wearing a red flashing LED at his ear.A stewardess or steward serving food sees the passenger.Fig. 3.We presented three different scenarios to our participants to get first insights into the acceptance of providing notifications on the earlobe.In the first scenario (left), a passenger is sleeping on an airplane, wearing a red flashing LED on the earlobe.The second scenario (middle) shows a person wearing an auditory ear clip that notifies the wearer when an egg was finished boiling.In the third scenario (right), a sleeping commuter on a train is wearing an ear clip with an attached display that shows "Please wake me up at Central Station." We asked the participants, if they were the stewardess or steward, what they thought the passenger might want to express.The participants answered the questions freely without predetermined answers to prevent bias.The answers were collected and clustered.Of the participants, 37 answered that they would not disturb the passenger, as they assumed that the red flashing light indicates that the passenger does not want to be disturbed.In contrast 9 participants thought that they should wake up the passenger, because the flashing could indicate a medical emergency or that the passenger wants to be woken up.The other 11 participants were undecided what they should do.
The participants were asked to state which of the following flashing patterns of the LED they would prefer in a scenario like this: always on, alternating on and off without fading, or alternating on and off with fading.The participants ranked alternating on and off with fading first, followed by always on, and alternating on and off without fading last.
Scenario 2: For this scenario, Figure 3 (middle) was presented to the participants.The participants should imagine using an auditory ear clip as a timer for boiling an egg.While the egg is boiling, the user can do other things like tidying up his or her room.When the boiling process is done, the wearer will be notified by an acoustic alert on the ear clip.
Forty participants could imagine using an ear clip with auditory feedback as an output channel for presenting this kind of information.We asked the participants, which of the following sounds would be best suited for receiving such notifications: single note, melody, beeping alarm sound and other (free text).21 participants answered that a single note would be best suited, followed by a melody, which was selected by 20 participants.The beeping alarm sound was chosen by 11 participants.The other 5 participants said that a custom sound would be suited best, so that the user can personalize the sound.
Scenario 3: A commuter is wearing a display-based ear clip on his daily train trip to work.For this scenario, we presented our participants Figure 3 (right).The commuter is sleeping and the ear clip shows: "Please wake me up at Central Station." 38 participants agreed that showing this information would would be useful in a commute scenario (possible answers were yes or no).We chose this scenario as it is common for commuters in cities like Tokyo to sleep on the train, which may result in missing the destination station [10].
Additionally, we asked our participants which representation they would prefer for showing such information on a display: text floating from left to right (automatic horizontal scrolling), words or phrases presented serially (Rapid Serial Visual Presentation, RSVP), or using just a few words so that the text is short enough to be displayed completely.We asked them to rank the three options.They preferred presenting short text over horizontal scrolling.RSVP was rated least.Furthermore, we asked our participants whether they could imagine other application scenarios for displays worn on the earlobe.The participants could reply with a free text answer.14 participants stated that the display could be useful for people with special needs to use it either to communicate over the display or to inform others about the special need.13 participants told us that they can only imagine using the display in the airplane scenario (Scenario 1). 3 participants would use the display for emergencies to show health or medical parameters to, e.g., paramedics and 14 participants found other scenarios like using the display at parties.
The results of this online survey show that the participants can imagine showing information to people nearby by wearing a device on the earlobe.Further, the participants indicated what information could be provided through devices worn on the earlobe and in which context.We also got first insights in how the information should be presented.These results were used as a basis for the lab study, which is reported next, regarding the presentation of public feedback.In the lab study, the public notifications are encoded according to the results of the survey: single note for audio feedback, alternating on and off with fading for LED notifications, and short text on the display.

LABORATORY STUDY
To evaluate the noticeability of the notification channels, we conducted a lab study, in which we measured the reaction times and error rates of the participants.In a real environment the user would likely perform an activity while a notification arrives.To simulate that the user is not fully concentrated on waiting for incoming notifications, we used a distraction task.The distraction task was the name matching task [5].We used a slightly modified version of the name matching task that is part of the ReactionTimeExperiment tool provided by MacKenzie [24].We reduced the minimum delay between two name matchings from 2000 ms to 500 ms to increase the cognitive load of the participants.The participant received a notification on a self-worn prototype for the private notification channels (see Figure 4, left) or worn by the colleague "Bob" for the public notification channels (see Figure 4, right).The participants were instructed to press a large button in our smartphone application, when they detected the notification.
The notifications were played for 20 s after a random delay of at most 50 s.Similar to NotiRing [37] a notification was alternately turned on and off every 500 ms.In contrast to NotiRing [37], the next notification started with a minimum delay of 1 s after the last played notification to avoid playing notifications directly after each other.The participants performed 10 trials for each notification channel.In total, each participant received 90 notifications.After each completed notification channel, the participants filled out a questionnaire.
We used screw-hinge clip-on earrings (as seen in Figure 1) to give participants without pierced earlobes the opportunity to participate.The participants tightened the ear clips themselves to ensure a comfortable wearing experience.The experimenter checked the correct positioning of the ear clips at the center of the earlobe.
Regarding the feedback we got from our preliminary online survey, we decided to use the text "HAMBURG HBF" for the display to simulate the scenario of showing information on the screen.For the sound notification channels, we used a single note, as preferred by the participants of the preliminary online survey.We also decided to use the color green for the LED to show an availability status as suggested by Stanke et al. [42].The alternating pattern we use for all notification channels also fits to the LED feedback we collected from the participants of our preliminary study.

