Microphone directivity-based single source point detection in multi-source direction-of-arrival estimation

We examine single source point (SSP) detection in multi-source direction-of-arrival (DOA) estimation, where SSPs are the time-frequency points in the array signal dominated by only one source. Focusing on a first-order ambisonics array, this study first proposes a phase-based SSP detection method using virtual array signals. By constructing virtual microphone signals and leveraging their phase relationships, SSPs can be effectively detected. Unlike physical arrays with fixed microphone directivity patterns, which may inherently have reception blind regions, the proposed virtual array allows dynamic adjustment of each microphone's directional gain, thereby improving both the strength of the source signal and the stability of its phase information under noise and reflections. This is beneficial for SSP detection based on the phase relationships of array signals. In addition, we propose a SSP detection method that utilizes the directional gain relationships between microphones. The proposed two methods are further integrated to achieve more robust SSP detection. Finally, we can extract the direction information of sources from detected SSPs and estimate the DOAs of all sources through kernel density estimation and peak search. In tests in simulated and real environments, the proposed method exhibits superior performance in DOA estimation compared to some existing methods.