Search

Abstract Binaural sound source localization is an important and widely used perceptually based method and it has been applied to machine learning studies by many researchers based on head-related transfer function (HRTF). Because the HRTF is closely related to human physiological structure, the HRTFs vary between individuals. Related machine learning studies to date tend to focus on binaural localization in reverberant or noisy environments, or in conditions with multiple simultaneously active sound sources. In contrast, mismatched HRTF condition, in which the HRTFs used to generate the training and test sets are different, is rarely studied. This mismatch leads to a degradation of localization performance. A basic solution to this problem is to introduce more data to improve generalization performance, which requires a lot. However, simply increasing the data volume will result in data-inefficiency. In this paper, we propose a data-efficient method based on deep neural network (DNN) and clustering to improve binaural localization performance in the mismatched HRTF condition. Firstly, we analyze the relationship between binaural cues and the sound source localization with a classification DNN. Different HRTFs are used to generate training and test sets, respectively. On this basis, we study the localization performance of DNN model trained by each training set on different test sets. The result shows that the localization performance of the same model on different test sets is different, while the localization performance of different models on the same test set may be similar. The result also shows a clustering trend. Secondly, different HRTFs are divided into several clusters. Finally, the corresponding HRTFs of each cluster center are selected to generate a new training set and to train a more generalized DNN model. The experimental results show that the proposed method achieves better generalization performance than the baseline methods in the mismatched HRTF condition and has almost equal performance to the DNN trained with a large number of HRTFs, which means the proposed method is data-efficient.

Head-related transfer function (HRTF) plays an important role in three-dimensional spatial sound system. However, the direct application of a large amount of original HRTF data would involve a great deal of computational burden, especially for high-spatial-resolution individual HRTF. To address this problem, we propose a novel compression method (called TT-Tucker) combining Tucker model with tensor train decomposition based on a 5-order HRTF tensor model developed in subspaces of an ear, subject, azimuth, elevation, and frequency. Lots of HRTF data can be decomposed into several low-parametric factors representing the key spectrum information of HRTF by capturing the hidden interactions among different subspaces. To evaluate the reconstruction performance, the numerical experiments were conducted on the CIPIC HRTF database. Under the same compression ratio of nearly 98%, the results suggest that the proposed method has a better performance in spectral distortion and signal-to-distortion ratio than that of the usual tensor method and the standard method principal component analysis (PCA). Moreover, the subjective listening test shows that the TT-Tucker method performs better in that, the compressed and reconstructed HRTF is closer to the original HRTF in the sound localization similarity.

The modern and future communication systems demand higher data rate exceeds current capabilities. Faster-than-Nyquist (FTN) signaling is a promising technique for the reason it can improve the spectral efficiency. This paper deals with the receiver design for FTN signaling equalization in nonlinear channels. To handle the correlated noise samples induced by FTN signaling, we use the auto-regressive process to model the colored noise. Moreover, due to the nonlinearities, the complexity of receiver becomes prohibitively high. To reduce the complexity, we linearize the Volterra model and formulate the considered system as linear system. Then, based on the factor graph and message passing framework, the FTN symbols are detected with low complexity. Simulation results show that the proposed low complexity receiver can achieve similar performance of Nyquist signaling while increase 25% data rate.

This paper deals with low-complexity joint channel estimation and decoding for faster-than-Nyquist (FTN) signaling over frequency selective fading channels. The inter-symbol interference (ISI) imposed by FTN signaling and the frequency selective channel are intentionally separated to fully exploit the known structure of the FTN-induced ISI. Colored noise due to the faster sampling rate than that of the Nyquist signaling system is approximated by autoregressive process. A Forney style factor graph representation of the FTN system is developed and Gaussian message passing is performed on the graph. Expectation propagation (EP) is employed to approximate the message from channel decoder to Gaussian distribution. Since the inner product between FTN symbols and channel coefficients is infeasible by belief propagation (BP), we propose to perform variational message passing (VMP) on an equivalent soft node in factor graph to tackle this problem. Simulation results demonstrate that the proposed low-complexity hybrid BP-EP-VMP algorithm outperforms the existing methods in FTN system. Compared with the Nyquist counterpart, FTN signaling with the proposed algorithm is able to increase the transmission rate by over 40%, with only negligible BER performance loss.

We consider the challenging problem of joint angle estimation and signal reconstruction for coherently distributed (CD) sources in massive multiple-input-multiple-output (MIMO) systems employing uniform rectangular arrays. A simplified method inspired by the two-dimensional (2-D) unitary estimating signal parameters via rotational invariance technique (ESPRIT) is proposed to estimate both the central angle and the angular spread without the need for a spectrum peak search and covariance matrix matching process. We first approximate the 2-D generalized steering vector expressed as a Schur-Hadamard product by a pair of one-dimensional generalized steering vectors. Then, we obtain two approximate rotational invariance relationships with respect to the central angles of the CD sources using a linear approximation of the individual generalized steering vectors of the azimuth and elevation subarrays. With the aid of this approximate decomposition, a new unitary ESPRIT-inspired algorithm is conceived to automatically pair the 2-D central angle estimations and a novel method capable of bypassing the high-complexity search process is proposed for angular spread estimation. Furthermore, the closed-form approximate Cramer-Rao lower bounds are derived for the estimators of both the central angles and the angular spreads. The complexity of the proposed estimator is also analyzed. Additionally, the orthogonality of the generalized steering vectors is proved, which enables us to propose a low-complexity method to reconstruct the CD signal matrix by replacing the inversion operator with the conjugate transpose operator. The simulation results demonstrate the efficiency of our proposed approach.

With the increasing demand for high quality audio, stereo audio coding has become more and more important. In this paper, a multi-frame coding method based on Principal Component Analysis (PCA) is proposed for the compression of audio signals, including both mono and stereo signals. The PCA-based method makes the input audio spectral coefficients into eigenvectors of covariance matrices and reduces coding bitrate by grouping such eigenvectors into fewer number of vectors. The multi-frame joint technique makes the PCA-based method more efficient and feasible. This paper also proposes a quantization method that utilizes Pyramid Vector Quantization (PVQ) to quantize the PCA matrices proposed in this paper with few bits. Parametric coding algorithms are also employed with PCA to ensure the high efficiency of the proposed audio codec. Subjective listening tests with Multiple Stimuli with Hidden Reference and Anchor (MUSHRA) have shown that the proposed PCA-based coding method is efficient at processing stereo audio.

This paper considers a 2-user multiple-input single-output (MISO) interference channel with confidential messages (IFC-CM), in which the Rician channel model is assumed. The coordinated beamforming vectors at the two transmitters have the similar parameterizations as those for perfect CSI, which could be optimized jointly and achieved by agreeing on the real parameters between the two users. Our main contribution is that a quadratic relationship between the two real-valued parameters can be derived for the Rician channel to reach the ergodic secrecy rate balancing point. Simulation results present the secrecy performance over the 2-user MISO IFC-CM scenario.