Newton Recursion Based Random Data-Reusing Generalized Maximum Correntropy Criterion Adaptive Filtering Algorithm.

Area of Science:

• Signal Processing
• Adaptive Filtering
• Machine Learning

Background:

• Gradient-based generalized maximum correntropy criterion (GB-GMCC) offers good performance in impulsive noise but suffers from slow convergence with correlated signals.
• Gradient methods rely on first-order derivatives, potentially leading to suboptimal convergence points (maxima, minima, or saddle points).

Purpose of the Study:

• To develop a robust adaptive filtering algorithm that overcomes the limitations of GB-GMCC, particularly in terms of convergence rate and accuracy.
• To enhance filtering performance in impulsive-noise environments through improved data utilization and update mechanisms.

Main Methods:

• Proposes the Newton recursion-based data-reusing GMCC (NR-DR-GMCC) algorithm, utilizing second-order derivative information via Newton recursion.
• Incorporates a data-reusing method, employing the latest M input vectors to boost convergence.
• Introduces the Newton recursion-based random data-reusing GMCC (NR-RDR-GMCC) by adding a random strategy to further leverage past M input vectors.

Main Results:

• Simulation results demonstrate that NR-RDR-GMCC outperforms existing algorithms in system identification and acoustic echo cancellation.
• The proposed NR-RDR-GMCC algorithm shows significant improvements in both filtering accuracy and convergence rate.
• The random strategy effectively extracts more information from historical data, enhancing overall performance.

Conclusions:

• NR-RDR-GMCC provides a superior robust adaptive filtering solution compared to traditional methods.
• The algorithm is effective in challenging environments characterized by impulsive noise and highly correlated input signals.
• The study validates the benefits of combining Newton recursion, data-reusing, and random strategies for adaptive filtering.