Please wait a minute...
浙江大学学报(工学版)
Journal of ZheJiang University (Engineering Science)  2023, Vol. 57 Issue (6): 1215-1223    DOI: 10.3785/j.issn.1008-973X.2023.06.017
    
Radar range super-resolution method based on deep neural network
Cheng-jin QIN1(),Jun-zheng JIANG1,2,*()
1. School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2. Satellite Navigation Positioning and Location Service National and Local Joint Engineering Research Center, Guilin University of Electronic Technology, Guilin 541004, China
Download: HTML     PDF(1871KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

Existing radar super-resolution algorithms are difficult to be effectively applied to real scenes. To resolve this problem, a range super-resolution method for frequency modulated continuous wave (FMCW) radar based on deep neural network (DNN) was proposed. The proposed method extrapolated the observation time of the radar signal by DNN to improve the resolution in frequency domain, so as to enhance the range resolution of the radar. Firstly, fast Fourier transform combined with discrete time Fourier transform (FFT+DTFT) algorithm was utilized to preprocess the intermediate frequency signal of radar for reducing the processing complexity of subsequent DNN. Then, the feature extraction of the input signal was realized by DNN with nonlinear fitting ability, and the development trend of the signal was predicted. Next, the predicted signal was recursively input to DNN to continuously extrapolate the length of the time domain signal. Accordingly, the high-resolution range profile could be obtained by performing fast Fourier transform on the time domain extrapolated signal. Finally, in order to eliminate clutter interference, constant false alarm rate (CFAR) detection was performed on the range profile to effectively extract the target range information. Simulation results show that the proposed method achieves range super-resolution, surpassing the limitation of radar bandwidth. Compared with the existing super-resolution methods, the proposed method has smaller error and is more suitable for processing radar signal in real scenes.

Key words： frequency modulated continuous wave (FMCW) radar      deep neural network (DNN)      range super-resolution      signal extrapolation      fast Fourier transform     
Received: 15 August 2022      Published: 30 June 2023
CLC:  TN 958.94  
Fund:  国家自然科学基金资助项目(62171146, 61761011); 广西创新驱动发展专项(桂科AA21077008)
Corresponding Authors: Jun-zheng JIANG     E-mail: 20022303111@mails.guet.edu.cn;jzjiang@guet.edu.cn
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Cheng-jin QIN
Jun-zheng JIANG
Cite this article:

Cheng-jin QIN,Jun-zheng JIANG. Radar range super-resolution method based on deep neural network. Journal of ZheJiang University (Engineering Science), 2023, 57(6): 1215-1223.

URL:

https://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2023.06.017     OR     https://www.zjujournals.com/eng/Y2023/V57/I6/1215


基于深度神经网络的雷达距离超分辨方法

现有的雷达超分辨算法难以有效地应用于现实场景, 为此提出基于深度神经网络(DNN)的调频连续波(FMCW)雷达距离超分辨方法. 所提方法通过DNN外推雷达信号的观测时间以提高频域分辨率, 进而提高雷达的距离分辨率. 为了降低后续DNN的处理复杂度，利用快速傅里叶变换结合离散时间傅里叶变换(FFT+DTFT)算法预处理雷达的中频信号. 采用具有非线性拟合能力的DNN对输入信号进行特征提取, 预测信号的发展趋势. 将预测信号递归输入DNN以不断外推时域信号的长度, 对时域外推后的信号进行快速傅里叶变换得到具有高分辨率的距离像. 为了排除杂波干扰, 对距离像进行恒虚警率(CFAR)检测以有效地提取目标的距离信息. 仿真实验结果表明, 所提方法可以超越雷达带宽的限制实现距离超分辨. 与现有的超分辨方法相比, 所提方法具有更小的误差且更适用于处理现实场景中的雷达信号.


