SIMO FSO通信における計算効率の高い自己注意支援信号検出方法と大気乱流

Atmospheric turbulence induces severe performance degradation in free-space optical (FSO) communication systems. While spatial diversity techniques can effectively mitigate turbulence-induced impairments, the selection of diversity combining and signal detection methods plays a critical role in determining system performance. Although deep learning (DL)-based combining approaches show promise in compensating for turbulence distortions, their practical implementation is constrained by excessive computational complexity. This paper presents a novel low-complexity self-attention neural network (SANN) for efficient signal combining and detection in single-input multiple-output (SIMO) FSO systems operating under turbulent channels. The proposed SANN employs lightweight self-attention mechanisms to adaptively focus on crucial signal components, achieving significant computational overhead reduction compared to the convolutional neural network (CNN)-based deep learning method and conventional combining methods with minimum mean-square error (MMSE) evaluation. To make the research findings convincing, we conducted experiments to measure intensity fluctuations for more realistic simulations. Results demonstrate that our solution maintains comparable or superior bit error rate (BER) performance while reducing computational complexity significantly by over 70% relative to the existing CNN-based detector. This work establishes a practical framework for computationally efficient signal detection in SIMO FSO systems where processing efficiency is paramount.