ASAP: An Azimuth-Priority Strip-Based Search Approach to Planar Microphone Array DOA Estimation in 3D

Ming Huang 1 , Shuting Xu 1 , Leying Yang 1 , Huanzhang Hu 1 , Yujie Zhang 1 , Jiang Wang 2 , Yu Liu 1 Hao Zhao 1 , and He Kong 1 · 2026

Read ASAP: An Azimuth-Priority Strip-Based Search Approach to Planar Microphone Array DOA Estimation in 3D by Ming Huang 1 , Shuting Xu 1 , Leying Yang 1 , Huanzhang Hu 1 , Yujie Zhang 1 , Jiang Wang 2 , Yu Liu 1 Hao Zhao 1 , and He Kong 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-25387.txt.

Introduction

Huang and eight colleagues put forth ASAP, an azimuth-first search for flat array DOA reads. Direction-of-arrival reads matter for mic arrays. The steered response power with phase transform way is widely used, but 3D reads need thousands of test directions, hurting live speed. This pinch is worst for flat arrays. Since azimuth reads tend to beat elevation reads for trust, ASAP shrinks azimuth strips to lock azimuth, then polishes elevation along a great-circle arc. Sims and tests back the way.

Catalog record

CREATORMing Huang 1 , Shuting Xu 1 , Leying Yang 1 , Huanzhang Hu 1 , Yujie Zhang 1 , Jiang Wang 2 , Yu Liu 1 Hao Zhao 1 , and He Kong 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24112
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.25387

