An Off-axis Pole Focusing Method for Robust DOA Estimation in Multi-source Environments

An off-axis pole focusing method for robust direction of arrival (DOA) estimation is presented in this paper. The roots (poles) of the root-MUSIC polynomial are generally evaluated over the unit circle in the z-domain. The DOAs can be computed from these poles using a simple geometric relation. Rather than computing the root-MUSIC polynomial over the unit circle, it can be estimated inside the unit circle at different radii. Such an off-axis estimation method along with averaging shows higher robustness in reverberant environments where the roots (poles) are displaced due to the effects of multipath. This robustness is result of the estimated poles migrating toward unit circle due to off-axis search. Experiments on source localization are conducted to evaluate the proposed method. The robustness of the proposed method under reverberation is illustrated using DOA confidence interval (CI) plots. Experiments on distant speech recognition are conducted under reverberant conditions on the subset of spatialized version of TIMIT (S-TIMIT) and MONC databases. The experimental results obtained for both source localization and distant speech recognition indicate a reasonable improvement over conventional sub-space and correlation based methods.