Signal estimation and filtering from quantized observations via adaptive stochastic resonance

Area of Science:

• Signal Processing
• Machine Learning
• Information Theory

Background:

• Single-bit quantizers are crucial in large-scale networks but face challenges in signal estimation and filtering.
• Traditional methods often struggle with noise management in complex nonlinear systems.

Purpose of the Study:

• To develop and evaluate a gradient-based learning algorithm for signal estimation and filtering in large-scale summing networks of single-bit quantizers.
• To investigate the role of adaptively updated, intentionally injected noise in improving signal processing performance.

Main Methods:

• A gradient-based learning algorithm was employed to adjust network weights and adaptively update the level of injected noise.
• The algorithm was tested in a large-scale summing network architecture.
• Mean-squared error was used as the primary metric for evaluating performance.

Main Results:

• Minimizing mean-squared error necessitates a non-zero optimal level of added noise.
• The adaptive optimization of noise levels leads to a form of stochastic resonance or noise-aided signal processing.
• The proposed method effectively enhances signal estimation and filtering in the studied network.

Conclusions:

• Adaptive optimization of injected noise is a viable strategy for improving signal processing in complex systems.
• The developed algorithm extends the application of adaptive stochastic resonance to challenging nonlinear signal processing tasks.
• This noise-aided approach offers a novel perspective for enhancing the performance of large-scale signal processing networks.