Performance weighted blended spectrogram

Area of Science:

• Signal Processing
• Time-Frequency Analysis
• Machine Learning

Background:

• Conventional spectrograms, based on tapered short-time Fourier transforms, involve a trade-off between resolution and interference suppression.
• Adaptive spectrogram algorithms offer improved performance by adjusting filter banks to incoming data, but often require significant computational resources.
• Existing adaptive methods present challenges for applications with limited computational power, such as autonomous systems.

Purpose of the Study:

• To develop an adaptive spectrogram algorithm suitable for computationally constrained applications.
• To enhance the separation of closely spaced tones and the detection of quiet signals.
• To reduce the computational cost associated with adaptive time-frequency analysis.

Main Methods:

• The performance weighted blended (PWB) spectrogram algorithm was developed.
• PWB combines outputs from conventional filter banks with varying tapers.
• Blend weights are adapted dynamically at each time-frequency point.

Main Results:

• The PWB spectrogram effectively separates loud, closely spaced tones.
• Quiet signals are successfully identified, even in the presence of noise.
• The algorithm demonstrates significantly lower computational requirements compared to other adaptive methods.
• Analysis of underwater glider data confirmed the PWB spectrogram's ability to detect a quiet chirp signal amidst vehicle noise.

Conclusions:

• The PWB spectrogram provides an efficient adaptive solution for time-frequency analysis in resource-limited environments.
• This algorithm offers a practical approach to improving signal detection and separation without prohibitive computational overhead.
• The PWB spectrogram is well-suited for applications like autonomous vehicles and underwater acoustics.