Kullback-Leibler divergence based angle estimation in low SNR
Huageng Liu, Xinliang Chen · IET conference proceedings. · 2024
Subspace-based and likelihood-based estimators are commonly used for direction of arrival (DOA) estimation. However, these methods exhibit performance degradation under low signal-to-noise ratio (SNR), known as the threshold effect. We propose a generalized maximum likelihood estimation (MLE) approach, named KLD-based estimation, which considers MLE as the projection of the empirical distribution onto the Gaussian distribution using the Kullback-Leibler divergence (KLD). Additionally, a beam domain KLD (BDKLD) method is proposed to reduce computational complexity by exploiting beam domain echoes. We compare the KLD-based estimation with several popular methods in simulation, and the results demonstrate that the KLD-based estimation can reduce the threshold SNR of MLE method by more than 15dB under low SNR conditions. Under a -30dB SNR condition, the KLD-based estimation achieves approximately 10 times higher estimation accuracy compared to the MLE.