Scholarlink

<p>In wavelength-division multiplexing (WDM) systems, linear and nonlinear crosstalk significantly degrade signal integrity, necessitating optimal detection threshold (ODT) estimation for minimizing bit error rate (BER). This study presents a machine learning-driven approach based on gradient descent and regression techniques to dynamically optimize ODT, eliminating the need for computationally intensive analytical methods. A WDM receiver is primarily modeled under the generalized framework of an optical network with linear crosstalk, and a closed-form BER expression incorporating a finite interferer is utilized. The proposed method establishes a direct correlation between the ODT and key system parameters, such as the number of sources of crosstalk and the level of crosstalk, allowing efficient threshold estimation. This approach is extensible to complex nonlinear crosstalk scenarios by integrating linear and nonlinear regression, ensuring adaptability to evolving network conditions. Performance validation using regression-based figures of merit (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\text {R}^{2}\)</EquationSource> </InlineEquation> values up to 0.9991 and root mean squared error, i.e., RMSE as low as 0.0003) demonstrates the model’s effectiveness, offering a scalable and computationally efficient solution for estimating adaptive thresholds in real time. These findings contribute to developing intelligent optical communication systems, enhancing the quality of transmission (QoT) and the overall network resilience.</p>

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

Intelligent detection threshold optimization in WDM systems: a machine learning approach for crosstalk mitigation

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Diptadip Chakraborty,
  • Pinakpani Mukherjee,
  • Santu Sarkar,
  • Nikhil R. Das

摘要

In wavelength-division multiplexing (WDM) systems, linear and nonlinear crosstalk significantly degrade signal integrity, necessitating optimal detection threshold (ODT) estimation for minimizing bit error rate (BER). This study presents a machine learning-driven approach based on gradient descent and regression techniques to dynamically optimize ODT, eliminating the need for computationally intensive analytical methods. A WDM receiver is primarily modeled under the generalized framework of an optical network with linear crosstalk, and a closed-form BER expression incorporating a finite interferer is utilized. The proposed method establishes a direct correlation between the ODT and key system parameters, such as the number of sources of crosstalk and the level of crosstalk, allowing efficient threshold estimation. This approach is extensible to complex nonlinear crosstalk scenarios by integrating linear and nonlinear regression, ensuring adaptability to evolving network conditions. Performance validation using regression-based figures of merit ( \(\text {R}^{2}\) values up to 0.9991 and root mean squared error, i.e., RMSE as low as 0.0003) demonstrates the model’s effectiveness, offering a scalable and computationally efficient solution for estimating adaptive thresholds in real time. These findings contribute to developing intelligent optical communication systems, enhancing the quality of transmission (QoT) and the overall network resilience.