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

ASAP: An Azimuth-Priority Strip-Based Search Approach to Planar Microphone Array DOA Estimation in 3D Thanks: Ming Huang and Shuting Xu contribute equally to this work. This work was supported by the National Key R & \& D Program of China under Grant No. 2024YFB4710900, the National Natural Science Foundation of China under Grant No. U24A20265, the Open Research Fund of the State Key Laboratory of Multimodal Artificial Intelligence Systems, the Shenzhen Science and Technology Program under Grant No. KQTD20221101093557010, and the Guangdong Science and Technology Program under Grant No.2024B1212010002. This research was conducted when Shuting Xu, Leying Yang, and Huanzhang Hu were visiting students at the Southern University of Science and Technology, Shenzhen, China. Corresponding author: He Kong (email: [email protected]). Ming Huang 1 , Shuting Xu 1 , Leying Yang 1 , Huanzhang Hu 1 , Yujie Zhang 1 , Jiang Wang 2 , Yu Liu 1 Hao Zhao 1 , and He Kong 1 Abstract Direction-of-arrival (DOA) estimation is an important task in microphone array processing and many downstream applications. The steered response power with phase transform (SRP-PHAT) method has been widely adopted for DOA estimation in recent years. However, accurate SRP-PHAT estimation in 3D scenarios requires evaluating steering responses over thousands of candidate directions, severely limiting real-time performance on resource-constrained platforms. This challenge becomes even more critical for planar arrays, which are widely used in robotics due to their structural simplicity. Motivated by the fact that azimuth estimation is usually more reliable than elevation estimation for most arrays, we propose ASAP, an A zimuth-priority S trip-based Se a rch a p proach to planar microphone array DOA estimation in 3D. In the first stage, ASAP performs coarse-to-fine region contraction within azimuthal strips to lock azimuth angles while retaining multiple maxima through spherical caps. In the second stage, it refines elevation along the great-circle arc between two close candidates. Extensive simulations and real-world experiments validate the efficiency and merits of the proposed method over existing approaches. I Introduction Direction-of-arrival (DOA) estimation for microphone arrays enables acoustic perception in many applications, including teleconferencing [ 1 , 2 , 3 , 4 ] , target tracking [ 5 , 6 , 7 , 8 ] , environmental monitoring [ 9 , 10 , 11 , 12 ] , robot audition [ 13 , 14 , 15 ] , and human–robot interaction [ 16 , 17 , 18 ] . Among the existing approaches, the steered-response power with phase transform (SRP-PHAT) method remains a strong baseline [ 19 ] . SRP-PHAT extends the generalized cross-correlation with phase transform (GCC-PHAT) algorithm by normalizing spectral magnitudes, ensuring equal frequency weighting and sharper correlation peaks for DOA estimation [ 20 , 21 ] . SRP-PHAT tolerates reverberation and noise with moderate modeling assumptions [ 22 , 23 ] , and can be easily integrated into more advanced frameworks [ 24 , 25 , 26 , 27 , 28 ] . Although SRP-PHAT is recognized for its robustness, it is computationally demanding, particularly in 3D scenarios. In particular, SRP-PHAT performs exhaustive evaluation over dense spherical grids, often comprising thousands of candidate directions, which can be prohibitive for embedded or mobile platforms with limited computing resources [ 13 , 11 , 15 , 29 , 30 ] . Moreover, planar arrays are preferred for their structural simplicity and ease of integration with the robotic hardware system [ 31 , 12 ] . For planar arrays, elevation estimates are usually less reliable than the azimuth estimates, leading to ambiguities and reduced uniformity in 3D [ 32 , 33 ] . Despite the wide applicability of planar microphone arrays, developing an efficient approach to 3D DOA estimation remains a critical challenge. To reduce the computational burden, prior studies have proposed two complementary families of methods. On the one hand, volumetric SRP and its variants evaluate regions rather than individual points, enabling coarser initial sampling without sacrificing robustness; a refinement step then recursively subdivides high-likelihood cells and re-evaluates them [ 34 , 35 , 36 ] . On the other hand, region contraction methods progressively shrink the candidate set around high-scoring maxima. Relevant methods include stochastic region contraction (SRC) [ 37 ] and its deterministic counterpart, coarse-to-fine region contraction (CFRC) [ 38 ] . Region contraction methods iteratively rescore only a small subset at increasing resolution, thus achieving substantial computational savings while maintaining localization accuracy. However, directly applying existing methods to planar arrays might be challenging. For example, full-grid SRP-PHAT is computationally prohibitive for real-time use. while recent methods such as CFRC treat the 3D space uniformly—failing to exploit the planar array’s better azimuth resolution than elevation resolution. To address these limitations, we propose ASAP, an A zimuth-priority S trip-based Se a rch A p proach to planar microphone array DOA estimation in 3D. Rather than exhaustively evaluating thousands of directions as in full-grid SRP-PHAT, or relying on unconstrained multi-stage heuristics such as CFRC [ 38 ] , we decompose the 3D search into two stages. In the first stage, ASAP performs a CFRC-style scan within azimuthal strips and applies spherical-cap filtering to concentrate evaluations near emerging peaks. The second stage then estimates elevation through a one-dimensional pass around the locked azimuth. Unlike existing sequential 2D methods [ 39 ] , ASAP integrates region contraction into the azimuth stage via strip scanning and spherical-cap filtering; by explicitly exploiting the planar array’s anisotropic reliability, it introduces the structured guidance necessary to effectively avoid full 3D scans. Extensive simulations and real-world experiments validate ASAP’s efficiency and performance improvement against SRP-PHAT and CFRC [ 38 ] . In simulations at the highest grid resolution, ASAP reduces runtime by 13.56% and RMSE by 5.87%, compared to CFRC [ 38 ] . Real-world tests with an 8-microphone uniform circular array (UCA) further show 13.98% faster runtime and 4.33% lower RMSE than CFRC [ 38 ] . Moreover, ASAP consistently achieves lower angular errors with faster runtime than full-grid SRP-PHAT across various resolution settings. To benefit the community, we open-source all the codes at https://github.com/AISLAB-sustech/ASAP/tree/main. Fig. 1: Overall architecture of the proposed ASAP framework. Stage 1 estimates a coarse azimuth \hat{\phi} through strip-constrained CFRC search. Stage 2 refines the estimates from Stage 1. II THE PROPOSED METHOD II-A Problem Formulation We consider a UCA consisting of M microphones arranged on a horizontal plane with radius R [ 31 ] . The position of the m -th microphone, for m=1,2,\ldots,M , can be expressed as \mathbf{r}_{m}=[R\cos\theta_{m},\;R\sin\theta_{m},\;0]^{T},\quad\theta_{m}=\tfrac{2\pi(m-1)}{M}. (1) For a sound source with azimuth angle \phi\in[-180^{\circ},180^{\circ}) and elevation angle \theta\in[0^{\circ},90^{\circ}] , we denote the received signal by the m -th microphone at time t as x_{m}(t,\theta,\phi) . As in [ 19 , 20 ] , we use P\left(\phi,\theta\right) to represent the steered response power with the use of the phase transform, namely, P\left(\phi,\theta\right)=2\pi\sum_{l=1}^{M-1}\sum_{m=l+1}^{M}R_{lm}\left(\Delta\tau_{lm}\left(\mathbf{u}\left(\phi,\theta\right)\right)\right), (2) where R_{lm} is the GCC-PHAT between x_{l}(t,\theta,\phi) and x_{m}(t,\theta,\phi) received by the l -th and the m -th microphone, respectively; \mathbf{u} is the unit direction vector and takes the form of \mathbf{u}(\phi,\theta)=\begin{bmatrix}\cos\phi\cos\theta\\ \sin\phi\cos\theta\\ \sin\theta\end{bmatrix}, (3) \Delta\tau_{lm}\left(\mathbf{u}\right)=\tau_{l}\left(\mathbf{u}\right)-\tau_{m}\left(\mathbf{u}\right) , in which \tau_{l}\left(\mathbf{u}\right) and \tau_{m}\left(\mathbf{u}\right) are the time delays occurring from the direction \mathbf{u} to the l -th and m -th microphones, respectively. Given the above system model, our primary objective is to efficiently and accurately estimate

Read the complete text