Multi-Antenna Radar Target Tracking Based on Dimension-Reduced Transport Mapping

Abstract

Target tracking is critical for applications such as automotive perception, unmanned aerial vehicle monitoring, and integrated sensing and communication. While target tracking can be achieved using multi-antenna radars, several challenges need to be addressed to ensure desirable accuracy. These challenges include the high dimension of the radar observations as well as the non-Gaussian noise in these observations. This paper proposes a dimension-reduced stochastic map filter for target tracking using multi-antenna radars. This method extracts an informative reduced-dimensional observation by amplitude-based fast-time subsampling and Fisher information matrix-based projection, and then performs transport map updating by conditional inversion. Simulation results demonstrate that the method achieves significant tracking accuracy improvement compared to unscented Kalman filtering and particle filtering under both Gaussian and K-distributed observation noise.