关键词： 调频连续波(FMCW)雷达,  深度神经网络(DNN),  距离超分辨,  信号外推,  快速傅里叶变换 
Fig.1 Time domain and frequency domain of extrapolated signal
Fig.2 Signal processing flowchart of radar range super-resolution method based on deep neural network
Fig.3 Radar front and measured scene
参数 数值
起始频率 ${f_{\rm{c}}}/{\rm{GHz} }$ 77
调频斜率 $ S/({\rm{MHz}} \cdot \text{μ} {{\rm{s}}^{ - 1}}) $ 25
调频时间 $T/\text{μs}$ 30
采样点数 $N$ 128
采样率 ${f_{\rm{s} } }/(10^{6}\;{\rm{s} }^{-1})$ 5
Tab.1 Parameter setting of radar measurement
Fig.4 Results of processing measured data 1 with different super-resolution methods
Fig.5 Results of processing measured data 2 with different super-resolution methods
Fig.6 Comparison of range-doppler map between original signal and extrapolated signal of measured data 1
Fig.7 Comparison of Range-doppler map between original signal and extrapolated signal of measured data 2
算法 ${{\rm{MAE}}_{ {\rm{M1} } }},{{\rm{RMSE}}_{ {\rm{M1} } }}$ ${{\rm{MAE}} _{\rm{M2}}},{\rm{RMSE}} _{\rm{M2} }$
$ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $ $ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $
EDFT 0.0847, 0.1083 0.2145, 0.2518 0.3456, 0.4010 0.3187, 0.4178 0.3486, 0.4211 0.2500, 0.3402
Fast-MUSIC 0.4640, 0.5551 0.4279, 0.5099 0.4121, 0.5291 0.4318, 0.5419 0.3731, 0.4599 0.3652, 0.4614
本研究 0.1926, 0.2134 0.1784, 0.2580 0.2426, 0.3280 0.2169, 0.2614 0.3165, 0.3513 0.2427, 0.2989
Tab.2 Error of range profile of measured data under different extrapolation lengths
算法 ${{\rm{MAE}}_ {\rm{S1} }}, {{\rm{RMSE}}_ {\rm{S1} }}$ ${{\rm{MAE}}_{\rm{S2} }}, {{\rm{RMSE}}_ {\rm{S2} } }$
$ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $ $ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $
AR模型 0.0228, 0.0386 0.0315, 0.0487 0.0559, 0.1016 0.1174, 0.1425 0.2591, 0.3135 0.4397, 0.4887
EDFT 0.0289, 0.0384 0.0369, 0.0421 0.0502, 0.0728 0.1844, 0.2236 0.2010, 0.2667 0.1758/0.2479
Fast-MUSIC 0.1275, 0.2742 0.1050, 0.2149 0.1107, 0.2280 0.2736, 0.3700 0.2235, 0.2903 0.1698, 0.2507
本研究 0.0631, 0.1042 0.0580, 0.0715 0.0559, 0.1042 0.1314, 0.1617 0.1651, 0.2295 0.1256, 0.1937
Tab.3 Error of range profile of simulation data 1 and simulation data 2 under different extrapolation lengths
算法 ${{\rm{MAE}}_{\rm{S3} }}, {{\rm{RMSE}}_{\rm{S3} } }$ ${{\rm{MAE}}_{\rm{S4} }}, {{\rm{RMSE}}_{\rm{S4} }}$
$ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $ $ \alpha = 2 $ $ \alpha = 3 $ $ \alpha = 4 $
AR模型 0.2142, 0.2733 0.2305, 0.2802 0.2849, 0.3410 0.1459, 0.1749 0.2203, 0.3063 0.2272, 0.3177
EDFT 0.2697, 0.3562 0.2483, 0.3312 0.2298, 0.3192 0.1544, 0.2024 0.2067, 0.2838 0.2172, 0.3069
Fast-MUSIC 0.2897, 0.3491 0.2877, 0.3512 0.2379, 0.3487 0.2782, 0.3642 0.2862, 0.3682 0.2901, 0.3805
本文方法 0.2693, 0.3357 0.1719, 0.2598 0.1482, 0.2225 0.1314, 0.1617 0.1651, 0.2295 0.1928, 0.2589
Tab.4 Error of range profile of simulation data 3 and simulation data 4 under different extrapolation lengths
[1]   NEEMAT S, KRASNOV O, VAN DER ZWAN F, et al Decoupling the doppler ambiguity interval from the maximum operational range and range-resolution in FMCW radars[J]. IEEE Sensors Journal, 2020, 20 (11): 5992- 6003
doi: 10.1109/JSEN.2020.2972152
[2]   LULU A, MOBASSERI B G High-resolution range-doppler maps by coherent extension of narrowband pulses[J]. IEEE Transactions on Aerospace and Electronic Systems, 2020, 56 (4): 3099- 3112
doi: 10.1109/TAES.2020.2965754
[3]   GIRGIS A A, HAM F M A quantitative study of pitfalls in the FFT[J]. IEEE Transactions on Aerospace and Electronic Systems, 1980, AES-16 (4): 434- 439
doi: 10.1109/TAES.1980.308971
[4]   GONG P C, LI J, GUO W, et al. A high resolution algorithm based on chirp Z-transform for FMCW radar [C]// 2015 IEEE International Conference on Communication Problem-Solving (ICCP). Guilin: IEEE, 2015: 482-484.
[5]   ZHENG C S, DING K, YANG Z J. Noise influence on frequency estimation accuracy from energy centrobaric correction method for discrete spectrum [C]// 2009 International Conference on Information and Automation. Zhuhai/Macau: IEEE, 2009: 1477-1481.
[6]   AI-QUDSI B, JORAM N, STROBEL A, et al. Zoom FFT for precise spectrum calculation in FMCW radar using FPGA [C]// Proceedings of the 2013 9th Conference on Ph. D. Research in Microelectronics and Electronics (PRIME). Villach: IEEE, 2013: 337-340.
[7]   IIZUKA T, TORIUMI Y, ISHIYAMA F, et al. Root-MUSIC based power estimation method with super-resolution FMCW radar [C]// 2020 IEEE/MTT-S International Microwave Symposium (IMS). Los Angeles: IEEE, 2020: 1027-1030.
