DuoTouch: Passive Binary Touch Control

• DuoTouch is a passive attachment that employs temporal binary encoding via two fixed conductive footprints to translate mechanical motion into digital commands.
• It leverages two decoding configurations—aligned for discrete commands and phase-shifted for continuous control—to minimize screen occlusion and wiring complexity.
• Experimental evaluations on devices like the iPhone 16 and Magic Trackpad 2 highlight its performance, design trade-offs, and efficacy across varied speeds and electrode geometries.

Searching arXiv for DuoTouch and closely related touch-interaction work. DuoTouch is a passive attachment for capacitive touch panels that adds tangible input while minimizing content occlusion and loss of input area. It operates on unmodified devices through standard touch APIs, uses only two fixed on-screen conductive footprints, and routes mechanical controls to those footprints through two conductive traces that encode motion as binary sequences over time. The system defines two paired configurations: an aligned configuration that maps fixed-length codes to discrete commands, and a phase-shifted configuration that estimates direction and distance from relative timing. Its reported implementations include a smartphone hand strap, a phone ring holder, and touchpad add-ons, with technical evaluation on an iPhone 16 and a Magic Trackpad 2 (Ikematsu et al., 20 Feb 2026).

1. Concept and design objectives

DuoTouch addresses a recurring problem in passive touch-panel accessories: many existing clip-on mechanisms require one on-screen footprint per input, increase visual occlusion, reduce usable touch area, and often depend on more complex wiring or non-public sensing interfaces. Its central design decision is to hold the number of on-screen contacts constant at two, regardless of how many commands or control states are encoded. The attachment therefore maintains constant low wiring complexity, described as $O(1)$ wiring, while shifting interaction complexity into temporal decoding of binary touch events (Ikematsu et al., 20 Feb 2026).

The approach is explicitly passive. It requires no power, batteries, or wireless pairing, and it relies on the already-running capacitive controller and the device’s normal touch APIs. The attachment changes only the conductive geometry presented to the panel: the device observes two touchpoints making and breaking contact over time, rather than a purpose-built electronic peripheral. This yields a tangible interaction channel that is mechanically robust, cheap to manufacture using PCBs and 3D printing, and deployable on existing phones, tablets, watches, or touchpads without OS, firmware, or controller modifications.

A key design goal is to move most of the interaction off the screen while keeping software access simple. DuoTouch therefore uses only OS-delivered touch events, including coordinates, identity, and timestamps. A plausible implication is that the system is intended to be compatible with application-layer deployment rather than privileged system integration, but the reported claim is specifically that it runs through standard touch APIs on unmodified devices (Ikematsu et al., 20 Feb 2026).

2. Physical architecture and binary sequence encoding

A DuoTouch mechanism consists of a main PCB and a user-actuated component. The main PCB carries two on-screen footprints, which are conductive pads that appear as two touch points, together with two conductive traces that extend from those footprints into the attachment body. These traces form the “sequence pattern”: alternating conductive segments and insulating gaps laid out along the path of a slider, rotary disk, button plunger, or related mechanism. The user-actuated component contains a small two-electrode PCB or conductive element that is electrically connected to the user’s body. When moved, it bridges segments of the traces and thereby grounds one or both footprints.

The touch controller does not sense the mechanical control directly. It senses only whether each footprint is currently grounded. At sampling frame $n$, the binary state is

$x_n = 1$

if the footprint is sensed as touched, and

$x_n = 0$

otherwise. Motion of the actuator across conductive and insulating regions therefore produces two binary time series. DuoTouch uses only on/off transitions and their timing; it does not use analog capacitance magnitude (Ikematsu et al., 20 Feb 2026).

Several geometric parameters govern the encoding. The footprints are square, about 8 mm on a side. Electrode segments in the evaluation patterns have width $w$, with tested values $1.5$, $2.0$, $2.5$, and $3.0$ mm. The gap between reference segments is set to $3w$, and the test-pattern height is 8 mm. The slider electrode width matches $n$0, so the geometric overlap per contact is about $n$1. These choices determine temporal dwell, robustness at higher speeds, and movement quantization for continuous control.

The binary-sequence model separates the two traces into a reference trace and an input trace. The reference trace is periodic and defines the timing grid. The input trace either stores a fixed pattern aligned to that grid, for discrete-command decoding, or adopts a spatial phase shift relative to the reference, for continuous direction-and-distance decoding. This division is the basis of DuoTouch’s two logical configurations (Ikematsu et al., 20 Feb 2026).

3. Decoder configurations

DuoTouch defines an aligned configuration for discrete commands and a phase-shifted configuration for continuous control.

