Research On Direction-of-arrival Estimation Using Acoustic Vector Sensor

Localizing the direction of arrival (DOA) of an acoustic source is one of the main research works inthe array signal processing field, and the DOA estimation technology plays an important role in manyapplications such as speech, seismology, sonar and radar system. The traditional sound source localizationsolution exploits the sound pressure sensor array which is spatially distributed, and combines with thesound pressure measurements estimation technique to accomplish the DOA estimation. The introduction ofacoustic vector sensor enriches the research contents of array signal processing area greatly, and theacoustic vector sensor has been a quite hot spot in the research field of DOA estimation.The manifold of a single acoustic vector sensor contains both the azimuth and the elevationinformation, and it enables2-D DOA estimation. Acoustic vector sensor employs a co-located sensorstructure which can measure the acoustic pressure as well as the particle velocity information at the sametime. Compared with the traditional pressure sensors, the main advantage of these vector sensors is thatthey are able to make better use of the available acoustic information to outperform the pressure arrays interms of estimation accuracy. Thus, the acoustic vector sensor can achieve a more accurate estimation at alower signal-to-noise (SNR) by using fewer sensor arrays.In order to solve the problem of sound source localization in the presence of non-uniform white noise,maximum power method (MP) employing the acoustic vector sensor was proposed by Dovid Levin andSharno Gannot. Although this method can approach the Cramer-Rao lower bond (CRLB) with a suitablechoice of the weighted parameter which depends on the noise power of monopole and dipole elements ofthe acoustic vector sensor, the choice procedure of the weighted parameter is quite complicated, so that itreduces the algorithm efficiency. Hence, in the scene of time varying, this MP method is unsuitable.Therefore, in order to avoid the troublesome choice of the weighted parameter, combining non-uniformnoise covariance estimator with the MP method, a new method is proposed based on noise-whiteningtechnique, which is weighted parameter fixed maximum power algorithm (FMP).Localizing the DOA of sound source in the scene of spatially colored noise, the maximum likelihoodDOA estimation algorithm using acoustic vector sensor was put forward in the past, although it has a higher resolution, similar to the MP algorithm, the problem of low algorithm efficiency also exists. On the otherhand, in some applications such as speech and audio, the spatially noise is colored noise, so at some degreethis method has limitation. Therefore in view of that the noise covariance matrix has an Inverse-Wishartdistribution, the prior distribution information of the noise covariance matrix can be taken use of, and thejoint DOA estimation algorithm which is based on the noise pre-whitening technique and noise covariancematrix iterative technique is proposed in this paper.Finally, the estimation precision and algorithm efficiency of the two new algorithms put forward inthis paper will be verified respectively through the simulation experiments.