Adaptive modulation and coding (AMC) is an effective technique for mitigating the severe dynamics of underwater acoustic (UWA) channels. However, its performance strongly relies on accurate channel state prediction, and prediction errors can cause significant performance degradation. Most existing approaches primarily focus on time-domain (TD) prediction, but the rapid variability of UWA channels makes long-term prediction highly challenging and limits their effectiveness. To overcome this limitation, we exploit the relative stability and sparsity of channel parameters in the delay–Doppler (DD) domain and propose a multi-scale convolutional long short-term memory (ConvLSTM) prediction framework. The DD-domain formulation reduces model complexity and supports multi-step forecasting, while the multi-scale architecture captures channel dynamics across different temporal scales and structural levels. The effectiveness of our method is validated through extensive experiments on real-world UWA measurements. The proposed framework achieves a normalized mean square error (NMSE) of 0.0818 in the DD-domain, significantly lower than the 0.2196 NMSE of direct TD prediction. Compared to conventional predictors, the proposed framework consistently reduces NMSE by approximately 70% across multiple datasets, thereby enhancing the robustness and feasibility of long-horizon channel forecasting for AMC systems.

Orthogonal time frequency space (OTFS) modulation has garnered significant interest for its robust performance in fast time-varying channels, making it suitable for mobile underwater acoustic (UWA) communication system. This article introduces OTFS modulation to the UWA system and proposes a low-complexity minimum mean-squared error (MMSE) turbo equalization method. Leveraging the characteristics of UWA channels in the delay-Doppler (DD) domain, the method employs symbol-level MMSE equalization. By focusing processing on signals within the DD domain’s interference range, it reduces the channel matrix size, thereby lowering complexity. Given the long delay spread and large Doppler shift of UWA channels, symbol-level MMSE equalization inherently involves high complexity. To mitigate this, we propose two methods to further reduce the computational load associated with matrix inversion. First, we utilize common blocks in the channel matrix and employ a block iterative matrix inversion algorithm to retain computational results, thereby avoiding repeated inversions of the large dimensional matrix. Additionally, we enhance the diagonal dominance property of the channel matrix using the discrete Fourier transform (DFT) matrix. Subsequently, we approximate the inversion using the second-order Neumann series decomposition, further lowering computational complexity. Simulation results and experimental validations at Danjiangkou Lake demonstrate the efficacy of the proposed low-complexity iterative equalization algorithm.

Faced with the dilemma of dense underwater collision avoidance near ports, the lack of small moving obstacle detection strategy can easily lead to the increase of crash risk. Hence, focusing on the application of active sonar, a short-time moving obstacle extraction algorithm based on joint conditional entropy synergy is proposed. On the basis of beamforming, this method combines low-rank matrix factorization (LMF) and variational modal decomposition (VMD) to rapid achieve single frame reverberation and dynamic background suppression. Aiming at screening out moving obstacles with weak echoes, a matching model based on joint conditional entropy similarity measure is derived, and the distance and direction of moving obstacles (such as divers) are extracted by matching the sparse matrix of reference frame and current frame. This method significantly reduces the computational overhead of reverberation suppression in multi-frame LMF, and effectively alleviates the issue of missing detection of small targets. Experimental results verify the reliability and accuracy of the proposed algorithm in quickly detecting small moving obstacles.

The time-reassignment method and time-synchrosqueezing transform show a good ability in impulse-like signal analysis. This kind of method calculates the group delay (GD) estimators at the spread TF locations first and then relocates the spread TF energy into the estimated GD trajectories to yield a high-concentration TF representation. However, computing the GD estimators for every TF energy point can lead to inaccurate location and energy diffusion. To address this issue, a new feature extractor called the second-order GD equation is proposed, which focuses only on the TF points on the GD to characterize the frequency-varying models withN-order amplitude and second-order phase. The theoretical analysis of the second-order GD equation is highlighted. By combining a fixed-point iterative algorithm with the extracting transform, we introduce a novel weighted transient-extracting transform based on the solutions of the second-order GD equation. This transform enhances TF distribution concentration while retaining reconstruction capability. Numerical simulations demonstrate that our proposed method improves the average performance for TF concentration by 3% across various noise levels and enhances the accuracy of GD location by over 0.2 within a SNR range of -1 dB to 20 dB, compared to current state-of-the-art TF analysis methods. Finally, the proposed TF transform is applied to analyze seismic data for low-frequency shadow attributes and thin layer characterization. The results clearly illustrate its effectiveness in seismic processing and interpretation.

