Best linear unbiased estimation (BLUE) filter is widely used in radar target tracking for its efficiency and robustness. Its estimation performance is deteriorated by the non-Gaussian converted measurement noises under the condition of large cross-range errors. To mitigate the problem

a modified converted measurement model with predicted bearing is proposed. The predicted bearing can help to alleviate the nonlinearity caused by bearing error trigonometric function and obtain pseudo-Gaussian converted measurements. Theoretic analysis of the proposed BLUE filter in the line-of-sight coordinates is derived. The constraints to use the predicted bearing is derived. The filtering diagram of the improved algorithm is given. The filtering model parameters for 3D radar tracking system are derived. The Gaussian level indexes of the converted measurements are provided. Simulation results show the converted measurements of the proposed BLUE are more closer to the Gaussian distribution

so its accuracy improves significantly with minor computation burden

which means it can provide real-time and accurate estimation for radar target tracking. The fundamental idea also applies to other occasions with relatively large nonlinear errors

thus providing some references for similar applications.