A Practical Study on Two-Dimensional Acoustic Direction-of-Arrival Estimation: Comparative Evaluation Using an Eight-Microphone Circular Array
Year: 2026
Type: article
Abstract: Joint estimation of source azimuth and elevation is sensitive to array geometry, room reflections, observation length, sampling frequency, and preprocessing. This study presents a controlled within-collection comparison of generalized cross-correlation with phase transform (GCC-PHAT), steered response power with phase transform (SRP-PHAT), multiple signal classification (MUSIC), and minimum variance distortionless response (MVDR) using a vertically mounted eight-microphone circular array in an acoustically untreated indoor room. Synchronized measurements were obtained at 220 labeled source coordinates and divided into five non-overlapping temporal excerpts per coordinate, yielding 1,100 multichannel evaluation units. All algorithms used the same recordings, measured microphone coordinates, angular convention, and error definitions. Observation duration, processing sample rate, energy-based active-frame selection, and channel-wise Wiener filtering were varied. Under the best observed within-collection configuration for each implementation, mean Euclidean error in the azimuth-elevation coordinate plane was 0.84° for SRP-PHAT, 0.85° for GCC-PHAT, 1.00° for MUSIC, and 1.18° for MVDR. Error generally decreased with observation duration, with diminishing improvement at longer durations. Preprocessing effects were estimator-dependent. The results provide an implementation-specific accuracy-runtime comparison rather than a universal ranking of the algorithms.
Authors: Berkay Bolat, Emirhan Coşkun, Ali Can Karaca
Institution: Yıldız Technical University
Language: English
Topic: Speech and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225792894