The sound speed profile (SSP) perturbations by oceanic dynamical processes are transferred to the sound field via acoustical models, introducing computational inaccuracies. Hence, an accurate estimation of the SSP holds significant importance. This letter introduces a method for SSP estimation based on the multipath delay structure of large depth vertical arrays. This approach capitalizes on the sensitivity of array’s multipath delay structure to SSP perturbations and estimates SSP using a single explosive charge signal, thus minimizing the reliance on extensive acoustic data. The method primarily comprises three components: the establishment of a sound speed perturbation model, construction of a synthetic dataset, and parameter estimation based on the genetic algorithm. This letter employs the proposed model to generate foundational samples and maximizes the coverage of authentic SSPs, thereby enhancing the precision of SSP estimation. The acoustic observational SSP is obtained from deep-sea regions of the Western Pacific to compare with estimated SSP, the result of traditional method and publicly available online datasets. The results indicate that the method proposed in this letter achieves commendable accuracy, and the root-mean-square error (RMSE) between the estimated SSP and the observed SSP is only 1.3 m/s.

In the rapidly advancing field of Underwater Acoustic Communication (UAC), channel prediction remains a major challenge, exacerbated by the complicated nature of ocean environments. This paper introduces an innovative Multi- Task Learning (MTL) framework for time-varying Underwater Acoustic (UWA) channel prediction. By decomposing the highdimensional Channel Impulse Response (CIR) prediction into interconnected tasks, the proposed framework leverages a Shared Feature Learning (SFL) layer, capturing intricate dependencies underlying UWA channels. To validate its efficacy, we conducted thorough evaluations, leveraging real-world data from two distinct at-sea experiments conducted in Wuyuan Bay, China. A comprehensive comparative study of various configurations for the SFL layer, ranging from commonly used Recurrent Neural Network (RNN)-based models to the more advanced transformer structure, further underscores the flexibility and broad applicability of our MTL framework for handling various challenging UWA environments.

Adversarial attack approaches to speaker identification either need high computational cost or are not very effective, to our knowledge. To address this issue, in this letter, we propose a novel generation-network-based approach, called symmetric saliency-based encoder-decoder (SSED), to generate adversarial voice examples to speaker identification. It contains two novel components. First, it uses a novel saliency map decoder to learn the importance of speech samples to the decision of a targeted speaker identification system, so as to make the attacker focus on generating artificial noise to the important samples. It also proposes an angular loss function to push the speaker embedding far away from the source speaker. Our experimental results demonstrate that the proposed SSED yields the state-of-the-art performance, i.e. over 97% targeted attack success rate and a signal-to-noise level of over 39 dB on both the open-set and close-set speaker identification tasks, with a low computational cost.

Although the security of automatic speaker verification (ASV) is seriously threatened by recently emerged adversarial attacks, there have been some countermeasures to alleviate the threat. However, many defense approaches not only require the prior knowledge of the attackers but also possess weak interpretability. To address this issue, in this paper, we propose anattacker-independentandinterpretablemethod, namedlearnable mask detector(LMD), to separate adversarial examples from the genuine ones. It utilizes score variation as an indicator to detect adversarial examples, where the score variation is the absolute discrepancy between the ASV scores of an original audio recording and its transformed audio synthesized from its masked complex spectrogram. A core component of the score variation detector is to generate the masked spectrogram by a neural network. The neural network needs only genuine examples for training, which makes it an attacker-independent approach. Its interpretability lies that the neural network is trained to minimize the score variation of the targeted ASV, and maximize the number of the masked spectrogram bins of the genuine training examples. Its foundation is based on the observation that, masking out the vast majority of the spectrogram bins with little speaker information will inevitably introduce a large score variation to the adversarial example, and a small score variation to the genuine example. Experimental results with 12 attackers and two representative ASV systems show that our proposed method outperforms five state-of-the-art baselines. The extensive experimental results can also be a benchmark for the detection-based ASV defenses.

