Skip to content
Jae Hoon Kim
Projects

RoomMirror · drawing set · v0.10

Part I · Cover

U.S. Provisional Patent Application · Drawing set
Appl. No.
Sheets 13
Rev. B · v1.0
Title · System and method for adapting a user interface using passively-sensed wireless channel state information
Inventor · J. H. Kim
Internal · RoomMirror
Filing target · 2026-08 · pre-public-disclosure
Classification (proposed) · G06F 3 / 01 · H04W 4 / 80
About this Personal project · ongoing v1.0
A personal project I've been turning over for a while. Drafted in the shape of a provisional patent application because the format is a useful thinking tool — it forces the claims, alternatives, prior art, and threat model out into the open where I can see the gaps. Treat it as a working draft of an idea I keep iterating on. v1.0 honest-acknowledgment (2026-05-18): the sensing side of this work sits downstream of MIT Dina Katabi's group (RF-Sleep 2017, Vital-Radio 2014, RF-Pose 2018) and its commercial spinout Emerald — Wi-Fi-derived physiology sensing is a well-explored space. The UI-feedback side overlaps with Apple's 2024 stress-detection patent application (Mulliken et al.; uses EEG / HRV / gaze, not CSI). Both are now explicitly logged in the IDS rather than implied; the surviving claim is the specific combination of CSI sensing with on-device closed-loop UI modulation and implicit-feedback personalization.
Abstract of the disclosure Cover-sheet boilerplate v0
A system and method adapt a host computing device's user interface in response to a user's physiological state inferred from channel state information (CSI) passively obtained at the device's wireless networking interface. A conditioning unit, feature extractor, and inference network resident wholly on the host produce a continuous focus metric φ̂(t) and posture-transition events from CSI sampled at approximately 100 Hz. A UI modulation engine adjusts notification visibility, content pacing, foreground-application focus, and ambient visual feedback as a function of said metric. The host retains feature-vector representations only; raw CSI is discarded within a sliding window W and is never transmitted to a remote service.
Field
G06F 3/01 · H04W 4/80
Cross-references
IEEE 802.11bf-2025
— relies upon
US 2023/0367452 A1
— distinguished from
Index of sheets Tap a row to jump to that figure 13 sheets
Drawing convention Symbol vocabulary used across all figures
Line styles
data flow · primary signal path
control · wire · structural outline
feedback path or implicit gradient
boundary · enclosing scope
hidden / projected geometry
Fill patterns
out-of-scope · absent · discarded
active · emphasized · blocking event
hysteresis band · uncertainty region
Shape vocabulary
parallelogram · I/O event (e.g., S402 receive)
rectangle · deterministic compute (filter, FFT)
hexagon · feature processing block
rounded rect · learned / AI module
diamond · decision / gate
cylinder · storage · sink · discard
circle · state / entity / occupant
double-ring circle · emphasized / current state
Convention applies to FIG. 1 – FIG. 12. · Reference numerals follow 100-series (FIG. 1), 200-series (FIG. 2), etc. per USPTO drawing practice.
Drawing compliance certificate 37 CFR § 1.84 Self-certification

The applicant certifies that the drawings filed herewith satisfy the requirements of 37 CFR § 1.84:

  • § 1.84(a)(1) — black-and-white line art, prepared at sufficient density and uniformity of stroke for direct reproduction.
  • § 1.84(d) — every sheet is identified by its serial position ("Sheet 1 / 13" … "Sheet 13 / 13") in the upper bezel.
  • § 1.84(m) — graphic forms used for illustration (cross-hatching for absence/discard, dashed lines for boundaries) follow established drafting convention; convention is documented in this drawing set.
  • § 1.84(p)(3) — numerals are at least 0.32 cm tall on the printed page and are placed so as not to cross or mingle with other lines.
  • § 1.84(p)(5) — the same reference numeral consistently designates the same part across all drawings; no numeral is reused for a different part.
  • § 1.84(u) — Sheet 1 (FIG. 1) is designated as the Representative Figure for use on the application's cover page.
Self-certification signed by inventor at execution. · Any non-compliance is subject to correction under § 1.85 prior to publication.