Study Design
The laboratory study has a within-subjects design with the notification channels as an independent variable.To overcome order effects, we used balanced Latin squares [24].The participants wore the prototypes alternately on the right and left earlobe to reduce effects that might occur due to fatigue.We also counterbalanced with which earlobe the participants started.Thus, one half of the participants started with wearing the first prototype on the right earlobe, the other half started with the left earlobe.During the trials, except for the sound modalities, the participants wore noise canceling in-ear headphones (Sony WF-1000XM3) playing white noise to drown out the mechanical sound of some of the actuators.The amplitude of the biphasic signals of the electrotactile feedback was calibrated for each participant individually before the study started to create a strong but still comfortable sensation.We used a constant pulse width of 50 µs and a constant frequency of 100 Hz.

Participants
We recruited 18 participants (3 female, 15 male, age 21-47 years, M = 25.9 years, SD = 5.6 years) of which 12 had participated in our preliminary online survey.Five participants had experience with wearing earrings, 3 of these also with ear clips.The mean width of the left earlobe was 18.9 mm (SD = 2.2 mm) and of the right earlobe 19.1 mm (SD = 2.3 mm).

Results
For the analysis of the reaction times, we used the data of correct trials and calculated the median time for each notification channel of each participant to get a representative value for each participant.We then calculated the mean for each notification channel over all participants using the median times.We counted a trial as an error if a user missed to respond to it within 20 s after the notification started.Table 2 summarizes the quantitative results, i.e., reaction times and error rates.
The reaction times of the private notification channels reveal two groups.The participants responded to the notification events on the channels Vibration (696 ms), Electrotactile (731 ms), Poke (823 ms) and Private Sound (1009 ms) on average in less than about 1 s (see Table 2).We denote these channels as the "fast" group.For notification channels in the second group, consisting of Thermal Warm (3358 ms) and Thermal Cold (4280 ms), our participants had a reaction time of more than 3 s.We denote these channels as the "slow" group.We conducted a Friedman test, which shows significant differences between the reaction times of the notification channels (χ 2 = 59.02, p < 0.001).A pairwise post-hoc Conover test with Bonferroni correction shows significant differences between the channels of the "fast" group and the channels of the "slow" group (p < 0.05).The error rates show that the participants missed 26 % of the notifications with the Thermal Cold channel, followed by 4 % error rate for Thermal Warm, and 1 % for Private Sound.For all other private channels, our participants did not miss any notification.These omission errors happened while the participants were distracted by the name matching task, as described above.We conducted a Friedman test for the public channels, which shows significant differences regarding the reaction times (χ 2 = 28.0,p < 0.001).A pairwise post-hoc Conover test with Bonferroni correction shows significant differences between Public Sound and Display (p < 0.001) as well as for LED and Display (p < 0.001).Our participants reacted significantly faster to Public Sound and LED than to Display (see Table 2).Further, the participants did not miss a single notification for Public Sound and LED.In contrast, the participants missed 22 % of the Display messages.