Owing to the strong reverberation and various obstacle echoes interference in harbor scenes, the avtive detection performance of tracking methods for weak targets will be seriously reduced. Moreover, the conventional dereverberation and tracking methods generally cannot effectively separate the weak target under the overlapping echoes of multiple scattering sources. Focusing on solving the above problems, the correlation residual cumulative clustering (CRCC) algorithm is proposed to extract scattering features of weak target motion. There are two innovations in this paper: Firstly, the complex cepstrum filter is improved to suppress strong reverberation. Secondly, the correlation residual accumulation of adjacent frames is extracted to detect the reduction matrix containing the target trajectory. Finally, the FCM objective function is optimized via spatial-temporal constraint, thus the weak target can be effectively separated from the overlapping giant interference. The experimental results indicate the strong anti-jamming, low detection loss and short-term accumulation of our proposed model, which are remarkably superior to the traditional posterior probability models.

Acoustic vector sensors have been widely used for direction-of-arrival (DOA) estimation in hydroacoustics and aeroacoustics. The orientation of the acoustic vector sensor must be measured, e.g., by an attitude sensor, before its application to DOA estimation. However, the measured orientation generally deviates slightly from its actual orientation because of imperfect attitude measurements. This paper investigates how random deviations in the orientation degrade the performance of DOA estimation. We derive a closed-form approximation of the hybrid Cramer-Rao bound for DOA estimation by modeling the deviations as zero-mean Gaussian variables with small variances. A relationship based on a determinant-differential formula is introduced to avoid a complicated brute-force solution, yielding a sufficiently simple expression with quantitative observations. Numerical results illustrate the high accuracy and effectiveness of the closed-form approximation.

A method based on delay-and-sum (DAS) of cross-correlation functions is proposed for multipath time-delay estimation of a broadband signal received by a vertical line array (VLA) in the reliable acoustic path (RAP) environment. The multipath arrivals are mainly concerned with the direct (D) and surface-reflected (SR) paths. Three types of multipath time-delays are obtained, namely, the average D-D time-delay and SR-SR time-delay between two adjacent hydrophones, and the D-SR time-delay at the first hydrophone. The proposed method has significant advantages for underwater applications. First, the D and SR arrivals are successfully separated from time-delay estimation point of view, outperforming the estimation of direction of arrival with conventional beamforming. Second, the estimations of the average D-D time-delay and SR-SR time-delay between two adjacent hydrophones occur at the diagonal line of the two-dimensional DAS outputs, facilitating the automatic detection and extraction of multipath time-delays. Third, the estimation is robust against noisy background noise due to the spatial accumulative effect. The method is applied to experimental data for explosive source at ~17 km range and 300 m depth, where it is shown that simple, stable, and fairly accurate multipath time-delay estimation can be made.

Currently, synthetic aperture radar (SAR) images are generally used for ship identification. However, the SAR can only be used to classify surface ships without any underwater target. In contrast, since sonar can receive radiated noises from both vessels and underwater targets, its images can be used to effectively identify targets at different depths. However, due to the shortage of underwater target data and the difficulty in modeling, sonar images are hardly used for deep learning (DL). To solve such problems, this letter proposes a compound convolutional neural network (CSDN) based on a shared latent sparse feature (SLS) and a deep belief network (DBN) to learn striation-based sonar images. This letter has three contributions. First, this letter uses striation images to overcome the lack of training data for CNNs. Second, the DBN is applied to optimize fuzzy or discontinuous fringes, and an SLS feature is proposed to represent the interference fringe. Finally, the two features are exploited to separately train CNNs, combined with weight, to enhance the accuracy of classifying water targets. The experimental results indicate that, compared with the other DL-based models-VGG, SSD, RFCN, and SCDAE-a CSDN is more stable for different datasets and has the highest accuracy of up to 93.34%.

Self-organizing map (SOM) method combined with the empirical orthogonal function was used to reconstruct the subsurface temperature field by using sea surface data in the Northwestern Pacific Ocean. In contrast to the traditional method, SOM method can extract nonlinear relations from the data and is more suitable for nonlinear dynamics in the ocean. Error statistics show that SOM method provides reconstructions of the subsurface temperature field with the majority of relative errors below 20% at 0-1000-m depth.