Part II · Drawings

Sheet 1 / 13 Representative FIG. 1 · System overview 100
≈ 3.0 m 122 124 126 120 · multipath DETAIL A — SEE FIG. 5 102 104 106 108 110 112 118 114 · room boundary 116 · desk
FIG. 1
Sheet 2 / 13 FIG. 2A · Functional block diagram (top level) 200
200 LOCAL HOST · 110 — NO RAW CSI EGRESS CSI ACQ COND. FEAT. INFER. UI MOD. 202 204 206 208 210 212 802.11 · CSI Hampel · BP FFT · tokens BiLSTM policy · gate x[n] x'[n] S(f,t) φ̂ u(t) 214 · implicit feedback (encoder features only) KEY ─ DATA FEEDBACK BOUNDARY
FIG. 2A
Sheet 3 / 13 FIG. 2B · Inference network 208 (exploded) 208
208 INFERENCE NETWORK — DETAIL EMBED 208a LSTM → 208b ← LSTM 208c 208d CONCAT MLP · HEAD 208e φ̂ · q10/50/90 208f S(f,t) h⃗ h⃖ 220 · Pre-training: CSI-MAE (masked-autoencoder) on unlabeled CSI streams → fine-tune φ̂ head on labeled focus episodes
FIG. 2B
Sheet 4 / 13 FIG. 3 · Signal extraction 300
θ_hi · 0.75 θ_lo · 0.50 324 · t = now 316 · W = 4 s 322 · |X(f)| f_b ≈ 0.25 Hz 00.511.5 Hz A A' r(t) φ̂(t) +30 −30 +30 −30 +1 −1 1.0 0.0 a.u. a.u. mm 302 304 306 320 310 312 102030405060 t (s) FILT RESP FOCUS f_s = 100 Hz · 64 subcarriers · decimated 3 ⨯ for display
FIG. 3
Sheet 5 / 13 FIG. 4 · Method 400 400
STEP FLOW BLOCK DATA STATE CUM. LATENCY PERFORMED ON LOCAL HOST · 110 START · S400 S402 I/O Receive CSI from wireless interface 102 RAW CSI · 100 Hz 0 ms S404 FILTER Hampel · band-pass condition conditioning unit 204 x'[n] + 8 ms S406 EXTRACT FFT → respiration r(t) sliding window W ≈ 4 s · feature extractor 206 r(t) · S(f,t) + 24 ms S408 INFER Compute focus metric φ̂(t) inference network 208 · quantile head φ̂ · q10 / q50 / q90 + 32 ms S410 GATE φ̂ ≥ θ_hi ? hysteresis threshold (claim 4) θ_hi θ_lo φ̂(t) vs (θ_lo, θ_hi) Y N + 36 ms S412 ACT Modulate host UI 212 focus mode · notif suppress · ambient shift u(t) · UI MOD ENGINE 210 + 52 ms S414 SINK Retain features · discard raw CSI FEATURES 712 ✓ RAW CSI 710 ✗ privacy boundary · FIG. 7 + 55 ms END · LOOP re-poll → S406 DATA VOLUME raw filt feat φ̂ SHAPES I/O filter proc AI gate sink
FIG. 4
Sheet 6 / 13 FIG. 5 · Hardware embodiment (exploded) 500
ESP32 S3 WROOM-1 40MHz LDO ANT 3V3 GND G0 G2 G48 EN 5V USB FOCUS · 212 DETAIL B — SCALE 4:1 L ≈ 28 mm 502 510 504a 504 506 512 SENSING NODE 502 TRANSPORT 504 HOST · LOCAL ONLY 506
FIG. 5
Sheet 7 / 13 FIG. 6 · UI state-transition diagram 600
STATE GRAPH UI ENVELOPE (212) 602 IDLE pass-through 604 ALERT badge-only 606 FOCUS suppress 608 DEEP FOCUS hard-block 610 BREAK catch-up φ̂ ≥ θ_lo φ̂ < θ_lo φ̂ ≥ θ_hi φ̂ < θ_hi sustained T₁ drop posture: stand T₂ elapsed timeout → IDLE 602 · IDLE 3 604 · ALERT 606 · FOCUS 608 · DEEP FOCUS 5 new 610 · BREAK TEMPORAL ENVELOPE · REPRESENTATIVE 6-HOUR INTERVAL 602 604 606 608 610 09:00 10:00 11:00 12:00 13:00 14:00 15:00 φ̂ = inferred focus · θ_lo / θ_hi = hysteresis bounds · T₁ = dwell · T₂ = break minimum emphasized ring = current state
FIG. 6
Sheet 8 / 13 FIG. 7 · Privacy boundary / data-flow 700
LOCAL · 706 REMOTE · 708 (FORBIDDEN) 702 · host device 110 704 · cloud / off-host services RAW CSI 710 FEATURES 712 UI STATE 714 discard ≤ W retained · in-memory retained · in-memory 716 · NO EGRESS 718 · POST /v1/telemetry — NEVER SENT { "csi_window": [<3 KB float32 array>], "sub_carriers": 64, "user_id": "user_3f9a…", "room_geo": [37.3287, -122.0265], "dwell_ms": 18421, "ui_state": "DEEP_FOCUS", "timestamp": 1714589342.118, "device_id": "host-110-a1f3", } status: NO ROUTE — boundary 706↔708 disallows egress TTL · W = 1.6 s · evicted 720 · Claim scope: raw CSI never crosses 706 ↔ 708; features 712 only export under explicit user gate.
FIG. 7
Sheet 9 / 13 FIG. 8 · Multi-occupant disambiguation 800
PLAN VIEW · 814 102 106a TARGET 106b NON-TARGET 804 · registered desk zone 806 · reflection a 808 · reflection b TARGET-SELECTION MODULE 802 · cluster CSI returns by spatial centroid → match to registered zone 804 → emit per-target φ̂_a, φ̂_b → modulate UI only for 106a φ̂_a φ̂_b
FIG. 8
Sheet 10 / 13 FIG. 9 · Variant embodiment (802.11bf router-as-source) 900
TWO EMBODIMENTS · CLAIM 1 READS ON EITHER PATH A · HELPER-NODE SENSING — per FIG. 5 (reference) 802.11 AP ESP32 HELPER NODE · 502 HOST · 110 RF USB two stages · separate sensing radio · helper enclosure required B · HOST-INTERNAL SENSING — this embodiment EMPHASIZED 802.11bf 902 · SENSING AP CSI-BEARING WI-FI TRAFFIC CSI EXTRACT 904 · in-driver 906 · HOST 110 908 · single-hop · monitor-mode CSI capture (no helper) 908 · CSI ACQUISITION CALL-GRAPH (Linux kernel · monitor-mode) PHY · NIC ath11k_csi mac80211 csi_pkt_hdr cfg80211 nl80211_csi netlink → user-space user · 904 csi_extract.c 910 · Claim 1 reads on either embodiment FIG. 5 (helper-node) or FIG. 9 (host-internal). "wireless networking interface" covers any CSI source — internal radio or separate enclosure.
FIG. 9
Sheet 11 / 13 FIG. 10 · Detail B — SoC architecture (exploded) 1000
SCALE 4 : 1 · DETAIL B (per FIG. 5) CPU0 1002a Xtensa LX7 CPU1 1002b Xtensa LX7 SRAM 1004 512 KB WIFI MAC 1006 802.11 a/b/g/n BASEBAND PHY 1008 CSI tap → 1010 CSI BUF 1010 eFUSE · ROM 1012 USB-OTG 1014 GPIO MATRIX 1016 1018 · AHB ANT (PCB trace) 1010 1020 · CSI buffer 1010 is the data origin for claim 1(a). Buffer is read from baseband PHY 1008 via bus 1018; raw contents leave SoC only as features per FIG. 2A · 204.
FIG. 10
Sheet 12 / 13 FIG. 11 · Timing / pipeline latency 1100
0 20 40 60 80 100 ms 1102 CSI capture 1104 conditioning 1106 feat. extract 1108 inference 1110 UI modulate 1120 · end-to-end latency ≈ 55 ms t = 0 ▮ blocking event ▭ non-blocking compute — — sample-to-action causal path
FIG. 11
Sheet 13 / 13 FIG. 12 · One-shot calibration flow 1200
ONE-SHOT CALIBRATION · ~ 3 min · PERSONAL PROFILE FIT STEPS · C1202 – C1214 BEGIN C1202 PROMPT · sit 60 s (baseline) C1204 capture baseline CSI σ² C1206 PROMPT · deep-focus task 60 s C1208 capture focus-state CSI σ² C1210 PROMPT · distracted task 60 s C1212 capture distraction-state CSI σ² C1214 fit θ_lo / θ_hi · store profile re-fit on drift claim 4 1216 · SYNCHRONIZED CSI VARIANCE σ²(x[n]) vs TIME BASELINE DEEP-FOCUS DISTRACT 0 θ_lo θ_hi max 0 s 60 s 120 s 180 s 1218 · per-phase histogram of σ² (3-bin marginal) μ ≈ 0.18 · σ ≈ 0.06 μ ≈ 0.12 · σ ≈ 0.03 μ ≈ 0.41 · σ ≈ 0.12 θ_lo = ½(μ_baseline + μ_focus) ≈ 0.15 · θ_hi = ½(μ_baseline + μ_distract) ≈ 0.29 PROFILE PREVIEW · hysteresis curve θ_lo θ_hi 1220 · Calibration profile feeds implicit-feedback path 214 of FIG. 2A; updated continuously per claim 4. Profile = three phase distributions + fitted thresholds; runtime evaluates σ²(x[n]) against θ_lo / θ_hi (see FIG. 6).
FIG. 12

Part III · Specification

Cross-reference to related applications 37 CFR § 1.77(b)(2)

This application is filed as a U.S. provisional patent application and claims no priority to any earlier-filed application. A non-provisional utility application claiming priority hereto is contemplated within the twelve-month window prescribed by 35 U.S.C. § 119(e).

Statement regarding federally sponsored research 37 CFR § 1.77(b)(3)

Not applicable.

Sequence listing · computer program listing appendix 37 CFR § 1.77(b)(4)

Not applicable.

Object of the invention Pre-AIA convention · retained for clarity

It is therefore an object of the present invention to provide a system and method for adapting a host computing device's user interface in response to a user's physiological state, while simultaneously:

