Deng Xiuqi, Bian Xin, Li Mingqi
Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China.
University of Chinese Academy of Sciences, Beijing 100049, China.
Sensors (Basel). 2021 Dec 30;22(1):257. doi: 10.3390/s22010257.
In recent years, Faster-than-Nyquist (FTN) transmission has been regarded as one of the key technologies for future 6G due to its advantages in high spectrum efficiency. However, as a price to improve the spectrum efficiency, the FTN system introduces inter-symbol interference (ISI) at the transmitting end, whicheads to a serious deterioration in the performance of traditional receiving algorithms under high compression rates and harsh channel environments. The data-driven detection algorithm has performance advantages for the detection of high compression rate FTN signaling, but the current related work is mainly focused on the application in the Additive White Gaussian Noise (AWGN) channel. In this article, for FTN signaling in multipath channels, a data and model-driven joint detection algorithm, i.e., DMD-JD algorithm is proposed. This algorithm first uses the traditional MMSE or ZFinear equalizer to complete the channel equalization, and then processes the serious ISI introduced by FTN through the deepearning network based on CNN or LSTM, thereby effectively avoiding the problem of insufficient generalization of the deepearning algorithm in different channel scenarios. The simulation results show that in multipath channels, the performance of the proposed DMD-JD algorithm is better than that of purely model-based or data-driven algorithms; in addition, the deepearning network trained based on a single channel model can be well adapted to FTN signal detection under other channel models, thereby improving the engineering practicability of the FTN signal detection algorithm based on deepearning.
近年来，超奈奎斯特（FTN）传输因其在高频谱效率方面的优势，被视为未来6G的关键技术之一。然而，作为提高频谱效率的代价，FTN系统在发射端引入了符号间干扰（ISI），这导致在高压缩率和恶劣信道环境下传统接收算法的性能严重恶化。数据驱动检测算法在高压缩率FTN信号检测方面具有性能优势，但目前相关工作主要集中在加性高斯白噪声（AWGN）信道中的应用。本文针对多径信道中的FTN信号，提出了一种数据与模型驱动的联合检测算法，即DMD-JD算法。该算法首先利用传统的最小均方误差（MMSE）或迫零（ZF）线性均衡器完成信道均衡，然后通过基于卷积神经网络（CNN）或长短期记忆网络（LSTM）的深度学习网络处理FTN引入的严重ISI，从而有效避免深度学习算法在不同信道场景下泛化能力不足的问题。仿真结果表明，在多径信道中，所提DMD-JD算法的性能优于纯基于模型或数据驱动的算法；此外，基于单信道模型训练的深度学习网络能够很好地适应其他信道模型下的FTN信号检测，从而提高了基于深度学习的FTN信号检测算法的工程实用性。
