Matrix Information Geometry for Spectral-Based SPD Matrix Signal Detection with Dimensionality Reduction.

Area of Science:

• Signal Processing
• Machine Learning
• Information Geometry

Background:

• Traditional signal detection methods struggle with complex signals at low signal-noise-ratio (SNR).
• Existing spectrogram analysis often extracts insufficient information, limiting detection performance.
• There is a need for advanced techniques to enhance feature extraction and discrimination in signal detection.

Purpose of the Study:

• To propose a novel signal detector leveraging matrix information geometric dimensionality reduction (DR).
• To improve signal detection performance, particularly for complex signals in low SNR conditions.
• To address the limitations of previous methods in extracting comprehensive information from spectrograms.

Main Methods:

• Received data are transformed into high-precision 2-D spectrograms using short-time Fourier transform (STFT).
• Global descriptors are employed to extract abundant features from spectrograms.
• Features are constructed as symmetric positive definite (SPD) matrices, enabling analysis on an SPD manifold.
• Information geometric DR techniques are applied to promote discrimination of heterogeneous samples on the SPD manifold.

Main Results:

• The proposed detector demonstrates satisfactory signal detection performance in low SNR scenarios.
• Validation was performed using both K distribution simulations and real-life sea clutter data.
• The method effectively handles complex signal detection tasks where traditional methods fall short.

Conclusions:

• The novel signal detector based on matrix information geometry and DR offers a significant advancement in signal detection.
• This technique is effective for improving detection in challenging low SNR environments.
• The proposed method has broad applicability in various signal detection fields.