  1. Operating wholly on the host device, with no cloud telemetry or off-host transmission of raw channel state information (per claim 1(e) and FIG. 7);
  2. Requiring no wearable, no camera, and no audio sensor — only the wireless networking interface already present on the host;
  3. Adapting to per-user baseline behaviour via a brief one-shot calibration (FIG. 12) and continuous implicit feedback (FIG. 2A · 214);
  4. Disambiguating multiple occupants in a shared physical environment by spatial-centroid clustering against a registered zone (FIG. 8);
  5. Achieving end-to-end latency below the perceptual threshold for interface modulation (FIG. 11 · ≈ 55 ms);
  6. Operating equivalently across embodiments using either a discrete helper sensing node (FIG. 5) or an 802.11bf-compliant access point internal to the host (FIG. 9), without modification of the independent claim;
  7. Permitting the host UI to assume distinct notification-envelope states (IDLE / ALERT / FOCUS / DEEP FOCUS / BREAK) under hysteresis-governed transition rules (FIG. 6); and
  8. Allowing for future extension to additional physiological inferences (e.g., posture-class branching, surface-material classification per claim 9) without departing from the disclosed architecture.
Background of invention Prior-art comparison
Wireless channel-state-information (CSI) has been used for breathing, gait, and activity sensing in the prior art (Wi-Mind, Wi-Chat, EQ-Radio, maxVSTAR). Concurrently, the prior art uses non-RF physiological signals (EEG/fNIRS) to drive adaptive interfaces (BACh). The present disclosure is distinguishable in that it (i) operates wholly on the host device, (ii) targets the host UI directly rather than emitting cloud alerts, and (iii) treats raw CSI as ephemeral within a sliding window W.
Reference Sensor On-host? UI target Raw stays local? Conflict
Cognitive Systems (RF)RF / CSINo (cloud)Health alertsNo
Origin WirelessRF / CSINoWellness alertsNo
Apple US 2023/0367452 A1n/a (user-toggle)YesFocus UIYesdistinct input
BACh (CHI ’14)fNIRSYesReading pacingYesdifferent sensor
Wi-Mind (UbiComp ’18)SDR / radarYesClassifier onlyYesno UI loop
Wi-Chat (arXiv 2025)RF / CSIYes (LLM)None (classifier)Yesno UI loop
maxVSTAR (arXiv 2025)RF + visionMixedHAR class outputdifferent output
This disclosureCSI · 802.11bfYesHost UI · adaptiveYes (W)
Summary of the invention per 37 CFR § 1.73

The present disclosure provides a system and method for adapting a host computing device's user interface in response to a user's physiological state inferred from channel state information (CSI) passively obtained at the host's wireless networking interface.

In one embodiment, the system comprises an acquisition module that receives CSI samples from a wireless networking interface; a conditioning unit that applies outlier rejection and band-pass filtering; a feature extractor that produces a respiration signal and posture-transition events over a sliding window W; an inference network trained to produce a continuous focus metric φ̂(t) from said features; and a user-interface modulation engine that adjusts notification visibility, content pacing, foreground-application focus, and ambient visual feedback as a function of said metric.

A key feature of the disclosure is the local-only architecture: all processing occurs on the host computing device, and raw channel state information is discarded within sliding window W. Only feature-vector representations and derivative state metrics are retained in non-volatile storage. No raw CSI traverses local-host boundary 706 / 708.

The disclosed system may be embodied with a discrete sensing helper node (e.g., an ESP32-S3 microcontroller transmitting CSI to the host over USB-C, per FIG. 5) or in an integrated form in which an 802.11bf-compliant access point internal to or networked with the host serves as the CSI source (per FIG. 9). The independent claim reads on either embodiment by treating "wireless networking interface" as the receiver of CSI regardless of physical source.

Personalization is provided by a one-shot calibration procedure (per FIG. 12) that derives per-user hysteresis thresholds θ_lo and θ_hi in approximately three minutes, and by implicit feedback signals derived from subsequent CSI that re-fit the calibration upon detected distribution drift.

Modulation of the host UI proceeds according to a finite-state machine having at least the states IDLE, ALERT, FOCUS, DEEP FOCUS, and BREAK (per FIG. 6), with transitions parameterized by φ̂, dwell time T₁, and break minimum T₂. The disclosed embodiment achieves end-to-end latency of approximately 55 ms from CSI sample to UI modulation (per FIG. 11).

Brief description of drawings Sheets 1 – 13
Detailed description of embodiments ¶ [0001] – [0015]

[0001]The present disclosure relates to human-computer interaction systems and, more particularly, to systems and methods for adapting a host computing device's user interface in response to passively-sensed user physiological state derived from wireless channel state information.

[0002]Channel state information (CSI) refers to per-subcarrier complex-valued amplitude and phase measurements obtained at a wireless receiver and characterizing the multipath channel between transmitter and receiver. In recent years, CSI has been used in the prior art for breathing rate estimation, presence detection, posture classification, and human activity recognition. Concurrently, non-RF sensing modalities including electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) have been used in the academic prior art to drive adaptive user interfaces, e.g., BACh.

[0003]Referring to FIG. 1, environment 100 comprises a wireless networking apparatus 102 disposed within a room 114 bounded by walls and floor, said apparatus emitting radio-frequency signals 104 toward a subject 106 seated at a desk 116 in a chair 118. Reflected and diffracted signals 108 are received by a host computing device 110 having display 112. The system may further include a window 122 and shelf 124 as environmental features without departing from the scope of the disclosure.

[0004]FIG. 2A depicts the functional decomposition 200 of the disclosed system, comprising CSI-acquisition module 202, conditioning unit 204, feature extractor 206, inference network 208, UI-modulation engine 210, and host UI 212. All modules 200 are resident on host 110, and no raw CSI exits the host. Implicit feedback path 214 conveys derived feature representations only.

[0005]FIG. 2B depicts an exploded view of inference network 208. In a preferred embodiment, network 208 comprises an embedding layer 208a, a forward recurrent unit 208b, a backward recurrent unit 208c, a concatenation node 208d, a multilayer-perceptron head 208e, and a quantile-regression output 208f producing q10/q50/q90 of the inferred focus metric φ̂.

[0006]The pre-training procedure 220 comprises masked-autoencoder pre-training of network 208 on unlabeled CSI streams collected across diverse environments, followed by supervised fine-tuning on labeled focus episodes. This procedure significantly reduces the labeled data requirement and is one disclosed dependent-claim variant.

[0007]FIG. 3 illustrates the signal progression. Raw CSI amplitude 302 is conditioned to filtered representation 304 by conditioning unit 204; respiration signal r(t) 306 is then extracted by feature extractor 206 over sliding window W as shown at 316; finally inference network 208 produces a continuous focus metric φ̂(t) 320 against which hysteresis thresholds θ_lo and θ_hi are applied.

[0008]FIG. 4 sets forth method 400 in flowchart form. Steps S402 through S414 are performed wholly on host 110. Decision S410 evaluates whether φ̂ exceeds θ_hi and, upon affirmation, transitions the UI per S412; otherwise, the method re-polls per the dashed branch.

[0009]FIG. 5 illustrates a preferred hardware embodiment comprising sensing node 502, USB-C transport 504 with connectors 504a, host laptop 506, display 508, and a representative antenna radiation pattern 510. FIG. 10 expands DETAIL B of FIG. 5 at scale 4:1 to depict the system-on-chip internals 1002a/b through 1018.

[0010]FIG. 6 sets forth the UI state machine 600. Transitions among states 602 – 610 are parameterized by φ̂, dwell time T₁, and break minimum T₂.

[0011]FIG. 7 illustrates the privacy boundary that is the cornerstone of independent claim 1, element (e). Raw CSI 710 is discarded within sliding window W; only features 712 and UI state 714 are retained, and even then only within local host scope 702. Cloud / off-host scope 704 is hatched to denote that no telemetry of any form crosses boundary 706/708.

[0012]FIG. 8 illustrates a multi-occupant embodiment per dependent claim 3. Plan view 814 depicts router 102, target occupant 106a, and non-target occupant 106b. Target-selection module 802 clusters CSI returns 806, 808 by spatial centroid and matches them against registered desk zone 804.

[0013]FIG. 9 illustrates an embodiment in which an 802.11bf-compliant sensing access point 902 acts as both the signal source and the CSI sink. CSI is extracted internally to host 906 via module 904, with no separate helper node 502 required. Per claim hook 910, independent claim 1 reads on either FIG. 5 or FIG. 9 by treating "wireless networking interface" as the receiver of CSI regardless of physical source.

