Radio Wave Abnormality Detection with Multidimensional Feature Masks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting abnormalities in radio wave emissions lack uniformity and multidimensional criteria, relying heavily on personal factors and failing to identify anomalies in features other than reception level.
Innovation Solution
An abnormality detection apparatus and method using unsupervised machine learning to generate an abnormality determination mask, setting threshold values for multiple feature amounts, including Amplitude Probability Distribution (APD), to uniformly and comprehensively detect radio wave abnormalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a threshold value is set for reception level to detect abnormalities, then detection can be performed, but personal factors contribute to creation and uniform determination cannot be made
Solution Approach 1:
The patent changes the parameter from single-reception-level threshold to multi-dimensional feature amounts including amplitude probability distribution. By expanding the detection parameters to include multiple statistical characteristics of radio wave amplitudes across different frequency bands, the system achieves objective, uniform abnormality determination independent of personal factors while maintaining high detection reliability.
2Device complexity
If only reception level threshold is used, then detection is simple, but abnormalities in other feature amounts cannot be detected
Solution Approach 1:
The patent applies dimensionality change by transitioning from one-dimensional reception level threshold to multi-dimensional feature space including amplitude probability distribution across multiple frequency bands. This dimensional expansion enables detection of abnormalities in various feature amounts (amplitude distribution, frequency characteristics) while maintaining manageable system complexity through structured feature extraction and unified threshold determination.
3Measurement precision
If supervised learning is used for periodic abnormality detection, then cause estimation can be performed, but failure prediction is difficult because abnormal failures cannot be predicted and classified
Solution Approach 1:
The patent inverts the approach by using unsupervised learning to determine abnormality thresholds based on normal operation characteristics rather than training on labeled abnormal data. By establishing thresholds that define normal behavior patterns and detecting deviations from these patterns, the system achieves both precise abnormality detection and adaptability to various failure modes without requiring pre-classified abnormal examples.
Data Source
AI summary
An abnormality detection apparatus according to an example embodiment includes a reception unit for receiving radio waves, a feature amount extraction unit for extracting a plurality of feature amounts in a predetermined frequency band from the received radio waves, a recording unit for recording the plurality of extracted feature amounts and the frequency band in association with each other, and a processing unit for acquiring a plurality of feature amounts in a predetermined range from the plurality of accumulated feature amounts, determining whether or not the acquired feature amounts fall within a preset normal range, and generating, based on a result of the determination, an abnormality determination mask, threshold values for the plurality of feature amounts, in order to detect an abnormality of the radio waves being set in the abnormality determination mask.