Normalized confidence factors for robust direction of arrival estimation

Estimation in presence of noise is a fundamental problem in signal processing. In time delay based direction of arrival estimation, a single outlier in time delays can produce a large error in the directional estimate. When signal-to-noise ratio is low, this problem becomes more acute because the probability of large errors in time delay estimation increases rapidly with noise power. This work presents a new confidence scoring method for far-field time delays. Using the score, erroneous time delays can be discarded. The score is developed by improving a previously proposed confidence scoring method. Improvements include easier parameterization and better scalability for different noise scenarios. The proposed method is tested with simulations and speech data experiments that demonstrate these improvements, as well as an overall improvement in estimation performance.