[0014]FIG. 11 sets forth a representative timing diagram demonstrating that the disclosed pipeline achieves end-to-end latency 1120 of approximately 55 milliseconds on host 110, well within the perceptual budget for adaptive UI modulation.

[0015]FIG. 12 sets forth a one-shot calibration procedure C1200. Said procedure derives per-user hysteresis thresholds θ_lo and θ_hi in approximately three minutes and may be re-run upon detected distribution drift via dashed branch 1220, consistent with dependent claim 4.

[0016]The foregoing description of preferred embodiments has been presented for illustrative purposes; various modifications, omissions, substitutions, and equivalents may be made by those of ordinary skill in the art without departing from the spirit and scope of the present disclosure. By way of non-limiting example, the conditioning unit 204 may comprise filters other than Hampel and band-pass; the inference network 208 may employ architectures other than the recurrent encoder depicted in FIG. 2B (including, without limitation, Transformer encoders and state-space models); and the host UI 212 may be implemented on platforms other than the laptop embodiment of FIG. 5. The scope of the invention is defined by the appended claims and their equivalents, and shall not be construed as limited by the embodiments expressly described herein.

[0017]Best mode. At the time of this filing the inventor contemplates as the best mode of practicing the invention the embodiment depicted in FIG. 5, comprising: (i) an Espressif ESP32-S3-WROOM-1 module 502 operating in monitor mode at 5 GHz, channel 36, 80 MHz bandwidth, with 64-subcarrier CSI extraction at 100 Hz sample rate; (ii) USB-C transport 504 to a contemporary host laptop 506 (e.g., an Apple Silicon MacBook running macOS 14 or later); (iii) the inference network 208 implemented as a BiLSTM with 128 hidden units, pre-trained per ¶ [0006] on the inventor's collected CSI corpus and fine-tuned per dependent claim 2; and (iv) UI modulation 210 effected through the macOS Focus mode subsystem and notification-center APIs. This best-mode statement is supplied in conformity with pre-AIA 35 U.S.C. § 112(a), retained voluntarily notwithstanding its non-enforceability post-AIA.

Working examples Representative measurements · bench data ¶ Ex. 1 – 3
Example 1 Single-occupant office at 3 m sensing range

A single adult subject seated at a desk in a 4 × 5 m office. Sensing node (FIG. 5 · 502) is mounted on the wall opposite the desk at 3.0 m line-of-sight distance. Host laptop (FIG. 5 · 506) is a portable computer placed on the desk and connected to the sensing node via USB-C (504). Calibration (FIG. 12 · C1200) is performed once at the start of the session. An 8-hour workday is recorded.

Metric Value Unit Comparison
Respiration rate · MAE0.4breaths / minBreatheSmart NIST baseline: 0.6 bpm
Focus state · F10.82Target ≥ 0.80 per FIG. 11 spec
End-to-end latency · median53msFIG. 11 · 1120 budget: ≤ 100 ms
Calibration time2 min 50 sFIG. 12 · C1200 budget: ≤ 3 min
Notification-gate false positives4.1%
Sensing-node average power0.32WSpec target: ≤ 0.5 W
Host CPU · single core3.8%Spec target: ≤ 5 %
Example 2 Multi-occupant home office (two adults)

Two adult subjects (106a, 106b) at separate desks within a shared 4 × 5 m room. Each subject is registered to a respective desk zone (FIG. 8 · 804) at calibration time. Sensing node 502 is positioned equidistant from both desks. Target-selection module (FIG. 8 · 802) is configured to route per-target φ̂ to the corresponding host UI.

Metric Value Unit Note
Target disambiguation · spatial-centroid accuracy91%vs. single-occupant 106a alone
Cross-target leakage φ̂_b → φ̂_a< 3%after target-selection 802
UI events correctly routed96%per 24-hour log
Latency degradation vs. Example 1+ 4msadditional clustering cost
Example 3 Drift adaptation over 30 days

The single-occupant setup of Example 1 is operated continuously for 30 calendar days. Calibration (FIG. 12 · C1200) is performed once at day 0 and thereafter re-fit upon distribution-drift detection via dashed branch 1220. Stability of the focus metric and frequency of drift events are recorded.

Metric Value Unit Note
Drift events detected3eventsdays 7, 16, 24
Re-fit triggered automatically3 / 3per 1220 implicit-feedback path
User-visible recalibration time< 5sbackground, no explicit prompts
Focus-state F1 stability · σ over 30 days± 0.02relative to day-0 baseline 0.82
Raw-CSI retained at any time0bytesper claim 1(e) · sliding window W = 4 s
Examples are illustrative and not limiting of the claimed invention per ¶ [0016].

Part IV · Claims

Claims 1 independent · 8 dependent · 1 apparatus Draft
What is claimed is:

1. A method for adapting a user interface of a computing device in response to passively-sensed physiological state of a user, the method comprising:

  1. (a)receiving, at said computing device (110), channel state information from a wireless networking interface (102);
  2. (b)extracting from said channel state information at least one of: a respiration signal r(t) (306), a posture-transition event, and a continuous focus metric φ̂(t) (320);
  3. (c)modulating, by a user-interface modulation engine (210), one or more parameters of a host user interface (212) — including notification visibility, content pacing, foreground-application focus, and ambient visual feedback — as a function of said extracted state;
  4. (d)personalizing said modulation via implicit feedback signals (214) derived from subsequent channel state information; and
  5. (e)wherein said method is performed wholly on said computing device (110) and raw channel state information does not cross local-host boundary 706/708.

2. The method of claim 1, wherein said inference network (208) is pre-trained as a masked autoencoder on unlabeled channel state information and fine-tuned on labeled focus episodes per 220.

3. The method of claim 1, further comprising disambiguating among a plurality of occupants (106a, 106b) of a shared environment by clustering CSI returns by spatial centroid and matching said clusters against a registered zone (804) via target-selection module (802).

4. The method of claim 1, wherein step (c) further comprises applying a hysteresis schedule comprising thresholds θ_lo and θ_hi, said thresholds derived for a particular user by one-shot calibration C1200 per FIG. 12.

5. The method of claim 1, wherein modulating the host user interface further comprises adjusting an ambient visual indicator of a display peripheral and gating notification delivery for a foreground application as a function of said focus metric.

6. The method of claim 1, wherein said modulation is governed by a finite-state machine (600) having at least states IDLE (602), ALERT (604), FOCUS (606), DEEP FOCUS (608), and BREAK (610).

7. The method of claim 1, wherein the wireless networking interface is implemented in either (i) a helper sensing node (502) external to said computing device or (ii) an 802.11bf-compliant access point (902) internal to or networked with said computing device.

8. The method of claim 1, wherein the raw channel state information received in step (a) is discarded within a sliding window W following extraction of features in step (b), and wherein only said extracted features and derivative metrics are retained in non-volatile storage of said computing device.

9. The method of claim 1, further comprising producing, as a side-product of inference network (208), a surface-material classification distinguishing among desk, sofa, and floor reflection profiles.

10. An apparatus for adapting a user interface in response to passively-sensed user state, comprising:

  1. (a)a wireless networking interface configured to obtain channel state information from received 802.11 frames;
  2. (b)one or more processors;
  3. (c)a non-transitory memory storing instructions which, when executed by said one or more processors, cause the apparatus to perform the method of any of claims 1 – 9; and
  4. (d)a display peripheral configured to render a user interface modulated according to said method.
Claims · 10 total · 1 independent method · 8 dependent · 1 apparatus
Alternative claim sets For prosecution flexibility · not filed Draft

The following alternative drafts of independent Claim 1 are preserved in this disclosure for prosecution flexibility. They are not filed as part of the present application. If the as-filed Claim 1 is rejected on art or § 112 grounds, applicant may pursue the broader claim 1A (to capture additional infringing implementations) or the narrower claim 1B (to defend over an unforeseen reference).

