Wireless Charger EMGesture

• Wireless Charger EMGesture is a contactless sensing technology that repurposes EM fields from wireless chargers to detect and classify mid-air gestures.
• It employs both Qi-compliant single-coil and multi-coil architectures along with advanced signal processing techniques, such as FFT and VMD, and machine learning classifiers to achieve high recognition accuracy.
• The technology offers robust, privacy-preserving control for smart home, automotive, and kiosk interfaces, providing low latency and cost-effective deployment.

Wireless Charger EMGesture refers to a class of contactless human-computer interaction techniques that repurpose the electromagnetic (EM) environment of wireless charging hardware for sensing and classifying mid-air gestures. These systems utilize the near-field EM perturbations produced by a user's hand or body in proximity to wireless charging coils and employ advanced signal processing and machine learning methods to extract, denoise, and interpret gesture-related features. EMGesture approaches encompass both single-coil, Qi-compliant charging pads for 3D mid-air gesture sensing and multi-coil configurations for high-precision surface gesture recognition, achieving recognition performance and usability competitive with or exceeding camera- and WiFi-based alternatives while maintaining advantages in privacy, cost, and environmental robustness (Wang et al., 21 Nov 2025, Zhang, 27 Jan 2025).

1. Fundamentals of Wireless Charger EMGesture

EMGesture systems leverage the fact that a wireless charger—typically operating in the 100–200 kHz range for Qi-compliant hardware—creates a strong, controllable near-field EM environment. When a user performs a gesture above or on the charger surface, their hand modulates the coupling between the transmitter (Tx) and receiver (Rx) coils (and, if present, additional sensing antennas or coils) via changes in mutual inductance and local field distributions. These perturbations manifest as amplitude and phase variations in the charger’s EM signals, which can be measured and digitized for subsequent analysis (Wang et al., 21 Nov 2025, Zhang, 27 Jan 2025).

Key characteristics:

• Contactless Sensing: Operates without requiring touch or line of sight.
• Privacy Preservation: No image or user-identifying data is acquired.
• Ubiquitous Hardware Base: Relies on existing or easily modified Qi and multi-coil charging modules.

2. System Architectures and Sensing Modalities

Qi-Based Single-Coil 3D Gesture Sensing

A typical system comprises:

• A Qi wireless charger as Tx, operating in closed-loop feedback mode to regulate resonant coil current.
• A receiver (usually a smartphone) magnetically coupled to the Tx coil.
• An external, wideband antenna positioned above the coil to sense field disturbances.
• Signal amplification and digitization, often using an SDR (e.g., HackRF One) at high sampling rates (20 MHz).
• Real-time or batch processing on a general-purpose host (Wang et al., 21 Nov 2025).

User gestures in the vertical column above the charging pad (12–30 cm) scatter and attenuate the field, producing measurable I/Q baseband shifts suitable for 3D, mid-air gesture recognition.

Multi-Coil Surface Gesture Tracking

Another modality employs a multi-coil array (e.g., a 4×4 or 5×5 grid) beneath the charging surface:

• Each planar induction coil provides localized sensing.
• Energy sensor modules extract real-time current and voltage from each coil.
• Coils are excited in sequence and time-multiplexed for spatial discrimination.
• An MCU digitizes and forwards all sensor signals to a central processor (Zhang, 27 Jan 2025).

Gestures performed above the pad modulate the EM response across the coil array, enabling 2D gesture trajectory reconstruction and finely localized mid-air or surface touchless interaction.

3. Signal Modeling and Feature Extraction

EM Environment Modeling

In the near field of a charging coil, the magnetic field $H(r, t)$ is approximated as

$$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$

under a magneto-quasi-static assumption ($<$1 MHz operation). Maxwell’s equations simplify to neglect displacement currents, focusing on the modulation of mutual inductance $M$ between Tx/Rx or Tx/sensing antenna driven by hand proximity and orientation (Wang et al., 21 Nov 2025).

Multi-coil systems model the current shift induced by a nearby conductive object as

$$\Delta I_i(t) \approx \frac{M_i}{|Z|}\frac{dI_{\rm drive}}{dt}$$

with the lateral coupling $k_i(x,y)$ parametrized as an exponential function of hand position (Zhang, 27 Jan 2025).

Signal Processing Pipelines

Qi EMGesture (Single-Coil, SDR-Based)

1. Complex-Signal Reconstruction: From I/Q SDR samples, form $s[n] = I[n] + jQ[n]$.
2. Windowed FFT Averaging: Segment $s[n]$ (0.5 s windows, $N=10^7$ samples) into short ($0.01$ s) sub-windows; compute and average per-window power spectra to yield a 200,000-dimensional APS feature vector.
3. Noise Reduction: Apply Variational Mode Decomposition (VMD) on APS, compute mode-wise spectral subtraction versus a noise baseline, then sum denoised modes for feature input (Wang et al., 21 Nov 2025).

