ROBUST SPHERICAL MICROPHONE ARRAY BEAMFORMING WITH MULTI-BEAM-MULTI-NULL STEERING, AND SIDELOBE CONTROL Haohai Sun
Shefeng Yan
U. Peter Svensson
sfyan@ieee.org svensson@iet.ntnu.no haohai.sun@iet.ntnu.no
Department of Electronics and Telecommunications,
Norwegian University of Science and Technology, 7491 Trondheim, Norway
ABSTRACT
A spherical harmonics domain microphone array beamforming approach is proposed. It unifies 3D multi-beam forming with tractable mainlobe levels, automatic multi-null steering, sidelobe control, and robustness control into one optimization framework, using a single spherical microphone array. The optimum array weights are designed by maintaining distortionless responses in multiple mainlobe directions and guaranteeing all sidelobes below given threshold values, while minimizing the beamformer output power. A weight vector norm constraint is also employed to improve the robustness of the beamformer. A convex optimization formulation is derived, and implemented by the second order cone programming (SOCP) method. Design examples demonstrate a satisfactory performance.
Index Terms—Array processing, spherical microphone array, multi-beam, sidelobe control, null steering.
1. INTRODUCTION
The spherical microphone array beamformer, a so-called wave field decomposition beamformer, has recently become an important research subject since flexible 3D beam pattern synthesis can be realized, and the array processing can be performed using the elegant mathematical framework in the spherical harmonics domain. Studies on such kinds of beamformer have been done for: regular beam pattern synthesis and steering [1], delay-and-sum method that shows high robustness with the compromise of low directivity at low frequencies [2], robust optimal beam pattern design [3], [4], [5], non-symmetrical beam pattern with fixed nulls steered to the interferences, to achieve better signal to noise ratio [6], etc. In video conferencing and tele-presence applications, microphone array beamformers are commonly used to capture and equalize the voices from all near-end speakers, and attenuate acoustic echo, interferences, and reverberation, to obtain good voice quality and full-duplex performance. So it is desired to have 3D multi-beam (mainlobes) capturing and equalizing all the desired near-end voices, and have low sidelobe with adaptive multi-null suppressing the multi-channel echo and interferences. In addition, robustness of the whole system should also be considered and included together with all other features into a single optimization framework, in order to obtain the global optimum solution for all these mutually correlated performance parameters. In this paper, we are trying to achieve this goal by formulating all these beamforming problems into convex optimization formulations, especially in the spherical harmonics
978-1-4244-3679-8/09/$25.00 ©2009 IEEE113
domain. The optimization criteria for performance parameters are presented, and the global optimum solution for array weights are solved by the SOCP method. The broadband beamformer can then be simply realized by decomposing the frequency band into narrower frequency bins and processing each bin with the narrowband beamformer.
2. BACKGROUND DESCRIPTION
Consider a sphere of radius r, whose surface is perfectly sampled by S microphones, and spatial aliasing can be negligible. A unit magnitude plane wave signal from direction 0, with wave number k, impings on each microphone position s, s=1, 2,…, S, on the sphere. The expressions of sound pressure at each microphone in both the space domain and the spherical harmonics domain are [2], [5] p(kr,Ω0,Ω∞n
m*ms)=
∑bn(kr)∑Y
n(Ω0)Yn(Ωs) (1)
n=0
m= n
pm*nm(kr,Ω0)=bn(kr)Yn(Ω0) (2) where =(θ,φ) is the position on the sphere in standard spherical coordinates [5], Ymn(Ω) the spherical harmonics of order n and degree m, and superscript * stands for the complex conjugation. bn(kr) depends on the sphere configuration, as given by [1], [2].
Since a wave field decomposition beamforming algorithm is based on an orthogonal set of spherical harmonics, which are obtained by the spherical Fourier transform [2], it is necessary to formulate the signal model into a squared integrable function on a unit sphere. In this case, suppose a compact spherical microphone array is placed in a room. All signal sources are assumed to be located in the far field of the array, and the early reflections in room are modeled as point image sources while the late reverberation is modeled as isotropic noise [5]. Then we assume that L signals of interest and D interferences impinge on the sphere from directions 1, 2,…, L, L+1, …, L+D, and in addition, noise is present. Then the space domain sound pressure for each microphone position can be written as
x(kr,Ωs)
=∑L R p(kr,Ω)+∑p(kr,ΩS l,Ωs)Sl(ωl,c,Ωs)αl,c(ω)l,c(ω)exp(jωτl,c)=1 +l=1 c(3) ∑L+D
R p(kr,Ω(ω)+ d,Ωs)Sd1 ∑p(kr,Ωd,c,Ωs)αd,c(ω)Sd,c(ω)exp(jωτd,c)d=L+c=1 +N(ω,Ωs), s=1, 2, ..., Swhere {SLa(ω)}a+=1
D
are the L+D source signal spectra,
2002] o On Robust Capon Beamforming and Diagonal...consider spherical uncertainty set: o Solution at ...Array steering vector with calibration errors where...
Robust adaptive beamforming based on iterative implementation...and the steering vector (SV) of the Capon ...spherical uncertainty set, and the WC beamformer ...
Robust adaptive beamforming based on iterative implementation...and the steering vector (SV) of the Capon ...spherical uncertainty set, and the WC beamformer ...
热门文档