Height estimation for automotive MIMO radar with group-sparse reconstruction

Abstract

A method is developed for sequential azimuth and height estimation of small objects at far distances in front of a moving vehicle using coherent or mutually incoherent MIMO arrays. The model considers phases and amplitudes of superposition of near-field multipath signals produced by specular non-diffusive ground-reflections. The reflection phase shift and power attenuation due to the interaction with the ground is assumed unknown and is estimated jointly. Groupsparsity allows combining measurements along the trajectory of the vehicle provided that the road is flat as well as measurements from multiple incoherent sensors at different locations in the vehicle. This model can be formulated for non-uniform sparse arrays in 2D and 3D with subsets of antennas at the same height and the resulting inverse problem can be approximated with efficient methods such as Block Orthogonal Matching Pursuit. It is shown in simulations that the proposed approach significantly increases estimation accuracy and decreases false alarms, both crucial for the detection of small objects at far distances.

A Novel 3D ArcSAR Sensing System Applied to Unmanned Ground Vehicles

“…The FFT algorithm does not require multiple snapshot signals, but its angle resolution is poor with a limited number of array antennas. Orthogonal Matching Pursuit (OMP) is a signal reconstruction algorithm that requires the number of targets to be known in advance [23,24]. IAA is a spectral estimation algorithm based on weighted least squares (WLS) estimation proposed by Li Jian at the University of Florida in 2010 [25][26][27], which has demonstrated good results in linear aperture experiments.…”

Performance Bounds for Displaced Sensor Automotive Radar Imaging

“…We formulate the non-coherent processing as a block-sparse recovery problem in a reduced-rate sensing framework [36]. Other recent works on single-sensor automotive [37] and MIMO imaging [38] harness block-sparsity of range profiles to mitigate the processing problem with massive data samples. Our approach exploits block sparsity across profiles from multiple sensors that are not perfectly synchronized.…”