A nested generalized sidelobe canceller for source counting, localization, and signal separation in reverberant...

Area of Science:

• Array Signal Processing
• Acoustic Signal Analysis
• Machine Learning for Audio

Background:

• Traditional array signal processing methods struggle with accurate source counting, localization, and separation in reverberant environments.
• Dominant sources and noise interference complicate the analysis of complex acoustic signals.
• Existing techniques like Minimum Description Length (MDL) and Multiple Signal Classification (MSC) have limitations in performance.

Purpose of the Study:

• To introduce a Nested Generalized Sidelobe Canceller (NGSC) for enhanced array signal processing.
• To improve the accuracy of source counting, direction-of-arrival (DOA) estimation, and signal separation.
• To evaluate the NGSC's performance against established baseline methods.

Main Methods:

• A nested structure of blocking matrices is employed to progressively eliminate dominant sources.
• Microphone signals are dereverberated using a multichannel weighted prediction error algorithm.
• Source counting uses blocked signal power; DOA estimation employs cosine similarity and golden section search; signal separation utilizes a Linearly Constrained Minimum Variance Beamformer with Postfiltering (LCMV-PF).

Main Results:

• The NGSC achieved at least 28.80% higher source counting accuracy compared to baselines.
• Root Mean Square (RMS) degree error for DOA estimation was 1.04° lower with NGSC.
• The LCMV-PF using NGSC resulted in a 1.52 dB higher signal-to-distortion ratio (SDR) than compared beamformers.

Conclusions:

• The proposed NGSC offers significant improvements in source counting, localization, and signal separation.
• NGSC demonstrates superior performance over traditional methods in complex acoustic scenarios.
• The NGSC framework provides a robust solution for advanced array signal processing applications.