Generalized likelihood ratio test for voiced-unvoiced decision in noisy speech using the harmonic model

In this paper, a novel method for voiced-unvoiced decision within a pitch tracking algorithm is presented. Voiced-unvoiced decision is required for many applications, including modeling for analysis/synthesis, detection of model changes for segmentation purposes and signal characterization for indexing and recognition applications. The proposed method is based on the generalized likelihood ratio test (GLRT) and assumes colored Gaussian noise with unknown covariance. Under voiced hypothesis, a harmonic plus noise model is assumed. The derived method is combined with a maximum a-posteriori probability (MAP) scheme to obtain a pitch and voicing tracking algorithm. The performance of the proposed method is tested using several speech databases for different levels of additive noise and phone speech conditions. Results show that the GLRT is robust to speaker and environmental conditions and performs better than existing algorithms.     

Source localization in a waveguide using polynomial rooting

Source localization in acoustic waveguides involves a multidimensional search procedure. We propose a new algorithm in which the search in the depth direction is replaced by polynomial rooting. Using the proposed algorithm, range and depth estimation by a vertical array requires a 1-D search procedure. For a 3-D localization problem (i.e., range, depth, and direction-of-arrival (DOA) estimation), the algorithm involves a 2-D search procedure. Consequently, the proposed algorithm requires significantly less computation than other methods that are based on a brute-force search procedure over the source location parameters. In order to evaluate the performance of the algorithm, an error analysis is carried out, and Monte-Carlo simulations are performed. The results are compared with the Cramer-Rao bound (CRB) and to the maximum likelihood (ML) simulation performance. The algorithm is shown to be efficient, while being computationally simpler than the ML or the Bartlett processors. The disadvantage of the algorithm is that its SNR threshold occurs in lower SNR than in the ML algorithm     

Direction of Arrival Estimation in the Presence of Noise Coupling in Antenna Arrays

The direction of arrival (DOA) estimation problem in the presence of signal and noise coupling in antenna arrays is addressed. In many applications, such as smart antenna, radar and navigation systems, the noise coupling between different antenna array elements is often neglected in the antenna modeling and thus, may significantly degrade the system performance. Utilizing the exact noise covariance matrix enables to achieve high-performance source localization by taking into account the colored properties of the array noise. The noise covariance matrix of the antenna array consists of both the external noise sources from sky, ground and interference, and the internal noise sources from amplifiers and loads. Computation of the internal noise covariance matrix is implemented using the theory of noisy linear networks combined with the method of moments (MoM). Based on this noise statistical analysis, a new four-port antenna element consisting of two orthogonal loops is proposed with enhanced source localization performance. The maximum likelihood (ML) estimator and the Cramer-Rao lower bound (CRLB) for DOA estimation in the presence of noise coupling is derived. Simulation results show that the noise coupling in antenna arrays may substantially alter the source localization performance. The performance of a mismatched ML estimator based on a model which ignores the noise coupling shows significant performance degradation due to noise coupling. These results demonstrate the importance of the noise coupling modeling in the DOA estimation algorithms.     

Source localization using vector sensor array in a multipath environment

Coherent signals from distinct directions is a natural characterization of the multipath propagation effect. This paper addresses the problem of coherent/fully correlated source localization using vector sensor arrays. The maximum likelihood (ML) and minimum-variance distortionless response (MVDR) estimators for source direction-of-arrival (DOA) and signal polarization parameters are derived. These estimators require no search over the polarization parameters. In addition, a novel method for "decorrelating" the incident signals is presented. This method is based on the polarization smoothing algorithm (PSA) and enables the use of eigenstructure-based techniques, which assume uncorrelated or partially correlated signals. The method is implemented as a preprocessing stage before applying eigenstructure-based techniques, such as MUSIC. Unlike other existing preprocessing techniques, such as spatial smoothing and forward-backward (FB) averaging, this method is not limited to any specific array geometry. The performance of the proposed PSA preprocessing combined with MUSIC is evaluated and compared to the Crame/spl acute/r-Rao Bound (CRB) and the ML and MVDR estimators. Simulation results show that the MVDR and PSA-MUSIC asymptotically achieve the CRB for a scenario with two coherent sources with and without an uncorrelated interference source. A sensitivity study of PSA-MUSIC to source polarization was also conducted via simulations.     