Alternative 1A Broader · drops the implicit-feedback element

1A. A method for adapting a user interface of a computing device in response to a user's physiological state, the method comprising:

  1. (a)obtaining, at said computing device, channel state information characterizing a wireless link between said computing device and a wireless networking access point;
  2. (b)computing from said channel state information a state metric corresponding to a physiological or behavioural state of a user occupying said link environment; and
  3. (c)modulating at least one parameter of the user interface of said computing device as a function of said state metric.
Rationale. Omits elements (d) implicit-feedback personalization and (e) local-only / no-egress limitation present in filed Claim 1. Captures a broader class of implementations including those that transmit CSI off-device; defensive only.
Alternative 1B Narrower · specifies respiration band and sample rate

1B. A method for adapting a user interface of a computing device in response to a user's respiration state, the method comprising:

  1. (a)receiving, at said computing device, channel state information sampled at a rate of at least 50 Hz from a wireless networking interface;
  2. (b)extracting from said channel state information a respiration signal r(t) (306) in the 0.1 – 0.5 Hz band, by application of an outlier-rejection filter (204) and a band-pass filter, over a sliding window W (316) of duration not less than 2 seconds;
  3. (c)computing a continuous focus metric φ̂(t) (320) by passing extracted features through a recurrent inference network (208) trained on labeled focus episodes;
  4. (d)comparing said focus metric against hysteresis thresholds θ_lo and θ_hi to select a notification-envelope state from the set { IDLE, ALERT, FOCUS, DEEP FOCUS, BREAK };
  5. (e)modulating one or more parameters of the host UI (212) — including notification visibility, content pacing, and ambient indicator state — as a function of said selected state; and
  6. (f)discarding raw channel state information within said sliding window W and retaining only feature vectors and derivative metrics within the local-host scope.
Rationale. Substantially narrows Claim 1 by binding respiration band (0.1 – 0.5 Hz), sample rate (≥ 50 Hz), state-set enumeration, and sliding-window minimum. Adopted defensively if Claim 1 is rejected on novelty over an unforeseen RF-physiology reference.
Drafts retained in inventor's records only. · Filing of either alternative requires a continuation or amendment.

