Scholarlink

<p>Non-orthogonal multiple access (NOMA) and index modulation (IM) are expected to play important roles in future wireless networks. In downlink transmission, multiple symbols are superimposed with different power levels and transmitted, which pose a major challenge for reliable symbol detection. The successive interference cancellation (SIC) method is commonly applied to recover symbols for the IM-NOMA system and it relies on interference cancellation to successively detect multiple symbols. However, the performance of the SIC detector depends heavily on precise knowledge of the channel model and channel state information (CSI), which is not always perfect at the receiver. To overcome this limitation, we introduce a deep learning-aided receiver called DeepIM-SIC which can perform a joint detection for both indices and conventional <i>M</i>-ary symbols in a data-driven manner to improve symbol detection performance in the presence of CSI uncertainty compared to conventional detectors. Our approach employs dedicated deep neural networks to replace the interference cancellation blocks of the traditional SIC algorithm, resulting in a data-driven implementation of the SIC algorithm. Thus, DeepIM-SIC can learn to detect IM-based superimposed symbols in the downlink of IM-NOMA systems while being less sensitive to CSI uncertainty compared to its model-based counterpart. Our numerical evaluations show that for linear channels with perfect CSI, DeepIM-SIC performs similarly to conventional detectors. More importantly, our DeepIM-SIC significantly outperforms model-based methods in the presence of CSI uncertainty.</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

Deep learning detector for downlink IM-NOMA

  • 期刊论文

  • 发表日期：

    2025年6月14日

查看全文
  • Toan Gian,
  • Ngoc-Hung Pham,
  • Van-Cuong Pham,
  • Tien-Hoa Nguyen,
  • Trung Tan Nguyen,
  • Thien Van Luong

摘要

Non-orthogonal multiple access (NOMA) and index modulation (IM) are expected to play important roles in future wireless networks. In downlink transmission, multiple symbols are superimposed with different power levels and transmitted, which pose a major challenge for reliable symbol detection. The successive interference cancellation (SIC) method is commonly applied to recover symbols for the IM-NOMA system and it relies on interference cancellation to successively detect multiple symbols. However, the performance of the SIC detector depends heavily on precise knowledge of the channel model and channel state information (CSI), which is not always perfect at the receiver. To overcome this limitation, we introduce a deep learning-aided receiver called DeepIM-SIC which can perform a joint detection for both indices and conventional M-ary symbols in a data-driven manner to improve symbol detection performance in the presence of CSI uncertainty compared to conventional detectors. Our approach employs dedicated deep neural networks to replace the interference cancellation blocks of the traditional SIC algorithm, resulting in a data-driven implementation of the SIC algorithm. Thus, DeepIM-SIC can learn to detect IM-based superimposed symbols in the downlink of IM-NOMA systems while being less sensitive to CSI uncertainty compared to its model-based counterpart. Our numerical evaluations show that for linear channels with perfect CSI, DeepIM-SIC performs similarly to conventional detectors. More importantly, our DeepIM-SIC significantly outperforms model-based methods in the presence of CSI uncertainty.