Direct Localization of UAV Targets in Dense Urban Areas Based on Sparse Bayesian Inference

‍ ‍With the increasing occurrence of unauthorized unmanned aerial vehicle （UAV） flights， commonly called “black flying”， there is a growing demand for effective localization approaches to implement countermeasures. However， in dense urban areas， the prevalence of low-altitude UAV flights results in their emitted signals being scattered off the surrounding buildings. The intercepted signals received by the antenna arrays form a superposition of multipath components， necessitating the establishment of a distributed source model rather than a simplistic point-source assumption. Conventional direct position determination （DPD） algorithms based on a point source model exhibit a sharp decline in performance and may even fail in scenarios characterized by distributed sources. This study proposes a DPD method based on the sparse Bayesian inference （SBI） technique under a coherently distributed source scenario， termed CD-SBL to overcome these challenges. Additionally， the proposed algorithm tackles the limitations of subspace-based methods in scenarios with limited snapshots and fewer array elements by incorporating the SBI technique for signal reconstruction， thereby facilitating effective target localization. The algorithm establishes a localization model for coherently distributed （CD） sources in scenes with multiple arrays. It constructs a sparse probability framework and applies prior information to sparse signals and noise within this framework. The SBI technique is then employed to iteratively update the estimated values of hyperparameters， obtaining power spectral values at each grid point （GP） in the region of interest （ROI）. The final step involves a multidimensional spectrum search to pinpoint the source’s position at the peak of the generated power spectrum. Furthermore， this study provides a detailed derivation of the Cramér-Rao lower bound for the DPD of CD sources in the time domain. Numerical simulation results demonstrate that the proposed algorithm exhibits higher localization accuracy and robustness than the traditional subspace-based algorithms in scenarios featuring a CD source model. In situations with more array elements， its performance approaches that of the maximum likelihood estimation algorithm.