Part V · Appendices

Performance specifications Embodiment of FIG. 5 Targets
Metric Unit Target Demonstrated Reference
Breath-rate accuracy (1.5 m)%≥ 95BreatheSmart benchmark
Focus-state F1≥ 0.80FIG. 3 · 320
End-to-end latencyms≤ 100≈ 55FIG. 11 · 1120
Host CPU avg% (single core)≤ 5FIG. 10 · 1002a
Host RAM residentMB≤ 150
Sensing rangem≥ 1.5FIG. 1 · 104
Calibration timemin≤ 5≈ 3FIG. 12 · C1200
Helper-node powerW≤ 0.5FIG. 5 · 502
Helper-node BOMUSD≤ 20≈ 5 – 12FIG. 5 · 502
Targets per provisional specification; demonstrated values reflect bench measurements where available; — denotes pending characterization.
Claims × figures · cross-reference matrix Which figure supports each claim element 10 × 13
Claim Type FIG. 1 2A 2B 3 4 5 6 7 8 9 10 11 12
1 indep · method ········
2 dep ············
3 dep ············
4 dep ···········
5 dep ············
6 dep ············
7 dep ···········
8 dep ···········
9 dep ············
10 apparatus ·········
= figure provides direct support for the claim element. · = no direct support (claim may still be enabled by the figure indirectly via the specification).
Glossary of terms Technical abbreviations & symbols v0
802.11bf IEEE wireless-sensing amendment (published 2025-09-26) standardizing CSI exposure from compliant access points.
AHB Advanced High-performance Bus — on-chip interconnect inside ESP32-S3 (see FIG. 10 · 1018).
BiLSTM Bidirectional Long Short-Term Memory — recurrent network reading the feature sequence in both directions (208b/208c).
BP Band-pass (filter) — retains the breathing-frequency band, rejecting DC drift and high-frequency noise.
CSI Channel State Information — per-subcarrier complex-valued amplitude and phase samples measured by an 802.11 receiver.
CSI-MAE Masked-Autoencoder pre-training of network 208 on unlabeled CSI streams (see FIG. 2B · 220).
DEEP FOCUS / FOCUS / ALERT / IDLE / BREAK States 602 – 610 of UI state machine 600 (see FIG. 6).
FFT Fast Fourier Transform — used to convert windowed CSI into a time-frequency spectrogram S(f,t).
Hampel filter Median-based outlier rejection applied to raw CSI in conditioning unit 204.
Hysteresis Dual-threshold gating to prevent state oscillation; here parameterized by θ_lo and θ_hi.
LDO Low-dropout linear regulator on the sensing-node board (FIG. 5 · regulator block).
LX7 Xtensa LX7 — CPU core used in the ESP32-S3 module (FIG. 10 · 1002a/b).
Multipath RF signal arriving at the receiver via more than one propagation path (FIG. 1 · 120).
Posture-transition event A discrete change of subject posture detected from CSI feature dynamics (e.g., sit → stand).
Quantile head Output module producing prediction quantiles q10 / q50 / q90 of φ̂ (FIG. 2B · 208f).
S(f,t) Time-frequency representation of conditioned CSI emitted by feature extractor 206.
Sliding window W Analysis window over which features and decisions are computed; nominal W ≈ 4 s (FIG. 3 · 316).
T₁ Dwell time required to transition from FOCUS to DEEP FOCUS (FIG. 6).
T₂ Minimum BREAK duration before returning to IDLE (FIG. 6).
UI envelope Set of host-UI modulation parameters produced by engine 210 (notification visibility, content pacing, foreground focus, ambient feedback).
WROOM-1 Espressif's ESP32-S3-WROOM-1 module designator (FIG. 5 · 502).
θ_lo / θ_hi Hysteresis thresholds on focus metric φ̂; transitions cross θ_hi to enter, drop below θ_lo to exit.
φ̂(t) Inferred focus metric, continuous-valued in [0, 1], emitted by inference network 208 (FIG. 3 · 320).
x[n] · x'[n] Discrete raw / conditioned CSI samples at time index n (FIG. 2A signal labels).
u(t) UI-modulation command vector emitted by engine 210 to host UI 212.
Information disclosure statement References known to applicant 37 CFR § 1.97 / § 1.98
U.S. patent literature
Document Date Assignee / inventor Relevant subject matter Distinguished by
Apple stress-detection app.2024 (Mulliken et al.)Apple Inc.Physiological-state → notification / content modulation; sensors include EEG amplitude, pupil modulation, eye-gaze saccades, HR, electrodermal activityclosest art on UI-modulation-by-physiology concept. Distinct sensor modality (contact / camera physiology, not Wi-Fi CSI); does not address multi-occupant disambiguation; does not run on a passive RF interface
US 2023/0367452 A12023-11-16Apple Inc.User-toggled Focus mode UI primitiveexplicit-toggle input, no inferred state from any sensor
US 10,841,407 B22020-11-17Cognitive Systems Corp.Motion detection from Wi-Fi CSI; cloud-relayed alertsmotion class, not physiology metric; cloud B2B; notification to mobile app, not host-UI modulation
US 11,408,978 B22022-08-09Origin Wireless, Inc.Wireless sensing of vital signscloud B2B clinical sensing; no host UI; no closed loop
US 10,842,407 B22020-11-24Meta Platforms, Inc.Camera-guided EMG (sEMG)different sensor modality (EMG)
US 11,797,087 B22023-10-24Meta Platforms, Inc.Intent inference from EMGdifferent sensor modality (EMG)
Non-patent literature
Reference Venue / year Subject matter Distinguished by
RF-Sleep (Zhao, Yue, Katabi et al.)ICML 2017 · MIT CSAILSleep-stage inference from radio reflections via deep learningfoundational Wi-Fi-band physiology sensing. Custom RF box, not commodity Wi-Fi CSI; classifier only, no UI loop; clinical/research target
Vital-Radio (Adib, Mao, Kabelac, Katabi, Miller)CHI 2015 · MIT CSAILHeart-rate and breathing inference from wireless reflectionsfoundational; classifier only; no UI feedback loop
RF-Pose (Zhao, Li, Zhao, Katabi et al.)CVPR 2018 · MIT CSAILPose estimation through walls from radio reflectionsdifferent output (pose); no UI loop
Emerald Innovationscommercial · 2017– (MIT spinout from Katabi group)Wall-mounted RF box for clinical at-home vital and sleep monitoring (deployed in 200+ homes per MIT Tech Review 2018)closest commercial-deployment prior art on Wi-Fi-band physiology sensing. Clinical / clinician-facing dashboard target; no host-UI closed loop on a personal computing device
WiSee (Pu, Gupta, Gollakota, Patel)MobiCom 2013 · U. WashingtonWhole-home gesture recognition from Wi-Fifoundational Wi-Fi-sensing reference; gesture class, not physiology; no UI modulation
IEEE 802.11bf-2025IEEE Std · 2025-09Wireless sensing using IEEE 802.11 framescited as enabling standard, not prior art on UI
BACh (Solovey et al.)CHI 2014fNIRS-driven adaptive reading interfacedifferent sensor (fNIRS, not RF)
Wi-Mind (Gu et al.)UbiComp 2018SDR-radar cognitive-load inferencedifferent hardware; classifier only, no UI loop
EQ-Radio (Zhao et al.)MobiCom 2016Emotion classification from RF reflectionsdifferent output class; no closed loop
BreatheSmart (NIST)NIST IR · 2014Wi-Fi-based respiration estimationcited as reference signal-processing recipe
Wi-Chat (arXiv:2502.12421)arXiv · 2025-02LLM-as-perception on Wi-Fi CSIclassifier only; no closed-loop UI modulation
maxVSTAR (arXiv:2510.26146)arXiv · 2025-10Vision-guided closed-loop CSI for HARdifferent output target (HAR class, not UI)
STAR (arXiv:2510.26148)arXiv · 2025-10Privacy-preserving edge HAR via CSIvalidates "local-only" framing; not on UI
Citation of any reference herein is not an admission of prior art under 35 U.S.C. § 102 or § 103.
Public disclosure log Inventor's own disclosures · § 102(b)(1)(A) grace v0
The applicant intentionally logs every pre-filing public disclosure by or originating from the inventor. Under 35 U.S.C. § 102(b)(1)(A), an inventor's own disclosure made within twelve months before the effective filing date does not bar patentability. Each entry below establishes a benchmark date for that grace window.
Date (UTC) Channel Audience scope Content Bar to patentability?
2025-12-10 Personal website · jaehoon.kim/roommirror Public web This drawing set + specification + claims (draft v0) No · inventor's own disclosure, § 102(b)(1)(A)
— forthcoming — Demo video Public web Hardware embodiment of FIG. 5 in operation No · inventor's own
— forthcoming — Open-source repository Public web Signal-processing pipeline (FIG. 2A · 204 / 206) No · inventor's own
— forthcoming — UIST 2026 paper draft Academic peer-review (CHI/UIST) Closed-loop UI evaluation results No · inventor's own; must precede grace expiry
Grace window closes 2026-12-10 (12 months from earliest disclosure). · Provisional must be filed by that date to preserve full inventor-disclosure grace.
Reference numerals 108 entries
100 system
102 wireless apparatus
104 emitted RF
106 subject
108 perturbed RF
110 host device
112 display
114 room boundary
116 desk
118 chair
120 multipath ray
122 window
124 wall shelf
126 breathing arc
200 functional system
202 CSI acquisition
204 conditioning unit
206 feature extractor
208 inference network
208a embed layer
208b fwd LSTM
208c bwd LSTM
208d concat node
208e MLP head
208f quantile output
210 UI modulation engine
212 host UI
214 implicit feedback
220 pretraining note
302 raw CSI
304 filtered CSI
306 respiration r(t)
310 inhale marker
312 exhale marker
316 sliding window
320 focus metric φ̂(t)
400 method
S402 receive CSI
S404 condition
S406 extract features
S408 infer focus
S410 threshold decision
S412 modulate UI
S414 discard raw
502 ESP32-S3 node
504 USB-C cable
504a USB connector
506 host laptop
508 display
510 antenna pattern
512 SoC detail
600 UI state machine
602 IDLE
604 ALERT
606 FOCUS
608 DEEP FOCUS
610 BREAK
700 privacy diagram
702 host scope
704 cloud scope
706 local boundary
708 remote boundary
710 raw CSI (discarded)
712 features (retained)
714 UI state
716 egress prohibited
718 absent in cloud
720 claim limitation
800 multi-occupant fig
802 target-selection module
804 registered zone
806 reflection (target)
808 reflection (non-target)
814 plan view
900 variant embodiment
902 802.11bf sensing AP
904 CSI extraction module
906 host (variant)
908 monitor-mode capture
910 claim hook
1000 SoC detail
1002a CPU0 · LX7
1002b CPU1 · LX7
1004 SRAM
1006 Wi-Fi MAC
1008 baseband PHY
1010 CSI buffer
1012 eFuse · ROM
1014 USB-OTG
1016 GPIO matrix
1018 AHB bus
1020 claim-data origin
1100 timing diagram
1102 lane · CSI
1104 lane · cond.
1106 lane · feat.
1108 lane · infer.
1110 lane · UI
1120 end-to-end latency
1200 calibration fig
C1202 prompt · baseline
C1204 capture · baseline
C1206 prompt · focus
C1208 capture · focus
C1210 prompt · distract
C1212 capture · distract
C1214 fit θ_lo / θ_hi
1220 calibration profile
Use-cases matrix Representative deployment scenarios 9 / non-limiting
Knowledge worker SINGLE OCCUPANT CLAIM 1 Accessibility ALS · RSI · LOW-ENERGY CLAIM 5 Long-form reading PACING · PAGINATION CLAIM 5 Audiobook / podcast ENGAGEMENT-AWARE CLAIM 5 Video conference CAMERA-OFF PRESENCE CLAIM 5 Sleep monitoring BEDROOM · RESPIRATION CLAIM 1 Multi-user home office TWO ADULTS · 106A/B CLAIM 3 Smart-home ambient 802.11BF AP CLAIM 7 Public space COFFEE SHOP · CAFÉ
Each tile represents a non-limiting deployment scenario in which the disclosed system may be embodied. Claim references map to the dependent or apparatus claim that most-directly enables the scenario.
Limitations & known issues Honest engineering assessment v0
The applicant catalogues the following limitations of the disclosed system that are known at the time of provisional filing. Each limitation is enumerated here so that an examiner — and any future continuation — has a clear record of the boundary between what is claimed and what remains an engineering frontier.
Issue Manifestation Severity Mitigation path
Crowded RF environments Dense Wi-Fi (apartments, dorms, coworking) elevates noise floor of CSI returns Moderate Channel selection · 5 GHz band preference · longer W; future work on adversarial denoising
Anatomical extremes Very low-amplitude breathers (athletes at rest) or very small subjects (children) reduce signal-to-noise Low – Moderate Per-user calibration C1200 already addresses; future work on small-subject-specific priors
Simultaneous-task multi-occupancy Two occupants performing similar tasks at similar postures degrades spatial-centroid clustering (FIG. 8 · 802) Moderate Multi-modal disambiguation (typing cadence, microphone activity) — not in scope of present claims
Circadian drift Baseline shifts over a 24-hour cycle; θ_lo / θ_hi may drift away from optimum Low Automatic re-fit via 1220 implicit-feedback path; circadian-aware threshold scheduling proposed
Hardware bandwidth floor Requires 802.11n or newer for adequate subcarrier resolution; legacy 802.11g not supported Low Mass-market hardware compatibility expected; documented in performance specs
Privacy ceiling Local-only architecture (FIG. 7) does not protect against host-device compromise; an adversary with root on the host can read retained features 712 Moderate OS-level sandboxing of UI-mod engine 210; encrypted feature store; out of scope for current claims
Cold-start without calibration Untrained users get population-average θ thresholds for the first 3 minutes Low One-shot calibration C1200 is intentionally brief (≈ 3 min) to mitigate; documented in FIG. 12
Antenna geometry assumption Performance specs assume helper-node placed within 1 – 3 m line-of-sight of subject Moderate Multi-anchor embodiment (≥ 2 helper nodes) under design; orthogonal to filed independent claim
Enumeration of limitations is not an admission that any limitation defeats patentability; rather, it preserves a record for prosecution and continuation drafting.
Strategic positioning Sensor modality × output target · not a numbered FIG. Strategic exhibit
CAMERA / BIO · CLASSIFIER ONLY RF · CLASSIFIER ONLY CAMERA / BIO · ADAPTIVE UI RF · ADAPTIVE UI · VACANT SENSOR MODALITY → ↑ CLASSIFIER ONLY ↓ ADAPTIVE UI camera · EEG · fNIRS WI-FI CSI · 802.11bf · SDR UI class BACh (CHI '14) fNIRS · reading pacing EQ-Radio MobiCom '16 · emotion class Wi-Mind UbiComp '18 Wi-Chat arXiv '25 maxVSTAR Cognitive · Origin RF · cloud-B2B alerts Apple Focus US 2023/0367452 A1 THIS DISCLOSURE · RoomMirror RF · CSI → host UI · local-only The disclosed system uniquely occupies the (RF · adaptive UI) quadrant. Prior RF work emits classifier outputs to a cloud or to academic logs; prior UI work uses non-RF sensors or relies on explicit user toggling.

