Qinghua Huang

Multi-beam steering for 3D audio rendering in linear phased loudspeaker arrays

In order to improve flexibility of surrounding stereo system, a novel system for 3D audio rendering is proposed in this paper based on multi-beam steering, which is composed of five beam units using linear loudspeaker arrays. By controlling accurately time delay or phase of loudspeakers, beam can be formed in a desired direction. The simulation results show that the proposed method is reasonable and effective.

Audio personalization using head related transfer function in 3DTV

In 3DTV, head related transfer function (HRTFs) can promote immersive feeling of listeners because it contains spatial information on sound source. Audio can be customized through using personalized HRTF. So, listening distortions are caused if HRTFs do not match anthropometric parameters concerning different listeners. In this paper, personalized method is proposed to customize individual HRTF based on non-negative matrix factorization (NMF) and support vector regression (SVR). The anthropometric parameters are selected and high dimensional HRTFs are decomposed into low dimensional matrix using NMF. Nonlinear regression model is derived between the selected anthropometric parameters and low dimensional matrix by SVR. Experimental results demonstrated that personalized HRTF has better performance than using the same HRTF for different listeners in 3DTV audio.

Two-Stage Decoupled DOA Estimation Based on Real Spherical Harmonics for Spherical Arrays

Spherical arrays have been widely used in direction-of-arrival (DOA) estimation in recent years. In this paper, we develop a unitary matrix to transform the complex spherical harmonics into real ones and obtain a real-valued covariance matrix after forward–backward (FB) averaging. Based on this transformation, a two-stage decoupled approach (TSDA) is proposed to decouple the estimation of the elevation and the azimuth. First, we propose a unitary spherical harmonics estimation of signal parameter via rotational invariance technique using one recurrence relation of real spherical harmonics to obtain the elevation estimation. Because of the limitation of using another recurrence relation to estimate the azimuth; second, we propose a unitary spherical harmonics root multiple signal classification (U-SHRMUSIC) to obtain the azimuth. Moreover, using the phase shift characteristic of real spherical harmonics, we also propose a low-complexity U-SHRMUSIC that can further decrease the computational load in the second stage. The proposed TSDA can not only achieve more accurate DOA estimation via FB averaging and two-stage estimation after decoupling but also reduce the computational complexity by exploiting real operations. Computer simulations, especially simulations of the real environment, validate the effectiveness of the proposed method.

Rotating spherical arrays for DOA estimation based on real-valued covariance matrix

The performance of direction-of-arrival (DOA) estimation based on spherical arrays is constrained by the number of the elements. In this paper, we propose a novel rotating spherical array, which can form more virtual elements. However, more elements will cause more cost of computation. In order to decrease the computation load, a real-valued covariance matrix is constructed by the unitary transformation. Therefore, a real-valued signal subspace can be obtained by the eigenvalue decomposition of the real-valued covariance matrix. Based on this signal subspace, we use covariance-assisted matching pursuit (CAMP) to obtain DOA estimation. Simulation results demonstrate the effectiveness of the proposed method.

Conditional and unconditional CRB of DOA estimation in spherical harmonics domain

The Cramer-Rao bound (CRB) is widely used in signal processing to characterize the estimation performance. In this paper, the conditional and unconditional CRBs are developed in spherical harmonics domain for direction-of-arrival (DOA) estimation of far-field sources using spherical arrays. The CRBs of azimuth and elevation decrease with the increasing of signal-to-noise ratios (SNRs) and snapshots. Moreover, the CRB of elevation is lower than that of azimuth.