基于深度学习的MIMO雷达一体化处理

Integrated Processing of MIMO Radar Based on Deep Learning

传统MIMO雷达采用基于模型驱动的级联信号处理流程,各处理流程对模型假设依赖严重,且互相依赖,引入了基于深度学习的数据驱动的方法,通过构建三维卷积神经网络检测模型,直接在接收端检测发射波形,实现空域、时域和距离维的联合处理。仿真结果显示:在简单目标环境下,深度学习方法同传统处理方案效果相当;在复杂目标环境下,深度学习方法要明显优于模型驱动的传统处理方法,表明了深度学习方法的有效性。

The traditional MIMO radar adopts a cascade signal processing flow based on model driving.Each processing flow depends heavily on the model assumptions and is interdependent.A data driven method based on deep learning is introduced.By building a three-dimensional convolutional neural network detection model,the transmission waveform is directly detected at the receiver,and the joint processing of space,time and distance dimensions is realized.The simulation results show that in the simple target environment,the deep learning method is equivalent to the traditional processing scheme;In the complex target environment,the deep learning method is obviously superior to the traditional model-driven processing method.The validity of deep learning in the integrated processing of MIMO radar signals is verified.