We propose a novel algorithm to improve the accuracy in estimating the angle-of-arrival (AoA) when the MUSIC (MUltiple SIgnal Classification) algorithm is used. An optimal precoder, with the objective to minimize the estimation errors of the AoAs, is first derived. However, to compute the optimal precoder requires the channel side information at the transmitter (CSIT) exclusive of the receiver array, which cannot be separately estimated practically. A more practical precoder design approach, which leverages on the feedback CSIT estimated at the receiver, is next proposed. We demonstrate that the practical precoder performs closely to the optimal precoder through simulation, and both precoders exhibit considerable performance improvement compared with the AoA estimation without precoder. Furthermore, the precoder technique is applied to a known AoA-based localization method, and the improvement on the accuracy of the location estimate is studied.