Telemonitoring of diaphragmatic electromyogram (EMGdi) signal in wireless sensor networks (WSNs) and Internet of Things (IoT) holds the promise to be an evolving direction in personalized medicine. The WSNs and IoT enable EMGdi information telemonitoring and communications technologies play important roles in the process of personal medical care, especially for the respiratory diseases. However, while designing such a system, one should consider the required functionality, miniaturization, energy efficiency, etc., to make fewer resources required in WSNs and IoT. Conventional methods of data acquisition cannot energy-effectively compress data with reduced device costs. Different from the traditional compression methods, compressed sensing (CS) takes promising steps toward these challenges. Unfortunately, EMGdi is not sparse in time domain. Hence, current CS algorithms are extremely difficult to use directly for recovering EMGdi. In order to satisfy the requirements of applications of personal medical care in WSNs and IoT, this study proposes an approximated ι0norm based method to search the solution via the gradient descent method, then projects the searched solution to the reconstruction feasible set. Meanwhile, this study adopts a new wavelet threshold based method to denoise the electrocardiographic interference. Experimental results are provided to testify the performance of the proposed methods.

Localized multiple kernel learning (LMKL) is an attractive strategy for combining multiple heterogeneous features with regard to their discriminative power for each individual sample. However, the learning of numerous local solutions may not scale well even for a moderately sized training set, and the independently learned local models may suffer from overfitting. Hence, in existing local methods, the distributed samples are typically assumed to share the same weights, and various unsupervised clustering methods are applied as preprocessing. In this paper, to enable the learner to discover and benefit from the underlying local coherence and diversity of the samples, we incorporate the clustering procedure into the canonical support vector machine-based LMKL framework. Then, to explore the relatedness among different samples, which has been ignored in a vector -norm analysis, we organize the cluster-specific kernel weights into a matrix and introduce a matrix-based extension of the -norm for constraint enforcement. By casting the joint optimization problem as a problem of alternating optimization, we show how the cluster structure is gradually revealed and how the matrix-regularized kernel weights are obtained. A theoretical analysis of such a regularizer is performed using a Rademacher complexity bound, and complementary empirical experiments on real-world data sets demonstrate the effectiveness of our technique.

This brief analyzes the effects of regularization variations in the localized kernel weights on the hypothesis generated by localized multiple kernel learning (LMKL) algorithms. Recent research on LMKL includes imposing different regularizations on the localized kernel weights and has led to varying formulations and solution strategies. Following the stability analysis theory as presented by Bousquet and Elisseeff, we give stability bounds based on the norm of the variation of localized kernel weights for three LMKL methods cast in the support vector machine classification framework, including vector -norm LMKL, matrix-regularized -norm LMKL, and samplewise -norm LMKL. Further comparison of these bounds helps to qualitatively reveal the performance differences produced by these regularization methods, that is, matrix-regularized LMKL achieves superior performance, followed by vector -norm LMKL and samplewise -norm LMKL. Finally, a set of experimental results on ten benchmark machine learning UCI data sets is reported and shown to empirically support our theoretical analysis.

Our objective is to train support vector machines (SVM)-based localized multiple kernel learning (LMKL), using the alternating optimization between the standard SVM solvers with the local combination of base kernels and the sample-specific kernel weights. The advantage of alternating optimization developed from the state-of-the-art MKL is the SVM-tied overall complexity and the simultaneous optimization on both the kernel weights and the classifier. Unfortunately, in LMKL, the sample-specific character makes the updating of kernel weights a difficult quadratic nonconvex problem. In this paper, starting from a new primal-dual equivalence, the canonical objective on which state-of-the-art methods are based is first decomposed into an ensemble of objectives corresponding to each sample, namely, sample-wise objectives. Then, the associated sample-wise alternating optimization method is conducted, in which the localized kernel weights can be independently obtained by solving their exclusive sample-wise objectives, either linear programming (for l1-norm) or with closed-form solutions (for lp-norm). At test time, the learnt kernel weights for the training data are deployed based on the nearest-neighbor rule. Hence, to guarantee their generality among the test part, we introduce the neighborhood information and incorporate it into the empirical loss when deriving the sample-wise objectives. Extensive experiments on four benchmark machine learning datasets and two real-world computer vision datasets demonstrate the effectiveness and efficiency of the proposed algorithm.

Our objective is to train SVM based Localized Multiple Kernel Learning with arbitrary$l_{p}$-norm constraint using the alternating optimization between the standard SVM solvers with the localized combination of base kernels and associated sample-specific kernel weights. Unfortunately, the latter forms a difficult$l_{p}$-norm constraint quadratic optimization. In this letter, by approximating the$l_{p}$-norm using Taylor expansion, the problem of updating the localized kernel weights is reformulated as a non-convex quadratically constraint quadratic programming, and then solved via associated convex Semi-Definite Programming relaxation. Experiments on ten benchmark machine learning datasets demonstrate the advantages of our approach.