Scenarios
We asked our participants in which scenarios they could imagine using the notification channels.For the private notification channels, we asked their assessment of 6 typical notifications that we aggregated from [39] (see Figure 5, top) and for the public notification channels we asked them about 5 notifications that a user might want to present to people nearby (see Figure 5, bottom) that are inspired by [30].
As shown in Figure 5, top, the majority of the participants could imagine using Vibration, Electrotactile, and Poke for the notifications Incoming Call, Instant Messaging, Appointment Reminder, Incoming Email, and Alarm.Private Sound was also rated as a possible notification channel for Low Battery, Instant Messaging, and Incoming Email.Our participants could imagine using Thermal Warm to indicate a Low Battery level or an Alarm.Thermal Cold was also favored for presenting a Low Battery level.
We tested the hypothesis of a correlation between the reaction times and the Likert scale ratings by performing Kendall Tau B correlation tests.We found significant negative correlations between the median reaction times of the participants and the Likert ratings for 5 of the 6 different example notification types: Incoming Call (r  = −0.31,p < 0.01), Instant Messaging (r  = −0.37,p < 0.01), Appointment Reminder in 30 min (r  = −0.28,p < 0.01), Incoming Email (r  = −0.41,p < 0.01), and Alarm (r  = −0.22,p < 0.01).This means that shorter reaction times are correlated with higher willingness to use the respective notification channel in the mentioned scenario.For the public notification channels, the Likert scale ratings are more divergent (see Figure 5, bottom).Our participants could imagine using Display and LED for indicating either if the wearer wants to be spoken to in public as well as in an office environment.In contrast, our participants could not imagine using the Public Sound notification channel in such scenarios.However, to indicate that the time is running out during a conversation, our participants could imagine using Public Sound, followed by LED, and Display.The channels Public Sound and LED are both favored to get attention from people nearby.
Like for the private notification channels, we analyzed the data for correlations between reaction times and Likert scale ratings.We found significant correlations for the following notification types: "Indicate if someone can talk to me in public space" (r  = 0.27, p < 0.01) and "Indicate if you have time for a conversation at work" (r  = 0.29, p < 0.01).This suggests that in scenarios in which our participants want to show an availability status, they prefer more subtle notification channels even though they may require a higher reaction time.Further, "Getting the attention from another person" has a significant negative correlation (r  = −0.25,p < 0.05), which shows that in a time-sensitive scenario the participants prefer a notification channel with a shorter reaction time.

User Experience and Subjective Preferences
Moreover, we asked our participants about their experience during the experiment with the private and public notification channels (see Figure 6).Overall, the feedback created by the presented notification channels was perceived as pleasant by our participants.The only exception is Thermal Warm, for which our participants were undecided regarding the pleasantness.All notification channels were rated as non exhausting, except for Display, which received a mixed rating.Display is also the only notification channel, for which our participants stated that they did not get used to it after a few tries.The participants found Vibration, Electrotactile, and Poke to be useful for daily life.Private Sound received mixed ratings.Thermal Warm and Thermal Cold received more negative ratings.The public notification channels received mixed ratings in terms of usefulness.
We asked our participants to rank the notification channels by their preference.We used the Schulze method [38,40] to calculate a ranking over all participants.According to the Schulze method, they ranked Electrotactile over Vibration, followed by Private Sound, Poke, Thermal Warm, and Thermal Cold for the private notification channels.For the public notification channels LED was ranked first, followed by Public Sound and Display.

FIELD STUDY
To ecological validity, we evaluated notification output on the earlobe in a follow-up field study.The study was conducted with one participant who wore an ear clip prototype during a supervised training session of about 30 minutes in a gym.During the training session, the ear clip gave both the fitness instructor and the exercising person relevant feedback about the level of exertion, based on the measured heart rate.We interviewed the instructor as well as the exercising person, to gather qualitative feedback on how the wearable device influenced the training session.Even though this is a small-scale field study, with just one exerciser and one fitness instructor, we found that it gives valuable insights in the practical viability of the proposed approach.

Ear Clip & Feedback Design
For the training session, we combined the LED and Electrotactile ear clips to provide public notifications via LED and private notifications via electrotactile feedback.In the lab study, both of these prototypes resulted in short reaction times and were favored by the participants.The ear clip prototype was controlled by the custom control board (see Figure 2).
In a training session, the heart rate is an important parameter for estimating training intensity.We classified the current training intensity into 5 heart rate zones, according to Garber et al. [13] and Riebe et al. [34] (see Table 3), by calculating the ratio of the actual and the maximum heart rate (%HR max ).We estimated the maximum heart rate as HR max = 220 − age, according to [11].The ear clip notifies the wearer when the recommended training zone has been reached.We chose 77 % of the maximum heart rate (zone 4) as the notification threshold for the desired vigorous zone.When passing this threshold, the clip plays an electrotactile feedback pattern (repeating 1 s on, 30 s off).This low repetition rate is designed to avoid annoying the exercising person by continuous tactile output.If the exerciser exceeds a heart rate of 96 % of the maximum heart rate (zone 5), the frequency of the alternating on and off pattern of the electrotactile feedback increases.Both the on and the off times are reduced to 250 ms to continuously indicate that the training effort should be reduced.This second pattern conveys a much higher sense of urgency and is unlikely to be unnoticed.As the training session is supervised, we decided not to inform the exercising person via electrotactile feedback about the other zones (1-3).The trainer sees the different colors on the ear-worn LED as shown in Table 3.When another heart rate zone is reached, the LED initially flashes once in the new color and then lights up constantly in that color until the zone changes again.