Configuration	Encoding principle	Output
Aligned	Fixed pattern aligned bit-by-bit to a periodic reference trace	Discrete command
Phase-shifted	Identical periodic traces with spatial offset	Direction and distance

In the aligned configuration, the reference and input traces are spatially aligned bit by bit along the motion direction. The reference provides the bit clock, and the input trace holds a code pattern. Moving in the opposite direction produces the time-reversed sequence, so direction-sensitive operation must exclude palindromic patterns and treat $n$2 and $n$3 as a reversible pair. After segmenting the timeline into reference-bit intervals $n$4, the decoder counts input samples in interval $n$5: $n$6 With $n$7, the decoded bit is

$n$8

Ties are resolved by duration-weighted overlap, using frame durations $n$9; persistent equality yields an ambiguous bit. The resulting pattern is matched against a predefined codebook, with ambiguous bits treated as wildcards, and a command is issued only if exactly one code matches. This mode is intended for one-shot actions, toggle switches, and directional navigation (Ikematsu et al., 20 Feb 2026).

In the phase-shifted configuration, both traces implement alternating bits such as $x_n = 1$0, but the input trace is shifted by $x_n = 1$1 relative to the reference. The method is explicitly analogous to quadrature encoders. Transitions are detected as $x_n = 1$2 or $x_n = 1$3 events between consecutive frames, and each transition is timestamped at the start time of the first frame after the change. For reference interval $x_n = 1$4, relative timing is measured by

$x_n = 1$5

and

$x_n = 1$6

A negative sign indicates that the input leads the reference, a positive sign that it lags, and zero that both events fall in the same sampling frame. If both offsets exist and have the same nonzero sign, that sign determines direction; weaker rules also allow one signed offset with the other neutral or undefined. Under interleaved geometry, spatial resolution is

$x_n = 1$7

where $x_n = 1$8 is the spatial pitch between successive reference transitions. With $x_n = 1$9 input transitions attributed to interval $x_n = 0$0 and direction sign $x_n = 0$1, the signed displacement estimate is

$x_n = 0$2

This mode is intended for sliders, dials, zoom controls, and scrolling (Ikematsu et al., 20 Feb 2026).

The two decoders trade robustness against information type. The aligned configuration is described as generally more robust because it relies on majority decisions within bit intervals. The phase-shifted configuration is less robust near the speed-sampling limit because it requires consistent timing-sign decisions, but it yields continuous distance estimates rather than discrete codes.

4. Sampling-limited analysis, implementation, and measured performance

DuoTouch formalizes reliability through a sampling-limited bound linking actuation speed $x_n = 0$3, electrode width $x_n = 0$4, and touch sampling rate $x_n = 0$5. With sampling period $x_n = 0$6, the dwell per bit is approximately

$x_n = 0$7

and the expected number of frames per bit is

$x_n = 0$8

The normalized speed is defined as

$x_n = 0$9

so that

$w$0

Decoding becomes unreliable when $w$1 drops below about one frame per bit, which yields the feasibility bound

$w$2

For phase-shifted decoding, the same inequality also emerges from the requirement that $w$3 in the worst-case sampling phase (Ikematsu et al., 20 Feb 2026).

The paper further defines an empirical 90% accuracy threshold $w$4 and aggregates it for $w$5 mm across $w$6 Hz. The median $w$7 yields design constants $w$8 for aligned decoding and $w$9 for phase-shifted decoding, leading to the design rule

$1.5$0

This converts the analysis into a direct geometric design criterion for a known device sampling rate and a target speed envelope.

Implementation uses an iPhone 16 with standard iOS UIKit touch events at 60 Hz and a Magic Trackpad 2 at 90 Hz. The software pipeline polls touch events, determines whether each footprint is touched, timestamps each frame, detects transitions, decodes according to the active configuration, and maps the result to either a command or a continuous parameter. The computations are described as light, consisting of integer and logical operations and table lookups at no more than 90 Hz (Ikematsu et al., 20 Feb 2026).

The technical evaluation uses a 3D printer, specifically a Bambu Lab A1 mini, as a motion controller. The printer mechanically moves the slider knob at constant speeds; the slider is connected to ground via copper tape. The test matrix spans 2 configurations, 2 devices, 4 widths, 10 speeds from 20 to 200 mm/s in 20 mm/s increments, and 10 repetitions, for a total of 1600 trials. In the aligned configuration, a trial is counted as correct only if every bit in the 6-bit pattern $1.5$1 is decoded correctly. In the phase-shifted configuration, each trial contains 3 cycles and each cycle is scored separately according to correct direction decision (Ikematsu et al., 20 Feb 2026).

Within the sampling-limited bound $1.5$2, the aligned configuration on the smartphone reports mean accuracy across speeds of 75.0% for $1.5$3 mm, 85.0% for $1.5$4 mm, 90.0% for $1.5$5 mm, and 93.3% for $1.5$6 mm. On the touchpad, the corresponding aligned accuracies are 92.9%, 86.7%, 94.0%, and 98.0%. For the phase-shifted configuration, smartphone mean per-cycle accuracy within the bound is 78.3%, 90.3%, 96.3%, and 94.4% for the same widths, while the touchpad reports 57.1%, 77.9%, 87.6%, and 97.7%. The reported failure modes are too few frames per bit, insufficient capacitive change per sample due to high speed and narrow width, and incomplete sequences when actuation stops mid-pattern (Ikematsu et al., 20 Feb 2026).

