Acta Optica Sinica, Volume. 33, Issue 5, 506002(2013)

Adaptive Blind Equalization Using Electrical Recurrent Neural Networks

Ruan Xiukai1,2、*, Li Chang1, Tan Yanhua1, and Zhang Yaoju1,2
Author Affiliations
  • 1[in Chinese]
  • 2[in Chinese]
  • show less

    To solve the special issue of electrical adaptive blind equalization for wireless spatial diversity optical coherent receivers, a novel electrical adaptive blind equalization method based on dynamically driven recurrent neural networks (DDRNNs) with in-phase/quadrature amplitude component continuous activation is proposed. A multi-threshold continuous amplitude sinusoidal type activation function is designed. This activation function is simple and flexible. How to select the parameters of the activation function is analyzed in detail and the new concepts of the approach point and leaving point are proposed. Amplification factor selection range of blind equalization is analyzed using DDRNN about electrical adaptive blind equalization from the perspective of activation function. Then, the applicability of detecting low-order quadrature amplitude modulation (QAM) signals using higher order QAM activation function is discoursed. Furthermore, the new energy function with the condition of the multi-threshold continuous activation function is proposed and proven. The design and analysis of the activation function and energy function are not only suitable for the blind equalization issues, but also can be extended to other fields.

    Tools

    Get Citation

    Copy Citation Text

    Ruan Xiukai, Li Chang, Tan Yanhua, Zhang Yaoju. Adaptive Blind Equalization Using Electrical Recurrent Neural Networks[J]. Acta Optica Sinica, 2013, 33(5): 506002

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Fiber Optics and Optical Communications

    Received: Oct. 29, 2012

    Accepted: --

    Published Online: Apr. 16, 2013

    The Author Email: Xiukai Ruan (ruanxiukai@wzu.edu.cn)

    DOI:10.3788/aos201333.0506002

    Topics