Procedure
We interviewed a professional fitness trainer before and after the training session in semi-structured interviews.In the first interview, we explained the trainer how the ear clip informs him about the heart rate state of the participant by showing him Table 3. Further, we collected demographic data from our exercising participant and asked him about his current training status.
We then prepared the training session and attached the ear clip to the earlobe of the exerciser.As we used screw-hinge ear clips, we could ensure that the ear clip was tight enough to not fall off during the training while also feeling comfortable on the ear.The electrotactile feedback was calibrated for our participant to have a comfortable but very noticeable feeling.
During the training session the participant wore a Huawei Watch 2. An Android application on the smartwatch measures and logs the heart rate and communicates with our custom control PCB via Bluetooth.The PCB controls the LED and the electrotactile feedback depending on the heart rate and the zones defined in Table 3.Before the study started, we temporarily activated the zones 4 and 5 (one after another) in the debug menu to familiarize the participant with the feedback.
We asked the trainer to instruct several exercises to reach different training intensities.He selected the following exercises: indoor cycling (with a recumbent bike; see Figure 7, middle), leg press, leg curl (see Figure 7, right), and circuit training (hydraulic chest press, hydraulic squat machine).For each exercise, the trainer instructed the exerciser which weight to choose and how fast to perform the exercise.After the training session, the prototype was powered off and the ear clip was taken off.The exerciser filled out a questionnaire and we interviewed the trainer a second time, followed by an interview of the exerciser.

Pre-Training Interview
The trainer completed a dual study program in fitness economics and has 6 years experience in a manager position in 5 different locations.Besides supervising people while exercising, he also performed managing tasks in the gyms.He did not specialize in any particular training area, but has 12 years of boxing experience.We asked him about the goals he usually aims to achieve with a person through the workout.As the goals of his clients are very individual, he always has a first interview to find out the personal motivation for training (e.g., rehabilitation training) and to gain initial insights into what the client would like to achieve.The goals for the training also depend on the current fitness level (untrained to athlete) and health status (recovery or healthy person).Typical goals include preventative training or muscle formation, and, more rarely, increasing endurance.Based on the goals, the trainer derives subgoals and determines parameters of the workout (e.g., intensity and number of repetitions).
Further, we asked about his personal approach to train someone, i.e., whether his training instructions are based on physiological parameters, on the current fitness level, or on his own feelings and experience.We also asked how he ensures that the exerciser trains at the right intensity.Again, he answered that this is very individual.Physiological parameters, like the heart rate, are only available at certain training devices and are necessary for special training, like cardio training.For healthy individuals, he observes the amount of handled repetitions and the facial expression of the exerciser during the workout.
We then asked which physiological parameters of the exerciser he would like to know during training.According to the trainer, the most interesting parameters are the lactate value and the heart rate.The latter is particularly interesting when dealing with COVID-19 patients or people with myocarditis or cardiac arrhythmias.We asked about the level of granularity that would be optimal when presenting these parameters to him.He stated that a coarse guiding value would be optimal.For example, knowing the exact heart rate of an exerciser is less relevant than knowing that he is currently in heart rate zone 3 or 4. Because of the individuality of each person, it is necessary to have one abstraction or interpretation step between the raw values and what the trainer sees.A heart rate of 170 bpm, for example, might be fine for a young healthy adult, but problematic for elderly people or recovery patients.Thus, physiological parameters should be interpreted in the context of the individual.A raw value representation would not be suitable in a training scenario.

Feedback from the Exerciser
Our male participant was 24 years old and had previous experience with electrical stimulation (Electrical Muscle Stimulation, EMS).He described himself as non-trained.A few years ago, he was regularly in the gym for being fit and looking athletic.In the 38-minute training session, the heart rate reached all zones except zone 5. His heart rate stayed 24.2 % of the time in zone 1, 28.0 % in zone 2, 32.4 % in zone 3 and 15.4 % in zone 4.He rated the exertion level of the training session on a 5-point scale (1 not straining, 5 very straining) with a 4. Thus, the training session felt physically demanding for our participant.
Apart from measuring the physical strain, we gathered qualitative feedback about wearing an interactive ear clip.On a 5-point Likert scale the participant answered both statements regarding privacy with a 1 (strongly disagree): I have concerns about sharing my heart rate zone in public via an ear clip and I have concerns about sharing my heart rate zones with an ear clip during training.Thus, he had no concerns showing his heart rate zone on an ear clip in a gym setting or in public space.On a 5-point Likert scale (1 strongly disagree to 5 strongly agree) he rated the electrotactile feedback with a 5 for each statement: I found the electrotactile feedback pleasant, I perceived the electrotactile feedback well, and I got used to the electrotactile feedback after a few tries.We asked the participant whether the feedback received in zone 4 corresponded to the experienced training intensity.The participant answered that at the exertion level, at which he received the electrotactile feedback, he had the impression of a "clear training effect." Regarding the electrotactile encoding, we asked the participant whether he would design the feedback differently.He proposed to play special feedback each time the heart rate zone changes, as well as to play special motivation patterns in the lower heart rate zones (1-3).Also he proposed other types of devices, such as a ring or wristband.

