Generation-Model Noise Waveform Removal for Wearable Movement Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Wearable devices struggle to accurately determine user movement states due to noise in movement waveform data caused by non-landing impacts, such as those from carrying the device in a pouch, which interferes with correct analysis.
Innovation Solution
A noise waveform removing device uses a trained generation model to determine a filter contribution rate and generate noise-free movement waveform data by applying low-pass filtering, while a model training device trains the generation model using a GAN to improve accuracy and avoid mode collapse.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If movement waveform data is collected from wearable devices, then user movement states can be detected, but noise from non-landing impacts reduces measurement precision
Solution Approach 1:
The patent segments the movement waveform data into different frequency components using Fourier transform, separating signal from noise. The generation model then processes these frequency components to identify and remove noise while preserving valid movement information.
Solution Approach 2:
The patent introduces a generation model as an intermediary between the raw movement waveform data and the final analysis. This model generates filtered waveform data by learning the characteristics of noise from training data, acting as a mediator that removes harmful factors while preserving useful information.
2Object-affected harmful factors
If traditional filtering methods are applied to remove noise, then noise reduction is achieved, but desired movement waveforms may be distorted
Solution Approach 1:
The patent changes the approach from fixed filtering parameters to dynamic parameter adjustment. The generation model learns optimal filtering parameters from training data and adapts them based on the input waveform characteristics, allowing effective noise removal while preserving waveform accuracy through parameter optimization.
Solution Approach 2:
The patent employs feedback mechanisms where the generation model is trained using the original waveform data as ground truth. The model continuously adjusts its filtering parameters based on the difference between generated and actual clean waveforms, ensuring that desired movement patterns are preserved while noise is removed.
3Object-affected harmful factors
If simple noise filtering is used, then processing is fast and simple, but noise removal effectiveness is insufficient
Solution Approach 1:
The patent performs preliminary action by pre-training the generation model offline using large amounts of training data. This preliminary training phase captures noise characteristics and filtering strategies, so that during actual use, the model can quickly process new waveform data without requiring complex real-time computations, balancing effectiveness with processing efficiency.
Data Source
AI summary
A noise waveform removing device includes one or more processors that are configured to obtain noise-containing movement waveform data; use a trained generation model, to determine a filter contribution rate based on the noise-containing movement waveform data; and generate noise-free movement waveform data, based on the determined filter contribution rate and the noise-containing movement waveform data.