Robust Personal Audio Geometry Optimization in the SVD-Based Modal Domain

Abstract

Personal audio generates sound zones in a shared space to provide private and personalized listening experiences with minimized interference between consumers. Regularization has been commonly used to increase the robustness of such systems against potential perturbations in the sound reproduction. However, the performance is limited by the system geometry such as the number and location of the loudspeakers and controlled zones. This paper proposes a geometry optimization method to find the most geometrically robust approach for personal audio amongst all available candidate system placements. The proposed method aims to approach the most 'natural' sound reproduction so that the solo control of the listening zone coincidently accompanies the preferred quiet zone. Being formulated in the SVD-based modal domain, the method is demonstrated by applications in three typical personal audio optimizations, i.e., the acoustic contrast control, the pressure matching and the planarity control. Simulation results show that the proposed method can obtain the system geometry with better avoidance of 'occlusion', improved robustness to regularization, and improved broadband equalization.

A comparison between two modal domain methods for personal audio reproduction

“…In our previous work (Zhu et al, 2019), we proposed using the SVDMD method for geometric optimization of sound zone systems. The SVDMD method retains the advantage of modal domain transfer function parameterization, and suffers fewer geometric limitations than the SHDMD.…”

“…(2), then the columns of L and Q are respectively the modes of the listening zone space and the quiet zone space, the columns of L and Q are respectively the modes of the loudspeaker space corresponding to the listening zone and the quiet zone, and the values of the diagonal elements in L and Q represent the amount of amplification or attenuation that the modes undergo for the transformation L and Q . The coefficients of the loudspeaker-space modes are obtained by (Zhu et al, 2019)…”

“…This SVD based modal domain (SVDMD) method has been previously studied for sound field reconstruction (Fazi, 2010), beamforming (Fazi et al, 2014) and ultrasound (Tanter et al, 2000), where the response matrix from an array to a set of control points is used to produce a desired field. When SVD is applied to personal audio reproduction (Zhu et al, 2019), the weights of the loudspeaker-space modes are obtained by optimizing the compatibility between the capabilities of the listening zone and quiet zone modes for creating a listening zone and a quiet zone.…”

Fast Generation of Sound Zones Using Variable Span Trade-Off Filters in the DFT-Domain

“…Third, ACC computes a matrix inversion that is required to solve the generalized eigenvalue problem; this computation can often be highly ill-conditioned, depending on the geometry of the loudspeaker array, the locations of the sound zones, and the frequency of interest. Either to improve the robustness or, in particular, to avoid the illconditioned inverse problem, a method that maximizes the energy difference between the bright and dark zones [22], an extensive study on regularization [23], and a geometry optimization problem [24] were investigated. Moreover, ACC was initially proposed in the frequency domain, which solves the problem one frequency at a time, although recently, the time-domain version, the so-called broadband ACC (BACC), was also proposed [25].…”

An experimental study on transfer function estimation using acoustic modelling and singular value decomposition

“…Apart from the spatial harmonic method, the singular value decomposition (SVD) method also represents an interior sound field with the basis fields through measurement over boundary samples, which has been applied to both loudspeaker weight design (Zhu et al, 2020b) and loudspeaker placement optimization (Zhu et al, 2018) in a sound zone reproduction system. In contrast to the SHD method, the number of basis functions used in the SVD method is no more than the number of sound sources used to generate a sound field.…”