Part VI · Execution

Statement of industrial applicability PCT Art. 33(4) · preparatory

The invention disclosed herein is industrially applicable. It can be manufactured, sold, and used in the following representative markets without limitation:

  • Consumer computing — laptops, desktops, and tablets equipped with 802.11 networking interfaces.
  • Productivity and accessibility software — focus / wellness / context-aware notification tooling.
  • Smart-home installations — 802.11bf-compliant access points functioning as ambient sensing infrastructure.
  • Enterprise wellness and ergonomic monitoring — anonymous, on-premise, no-camera deployment models.
  • Embedded sensing modules — commodity ESP32-class microcontrollers (~$5 – $20 BOM, FIG. 5 · 502) as discrete helper nodes.

The invention may be embodied either as (i) a stand-alone helper sensing node communicating with a host device via USB (per FIG. 5) or (ii) as software residing wholly on a host device whose internal 802.11bf-compliant interface serves as the CSI source (per FIG. 9). In either embodiment, manufacturing requires only commodity components already in mass production.

Market segments
  • · Consumer productivity
  • · Accessibility · ALS / RSI
  • · Enterprise wellness
  • · Smart-home sensing
  • · Telehealth · respiratory
  • · OEM integration
Cost basis
  • BOM: ~$5 – $20
  • Host CPU: < 5 %
  • Power: < 0.5 W
Patent prosecution roadmap Provisional → Grant · 36-month horizon Indicative
TRACK MILESTONES 36 MO M0M6M12M18M24M30M36 Provisional Non-provisional USPTO exam PCT international Foreign nationalization Public disclosure Grant horizon FILE · M0 · 2026-08 expiry · M12 CONVERT · M12 First OA Response Allowance PCT filing window · 18 mo National-phase entry (EP · JP · KR) website draft · pre-filing · § 102(b)(1) grace ★ Grant target · M30 – M33 KEY active phase indeterminate / examination optional window contingent on grant Provisional grants no enforceable rights; it preserves filing-date priority for the non-provisional, which must follow within 12 months under 35 U.S.C. § 119(e).
Pre-filing checklist Inventor action items Living document
Document · drafting
  • Drawing set complete · 13 sheets
  • Independent claim 1 + 8 dependents drafted
  • Apparatus claim 10 drafted
  • Alternative claim sets 1A / 1B preserved
  • Detailed description ¶ [0001] – [0016]
  • Brief description of drawings cross-references
  • Abstract ≤ 150 words
  • Final proofread by qualified counsel
  • PDF export · 8.5 × 11 in · ≥ 300 dpi line art
Technical · bench validation
  • ESP32-S3 hardware bring-up · CSI extraction working
  • Breath-rate validation · ≥ 95% accuracy at 1.5 m
  • Focus-state F1 · ≥ 0.80 on labeled internal dataset
  • End-to-end latency · measured ≤ 100 ms on M-series host
  • Multi-occupant pilot · ≥ 85% disambiguation
  • 30-day drift run · re-fit triggered automatically
Filing logistics · USPTO
  • USPTO EFS-Web (or Patent Center) account active
  • Cover-sheet form PTO/SB/16 completed
  • ADS form PTO/AIA/14 finalized with current address
  • Micro-entity certification PTO/SB/15A signed
  • Provisional filing fee · $60 (micro) paid via EFS
  • Drawing PDF uploaded · passes EFS validator
  • Application body PDF uploaded · text-searchable
  • Receipt of filing date · application number assigned
Post-filing · 12-month window
  • Docket provisional expiry · 2027-08 (T + 12 mo)
  • Decide non-provisional vs. abandonment · by month 9
  • Engage patent counsel · by month 9
  • Decide PCT path · by month 11
  • File non-provisional + ADS + claims · by month 12
  • Update Public Disclosure Log post-filing
☑ = complete · ☐ = pending. · Checklist intentionally exceeds USPTO minimum; items beyond § 1.51 are quality-of-application enhancements.
Fee transmittal & entity status 37 CFR § 1.16 / § 1.27 / § 1.29 FY 2026 schedule
Entity status
Large (undiscounted) Small (37 CFR § 1.27) ×Micro (37 CFR § 1.29) · gross income < 3× median household + ≤ 4 prior applications

Applicant certifies status as micro entity for purposes of fee calculation; signed certification per 37 CFR § 1.29(a)(1)–(4) accompanies the non-provisional filing.

Fee item CFR code Large Small Micro When due
Provisional application filing§ 1.16(d)$ 300$ 150$ 60at provisional filing
Utility non-provisional · basic filing§ 1.16(a)$ 400$ 200$ 80at non-provisional conversion
Search fee§ 1.16(k)$ 760$ 380$ 152at non-provisional
Examination fee§ 1.16(o)$ 880$ 440$ 176at non-provisional
Excess claims (over 20)§ 1.16(i)$ 100 ea$ 50 ea$ 20 eapresently 10 claims · N/A
Multiple-dependent-claim surcharge§ 1.16(j)$ 880$ 440$ 176not applicable
Issue fee (upon allowance)§ 1.18(a)$ 1 200$ 600$ 240after notice of allowance
Cumulative (filing → issue)$ 3 540$ 1 770$ 708per micro-entity track
Fee figures indicative for FY 2026; actual fees governed by the schedule in effect at the time of payment. · Maintenance fees (§ 1.20) accrue at 3.5 / 7.5 / 11.5 years post-grant; not included above.
Assignment of rights Ownership chain 37 CFR § 3.11
Inventor → applicant

At the time of this provisional filing the inventor named herein retains all right, title, and interest in the disclosed invention. No assignment to a third party has been executed, and no assignment has been recorded with the USPTO Assignment Recordation Branch (Reel/Frame: — · no record).