Post-Training Interview
After the training session we interviewed the trainer again.We first wanted to know, how well the colors of the LED ear clip matched his estimation of the exertion level of the exerciser.The trainer found that the visual encoding of the heart rate zones reflected the exertion level of the exerciser well and clearly indicated the heart rate zones.
We then asked if he could imagine using such a system as a support device during training.He agreed and mentioned the following use cases in which it would be helpful.Some exercisers simulate during training and with such a device he could see that.He found our ear clip especially useful for exercisers after illness, for re-entry, for people with cardiac muscle disorder or after an operation.The system can warn the exercising individual and inform the trainer, so that the exertion level is not too high.Further, some exercisers could benefit from such a device in special cases like circuit training, which is very intense.From his own experience he mentioned that sometimes people strain their body too much and collapse.Such a wearable device could help them to increase their pause time between two exercise machines or give the trainer an opportunity to intervene when he sees an ear clip flashing red.
Additionally, we wanted to know for which users such a prototype might be particularly valuable.He strongly stated that individuals who fall under the above mentioned criteria would benefit from the features of the ear clip during an individual workout.He also deemed the ear clip beneficial for training groups to get an overview of the exertion level of each individual in the group.Based on the individual exertion level he further sees advantages for competitive athletes (e.g., for the selection process in an assessment center), but not for amateurs performing popular sports as a hobby.
At the end, we asked him how he would improve the presented feedback to better match the needs of a trainer and an exerciser.The trainer argued that for the most cases 3 colors (green, orange, red) would be sufficient.To achieve a target training intensity, he considered the 5-color model as fine.He suggested to provide more details for a specific target group, like professional athletes.For the location of presenting the feedback, he mentioned that many exercisers wear headphones and a color display on the outside would be a good position (as proposed in [42]).He could also imagine presenting this information on an armband worn on the users upper arm.

DISCUSSION
In the following, we discuss the results of our study in terms of the suitability of the earlobe as a location for everyday-life wearables that provide notifications before we discuss the results of our field study.As mentioned above, we see unique advantages of the placement of a wearable on the earlobe in that it is suitable for both private and public notifications.This makes the earlobe different from most other placements of wearables.Even the smartwatch, although it has been suggested as a public display [30], is limited here, since its display is only visible to another person in certain postures and is typically not at eye level.A display on the earlobe, in contrast, is roughly at eye level of a nearby person and is thus compatible with being used during a conversation, either in a dialogue between two people or within a small group of people.
As public notifications on the earlobe are perceivable by and even directed to people around the wearer, new challenges arise.Not only the attention of the wearer has to be managed, but also the attention of those around the wearer, who can perceive the output.This turns output on the earlobe into a potential component of interpersonal communication.Design has to be careful with regard to social acceptability, opportune moments, and distractions.The initial online questionnaire showed that the participants preferred cautious and nonobtrusive public notifications.In Scenario 1, the participants favored a continuously fading light pattern instead of abrupt change or a bright, steady light.In Scenario 2, the participants rated a single note better than a melody.In the Display scenario, the participants preferred static text over RSVP and horizontal scrolling.The preferred variants of the public channels have in common that they are less obtrusive then their alternatives.

