Aeroacoustic array is an important testing tool in noise identification for industry applications. However, due to its extensive computational requirements the typical beamforming has to be conducted off-line and is therefore not capable of real-time processing. An alternative algorithm in the form of a state observer is proposed to solve the problem. The theoretical background of this algorithm, derivation of its stability, array construction methods as well as experimental results based on the array are discussed in this study. The algorithm has a form similar to a classical state observer and the array processing is performed in the frequency domain. An ideal single-point noise source and a landing gear model tests both confirm the validity and effectiveness of this algorithm. This observer-based algorithm can be conducted recursively over each sampling, thus possessing real-time capability. It can also be quite useful in experiments by observing any potential test defects on site.