A Transformer Based Signal Detection Method for Orthogonal Time Frequency Space Communication Systems

Orthogonal Time Frequency Space (OTFS) is an emerging modulation technique designed for high-speed mobile communication environments, which can effectively process signals in the delay-Doppler domain. However, OTFS faces some challenges in practice. One challenge which is emergent to address is the low robustness and high complexity of the existing signal detection methods. To this end, this paper proposes a joint channel estimation and signal detection scheme based on transformer neural networks. The proposed method achieves channel estimation and signal detection by fitting the input-output relationship of delay-Doppler (DD) domain signals without explicit channel state information (CSI). Simulation results show that the proposed transformer based approach can outperform existing methods, especially at high Doppler frequency shifts and multipath effects, which demonstrates the potential of deep learning for signal detection in OTFS systems.

I. Introduction

With the rapid development of highways and high-speed railways, the demand for communication in high-speed mobile scenarios is growing. [1]. This poses a serious challenge to tra-ditional communication modulation schemes. In these scenar-ios, higher speeds often result in higher Doppler shifts, making current 5G technologies appear inferior. The staple modulation technique utilized in modern communication systems is Or-thogonal Frequency Division Multiplexing (OFDM). OFDM achieves high spectral efficiency by exploiting orthogonality between subcarriers [2], [3]. However, in high-speed mobile scenarios, the Doppler shift destroys the orthogonality between subcarriers, thus causing inter-code interference. Orthogonal Time Frequency Space (OTFS) was created to address this issue [4]. The technique's core concept involves processing the signal in the delay-Doppler (DD) domain to effectively address the multi-path delay and Doppler frequency shift issues in fast time-varying channels [5].