Precise and reliable indoor localization is a demanding task with a variety of possible applications. Besides other approaches, Angle of Arrival (AoA) based positioning systems evolved over the past decades. The introduction of Bluetooth 5.1 added support for AoA-based localization to the Bluetooth Core Specification and hence promised the development of low-cost, accurate, and highly-available location services. However, due to pseudo-random phase offsets between measurements on different Bluetooth channels, state-of-the-art AoA estimation algorithms cannot be applied in this context. In this work a phase normalization scheme and a corresponding data model are proposed that resolve these problems. For an experimental evaluation, a study of the capabilities of a Bluetooth localization system in a 100 square meter real-world testbed is presented. It is shown that the localization accuracy could be significantly improved in comparison with existing techniques suitable for Bluetooth indoor positioning. Using the proposed method, a median localization accuracy of 18 cm was achieved.