【作者】 封泽运； 张慧； 林凤泰； 徐刚；

【Author】 Feng Zeyun;Zhang Hui;Lin Fengtai;Xu Gang;State Key Laboratory of Millimeter Waves, Southeast University;

【机构】 东南大学毫米波国家重点实验室；

【摘要】 毫米波雷达由于其全天时全天候且体积小的特点,在智能驾驶中成为车载传感器不可或缺的一部分。基于多输入多输出(Multiple-Input and Multiple-Output, MIMO)技术的毫米波雷达,可以通过多发多收形成更大的虚拟阵列,从而提高角度分辨率。基于时分多址多输入多输出(Time Division Multiple Access MIMO, TDMA-MIMO)体制设计,本文提出了一种车载毫米波雷达单帧高分辨点云成像算法。结合时分发射波形设计,研究了接收阵列时分误差校正相干化处理方法。在阵列测角中,针对车载场景中单帧数据实时处理的局限性问题,结合深度学习技术,提出了一种基于多层深度卷积网络(Deep Convolution Network, DCN)的单帧到达角(Direction of Arrival, DOA)估计算法。与经典的多信号分类(Multiple SignalClassification,MUSIC)算法相比,具有更低的计算复杂度,能够满足车载应用的实时性能要求。同时,该算法在低信噪比(Signal-to-Noise Ratio, SNR)情况下具有更优的超分辨性能,可以提升点云密度,使点云图像轮廓更加清晰。最后,通过仿真以及实测实验有效验证了本文所提方法的可行性。更多还原

【Abstract】 Millimeter wave radar has become an indispensable part of vehicle sensors in intelligent driving because of its allweather and small size. Millimeter-wave radar based on Multiple-input and Multiple-output(MIMO) technology can form a larger virtual array through multiple transmitting and receiving, thus improving the angular resolution. Based on the design of Time Division Multiple Access MIMO(TDMA-MIMO) system, this paper proposes a single frame high resolution point cloud imaging algorithm for vehicle-mounted millimeter wave radar. Combined with the design of time division transmitting waveform, the coherent processing method of receiving array time division error correction is studied. In array Angle measurement, aiming at the limitation of real-time processing of single frame data in vehicle-mounted scenes, and combined with deep learning technology,a single frame Direction of Arrival(DOA) estimation algorithm based on multi-layer Deep Convolution Networks(DCN) is proposed. Compared with the classical Multiple Signal Classification(MUSIC) algorithm, it has lower computational complexity and can meet the real-time performance requirements of on-board applications. At the same time, the algorithm has better superresolution performance under the condition of low Signal-to-Noise Ratio(SNR), which can improve the density of point cloud and make the contour of point cloud image clearer. Finally, the feasibility of the proposed method is validated through simulation and measurement experiments.更多还原