Private Notification Channels
The results of our study show the usefulness of the earlobe as a location for notification channels.We found no significant differences in the reaction times between the four notification channels with the fastest reaction times: Vibration, Electrotactile, Poke, and Private Sound.Only the thermal notification channels were significantly slower and less reliable due to relatively high error rates.Overall, Vibration and Electrotactile showed the best results in terms of recognition times, error rates, and qualitative feedback.They both received a high ranking (1 st and 2 nd place) and were judged as pleasant and non-exhausting.Moreover, the participants quickly got used to them and agreed that Vibration and Electrotactile could be useful for obtaining information in everyday life.
We are aware that the Vibration feedback may have been positively biased in the rankings, because the participants are likely already familiar with it as a common notification method on mobile devices.P13 stated: "Vibration in general is very well-known, because you are already used to it on the smartphone." P8 found it "very pleasant." In contrast P4 mentioned: "It made an audible noise that could possibly be disturbing." The Electrotactile feedback is less common than Vibration feedback.However, our participants mentioned several advantages of the Electrotactile channel, e.g.P4: "[It] is surprisingly pleasant and feels more like vibration feedback without the disadvantage of producing noise."P15 stated: "[It] combines the advantages of vibration and poke -very present feedback -with the advantage of low weight compared to the other two modalities.[It] feels much more comfortable than poke and vibration." The Private Sound feedback had fast reaction times and received positive qualitative ratings.Our participants were undecided if the feedback is useful for everyday usage.11 participants stated that the feedback is too quiet, e.g.P12: "Was a bit too quiet for me.On the street with loud traffic you could miss it, better usable in the library." We believe that the volume could be dynamically adjusted depending on the situation.In the study, we simulated a quiet office environment, in which the volume of the feedback was sufficient.Because the sound is played on the earlobe instead of the ear canal, the volume of this feedback is a tradeoff between a volume loud enough to be noticeable by the user and quiet enough not to be noticed by people nearby.To play more complex sounds or melodies, a loudspeaker would be possible on the earlobe.We decided to use a piezo element, as the results of our online survey showed that playing a single note instead of a melody is sufficient for notification purposes and a piezo element has a lower weight compared to a loudspeaker.
Both thermal channels have, compared to the other private notification channels, higher mean reaction times at 3.3 s for the Thermal Warm feedback and 4.3 s for the Thermal Cold feedback.4 % of the trials of Thermal Warm and 26 % of the trials of Thermal Cold were missed by our participants.P1, P2, and P4 mentioned that they were not able to detect the Thermal Cold feedback.P3, P6, P12, and P14 mentioned that they felt the Thermal Cold feedback, but that the effect was weak.Further, Thermal Warm was perceived as too warm by P1, P5, P8, P14, and P18.Overall, the thermal channels received mixed qualitative results from our participants.The results indicate, that the thermal feedback needs to be calibrated precisely and controlled in a closed loop.In comparison to NotiRing [37], our measured mean reaction times for the thermal feedback were faster, as we used Peltier elements instead of resistors to create the sensation.The error rate of 26 % for Thermal Cold is comparable to the error rate of Thermal Warm of NotiRing [37].However, the error rate for Thermal Warm in our experiment is much lower at 4 %.Comparing our results with "Baby It's Cold Outside" [14] shows that the error rates for Thermal Cold at 26 % in an office environment with a temperature of about 21 °C are close to the error rates of about 20 % in "Baby It's Cold Outside" [14].In contrast, our error rates for Thermal Warm are much lower.Producing cold feedback is even harder, because the additional temperature caused by the electric power dissipation has to be discharged.We measured the temperature change rates of -1.5 °C per second for cooling and 2.5 °C per second for heating.When comparing the reaction times with NotiRing [37], it is striking that their participants reacted quite fast to Vibration, Sound, and Poke, like we observed in our study, even if the absolute times are not directly comparable as they depend on the apparatus.The reaction times of 3-4 s of the thermal feedback of our study are comparable with the measured 3-4 s in "Baby It's Cold Outside" [14].

Public Notification Channels
First of all, the Public Sound and LED notification channels had error rates of 0 % with recognition times of 738 ms and 828 ms, respectively.The recognition times and error rates are comparable with the times of the "fast" group of private notification channels (Vibration, Electrotactile, Poke, and Private Sound).The mean time for Public Sound is 271 ms faster than the Private Sound condition, which could result from the higher volume of the feedback.Five participants mentioned that they did not like the tone of the single note (a 1kHz sine wave) that we played.Further, they argued that the sound level should be increased in loud surroundings, e.g. in a crowd of people, that should be considered when using it in arbitrary public spaces.
For the LED, P14 mentioned: "Different information can be conveyed by different colors.However, it is difficult to interpret the meaning of light in everyday contexts." Our participants also mentioned that they see problems in the interpretation of the feedback as the presented information is limited to colors and flashing patterns.While our participants of the preliminary online survey favored alternating on and off with fading as pattern for the given scenario there are various pattern possible [15].The participants were undecided for all public notification channels regarding the usefulness for everyday use.
The Display notification channel had significantly higher reaction times at an error rate of 22 %.This corresponds to the qualitative feedback of our participants.Overall, they ranked Display on the last position.Further, Display received mixed qualitative ratings (see Figure 6) compared to the other public notification channels, which were Proc.ACM Interact.Mob.Wearable Ubiquitous Technol., Vol. 7, No. 3, Article 123.Publication date: September 2023.rated positively overall.13 participants mentioned different problems with the Display prototype.P6 stated that the information on the display was "hardly visible".P11 said: the "font on the display is much too dark".P18 stated: the "feedback disappears too quickly".Although Display received lower ratings from our participants, it offers the opportunity to present more information than the LED channel.This is supported by the participants who agreed more that the display could show information that enriches a conversation compared to the LED.However, in the scenario of getting the attention of a person nearby, the display fails because of the poor visibility of the content.An initial full-screen flashing pattern on the display, before the actual content is shown, could draw the users' attention and thus reduce this problem.But there will be always a tradeoff between the amount of information that is presented and keeping the presentation recognizable in the near vicinity.Concepts suggested for headphone mounted displays [42] could extend the usage of earlobe displays beyond text.

