Jae Hoon Kim
Projects

PalmEcho · drawing set · v1.0 ▸ NEXT EXPERIMENT

Part I · Cover

U.S. Provisional Patent Application · Drawing set
Appl. No. —
Sheets 4
Rev. B · v1.0
Title · System and method for mapping taps on a host device's own chassis to input events using built-in acoustic transducers and impulse-response classification
Inventor · J. H. Kim
Internal · PalmEcho
Filing target · pending W1 gate
Classification (proposed) · G06F 3/043 · G06F 3/16
About this Personal project · v1.0 skeleton Gate-pending
Draft cover sheet and drawings for PalmEcho. The W1 experiment lives at the companion W1 experiment; this page upgrades from v1.0 skeleton to full provisional draft once the kNN(3) leave-one-out gate at 70 % accuracy clears on a four-class chassis-tap task. The format is a patent draft because it forces the claims, alternatives, and prior art out into the open before the experiment runs. v1.0 honest acknowledgment (2026-05-18): the broad concept of "tap on the device's own chassis → input event" is anticipated by Apple US 9,086,738 (Tsudik, 2013, "Fine-tuning an operation based on tapping") and by Apple's commercial Back Tap feature (iOS 14, 2020) — both detect taps on a device's non-touchscreen surfaces and map them to actions, using accelerometer + gyroscope rather than acoustic ranging. The closest active-acoustic-sensing academic prior art is Touch & Activate (Ono et al., UIST 2013) which uses an attached vibration speaker and piezo microphone to detect taps on existing objects. The surviving novelty is the combination: built-in commodity microphones and loudspeakers (no instrumentation), inaudible swept-sine chirp + matched-filter deconvolution to recover the chassis impulse response, and a perturbation-pattern classifier that distinguishes among palmrest / lid / bezel regions — which neither the accelerometer-based Apple art nor the attached-transducer Touch & Activate art teaches.
Abstract of the disclosure Cover-sheet boilerplate v1.0
A system and method detect finger taps on a host computing device's own chassis — palmrest, lid, or bezel — using the device's built-in loudspeaker(s) and microphone(s) as acoustic ranging transducers, without external sensors or surface instrumentation. The device emits a brief inaudible swept-sine chirp in the 18–22 kHz band and recovers the room-and-chassis impulse response by matched-filter deconvolution of the received microphone signal. A finger tap on a chassis region perturbs the impulse response over a short post-onset window; the perturbation pattern is mapped to one of a small set of chassis regions by a lightweight classifier and converted to an input event. No audio content is retained; only impulse-response features are extracted, and feature extraction is performed on-device.
Field
G06F 3/043 · acoustic input
Acknowledged prior art (tap-on-chassis concept)
  • Apple US 9,086,738 (Tsudik 2013) — chassis taps via accelerometer/gyroscope
  • Apple Back Tap (iOS 14, 2020) — commercial, accelerometer-based
  • Touch & Activate (Ono UIST '13) — active-acoustic, attached transducer
Distinguished from (surviving novelty)
  • Apple Back Tap uses accelerometer · this uses acoustic impulse-response
  • Touch & Activate attaches transducer · this uses built-in commodity I/O
  • EclipseTouch (UIST '25) — external surface, worn IR
  • UbiTap / MM-Tap — external table
  • Apple Force Touch — trackpad pressure only
Index of sheets Tap a row to jump 4 sheets

