1. Introduction
The demand for natural and intelligent Human-Computer Interaction (HCI) is rapidly growing, driven by applications in gaming, smart homes, and automotive interfaces. However, conventional methods face significant limitations: touchscreens fail in wet/oily environments, cameras raise privacy concerns and have high power consumption, and voice control struggles with complex commands and privacy. The global HMI market is projected to reach USD 7.24 billion by 2026, underscoring the need for better solutions.
This paper introduces EMGesture, a novel contactless interaction technique. It repurposes the ubiquitous Qi-standard wireless charger as a gesture sensor by analyzing the electromagnetic (EM) signals emitted during charging. These signals are perturbed by hand movements, carrying rich gesture-related information. EMGesture proposes an end-to-end framework to capture, process, and classify these perturbations, offering a practical, low-cost, and privacy-conscious alternative for pervasive interaction.
97%+
Recognition Accuracy
30
Participants
10+5
Devices & Chargers Tested
2. Methodology & System Design
EMGesture transforms a standard Qi wireless charging pad into a gesture-sensing platform. The system does not require hardware modification but uses a software-defined radio (SDR) or integrated sensor to monitor the charger's EM field.
2.1. EM Signal Acquisition & Preprocessing
The core signal is the electromagnetic field generated by the charger's power transmission coil, operating at frequencies around 100-205 kHz for Qi. When a user's hand performs a gesture near the charger, it acts as a conductive medium, perturbing this field. These perturbations are captured as time-series voltage data.
Preprocessing involves:
- Noise Filtering: Applying band-pass filters to isolate the relevant Qi frequency band from environmental noise.
- Normalization: Scaling signals to account for different device/charger pairings and baseline power levels.
- Segmentation: Isolating the signal window corresponding to a single gesture instance.
2.2. Feature Extraction & Gesture Classification
The preprocessed signal is analyzed to extract discriminative features. Given the sequential nature of gestures, features are likely drawn from both time and frequency domains:
- Time-domain: Signal amplitude, zero-crossing rate, energy.
- Frequency-domain: Spectral centroid, bandwidth, Mel-Frequency Cepstral Coefficients (MFCCs) adapted for EM signals.
- Time-Frequency: Features from Short-Time Fourier Transform (STFT) or wavelet transforms to capture evolving patterns.
A robust machine learning model (e.g., a Support Vector Machine (SVM), Random Forest, or a lightweight neural network like a 1D-CNN or LSTM) is trained on these features to classify gestures (e.g., swipe left/right, tap, circle). The model's robustness is key to handling variability across users and hardware.
3. Experimental Results & Evaluation
3.1. Recognition Accuracy & Performance
The authors conducted comprehensive experiments with 30 participants, 10 different mobile devices, and 5 Qi chargers. The system demonstrated a remarkable over 97% recognition accuracy for a defined set of gestures (e.g., directional swipes, circles, taps). This high accuracy was maintained across different device-charger combinations, proving the generalizability of the approach.
Chart Description (Inferred): A multi-bar chart likely shows accuracy percentages (y-axis) for different gesture types (x-axis) such as Swipe Left, Swipe Right, Circle, Tap, and Push. Each bar is subdivided to show performance across different test conditions (e.g., User 1-10, Device A-E). A line overlay indicates the overall average accuracy hovering consistently above the 97% mark.
3.2. User Study & Usability Assessment
Beyond accuracy, user studies were conducted to evaluate practicality. Participants reported:
- High Convenience: Leveraging an existing, ubiquitous device (charger) eliminated the need for additional sensors.
- Strong Privacy Perception: Unlike cameras, the system does not capture visual or biometric data, only abstract EM perturbations.
- Ease of Use: Gestures were found to be intuitive and easy to perform in contexts like a desk or bedside table.
The study positions EMGesture as not just technically viable but also user-acceptable.
4. Technical Analysis & Framework
4.1. Mathematical Foundation & Signal Processing
The perturbation of the EM field by a conductive object (the hand) can be modeled through changes in mutual inductance and induced eddy currents. The received signal $s(t)$ can be considered as:
$s(t) = A(t) \cdot \sin(2\pi f_c t + \phi(t)) + n(t)$
where $A(t)$ is the time-varying amplitude, $f_c$ is the carrier frequency (~110-205 kHz), $\phi(t)$ is the phase, and $n(t)$ is noise. Gestures modulate $A(t)$ and $\phi(t)$. Feature extraction often involves calculating the signal's envelope $E(t)$:
$E(t) = |s(t) + j \cdot \mathcal{H}\{s(t)\}|$
where $\mathcal{H}\{\cdot\}$ is the Hilbert transform, used to obtain the analytic signal for envelope detection.
