Additive and multiplicative noise reduction by back propagation neural network

Area of Science:

• Signal Processing
• Artificial Intelligence
• Biomedical Engineering

Background:

• Noise significantly degrades signal quality in various applications.
• Existing filters often struggle to balance noise reduction with signal preservation.
• Traditional filters may exhibit performance degradation with increasing noise levels.

Purpose of the Study:

• To introduce a novel filter based on a back propagation neural network (BPNN) ensemble.
• To evaluate the filter's effectiveness in reducing additive and multiplicative white noise.
• To compare the proposed filter's performance against existing methods and analyze its behavior under varying noise conditions.

Main Methods:

• Utilizing a back propagation neural network (BPNN) ensemble with identical noisy and reference signals.
• Implementing the filter for processing simulated and real-world electroencephalogram (EEG) signals.
• Analyzing the relationship between noise reduction efficacy and noise bandwidth.

Main Results:

• The NN ensemble filter effectively reduces both additive and multiplicative white noise while preserving signal characteristics.
• Performance surpasses improved nonlinear filters and single NN filters, especially at higher noise power levels.
• The degradation in noise reduction capability with increased noise power is significantly suppressed compared to the improved nonlinear filter.

Conclusions:

• The proposed NN ensemble filter offers superior noise reduction capabilities, particularly in challenging noisy environments.
• The filter demonstrates robustness and maintains signal integrity, making it suitable for complex signal processing tasks.
• This approach provides a valuable advancement in signal denoising, with demonstrated applications in EEG signal analysis.