Scholarlink

In numerous studies, orthogonal time-frequency space (OTFS) has been utilized in satellite-terrestrial communication systems with high mobility due to its strong adaptability to Doppler shifts and delays. Signal detection, a crucial technology influencing OTFS performance, is often hindered by high complexity or suboptimal detection capabilities. To address this, we propose a Kolmogorov-Arnold Networks (KANs)-based OTFS signal detection method for low Earth orbit satellite (LEO-Sat) communication systems. KANs employ learnable activation functions instead of traditional learnable linear weights between network nodes, enabling dynamic activation functions to enhance model accuracy. We integrate KANs, trained offline, into the LEO-Sat system to recover distorted signals at the receiver, facilitating effective signal detection. Compared to conventional detection algorithms, the proposed method demonstrates superior Bit Error Rate (BER) performance. Additionally, KANs are more parameter-efficient than DNNs, using significantly fewer parameters.

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

Kolmogorov-Arnold Networks Based Signal Detection for OTFS Systems in LEO Satellite Communications

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yakai Zhang,
  • Jiayi He,
  • Zhiyong Liu

摘要

In numerous studies, orthogonal time-frequency space (OTFS) has been utilized in satellite-terrestrial communication systems with high mobility due to its strong adaptability to Doppler shifts and delays. Signal detection, a crucial technology influencing OTFS performance, is often hindered by high complexity or suboptimal detection capabilities. To address this, we propose a Kolmogorov-Arnold Networks (KANs)-based OTFS signal detection method for low Earth orbit satellite (LEO-Sat) communication systems. KANs employ learnable activation functions instead of traditional learnable linear weights between network nodes, enabling dynamic activation functions to enhance model accuracy. We integrate KANs, trained offline, into the LEO-Sat system to recover distorted signals at the receiver, facilitating effective signal detection. Compared to conventional detection algorithms, the proposed method demonstrates superior Bit Error Rate (BER) performance. Additionally, KANs are more parameter-efficient than DNNs, using significantly fewer parameters.