Multiple-input-multiple-output (MIMO) radar is well-known for providing high-resolution direction-of-arrival (DOA) estimation by forming a large-scaled sum coarray utilizing waveform diversity. However, the sacrifice is that a large number of snapshots are required to estimate the sample covariance matrix. When the number of training snapshots is limited, the performance of subspace-based DOA estimation method, such as multiple signal classification (MUSIC), deteriorates due to the distortion of noise subspace. In order to improve the accuracy of DOA estimation using MIMO radar in the case of few snapshots, we propose a method to refine the covariance matrix iteratively. The sampled covariance matrix is iteratively refined by subtracting cross-correlation terms using generalized inner product based on the previous DOA estimates. Finally, the MUSIC algorithm is implemented based on the refined sample covariance matrix to update the DOA estimates until achieving termination condition. Simulation results demonstrate that the additional covariance matrix refinement step enhances the accuracy of DOA estimation using MIMO radar in the case of limited snapshots significantly.

MIMO radar; DOA Estimation; covariance matrix refinement; generalized inner product; generalized norm