Part II · Drawings

Sheet 1 / 4 Representative FIG. 1 · System overview · chassis ROIs + ToF arcs 100
100 · chassis 130 · hinge 116 · keyboard · out-of-scope 102 104 110 · mic mic · 112 114 · mic 118 · trackpad R1 · left palmrest R2 · right palmrest R3 · lid / bezel 122 · Δt₁ 124 · Δt₂ 126 · t = 0 120 · approach 140 · ≈ 320 mm · 14-inch chassis 150 · LEGEND speaker mic ROI direct reflected out-of-scope
FIG. 1
Sheet 2 / 4 FIG. 2 · Acoustic ToF pipeline · chirp → IR → features 200
202 · chirp · swept-sine frequency (kHz) 16 18 22 24 18 kHz · audible time (s) 0 0.5 1.0 204 · impulse response |h(τ)| |h(τ)| delay τ (ms) 0 15 30 210 · peak-pick window · first 30 ms 206 · Δt₁ direct 208 · Δt₂ reflected 212 → 214 → 216 · classify τ₁ τ₂ Δf ρ E₁ E₂ E₃ c 212 · 8-dim kNN k = 3 214 · classify R1 R2 R3 ∅ 216 · p(class) 218 · TIMING BUDGET · per chirp on Apple Silicon emit chirp ≈ 1000 ms capture + deconvolve ≈ 2 ms feature extract ≈ 1 ms kNN classify ≈ 1 ms dispatch ≤ 1 ms
FIG. 2
Sheet 3 / 4 FIG. 3 · Functional block diagram · transducers + classifier 300
322 · EMIT PATH · 48 kHz PCM → 18–22 kHz speakers chirp generator 302 · sw CoreAudio out 304 · DAC speakers · L, R 306 · hw PCM 48k analog 326 · SENSE PATH · 96 kHz mic capture → deconvolve → classify → dispatch mic array 308 · hw CoreAudio in 310 · ADC deconvolve + IR window 312 · sw features 314 · 8-dim kNN classify 316 · k = 3 event dispatch 318 · sw x[n] · 96k buffer h(τ) 8-dim p(class) 324 · in-air · chassis-borne acoustic ref · matched filter OS / shortcuts 320 shortcut id 328 · CLOCK DOMAINS · 330 · LEGEND EMIT · 48 kHz PCM DAC → speaker driver SENSE · 96 kHz capture ADC → ring buffer BAND · 18–22 kHz inaudible operating band hardware software data flow control
FIG. 3
Sheet 4 / 4 FIG. 4 · Tap-event state machine 400
400 · TAP EVENT STATE MACHINE FIG. 4 412 · start IDLE 402 idle · pre-chirp LISTEN ★ 404 · primary active capture CLASSIFY 406 feature + kNN EMIT EVENT 408 shortcut dispatch chirp emit / start capture 414 · |peak| below τ_lo / skip |peak| ≥ τ_hi / window IR p(class) ≥ τ_p / map shortcut dispatched / reset 416 · p(class) below τ_p / drop chirp playing 410 · LEGEND · thresholds and arrow conventions τ_hi high IR-peak entry τ_lo low exit (hysteresis) τ_p classifier confidence ★ primary state nominal flow skip / drop initial state
FIG. 4

Part III · Specification

Background of invention Prior-art context

Prior systems for acoustic touch input on commodity hardware include VibSense (SECON 2017), UbiTap and S-UbiTap (SenSys 2018; TMC 2023), MM-Tap (TMC 2023), and most recently EclipseTouch (UIST 2025). All such systems treat an external surface — a tabletop, wall, or ad-hoc plane — as the input region, and require the user to set down or position the host device adjacent to that surface.

The disclosed system inverts the assumption: the host computing device's own chassis surfaces (palmrest, lid, bezel) serve as the input region, using the device's built-in microphones and loudspeakers as acoustic ranging transducers. No external surface is needed; no setup ritual is performed; no environmental priors are required. The novelty lies in where the tap is detected (on the device that hosts the input pipeline), not in the acoustic-ToF physics, which builds on the cited prior art.

Capacitive trackpad input (e.g. Apple Force Touch, US 10,162,447 B2) covers a restricted spatial region of the chassis (the trackpad surface) and depends on a dedicated force sensor. The present disclosure expands the input region to substantially the entire upper chassis without additional hardware.

The broader concept of mapping a tap on a device's non-touchscreen chassis surfaces to an input event is anticipated by Apple US 9,086,738 (Tsudik, 2013, "Fine-tuning an operation based on tapping"), which discloses detection of taps on a device's sides and other non-touchscreen portions via the device's accelerometer and gyroscope, mapped to granular on-screen controls. Apple's Back Tap feature (iOS 14, 2020) is the commercial embodiment, detecting double- and triple-tap gestures on the back glass of an iPhone via accelerometer pattern matching. Both are acknowledged prior art on the chassis-as-input-region concept; PalmEcho is distinguished by sensor modality (active acoustic ranging via built-in speakers and microphones, recovering an impulse-response perturbation) rather than passive accelerometer pattern matching. Touch & Activate (Ono et al., UIST 2013) is the closest active-acoustic prior art and is similarly distinguished: it requires an attached vibration speaker and piezoelectric microphone affixed to the target object, whereas the disclosed system reuses the host device's built-in loudspeakers and microphones without any added transducer.

