Machine Learning Wavelet Generation for Signal Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for time-series signal analysis, such as FFT and STFT, face limitations in capturing time-frequency information accurately, especially for sudden phenomena, and wavelet analysis with neural networks struggles with low time-frequency resolution and complex configuration conditions.
Innovation Solution
An information processing device employs a machine learning model to generate a local waveform for continuous wavelet transform, using unsupervised learning to optimize the generation function, enabling more accurate time-frequency analysis by integrating neural networks and specific function forms that satisfy allowable conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If wavelet analysis with neural networks is used, then time-frequency analysis capability is improved, but time-frequency resolution and configuration simplicity deteriorate
Solution Approach 1:
The patent changes the parameters of the wavelet function by learning optimal parameters (such as shape parameters, scale parameters) through machine learning models. This allows the wavelet analysis to adapt to different signal characteristics dynamically, improving time-frequency analysis capability while the learned parameters simplify the configuration process compared to manual tuning of complex wavelet parameters.
Solution Approach 2:
The patent replaces the traditional manual configuration and fixed wavelet function selection with an automated machine learning-based parameter learning system. The neural network automatically learns optimal wavelet parameters from data, substituting the mechanical process of manual wavelet selection and configuration with an intelligent adaptive system that simplifies usage while enhancing analysis capability.
2Speed
If traditional FFT and STFT methods are used, then computational speed is maintained, but time-frequency information accuracy for sudden phenomena deteriorates
Solution Approach 1:
The patent introduces dynamic adaptability to wavelet analysis by using machine learning to automatically adjust wavelet parameters based on the input signal characteristics. Unlike fixed-window STFT or basic wavelet transforms, this dynamic parameter adjustment allows the system to optimize time-frequency resolution for different signal types and sudden phenomena, significantly improving detection accuracy while maintaining computational efficiency through efficient learning model inference.
3Ease of manufacture
If fixed window width is used in STFT, then computational simplicity is maintained, but time-frequency analysis accuracy across multiple frequency domains deteriorates
Solution Approach 1:
The patent dynamically changes the wavelet parameters (including effective window width equivalent) based on the frequency domain of interest and signal characteristics. The machine learning model learns to adjust these parameters automatically, providing high time-frequency resolution for multiple frequency domains without requiring complex manual configuration of different window widths for different analyses.
Data Source
AI summary
According to one embodiment, an information processing device includes one or more processors. The one or more processors are configured to generate a generation function for generating a local waveform used for a continuous wavelet transform, at least part of the generation function being expressed by a machine learning model; and learn the generation function by using learning data including a first input signal.