[8]   LIEPINS V. Extended Fourier analysis of signals [EB/OL]. [2022-07-15]. https://doi.org/10.48550/arXiv.1303.2033.
[9]   YU J, KROLIK J. Multiband chirp synthesis for frequency-hopped FMCW radar [C]// 2009 Conference Record of the Forty-Third Asilomar Conference on Signals, Systems and Computers. Pacific Grove: IEEE, 2009: 1315-1319.
[10]   KIM S, LEE K K Low-complexity joint extrapolation-MUSIC-based 2-D parameter estimator for vital FMCW radar[J]. IEEE Sensors Journal, 2018, 19 (6): 2205- 2216
[11]   ZENG W J, SO H C, HUANG L ${\ell _p}$-MUSIC: robust direction-of-arrival estimator for impulsive noise environments [J]. IEEE Transactions on Signal Processing, 2013, 61 (17): 4296- 4308
doi: 10.1109/TSP.2013.2263502
[12]   姚昕彤, 王玉文, 刘奇, 等 基于MUSIC及其改进算法的DOA估计研究[J]. 通信技术, 2021, 54 (6): 1363- 1369
YAO Xin-tong, WANG Yu-wen, LIU Qi, et al Research on DOA estimation based on MUSIC and its improved algorithm[J]. Communications Technology, 2021, 54 (6): 1363- 1369
doi: 10.3969/j.issn.1002-0802.2021.06.012
[13]   PAM M, LIU A, YU Y, et al Radar HRRP target recognition model based on a stacked CNN–Bi-RNN with attention mechanism[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 5100814
[14]   CHEN X, HUANG W Spatial–temporal convolutional gated recurrent unit network for significant wave height estimation from shipborne marine radar data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 4201711
[15]   HAN L, SUN J, ZHANG W Convolutional neural network for convective storm nowcasting using 3-D doppler weather radar data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58 (2): 1487- 1495
doi: 10.1109/TGRS.2019.2948070
[16]   KUMCHAISEEMAK N, CHATNUNTAWECH I, TEERAPITTAYANON S, et al Toward ant-sized moving object localization using deep learning in FMCW radar: a pilot study[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5112510
[17]   LIU Q Y, SUN Y X, ZHANG Q, et al. Multi-target parameter estimation based on velocity compensation in FMCW radar [C]// IET International Radar Conference 2015. Hangzhou: IET, 2015.
[18]   CADZOW J A Spectral estimation: an overdetermined rational model equation approach[J]. Proceedings of the IEEE, 1982, 70 (9): 907- 939
doi: 10.1109/PROC.1982.12424
[19]   HECHT-NIELSEN R. Theory of the backpropagation neural network [C]// Neural Network. Washington, DC: IEEE, 1989.
[20]   刘洁怡, 公茂果, 詹涛, 等 一种深度神经网络多站雷达系统干扰鉴别方法[J]. 西安电子科技大学学报, 2021, 48 (2): 133- 138
LIU Jie-yi, GONG Mao-guo, ZHAN Tao, et al Method for discrimination of false targets in multistation radar systems based on the deep neural network[J]. Journal of Xidian University, 2021, 48 (2): 133- 138
doi: 10.19665/j.issn1001-2400.2021.02.017
[21]   DAS O, ABEL J S, SMITH III J O. Fast MUSIC–an efficient implementation of the MUSIC algorithm for frequency estimation of approximately periodic signals [C]// Proceedings of the 21st International Conference on Digital Audio Effects. Aveiro: [s.n.], 2018: DAFX1-DAFX7.
[22]   白炳潮. 毫米波雷达多径环境超分辨测角及测距[D]. 成都: 电子科技大学, 2020.
BAI Bing-chao. High resolution direction of arrival and range estimation in multipath environment of millimeter wave radar [D]. Chengdu: University of Electronic Science and Technology of China, 2020.
[1] Yang LUO,Wei ZHANG. Hardware efficient FFT design based on improved rotation factor[J]. Journal of ZheJiang University (Engineering Science), 2021, 55(6): 1199-1207.
[2] Shu-guang SUN,Peng TIAN,Tai-hang DU,Jing-qin WANG. Dense frequency harmonic/interharmonic detection based on apFFT-AMD[J]. Journal of ZheJiang University (Engineering Science), 2020, 54(1): 178-188.
[3] ZHANG Xin-yi, YANG Jia-qiang, ZHANG Xiao-jun. Dynamic economic dispatch incorporating multiple wind farms based on FFT simplified chance constrained programming[J]. Journal of ZheJiang University (Engineering Science), 2017, 51(5): 976-983.
[4] LOU Wen-juan, WANG Jia-wei, YANG Lun, CHEN Yong. Simulation of three-dimensional fluctuating wind velocity field #br# upon thunderstorm downburst[J]. Journal of ZheJiang University (Engineering Science), 2014, 48(7): 1162-1169.
[5] ZHANG Jin-gao, LU Qin-fen, WANG Li, YE Yun-yue, Hong Wei-ming. Measurement of power factor angle of linear metro’s traction
motor based on FFT[J]. Journal of ZheJiang University (Engineering Science), 2012, 46(11): 1981-1984.