5. Form factors and application space

DuoTouch is presented as a general mechanism rather than a single device. Because the traces can be laid out in straight, circular, or other paths, the same architecture supports sliders, rotary dials, push buttons, and cross-keys. The paper demonstrates this versatility through several working prototypes (Ikematsu et al., 20 Feb 2026).

The smartphone hand strap prototype places a sliding clasp on the back of a phone case. The upper part of its path uses an aligned pattern for discrete mode toggling: pulling in one direction yields a code such as $1.5$7 to enter one-handed mode, while returning generates the reverse code to exit. A lower phase-shifted segment supports continuous camera zoom adjustment. Both decoders are active concurrently. For this prototype, using $1.5$8 mm on the phone, 96.6% of sequences had each input transition aligned to its intended reference transition within the same sampling frame, enabling early classification for some codes.

A phone ring holder prototype embeds a circular phase-shifted pattern in the ring base. Rotating the ring moves a conductive element along the circular path, producing quadrature-like sequences that drive volume, zoom, or scrolling. The reported benefit is natural one-handed use together with eyes-free operation and elimination of an on-screen slider. A touchpad add-on places a slim macropad along the edge of a Magic Trackpad 2. It contains eight physical buttons, all routed to the same two footprints through distinct aligned patterns. In the demonstration, these are mapped to 3D CAD shortcuts such as copy, paste, redo, undo, and align, while leaving most of the touchpad area available for normal pointing and gestures (Ikematsu et al., 20 Feb 2026).

Additional concepts implemented from participant ideas include a circular menu dial on a tablet, a surface length measurer that counts transitions while being rolled along a physical surface, and a camera-shaped controller with a front dial for zoom and a top button for shutter. These examples show that DuoTouch’s binary-sequence mechanism is not tied to a particular host device shape. A plausible implication is that the two-footprint abstraction is intended to function as a reusable touch-side interface primitive across multiple tangible form factors, but the reported contribution is specifically the demonstration of those form factors and the supporting analytical framework (Ikematsu et al., 20 Feb 2026).

6. Limitations, design trade-offs, and research context

DuoTouch’s main limitations are presented as consequences of its passive, sampling-based design. The bound $1.5$9 is a hard limit imposed by the host device’s touch controller. The mechanism depends on grounding through the user’s body, so detection may fail with very dry fingers or gloves unless glove mode or chassis grounding is used. The decoder assumes only one mechanism is actuated at a time; simultaneous use of multiple devices sharing the same footprints produces overlapping sequences that cannot be disentangled. Recognition depends on transitions rather than static contact, so some minimal motion is always required. Fabrication requires PCB manufacturing and 3D-printed parts, and participants raised concerns about bulk and mass in the strap and dial prototypes (Ikematsu et al., 20 Feb 2026).

The design trade-offs are equally explicit. Longer codes or narrower segments increase information density or continuous-control resolution, but they require more motion and move the mechanism closer to the sampling limit. Combining aligned and phase-shifted segments in one control, as in the strap, increases functional richness but can make usage less straightforward. Remaining fully passive preserves cost and maintenance advantages but sacrifices on-the-fly reconfigurability; the geometry is fixed. The paper identifies several future directions: broader material and wear studies, co-fabrication tools for integrated mechanical and conductive design, improved decoders using prefix-free codebooks and timeouts, task-level evaluations against direct touch or other accessories, evaluation on higher-$2.0$0 panels, and accessibility-focused designs (Ikematsu et al., 20 Feb 2026).

Within the broader arXiv touch-interaction literature, DuoTouch occupies a distinct position. It differs from Lamb-wave tactile sensing systems that localize one or two fingers on a thin copper plate through acoustic absorption and nearest-neighbor matching (Liu et al., 2009). It also differs from optical multi-touch pipelines that detect fingertips and cluster them into hands using Maximally Stable Extremal Regions on rear-projected tabletops (Ewerling, 2013). In robotics, “Bi-Touch” denotes a dual-arm tactile manipulation platform based on TacTip sensors and deep reinforcement learning rather than a capacitive-screen attachment (Lin et al., 2023). Likewise, “BiFingerPose” addresses bimodal finger-pose estimation using a capacitive image and a fingerprint patch from an under-screen fingerprint sensor (Guan et al., 21 Nov 2025). Against these adjacent lines of work, DuoTouch’s specific novelty is the fixed two-footprint passive architecture, the use of binary temporal encoding under public touch APIs, and the analytical treatment of decoding reliability through the sampling-limited bound (Ikematsu et al., 20 Feb 2026).