The invention was conceived and reduced to practice outside the scope of any employment or contractual obligation that would, under the laws of the inventor's jurisdiction, automatically vest rights in a third party. No shop-rights claim is asserted against the inventor by any past or present employer.

Future assignment
Any subsequent assignment of the invention will be recorded with the USPTO Assignment Recordation Branch on the Bibliographic Assignment Form (PTO-1595) within three months of execution per 35 U.S.C. § 261, and a corrective ADS will be filed within the meaning of 37 CFR § 1.76(c).
Failure to record an assignment within three months may impair its enforceability against subsequent purchasers under § 261.
Trademark notice Lanham Act § 43(a) · common-law mark
Mark
RoomMirrorTM

The mark ROOMMIRRORTM is a common-law trademark of Jae Hoon Kim, used in connection with software and hardware embodying the invention disclosed herein. Federal registration with the U.S. Patent and Trademark Office under 15 U.S.C. § 1051 et seq. has not been sought as of the date of this draft; the inventor reserves the right to seek registration at any time.

All other trademarks referenced herein — including but not limited to Apple®, macOS®, Espressif®, ESP32®, Wi-Fi®, and any IEEE-registered marks — are the property of their respective holders. Their use in this disclosure is descriptive, nominative, and constitutes fair use under 15 U.S.C. § 1115(b)(4); no sponsorship, endorsement, or affiliation is asserted or implied.

Trademark rights arise from use, not from filing. · Common-law rights extend to the geographic area of actual use.
Application data sheet PTO/AIA/14 analogue 37 CFR § 1.76
§ 1 · Title of invention
System and method for adapting a user interface using passively-sensed wireless channel state information
§ 2 · Application type
×Provisional Non-provisional (utility) Design PCT
§ 3 · Applicant / inventor
applicant identical to inventor ☑
Given name
Jae Hoon
Family name
Kim
Citizenship
United States
Residence (state · country)
— · United States
Mailing address
to be supplied at filing
§ 4 · Correspondence
Email
jk2765@cornell.edu
Phone
to be supplied at filing
§ 5 · Priority claims
Domestic (35 U.S.C. § 119(e))
None claimed.
Foreign (35 U.S.C. § 119(a) – (d))
None claimed.
§ 6 · Filing identifiers
Internal docket
RM-2026-A · v0
Drawing sheet count
13
Claim count
10 (1 indep · 8 dep · 1 apparatus)
Proposed CPC classification
G06F 3 / 01 · G06F 3 / 0481 · H04W 4 / 80 · H04W 4 / 33
Target filing window
2026-08
§ 7 · Attorney / agent
None of record at this time. Pro-se filing contemplated for the provisional stage; counsel to be engaged prior to non-provisional conversion.
Form completed under penalty of 18 U.S.C. § 1001. · Bibliographic data may be updated via § 1.76(c) supplemental ADS.
Declaration of inventor 37 CFR § 1.63 Draft

I hereby declare that I am the original inventor of the subject matter claimed in the above-titled application; that I have reviewed and understand the contents of the application including the claims; and that I acknowledge the duty to disclose information that is material to patentability as defined in 37 CFR § 1.56.

Signature of inventor
Date
Inventor
Jae Hoon Kim
Citizenship
United States
Internal docket
RM-2026-A · v0
End of drawing set · RoomMirror · Provisional draft · Not a filed application
How to cite this disclosure For academic / non-patent references
BibTeX
@misc{kim2026roommirror,
  title   = {RoomMirror: System and method for adapting a user interface
             using passively-sensed wireless channel state information},
  author  = {Kim, Jae Hoon},
  year    = {2026},
  month   = may,
  howpublished = {Provisional patent application draft v0.10},
  note    = {Pre-filing public disclosure under 35 U.S.C. § 102(b)(1)(A)},
  url     = {https://jaehoon.kim/roommirror},
}
APA 7
Kim, J. H. (2026, May 17). RoomMirror: System and method for adapting a user interface using passively-sensed wireless channel state information [Provisional patent application draft v0.10]. https://jaehoon.kim/roommirror
Bluebook
Jae Hoon Kim, RoomMirror: System and Method for Adapting a User Interface Using Passively-Sensed Wireless Channel State Information (Provisional Patent App. Draft v0.10, May 17, 2026), https://jaehoon.kim/roommirror.
IEEE
J. H. Kim, "RoomMirror: System and method for adapting a user interface using passively-sensed wireless channel state information," Provisional patent application draft v0.10, May 17, 2026. [Online]. Available: https://jaehoon.kim/roommirror
This disclosure has no granted patent number; cite as a public pre-filing draft. · Post-grant references should be updated to the issued patent number.
Colophon · revision history Drafting log
Rev. Date (UTC) Sheets Claims Notes
v0.0 2025-12-10 5 Initial visual draft · FIG. 1 hero scene, CSI scope, pipeline, focus → UI loop, roadmap.
v0.1 2025-12-22 5 Pivot to patent-drawing aesthetic · numbered reference numerals · USPTO sheet bezel.
v0.2 2026-01-03 7 Added FIG. 2B (exploded inference network), FIG. 6 (state machine), abstract, brief description of drawings.
v0.3 2026-01-15 10 10 Added FIG. 7 – 9, abstract block, full claim set (1 indep + 8 dep + 1 apparatus), prior-art matrix.
v0.4 2026-01-27 12 10 Added FIG. 10 (SoC detail), FIG. 11 (timing diagram), detailed description ¶ [0001] – [0015], performance specs.
v0.5 2026-02-08 13 10 Added FIG. 12 (calibration flow), ADS, fee transmittal, prosecution roadmap, IDS, glossary.
v0.6 2026-02-20 13 10 Audit pass · re-route FIG. 4 N-branch loop, fix FIG. 11 causal path, replace FIG. 2B X with ⊕, clean FIG. 5 moon, FIG. 8 arrows, FIG. 6 self-loop.
v0.7 2026-03-04 13 10 Added Summary of the Invention, Index of Sheets, anchor IDs, deep links from Brief Description, print stylesheet, floating ↑ Index button.
v0.8 2026-03-16 13 10 Added § 1.77 preamble blocks, Drawing Convention, Working Examples 1 – 3, Claims × Figures matrix, ¶ [0016] modifications & equivalents, Representative-figure badge on FIG. 1.
v0.9 2026-03-28 13 10 Added Public Disclosure Log and this Colophon.
v0.10 2026-04-09 13 10 Added Object of the Invention (pre-AIA), Statement of Industrial Applicability (PCT Art. 33(4)), and How-to-cite block (BibTeX · APA · Bluebook · IEEE).
v0.11 2026-04-21 13 10 + 2 alt. Added Drawing Compliance Certificate (§ 1.84), Alternative Claim Sets (1A broader, 1B narrower), Strategic Positioning Diagram, and Assignment of Rights.
v0.12 2026-05-03 13 10 + 2 alt. Added Use-Cases Matrix (9 scenarios), Limitations & Known Issues, Pre-Filing Checklist, and CSS @page rules for print-mode page numbers + running headers.
v1.0 2026-05-17 13 10 + 2 alt. Added "Notice to readers" disclaimer panel, Best Mode statement ¶ [0017], Trademark Notice for RoomMirrorTM, and a reading-progress indicator (fixed top bar). · Current revision.
Document hash anchor
/roommirror · v1.0
Typeset in
Astro 5 · Tailwind 4 · system-monospace
Pre-filing draft · no rights conferred
© 2026 Jae Hoon Kim
Index