Scholarlink

<p>Photonic reservoir computing (RC) based on semiconductor lasers (SLs) has gained significant attention due to its simplicity, fast training, and low energy consumption. However, the reliance on offline masking at the input layer limits its application in real-time tasks. In this work, we propose and numerically investigate a real-time masking method for photonic RC system using an SL subjected to optical feedback and sinusoidal modulation. The mask signal is injected into a Mach–Zehnder modulator alongside input data, enabling real-time masking. We analyze the nonlinear dynamics of the mask-SL and identify optimal parameters for mask signal generation. The system demonstrates robust predictive performance for the Santa Fe time series (minimum NMSE = 0.033) and high memory capacity (MC), with MC values reaching up to 20 in specific parameter regions. Although the real-time masking RC system exhibits slight performance degradation compared to conventional offline masking (NMSE = 0.01, MC = 23) and requires additional parameter tuning for mask generation, it provides a viable solution for real-time data processing in photonic RC implementations.</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

Real-time masking in photonic reservoir computing based on semiconductor lasers

  • 期刊论文

  • 发表日期：

    2025年6月30日

查看全文
  • Jiahao Chen,
  • Yushuang Hou,
  • Xiaoyu Guo,
  • Qiudi Li,
  • Siyu Li,
  • Chuanlong Guo,
  • Dianzuo Yue

摘要

Photonic reservoir computing (RC) based on semiconductor lasers (SLs) has gained significant attention due to its simplicity, fast training, and low energy consumption. However, the reliance on offline masking at the input layer limits its application in real-time tasks. In this work, we propose and numerically investigate a real-time masking method for photonic RC system using an SL subjected to optical feedback and sinusoidal modulation. The mask signal is injected into a Mach–Zehnder modulator alongside input data, enabling real-time masking. We analyze the nonlinear dynamics of the mask-SL and identify optimal parameters for mask signal generation. The system demonstrates robust predictive performance for the Santa Fe time series (minimum NMSE = 0.033) and high memory capacity (MC), with MC values reaching up to 20 in specific parameter regions. Although the real-time masking RC system exhibits slight performance degradation compared to conventional offline masking (NMSE = 0.01, MC = 23) and requires additional parameter tuning for mask generation, it provides a viable solution for real-time data processing in photonic RC implementations.