Scholarlink

In order to make use of the system's sparseness and robustness to outliers, two new algorithms—the zero attraction-based logistic distance metric adaptive filter (ZA-LDMAF) and re-weighted zero attraction-based LDMAF (RZA-LDMAF)—were recently created. However, there is no controllable variable that can properly adapt the norm penalty to the system's unknown sparse finite impulse response. This motivated us to develop a non-uniform constraint LDMAF (NNC-LDMAF) which is realized by integrating a \(p\) -norm-like constraint within the LDMAF algorithm’s cost function and serves as a combination of the \({l}_{0}\) and \({l}_{1}\) -norm. This adjustment is equal to imposing on the iteration a sequence of \({l}_{0}\) and \({l}_{1}\) -norm zero attraction elements by considering the relative data of individual filter coefficients across all entries. Numerical simulations illustrate the superiority of the suggested method to varying sparsity levels and lower normalized mean square deviation value.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LDMAF Algorithm with Non-uniform Constraint for Sparse System Identification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rosalin,
  • Ansuman Patnaik,
  • Deepak Kumar Rout,
  • Satish Kumar Gannamaneni

摘要

In order to make use of the system's sparseness and robustness to outliers, two new algorithms—the zero attraction-based logistic distance metric adaptive filter (ZA-LDMAF) and re-weighted zero attraction-based LDMAF (RZA-LDMAF)—were recently created. However, there is no controllable variable that can properly adapt the norm penalty to the system's unknown sparse finite impulse response. This motivated us to develop a non-uniform constraint LDMAF (NNC-LDMAF) which is realized by integrating a \(p\) -norm-like constraint within the LDMAF algorithm’s cost function and serves as a combination of the \({l}_{0}\) and \({l}_{1}\) -norm. This adjustment is equal to imposing on the iteration a sequence of \({l}_{0}\) and \({l}_{1}\) -norm zero attraction elements by considering the relative data of individual filter coefficients across all entries. Numerical simulations illustrate the superiority of the suggested method to varying sparsity levels and lower normalized mean square deviation value.