A Novel Angle Estimation for mmWave FMCW Radars Using Machine Learning
The use of mmWave Radar modules in conjunction with other sensors such as ultrasound, infrared, optical, a global positioning system (GPS), and light detection and ranging (LiDAR) enables the development of highly reliable, fully autonomous systems. The highly integrated modules enhance the reliability of unmanned ground vehicles (UGV) and unmanned aerial vehicles (UAV) for traffic management of autonomous vehicles. Radars are vital in managing and monitoring the dynamic activity associated with such autonomous operations.
The mmWave radars are frequency-modulated continuous wave (FMCW) radars. They are ultra-high resolution, low power, and single-chip radar sensors that can detect the range, velocity, and angle of arrival (AoA) of the objects in their field of view (FoV). The mmWave radars have several advantages over their counterparts, such as cameras and LiDARs. The mmWave radars are also resistant to adverse weather conditions such as heavy rain, fog, dust, and snow. The mmWave radars operating in the frequency range of GHz can detect the range, velocity, and angle of arrival (AoA) of targets with a field of view (FoV) of up to 120°.
Target range and velocity can be estimated more accurately than AoA, even with a limited number of transceivers. Resolution and Accuracy of AoA estimation improve as the number of transceivers increases. In addition, due to the limited number of transceivers, FoV in mmWave radars can only be maximized either in azimuth or elevation. Furthermore, conventional AoA estimation requires at least one transmitting antenna and two receiving antennas to estimate the AoA of a single target in the FoV.
The researchers developed novel machine learning-based AoA estimation and FoV enhancement techniques to address the aforementioned issues. Measurements have validated these techniques in a suitable outdoor multi-target environment. The field of view is enhanced in both azimuth and elevation. Elevation FoV enhancement is achieved by keeping antenna elements oriented in elevation. In this orientation, radar focuses the beam vertically, increasing the elevation FoV.
The azimuth FoV is increased by mechanically rotating the radar horizontally, with antenna elements in elevation. The researchers also proposed a novel algorithm for estimating AoA by locating the peaks of clusters around the targets. It was then used as a polynomial regression method with a degree of 4 to optimize the results further. The proposed approach was evaluated by conducting extensive real-world experiments. A variety of objects, including five humans with varying physical characteristics, a car, and a drone, were used to demonstrate the robustness of the approach. The proposed AoA estimation technique achieves a root mean square error (RMSE) of 2.5 degrees for such rotating radars.
Using machine learning, a novel angle estimation for mmWave FMCW radars will be highly useful in small-scale and low-cost ground station monitoring radar systems. In the future, the resolution limitation experienced currently can be resolved by separating the clusters formed for the objects in range-angle plot using machine learning techniques.