An Adaptive S-Method to Analyze Micro-Doppler Signals for Human Activity Classification

Area of Science:

• Signal Processing
• Radar Systems Engineering
• Machine Learning Applications

Background:

• Time-frequency analysis is crucial for radar signal processing.
• Traditional methods struggle with cross-term interference in multi-component signals.
• Accurate signal separation is needed for advanced radar applications.

Purpose of the Study:

• To introduce the multiwindow Adaptive S-method (AS-method) for enhanced radar signal time-frequency analysis.
• To improve the suppression of cross-terms and concentration of auto-terms.
• To enable effective micro-signal extraction for subsequent classification.

Main Methods:

• Utilized orthogonal Hermite functions for high time-frequency resolution.
• Implemented a multiwindow approach with adaptive window length variation.
• Applied threshold segmentation and envelope extraction for signal isolation.
• Employed a Support Vector Machine (SVM) classifier for motion state recognition.

Main Results:

• The AS-method effectively suppresses cross-terms while maintaining auto-term concentration.
• Achieved successful separation of multi-component radar signals.
• Extracted micro-signals for feature-based classification.
• The trained SVM classifier demonstrated 95.4% accuracy in detecting human subjects in two interference-free scenarios.

Conclusions:

• The proposed multiwindow AS-method offers a superior compromise for time-frequency analysis of radar signals.
• This technique facilitates multi-component signal separation and micro-signal extraction.
• The method shows significant potential for accurate human motion detection and classification using radar data.