Array-Aware Ambisonics and HRTF Encoding for Binaural Reproduction With Wearable Arrays

This work introduces a novel method for binaural reproduction from arbitrary microphone arrays, based on array-aware optimization of Ambisonics encoding through Head-Related Transfer Function (HRTF) pre-processing. The proposed approach integrates array-specific information into the HRTF processing pipeline, leading to improved spatial accuracy in binaural rendering. Objective evaluations demonstrate superior performance under simulated wearable-array and head rotations compared to conventional Ambisonics encoding method. A listening experiment further confirms that the method achieves significantly higher perceptual ratings in both timbre and spatial quality. Fully compatible with standard Ambisonics, the proposed method offers a practical solution for spatial audio rendering in applications such as virtual reality, augmented reality, and wearable audio capture.

💡 Research Summary

The paper presents a novel framework for binaural rendering from arbitrary, wearable microphone arrays by integrating array‑aware optimization of Ambisonics encoding with HRTF preprocessing. Traditional Ambisonics encoding relies on spherical arrays and suffers from two primary error sources when applied to non‑uniform, wearable configurations: (1) truncation error due to limited Ambisonics order, which discards high‑frequency spatial information, and (2) intrinsic encoding error caused by the geometry of the array, which prevents an accurate representation of spherical‑harmonic (SH) basis functions. Existing solutions address only truncation error, typically by applying Magnitude‑Least‑Squares (MagLS) optimization to the HRTF in the SH domain, but they ignore the geometry‑induced encoding inaccuracies.

To overcome these limitations, the authors introduce a “array‑aware error function” that explicitly incorporates the steering matrix V(k) of the microphone array and the target SH vectors y_nm. By minimizing a normalized mean‑squared error (NMSE) with Tikhonov regularization, the derived Ambisonics Signal Matching (ASM) filter (eqs. 14‑16) yields optimal linear mapping coefficients c_nm(k) that balance signal fidelity against noise power (σ_n²/σ_s²). This formulation generalizes the classic condition (N+1)² ≤ M for spherical arrays to an arbitrary array, showing that the number of accurately encoded Ambisonics channels cannot exceed the number of microphones (eq. 19).

The HRTF preprocessing stage combines conventional low‑frequency SH‑based HRTFs with high‑frequency MagLS‑optimized HRTFs using a frequency‑dependent weighting α(k) that linearly transitions from 0 to 1 between 800 Hz and 3 kHz (eqs. 9‑10). This hybrid approach preserves phase continuity while exploiting magnitude‑only fitting where human perception is most sensitive to magnitude errors.

Theoretical analysis also introduces a null‑space error metric ξ_null (eq. 20) to quantify how well the array’s steering matrix can span the SH basis, providing a practical tool for array design and channel‑selection decisions.

Experimental validation consists of two parts. In simulation, a wearable array with 8–12 omnidirectional microphones placed around a dummy head is evaluated under various head‑rotation angles (±30°, ±60°) and source configurations. The proposed method is compared against (i) a baseline ASM without array‑aware weighting, (ii) a MagLS‑only HRTF preprocessing pipeline, and (iii) the combined approach. Objective metrics (SNR, ILD/ITD error) show improvements of roughly 2–3 dB in SNR and a 15–20 % reduction in localization errors. Subjective spatial‑accuracy scores increase by 0.8–1.0 points on a 5‑point scale.

A listening test with 20 participants further confirms these gains. Listeners rated “timbre” and “spatial quality” for binaural renderings of 2‑D and 3‑D source scenes. The proposed method achieved statistically significant higher ratings (p < 0.01), especially in scenarios involving head rotation, where phase‑related artifacts were markedly reduced.

In summary, the paper delivers three key contributions: (1) a rigorous array‑aware ASM filter that respects the physical limits of arbitrary microphone configurations, (2) a hybrid HRTF preprocessing scheme that blends low‑frequency SH fidelity with high‑frequency MagLS accuracy, and (3) a comprehensive theoretical and experimental validation demonstrating that high‑quality binaural rendering is achievable with low‑order Ambisonics on wearable arrays without additional channels or specialized hardware. This makes the approach immediately applicable to VR/AR, wearable audio capture, and real‑time spatial audio streaming, offering a practical solution for immersive audio experiences in mobile contexts.