DD-TransNet-Based CSI Feedback in Imperfect Channel Estimation in MIMO-OTFS Systems for V2I Scenarios

With the development of 5th generation (5G) wireless communication technology and 6th generation (6G) wireless communication technology, ultra-reliable low-latency communication (URLLC) in high-speed mobile environments becomes a key challenge. Focusing on vehicle-to-infrastructure (V2I) communication, this scenario plays a crucial role in enabling real-time interactions between vehicles and roadside infrastructure. Orthogonal time frequency space (OTFS) modulation technique has emerged as a promising waveform for 6G standards because of its efficient mitigation of delay spread and Doppler impact. When multiple input multiple output (MIMO) and OTFS are integrated, the MIMO-OTFS system can enhance communication performance in dynamic scenarios. However, channel state information (CSI) feedback is a fundamental technology in a frequency division duplex (FDD) MIMO-OTFS system, and currently confronts issues of large feedback overhead and low precision. This paper proposes a deep learning-based CSI feedback method for MIMO-OTFS systems to enhance the reliability of ultra-high speed mobile communication systems. It leverages the global attention mechanism advantage of Transformer architecture and offers an effective CSI compression and reconstruction method that assists communication systems adapt more effectively to the channel characteristics in high-speed mobile environment. Firstly, this work presents an introduction to the concept of CSI feedback and its significance in MIMO-OTFS systems. Subsequently, we provide a comprehensive description of the network architecture design of Transformer-based delay-Doppler (DD) domain CSI feedback network (DD-TransNet), encompassing the precise implementations of the encoder and decoder, as well as the construction of the Transformer layer within this architecture. A diverse channel dataset in the DD domain is generated and analyzed in system simulation using the spatial channel model (SCM). This dataset contains three channel estimation algorithms: impulse response, Bernoulli orthogonal matching pursuit (OMP), and Bernoulli structured OMP. The purpose of this dataset is to simulate the non-perfect channel estimation conditions that correspond to real-world operation scenarios. Ultimately, an experimental system is used to assess how effectively DD-TransNet performs in comparison to the current approaches (CsiNet, CsiNet+, and BCsiNet). The algorithm presented in this article demonstrates greater robustness and adaptability by maintaining high CSI feedback accuracy and low feedback overhead under various channel estimation algorithms and high-speed mobile situations, in comparison to these existing methods.