Relationships between adaptive minimum variance beamforming andoptimal source localization

For many years, the popular minimum variance (MV) adaptive beamformer has been well known for not having been derived as a maximum likelihood (ML) estimator. This paper demonstrates that by use of a judicious decomposition of the signal and noise, the log-likelihood function of source location is, in fact, directly proportional to the adaptive MV beamformer output power. In the proposed model, the measurement consists of an unknown temporal signal whose spatial wavefront is known as a function of its unknown location, which is embedded in complex Gaussian noise with unknown but positive definite covariance. Further, in cases where the available observation time is insufficient, a constrained ML estimator is derived here that is closely related to MV beamforming with a diagonally loaded data covariance matrix estimate. The performance of the constrained ML estimator compares favorably with robust MV techniques, giving slightly better root-mean-square error (RMSE) angle-of-arrival estimation of a plane-wave signal in interference. More importantly, however, the fact that such optimal ML techniques are closely related to conventional robust MV methods, such as diagonal loading, lends theoretical justification to the use of these practical approaches     

Three-dimensional source localization in a waveguide

This paper deals with three-dimensional (3-D) passive localization of a narrowband point source in a 2½-dimensional waveguide using an array of sensors. Two different maximum likelihood (ML) procedures for estimating the source range, depth, and direction-of-arrival (DOA) based on the normal mode representation of the received data are studied. In the first procedure, ML estimation of range and depth is applied on the data collected by a vertical array, and DOA is estimated using the ML algorithm on the data received by a separate, horizontal array. In the second procedure, the ML algorithm is applied on the data received by a two-dimensional (2-D), hybrid array for simultaneously estimating of all three source location parameters. Our study shows that although a horizontal array is sufficient for 3-D localization, to reduce sensitivity of the localization algorithm, a 2-D array should be used. The presented performance analysis of the two algorithms enables one to determine the performance losses in using the stage-wise, suboptimal algorithm relative to the optimal one in any given scenario. Numerical examples with channel parameters, which are typical to shallow water source localization, show performance losses of 0-3 dB. Simulation results of the two ML algorithms and their comparison with the Cramer-Rao bound (CRB) support the theory     

Detection of environmental mismatch in a shallow water waveguide

Performance of source localization, particularly using matched-field processing techniques, is very sensitive to modeling mismatch. This paper presents a novel test for detecting modeling mismatch in a bounded propagation medium, such as shallow water. The test is based on the fact that when there are no model uncertainties, the modal spectrum of the received signal is strictly limited to an a priori known set of modes. Mismatch in the modal eigenfunction of the assumed model causes the modal spectrum out of this band to be nonzero. The proposed test requires no prior knowledge about the radiating sources. The performance of the proposed test is evaluated via computer simulations of a benchmark shallow water channel. It is shown that the test is sensitive to mismatch in the channel depth and in bottom sound speeds. The ability of the test to identify array tilt mismatch is also presented. An application of the test for validation of maximum-likelihood localization results is also discussed     

Electrical Dispersion Compensation Equalizers in Optical Direct- and Coherent-Detection Systems

We study the performances of several electrical dispersion compensation (EDC) equalizers in the presence of chromatic dispersion (CD) and polarization mode dispersion (PMD) for optical coherent and direct detection on-off keying systems. The EDCs that are analyzed include the decision-feedback equalizer, linear equalizer, and maximum-likelihood sequence estimator (MLSE). We present an inclusive quantitative analysis of the performance difference between the various techniques. The MLSE gives a good indication of the best possible performance     

MIMO-AR System Identification and Blind Source Separation for GMM-Distributed Sources

The problem of blind source separation (BSS) and system identification for multiple-input multiple-output (MIMO) auto-regressive (AR) mixtures is addressed in this paper. Two new time-domain algorithms for system identification and BSS are proposed based on the Gaussian mixture model (GMM) for sources distribution. Both algorithms are based on the generalized expectation-maximization (GEM) method for joint estimation of the MIMO-AR model parameters and the GMM parameters of the sources. The first algorithm is derived under the assumption of unstructured input signal statistics, while the second algorithm incorporates the prior knowledge about the structure of the input signal statistics due to the statistically independent source assumption. These methods are tested via simulations using synthetic and audio signals. The system identification performances are tested by comparison between the state transition matrix estimation using the proposed algorithms and the well-known multidimensional Yule-Walker solution followed by an instantaneous BSS method. The results show that the proposed algorithms outperform the Yule-Walker based approach. The BSS performances were compared to other convolutive BSS methods. The results show that the proposed algorithms achieve higher signal-to-interference ratio (SIR) compared to the other tested methods.     

Theoretical performance limits on tropospheric refractivityestimation using point-to-point microwave measurements

Ducted propagation above the ocean surface can seriously impact shipboard radar and communications. Point-to-point microwave measurements have been proposed as a means of estimating tropospheric refractivity for the purposes of characterizing surface-based ducts. This paper addresses the theoretical performance of refractivity estimates that can be made by combining field measurements at different frequencies with prior statistics of refractivity variation. Parameterizing the refractivity profile using empirical orthogonal functions derived from a historical database, both Cramer-Rao performance bounds and the maximum a posteriori (MAP) estimate are discussed using coherent or incoherent signals. Results obtained using a realistic model of refractivity conditions off Southern California suggest that multifrequency propagation measurements can significantly improve the estimation of refractivity and propagation loss profiles