基于注意力机制和复数卷积循环网络的汽车雷达干扰抑制  

Automotive Radar Interference Suppression Based on Complex Convolution Recurrent Network with Attention Mechanism

作　　者：吴秋雨 高勇[1] WU Qiuyu;GAO Yong(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《无线电工程》2024年第1期63-70,共8页Radio Engineering

基　　金：四川大学科研项目资助(0020505516013)。

摘　　要：随着自动驾驶技术的发展,越来越多的汽车装载车载雷达,不同车辆的车载雷达之间会产生相互干扰,导致虚假目标的出现或基底噪声的增加,降低检测性能。针对汽车雷达之间的相互干扰问题,提出了一种基于注意力机制的深度复数卷积循环网络(Deep Complex Convolution Recurrent Network with Attention,DCCRN-Attention),在频域实现干扰抑制。模型使用复数网络将信号的实部和虚部联合起来进行特征学习,能同时预测干扰抑制后目标的幅度和相位,并在跳跃连接中引入注意力机制聚焦于更重要的特征信息,抑制无关信息。实验结果表明,所提模型能有效抑制干扰、提高目标的信噪比(Signal to Noise Ratio,SNR),在评价指标上均优于基线方法。With the development of autonomous driving technology,more and more vehicles are equipped with on-board radars.But the on-board radars of different vehicles may interfere with each other,leading to the appearance of false targets or the increase of floor noise,which reduces the detection performance.A Deep Complex Convolution Recurrent Network with Attention(DCCRN-Attention)is proposed to solve the problem of mutual interference between automotive radars,and interference suppression is realized in frequency domain.The proposed model uses complex network to combine the real and imaginary parts of the signal for feature learning,which can simultaneously predict the amplitude and the phase of the target after interference suppression.And the model introduces the attention mechanism in the skip connection to focus on the more important feature information and suppress the irrelevant information.The experimental results show that the proposed model can effectively suppress the interference,improve the Signal to Noise Ratio(SNR)of the targets,and outperform the baseline methods in the evaluation indexes.

关 键 词：汽车雷达 干扰抑制 深度复数卷积循环网络 注意力机制 

分 类 号：TN958.94[电子电信—信号与信息处理]