Earphone Sensor Data Transformation Model for Wear Position Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Smart mobile accessories require re-collection and re-processing of data and re-training of identification models when design changes occur, such as changes in the position or orientation of inertial sensors, leading to inefficiencies and inaccuracies in motion identification.
Innovation Solution
An earphone system with an inertial sensor and processor that uses a pre-stored data transformation model to convert sensed data into a format compatible with existing identification models, allowing for direct application without re-training, and includes features to detect and correct incorrect wear positions through vector analysis and prompting messages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the installed position or orientation of the inertial sensor is changed in a new design, then the device can be improved with new design flexibility, but the existing identification model cannot be used and must be re-trained
Solution Approach 1:
The patent introduces a data transformation model as an intermediary between the inertial sensor and the identification model. This transformation model converts sensed data from new sensor positions/orientations into a format compatible with existing identification models, eliminating the need to re-train models when sensor configuration changes. The intermediary layer decouples the sensor hardware from the analysis algorithm, allowing independent optimization of each component.
Solution Approach 2:
The patent applies parameter changes by transforming the parameters (sensed data) from the new sensor configuration into equivalent parameters that match the original training data distribution. The data transformation model adjusts acceleration and angular velocity vectors through mathematical transformations, changing the parameter representation while preserving the underlying motion information, thereby maintaining compatibility with existing identification models.
2Ease of manufacture
If the installed position or orientation of the inertial sensor is changed, then design updates can be made, but data re-collection and re-processing are required
Solution Approach 1:
The data transformation model serves as an intermediary that handles the adaptation to new sensor configurations automatically. When the sensor position or orientation changes, the transformation model is adjusted accordingly, but the existing identification model remains unchanged. This intermediary layer absorbs the complexity of adaptation, allowing rapid design updates without sacrificing productivity in motion identification.
Solution Approach 2:
The patent implements preliminary action by pre-storing multiple data transformation models corresponding to different sensor installation positions and orientations. Before actual motion identification occurs, the appropriate transformation model is selected and applied to convert the sensed data into the correct format. This preliminary transformation ensures that the existing identification model can process data from various sensor configurations without requiring re-training, thereby maintaining high productivity.
3Loss of time
If a data transformation model is introduced to avoid re-training, then time is saved, but the system complexity increases
Solution Approach 1:
The data transformation model implements parameter changes through mathematical transformations of the sensed data. Rather than adding complex hardware components, the solution transforms the existing data parameters (acceleration and angular velocity vectors) into a format compatible with existing identification models. This software-based parameter transformation achieves the goal of avoiding re-training while adding minimal system complexity.
Data Source
AI summary
A system for transforming sensed data is provided. The system includes an inertial sensor and a processor. The inertial sensor generates first sensed data by sensing a motion or behavior of a user. The processor is communicatively connected to the inertial sensor and pre-stores a data transformation model, wherein the data transformation model converts an acceleration vector and an angular velocity vector corresponding to the first sensed data into an acceleration vector and an angular velocity vector corresponding to second sensed data, and outputs the acceleration vector and the angular velocity vector corresponding to the second sensed data to an identification model pre-stored in the processor.