Multi-Coil Pad EMGesture

1. Noise Smoothing: Moving-average smoothing per channel.
2. Temporal Sorting: Ensures temporal coherence across the coil array.
3. High-Pass Filtering: Removes low-frequency drift with a first-order digital filter.
4. Windowing: Segments filtered data into overlapping vectors for recognition (Zhang, 27 Jan 2025).

4. Machine Learning and Classification

Random Forests for Spectral Classifiers

For Qi EMGesture, the feature vector (denoised APS) is input into a random forest (RF):

• 100-tree ensemble, no depth constraint, node splits by $$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$0 random features ($$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$1).
• No further dimensionality reduction or explicit regularization is required, due to the RF’s implicit feature selection.
• RF outperforms SVM (RBF kernel with 100 PCA components) and KNN, yielding up to 97.59% accuracy on a held-out test set—relative improvements of over 10% to 30% versus classical baselines (Wang et al., 21 Nov 2025).

Probabilistic Tracking in Multi-Coil Pads

• Bayesian Network: Models gesture state transitions and observation likelihoods via learned CPTs and inference by forward filtering.
• Particle Filter: Tracks continuous gesture trajectories such as $$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$2, using state-space prediction and measurement updates with online resampling.
• The joint PF+BN framework achieves 94.6% accuracy, exceeding the performance of SVM (54.3%), MLP (62.1%), and CNN (68.5%) classifiers by a 0.73 relative margin (Zhang, 27 Jan 2025).

5. Experimental Results and Performance

Qi EMGesture (Single-Coil)

• 30 participants, 10 smartphones, 5 Qi chargers, 9 gesture classes (8 hand shapes + “no gesture”).
• Training set: 414 samples from 0.5 s windows; test set: held-out 20%.
• Overall RF accuracy: 97.59%; per-class accuracy $$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$395% (confusion only between similar shapes).
• Robustness: Mean user accuracy, 94.5% (min 88.7%); mean phone accuracy, 92.6%; charger model accuracy, 93.7–95.4%; real-world environments, $$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$494%.
• Ablation: Denoising (VMD+subtraction) improves accuracy from 90.77% to 99.2%.
• Operational range: experimentally 12 cm to 2 m, with optimal usability in the 0–30 cm region (Wang et al., 21 Nov 2025).

Multi-Coil Pad

• 10 participants, 16-coil (4×4) pad, 8 gestures (swipes, taps, circles), total 4000 gesture instances.
• 5-fold cross-validation: 94.6% (PF+BN), 73% (BN-only), error rate $$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$55% after 50 iterations.
• Works in darkness and with occlusions; limited to surface or near-surface interaction area ($$H(r, t) \approx \frac{I(t) \cdot \hat{a} \times r}{2\pi |r|^2}$$620 cm²) (Zhang, 27 Jan 2025).

6. Usability, Deployment, and Applications

• Usability: 80% of Qi EMGesture users rated it novel and easy to use; 90% learned gestures in under 2 minutes. Over 50% described the interaction as very natural; 75% would use it in smart home/in-car settings (Wang et al., 21 Nov 2025).
• Latency: End-to-end windowing, FFT, and RF inference produce control latencies of tens of milliseconds, supporting responsive interfaces.
• Deployment: No perceptible power draw on the charger; all sensing elements can be co-located in the charger housing for an integrated product.
• Application Domains: Smart homes (HVAC/light control), vehicular interfaces, public kiosks, and office desks for touchless command and menu navigation. Environmental adaptability is high, as functionality is robust to lighting and visual obstructions (Zhang, 27 Jan 2025).

7. Limitations and Future Perspectives

Key limitations include:

• For Qi EMGesture, external SDRs are currently required, and the gesture lexicon is constrained by the hardware geometry and spectral separability.
• Multi-coil systems are inherently limited in spatial coverage (surface area) and vertical (z-axis) discrimination.

Research directions include:

• Integration: Miniaturizing and integrating antennas, amplifiers, ADC, and embedded classifiers within the charging hardware itself (Wang et al., 21 Nov 2025).
• Gesture Set Expansion: Ergonomic design of larger lexicons while mitigating user fatigue.
• Modeling: Exploring deep denoising networks, attention-based architectures, and richer transformer-type models for feature learning, subject to resource constraints.
• Bidirectional EM Signaling: Modulating receiver (e.g., smartphone CPU) load to inject frequency-coded tones into the Qi EM field for active feedback or more complex HCI (Wang et al., 21 Nov 2025).
• Generalization: Extensive user studies, diverse environments, and intricate scenes to probe real-world generalizability.
• Multi-Coil Extensions: Improvements in signal processing (filtering, CPT structure), higher coil densities, and frequency-multiplexed channels for finer gesture granularity (Zhang, 27 Jan 2025).

The Wireless Charger EMGesture paradigm demonstrates that ambient EM signals from ubiquitous wireless chargers can be harnessed for robust, privacy-preserving, and cost-effective gesture-based interaction in pervasive computing environments (Wang et al., 21 Nov 2025, Zhang, 27 Jan 2025).