面向高可靠车联网通信的稀疏干扰消除方法研究

The furious contradiction between the nonstationary non-Gaussian and nonlinear time- and frequency- domain interference in the complicated time-variant vehicular communication systems, and the ultra-reliable and low latency (URLLC) prospects for vehicular networks, has brought tremendous challenge to traditional technologies of communications and signal processing. In order to solve the key bottleneck problem and guarantee the reliability and robustness of vehicular communications, this project introduces the recently emerging theories of machine learning and compressed sensing. The key technology architecture of sparse interference cancellation for ultra-reliable vehicular communications is formulated. The essential techniques including compressed sampling and modeling, compressed sensing based spectrum-efficient and accurate multi-dimensional recovery, and machine learning based low-latency adaptive intelligent inference, of the sparse interference, are proposed. This project overcomes many difficulties of conventional methods in passively mitigating the interference, and further breaks out the limitation of compressed sensing based methods in vehicular communications scenarios. The estimation performance of the sparse interference can be significantly improved. Meanwhile, the cost in time and frequency resources, the algorithm delay in vehicular terminals, and the implementation complexity can be reduced. Moreover, the reliability of the vehicular communications system in the presence of severe noise and interference can be enhanced. The research of this project will provide theoretical basis and technological preparation for the research on the next-generation intelligent URLLC vehicular communications and networks.

复杂多变的车联网通信系统中存在的非平稳、非高斯、非线性时频稀疏干扰，与低延时、高可靠的车联网愿景形成尖锐矛盾，对传统通信与信号处理技术构成巨大挑战。为解决这一瓶颈问题，保障车联网通信的可靠性和鲁棒性，本项目引入新兴的机器学习与压缩感知理论，形成面向高可靠车联网通信的稀疏干扰消除关键技术体系，提出稀疏干扰压缩采样建模、基于压缩感知的高谱效高精度稀疏干扰多维联合重构、基于机器学习的低延时自适应稀疏干扰智能预测等关键技术。本项目克服传统方法被动抑制稀疏干扰的诸多缺陷，进一步克服压缩感知方法在车联网场景中的局限，能够显著提升稀疏干扰估计性能，并降低时频资源开销、车载终端算法时延和实现复杂度，增强车联网通信系统在复杂恶劣噪声干扰下的可靠性，有望为低延迟、高可靠、智能化的下一代车联网通信与网络研究提供理论基础和技术储备。

复杂多变的车联网通信系统中存在的非平稳、非高斯、非线性时频稀疏干扰，车载环境无线信道条件较为恶劣，具有复杂高速移动、时变和稀疏等特征。复杂恶劣的无线信道环境和特殊的干扰，与低延时、高可靠的车联网愿景形成尖锐矛盾，对传统通信与信号处理技术带来挑战。为此，本项目引入了压缩感知理论和新兴的深度神经网络与机器学习等人工智能理论技术，提出了基于多维压缩感知的车载通信时频稀疏干扰重构方法，进而构建了基于深度神经网络和机器学习的稀疏干扰智能预测理论技术框架，围绕着提升车联网、物联网等无线通信网络中系统传输的性能、效率、鲁棒性、可靠性和安全性的目标，主要开展了稀疏信道智能估计、干扰对抗与控制、稀疏信号感知与恢复等三个方面的代表性研究工作。具体地，提出了基于深度压缩感知和稀疏学习的稀疏信道估计方法，有效提高了车联网通信大规模多天线系统稀疏信道估计精度，降低了资源开销和时间复杂度；提出了基于强化学习的无线通信干扰对抗与控制技术，提升了无人车、无人机等智能车辆在通信干扰环境下的宽带传输质量和性能，提高了超密集无线网络的容量和频谱效率、能量效率；提出了面向物联网的高效安全稀疏信号传感与传输技术，提升了车联网、医疗物联网等稀疏传感信号的传输效率，提高了信号传输的保密性和可靠性。通过本项目研究所提出的相关技术，可以帮助缓解车联网通信中时频稀疏干扰非平稳、非高斯、非线性的恶劣条件与车联网对鲁棒性、可靠性、实时性需求之间的尖锐矛盾，克服传统稀疏干扰抑制方法的瓶颈，降低车载通信终端算法时频资源开销，保障车联网等无线通信网络在恶劣干扰条件下的低延迟、高可靠、鲁棒性和安全性需求。项目研究技术有望应用于车联网通信、新一代移动通信、无线网络抗干扰、物联网等技术领域。