Field Study: Mixing Public and Private Notification Channels
The results of the field study show that ear clips with the ability to simultaneously provide private and public notifications are promising.The use case of fitness training shows the benefits of the earlobe as a location for feedback.Although the color encoding of the LED feedback was visible for the trainer during the whole training session and is, according to him, suitable, the trainer suggested different encodings for slightly different use cases.The exerciser also suggested different encodings for the private notifications.The trainer needs less and more aggregated information, but adapted to the target group, and the exerciser is interested in more details through additional feedback, such as an indication of level changes or motivational feedback when the level is too low.This contradicts our assumptions that the trainer may want to get as many details as possible and the user may want to focus on the main task.Based on the fact that one trainer supervises dozens of exercisers at the same time, the suggested simplification is reasonable.In a non-supervised training session, the ear clip could give additional feedback to the exerciser to encourage staying in a specific zone for a given duration or to lower the intensity rate for a recovery interval.The feedback could even guide the exerciser through a mixed set of exertion levels to reach the desired training goal.The LED encoding in an unsupervised session could still display the heart rate zone.Trainers who notice someone training in the red zone could then suggest this person to reduce the training load.However, in the suggested color coding, red-green color blindness has to be taken into account.
The trainer and the exerciser both mentioned different locations and/or wearables for presenting private and public notifications.This might be because the mentioned locations and wearables are well accepted in contrast to our new approach.Further, our prototype is limited by the fact that the hardware to present the feedback is not yet completely integrated and miniaturized in the ear clip.The current setup requires cables to the custom PCB that cause a force pulling at the ear clip, which could interfere with the usage of some training machines.This problem was mentioned by both the trainer and the exerciser.As this was only a small field study the results may not be generalizable.

LIMITATIONS
As already mentioned, the results of this work are limited by the study setup.We tested all notification channels in a quiet office environment under controlled lab conditions and only the most promising ones, Electrotactile and LED, in a field study.Especially the results of the thermal channels may differ in the wild as the feedback depends on the ambient temperature in the environment.Warm feedback might perform better in cold environments and cold feedback might perform better in warm environments [14].The auditory notification channels also benefit from the controlled environment and would require an adaptive playback volume based on the current noise level in the environment.
A technical limitation of our implementation of the Poke notification channel is the weight and size of our prototype.Six of our participants complained about this, e.g.P1: "the ear clip was quite heavy".However, our participants reached fast reaction times with our implementation of the Poke feedback and rated it overall positive.Only four participants stated that the Poke feedback had an unpleasant feeling.A feedback similar to Poke as shown by Pece et al. [31] is more lightweight and has a smaller size, but comes with the flaw that it needs 24 V and 140 W for the actuation pulses.In comparison, the solenoid we use needs 3.3 V and 1.9 W. Further, even smaller and less power consuming solenoids, as used in PokeRing [17], consume considerably more power (1.65 W) than Vibration (0.21 W) or even Electrotactile feedback (0.045 W).This is factor of 7.9 compared to Vibration and even a factor of 36.7 compared to Electrotactile.Electrotactile only requires 21 % of the power of Vibration.
For a real deployment of wearables on the earlobe several technical problems have to be solved, among them the minimization of energy consumption, size, and weight.Moreover, the aesthetics of the devices have to be improved and deliberately designed.In this regard, much can probably be learned from the design of hearing aids [8], which have become very advanced technical devices [28] that focus on the private auditory channel and whose role is typically understood as fixing an impairment problem [7].We argue that an extension to other modalities offers many opportunities and opens a space for new forms of interpersonal communication that is enriched by providing information on the earlobe.In particular, if the wearable becomes "smart" and, e.g., can follow the conversation of two people or can recognize the current context, new applications become possible.