Summary of the invention per 37 CFR § 1.73

The disclosed system converts the host computing device's own chassis surfaces into an input region, using only the loudspeakers and microphones already integral to the device. No external sensor, no piezoelectric or accelerometer surface instrumentation, and no surface attachment is required.

A host computing device comprising at least one loudspeaker (102), at least one microphone, and one or more processors, is programmed to emit a periodic inaudible swept-sine chirp (202) in the 18–22 kHz band, receive the resulting acoustic signal at the microphone, recover an impulse-response estimate by matched-filter deconvolution, and classify a region (R1, R2, R3, none) of the device chassis at which a finger tap (106) has occurred within the impulse-response window.

Region classification is performed by a small classifier (208) operating on an eight-dimensional feature vector (206) extracted from the impulse-response window (204). The classifier outputs are converted to host-device input events (e.g. application-launch, focus-toggle) by an event dispatch module (312). All processing is on-device; raw audio samples are not retained after feature extraction.

Brief description of drawings Sheets 1 – 4

Part IV · Claims

Claims 1 independent · 3 dependent · 1 apparatus Draft v1.0
What is claimed is:

1. A method for detecting a tap event on a chassis surface of a host computing device, comprising:

  1. (a)emitting, from at least one loudspeaker (102, 104) integral to said host device, an inaudible swept-sine acoustic signal (202) in a frequency band above 18 kHz;
  2. (b)receiving, at at least one microphone integral to said host device, an acoustic response to said emitted signal;
  3. (c)recovering an impulse-response estimate by matched-filter deconvolution of said received signal against said emitted signal;
  4. (d)extracting a feature vector (206) from a temporal window (204) of said impulse-response estimate; and
  5. (e)classifying said feature vector to one of a plurality of regions of said chassis (R1, R2, R3) or to a no-tap state, and emitting a host-device input event (410) as a function of said classification;
  6. (f)wherein the loudspeaker(s) of (a) and the microphone(s) of (b) are integral to said host device and serve as the sole acoustic transducers for said tap detection, with no external sensor, no piezoelectric or accelerometer surface instrumentation, and no surface attachment recruited for said detection; and
  7. (g)wherein raw audio samples received at said microphone are not retained on-host beyond feature extraction, and no raw audio, no impulse-response sample stream, and no per-user identifier is transmitted off-host.

2. The method of claim 1, wherein said feature vector (206) comprises lag and amplitude of the two strongest impulse-response peaks within said temporal window, energy in three temporal sub-bands, and a temporal centroid.

3. The method of claim 1, wherein said classification is performed by a k-nearest-neighbor classifier (208, 310) trained on a small per-user calibration set covering each of said plurality of regions and a no-tap baseline.

4. The method of claim 1, wherein said host-device input event (410) is mapped via a configurable map to one of an application-launch command, a window-focus command, a media-control command, and an accessibility command.

5. A host computing device, comprising:

  1. (a)at least one loudspeaker (102, 104) and at least one microphone integral to said device;
  2. (b)one or more processors implementing a chirp generator (308), deconvolution and feature extractor (306), classifier (310), and event dispatch (312); and
  3. (c)a non-transitory memory storing instructions which, when executed by said processors, cause said device to perform the method of any of claims 1 – 4.
Claims · 5 total · 1 independent · 3 dependent · 1 apparatus

Part V · Appendices

Prior-art bibliography Selected; not exhaustive

Part VI · Execution

Version history Draft · not filed

The applicant retains this draft in personal records. No filing has been made; no priority claim is asserted. Promotion of this page to a full provisional-application draft (in the form of /echocast or /roommirror) is conditioned on the W1 gate reaching ≥ 70 % kNN(3) leave-one-out accuracy on a four-class chassis-tap task.

/palmecho · v1.0 · drawing-stage
Index