4.2. Analysis Framework: A Non-Code Case Study
Scenario: Controlling a smart desk lamp (on/off, dim up/down) using gestures over its integrated wireless charging pad.
- Signal Flow: User performs a "circle" gesture. The hand movement alters the local EM field of the charging coil.
- Data Pipeline: An ADC on the charger's control board samples the coil's current/voltage feedback (data already monitored for charging control).
- Feature Vector Creation: The sampled 500ms window is processed. A 1D-CNN model extracts spatial-temporal features: e.g., a spike in low-frequency spectral power followed by a cyclic amplitude pattern.
- Classification & Action: The model matches the feature vector to the "circle" class with 98% confidence. The system translates this to the command: "Cycle through lamp color temperatures."
- Robustness Check: The system ignores minor perturbations (like a phone being placed on the pad) by checking if the signal pattern matches a known charging device signature before enabling gesture mode.
This framework highlights the seamless integration of sensing into an existing function.
5. Discussion & Future Directions
Core Insight: EMGesture isn't just another gesture tech—it's a masterclass in infrastructure repurposing. The authors have identified a pervasive, silent data source (the Qi EM field) and turned a power delivery component into a contextual sensor. This moves beyond adding sensors to leveraging what's already there, a principle crucial for sustainable and scalable ubiquitous computing, as championed by Mark Weiser's original vision.
Logical Flow & Comparison: The argument is compelling: cameras are intrusive and power-hungry, touch fails in messy environments, voice is noisy. EM signals are always-on, low-power, and abstract. Compared to other RF-based methods like Wi-Fi or radar (e.g., Google's Soli), EMGesture's strength is its constrained, predictable environment (the near-field of a coil), which simplifies signal processing and boosts accuracy, as evidenced by the 97%+ result—often higher than early Wi-Fi sensing work reported in publications like ACM MobiCom.
Strengths & Flaws: The killer app is its privacy-by-design and zero marginal hardware cost for devices with Qi charging. However, let's be critical: The range is severely limited (a few cm), making it a "desk-top" or "bedside" interaction, not a room-scale one. The gesture vocabulary is likely small and simple. It also depends on the charger being active, which may not always be the case. There's a potential conflict between optimal charging alignment and gesture ergonomics.
Actionable Insights & Future Directions: 1. Standardization Push: The real win is getting Qi 2.0 or future standards to include a dedicated, low-bandwidth sensing channel alongside power transfer. Chipmakers like NXP and IDT should take note. 2. Context-Aware Fusion: Future systems shouldn't rely on EM alone. Fusing its intent signals with a device's accelerometer (for "pick-up" detection) or microphone (for a voice confirmation) could create robust, multi-modal commands. 3. Expanded Vocabulary: Research should explore more complex, 3D gestures using multi-coil charger arrays, potentially enabling sign language alphabets over a charging pad. 4. Biometric Side-Channel: Could the unique capacitive coupling of a user's hand provide a passive, continuous authentication signal while the phone charges? This merges interaction with security.
In conclusion, EMGesture provides a brilliantly pragmatic path forward. It won't replace cameras or touchscreens, but it carves out a vital niche for ambient, casual, and private interaction in the personal device sphere, turning a mundane act—charging—into an opportunity for connection.
6. References
- Wang, W., Yang, L., Gan, L., & Xue, G. (2025). The Wireless Charger as a Gesture Sensor: A Novel Approach to Ubiquitous Interaction. In Proceedings of CHI Conference on Human Factors in Computing Systems (CHI '26).
- National Highway Traffic Safety Administration (NHTSA). (2023). Distracted Driving Fatality Data.
- Zhang, C., et al. (2020). A Survey on Vision-Based Human Activity Recognition. Image and Vision Computing, 103.
- Grand View Research. (2023). Human Machine Interface Market Size Report, 2023-2030.
- Malkin, N., et al. (2019). Privacy and Security in Voice-Based AI. IEEE Security & Privacy.
- Zhu, H., et al. (2021). Touchscreens in Wet Conditions: A Review. International Journal of Human-Computer Studies.
- Weiser, M. (1991). The Computer for the 21st Century. Scientific American.
- Pu, Q., et al. (2013). Whole-Home Gesture Recognition Using Wireless Signals. In Proceedings of ACM MobiCom.
- Wireless Power Consortium. (2023). Qi Wireless Power Transfer System Specification.