FUTURE WORK
Although we focused on evaluating and discussing private and public channels separately to explore the opportunities on the earlobe, we showed in our field study that mixing them is also possible.This opens up a wide design space of how notifications can be delivered and also to whom and in which manner.Apart from the presented fitness training scenario of the field study, other aspects, like social, temporal (synchronizing public with private feedback) and other scenarios could be evaluated.For example, during a conversation with colleagues Bob's ear clip starts flashing red at an increasing rate.Simultaneously, Bob himself feels a prickling on his earlobe, generated by electrotactile feedback, which increases slowly and gets stronger over time.He suddenly remembers that he has to pick up his children today from the kindergarten, because his wife has an interview with a potential employer.From the LED signal, Bob's colleagues infer that Bob has some important task to do.They end the conversation quickly to give Bob the chance to pick up his children.
Based on our findings from the lab study, on the miniaturization potential, and on the results of our field study, we recommend Vibration or Electrotactile feedback for the private part of mixed output.Poke feedback, based on solenoids, as well as Thermal feedback are difficult to miniaturize, as they require large electrical currents and thus large batteries.The audio channel shows good results, but as mentioned by the participants, it may fail in many in-the-wild situations.Private Sound has to be quiet enough not to be noticeable in the vicinity, but loud enough to attract the attention of the wearer.Public Sound should be loud enough to be noticeable in the nearer vicinity, but not so loud as to cause hearing damage.In addition, the auditory sense is typically engaged with many sound sources in an everyday environment.For the public channel we therefore only recommend the LED or Display channels.Display is able to present more information than LED, but a LED can be built extremely small.Moreover, the energy consumption of a LED can be minimized.The Display channel typically consumes more energy, but does not need to be active all the time.

CONCLUSION
Our work shows that the earlobe is a well suited location for presenting notifications through various kinds of output channels.The low mean reaction time (< 1 s) and low error rate of under 1 % for the "fast" group of private notification channels (Vibration, Electrotactile, Poke, and Private Sound) make them all viable for presenting notifications on the earlobe.Each notification channel offers special advantages.The Electrotactile and Vibration channels are small and very private.The Electrotactile feedback offers completely silent tactile feedback and is very energy efficient in contrast to Vibration and Poke feedback.However, the Poke prototype was able to create a very strong tactile sensation.The Sound feedback could be used for either private or public feedback, but needs volume adjustment based on the ambient noise level.The Thermal feedback produced poorer quantitative and qualitative results.Nevertheless, a specific calibration of the Thermal channels would presumably improve these results.The Thermal Warm feedback condition produced slightly better results than Thermal Cold, but Thermal Cold was rated as more pleasant.
The evaluation of the public notification channels showed on the one hand that a bright light LED can gather enough attention to be noticed by others.On the other hand, it creates the problem of interpreting the colors as it is not clear a priori what they indicate.For that purpose a display could be used in a two-staged interaction as an alternative.First the display would change the color of a larger number of pixels to draw attention.After that the display would convey detailed information to express what the notification means.However, we showed in our field study that in a scenario like training in a gym, the LED provides suitable feedback to support the exerciser and the trainer during the training.
Follow-up research should investigate the earlobe further for presenting notifications in the wild as well as mixed presentation of public and private notifications.We recommend building a multi-channel notification ear clip like we used in our field study.For such a multi-channel ear clip, we suggest the usage of either Vibration or Electrotactile for private notifications and LED as well as Display for presenting public notifications.A specific combination of these notification channels should be suitable in many situations in which either highly detailed information is required or coarse guidance is sufficient.
A promising direction is also to embed wearables on the earlobe into social situations, such as conversations between people to enrich the conversation, provide additional information that is relevant to the conversation, or to provide information about the context.Wearables worn on the earlobe have the distinct advantage that they are available at eye level during a conversation, so that eye contact can be maintained most of the time and the wearables can be used as micro displays.In this way such micro displays can serve as output possibilities in future augmented reality scenarios.

Fig. 2 .
Fig. 2. The custom board to control the different ear clip prototypes with connectors to each of the prototypes.An Adafruit Feather M0 Bluefruit is stacked on top.

Fig. 4 .
Fig. 4. Presentation of private notifications (left) and public notifications (right).The colleague "Bob" (right) was placed in front of the participant (1.5 m distance between the participant and "Bob").

Fig. 5 .
Fig. 5. Participants rated on a 5-point Likert scale if they can imagine to use one of the private (top) and public (bottom) notification channels for different notification types, e.g. an incoming call or getting attention from others.The notification channels are ordered by mean reaction time.

Fig. 6 .
Fig.6.User ratings regarding the sensation of the notification channels.The notification channels are ordered by mean reaction time.

Fig. 7 .
Fig. 7.The ear clip was attached to the left earlobe of our participant (left).During the training session, the trainer monitored the exerciser and his heart rate via the ear clip (middle; showing the view of the trainer).The participant performed several exercises as instructed by the trainer, such as indoor cycling (middle) and leg curl (right).

Table 1 .
Electronic parts used to build the notification ear clips.

Table 2 .
Quantitative results for the notification channels (private channels above and public channels below the divider), ordered by mean reaction time.

Table 3 .
[13,34]field study, different heart rate zones[13,34]are distinguished.The public notification channel is encoded via a color LED and the private notification channel via Electrotactile feedback.