Acoustic Interaction for Compact Wearables

🚀 Overview

I led the research and prototyping of a novel input mechanism for next-generation wearable devices. The project focused on enabling discrete interaction on ultra-compact form factors without relying on physical buttons or traditional touch interfaces.

• Status: Awarded Samsung A2 Patent Grade (Internal Innovation Award) & Selected for Patent Filing.
• Role: Lead Researcher & Software Engineer.

❓ Why?

As wearables shrink, display real-estate vanishes. I realized that current ultra-compact devices generally rely on voice (not private) or active buttons (bulky).

I observed that people naturally tap surfaces to fidget or signal. I wondered: Could we repurpose the device’s internal acoustic sensors—not to hear voice, but to ‘feel’ these mechanical activities?

💡 The Approach

Due to Samsung Research IP regulations, specific technical implementations and sensor architectures are omitted.

I hypothesized that structure-borne signals could serve as a proxy for user intent. Instead of adding new hardware, I developed a software-defined sensing pipeline that repurposes existing embedded sensors to detect micro-mechanical interactions (e.g., taps, gestures) on the device chassis.

My contribution focused on the end-to-end pipeline:

• Data Strategy: Designed protocols to collect samples of intended user interactions vs. environmental noise.
• Edge AI: Developed and optimized a lightweight classification model to run locally on the device, ensuring privacy and low latency.

🏆 Impact

• Innovation: Validated that software-only solutions can establish a robust input channel on constrained surfaces.
• Recognition: The project passed internal novelty searches and was awarded an A2 Innovation Grade by the Samsung.
• Feasibility: Achieved high detection accuracy in controlled user studies.

🧠 Retrospective: Key Learnings

This project represented a pivotal evolution in my skillset, bridging the gap between theoretical ML and embedded reality.

• Technical Upskilling: Initially limited in signal theory, I aggressively upskilled in spectrogram analysis and MFCC feature extraction, eventually mentoring my team on these implementations.
• Edge Optimization: Faced with extreme power constraints, I moved beyond standard training to master knowledge distillation and network pruning, optimizing our models for low-powered wearable chipsets.
• Strategic Growth: Beyond technical execution, I developed critical IP strategy skills—learning to defend the commercial viability of novel input modalities against scrutiny from the Patent Council, which ultimately secured the project’s A2 Grade.