Orthogonal frequency division multiplexing (OFDM) provides a promising modulation technique for underwater acoustic (UWA) communication systems. It is indispensable to obtain channel state information for channel estimation to handle the various channel distortions and interferences. However, the conventional channel estimation methods such as least square (LS), minimum mean square error (MMSE) and back propagation neural network (BPNN) cannot be directly applied to UWA-OFDM systems, since complicated multipath channels may cause a serious decline in performance estimation. To address the issue, two types of channel estimators based on deep neural networks (DNNs) are proposed with a novel training strategy in this paper. The proposed DNN models are trained with the received pilot symbols and the correct channel impulse responses in the training process, and then the estimated channel impulse responses are offered by the proposed DNN models in the working process. The experimental results demonstrate that the proposed methods outperform LS, BPNN algorithms and are comparable to the MMSE algorithm in respect to bit error rate and normalized mean square error. Meanwhile, there is no requirement of prior statistics information about channel autocorrelation matrix and noise variance for our proposals to estimate channels in UWA-OFDM systems, which is superior to the MMSE algorithm. Our proposed DNN models achieve better performance using 16QAM than 32QAM, 64QAM, furthermore, the specified DNN architectures help improve real-time performance by saving runtime and storage resources for online UWA communications.

In this paper, we present a novel lightweight elliptic curve scalar multiplication architecture for random Weierstrass curves over prime field ${\mathbb{F}}_p$. The elliptic curve scalar multiplication is executed in Jacobian coordinates based on the Montgomery ladder algorithm with (X,Y)-only common Z coordinate arithmetic. At the finite field operation level, the adder-based modular multiplier and modular divider are optimized by the pre-calculation method to reduce the critical path while maintaining low resource consumption. At the group operation level, the point addition and point doubling methods in (X,Y)-only common Z coordinate arithmetic are modified to improve computation parallelism. A compact scheduling method is presented to improve the architecture’s performance, which includes appropriate scheduling of finite field operations and specific register connections. Compared with existing works, our design is implemented on the FPGA platform without using DSPs or BRAMs for higher portability. It utilizes 6.4~6.5k slices in Kintex-7, Virtex-7, and ZYNQ FPGA and executes an elliptic curve scalar multiplication for a field size of 256-bit in 1.73 ms, 1.70 ms, and 1.80 ms, respectively. Additionally, our design is resistant to timing attacks, simple power analysis attacks, and safe-error attacks. This architecture outperforms most state-of-the-art lightweight designs in terms of area-time products.

For K-distributed sea clutter, a constant false alarm rate (CFAR) is crucial as a desired property for automatic target detection in an unknown and non-stationary background. In multiple-target scenarios, the target masking effect reduces the detection performance of CFAR detectors evidently. A machine learning based processor, associating the artificial neural network (ANN) and a clustering algorithm of density-based spatial clustering of applications with noise (DBSCAN), namely, DBSCAN-CFAR, is proposed herein to address this issue. ANN is trained with a symmetrical structure to estimate the shape parameter of background clutter, whereas DBSCAN is devoted to excluding interference targets and sea spikes as outliers in the leading and lagging windows that are symmetrical about the cell under test (CUT). Simulation results verified that the ANN-based method provides the optimal parameter estimation results in the range of 0.1 to 30, which facilitates the control of actual false alarm probability. The effectiveness and robustness of DBSCAN-CFAR are also confirmed by the comparisons of conventional CFAR processors in different clutter conditions, comprised of varying target numbers, shape parameters, and false alarm probabilities. Although the proposed ANN-based DBSCAN-CFAR processor incurs more elapsed time, it achieves superior CFAR performance without a prior knowledge on the number and distribution of interference targets.

Multiple Signal Classification (MUSIC) is a high-performance Direction of Arrival (DOA) estimation algorithm, which has been widely used. The algorithm needs to calculate the covariance matrix, eigenvalue decomposition and spectral peak search. In the paper, the hardware structure of the existing Jacobi algorithm for Hermitian matrices is proposed. On this basis, a novel hardware acceleration of the MUSIC algorithm for sparse arrays and uniform linear arrays is proposed, and the sparse array is a nested array. There are two designs, Design 1 supports 1~10 nested array elements or 1~32 uniform linear array elements, distinguishes 1~32 sources, configures snapshots 1~2048, and the maximum number of iterations and iteration accuracy of the complex Jacobi algorithm. Design 2 only needs $101.8~\mu $ s to complete a DOA estimation when the number of array elements is 8, the number of sources is 1, and the snapshots is 128. In more detail, the Root Mean Squared Error (RMSE) of both can reach 0.03°. The logic resources on the Zynq-7000 development board are 14,761 and 28,305 Look-Up Tables (LUTs), respectively.

In reconfigurable intelligent surface (RIS)-assisted wireless communication systems, channel estimation is of paramount importance to unleash the full advantage of the high passive beamforming gains from RIS. However, it is particularly challenging due to the massive number of RIS reflecting elements which lack signal perception and processing capabilities, thus incurring prohibitively high pilot overhead between transceivers. To address this challenge, we focus on an RIS-assisted multiuser multiple-input single-output orthogonal frequency division multiplexing (MU-MISO-OFDM) system, and propose a practical transmission protocol with non-uniformly spaced comb -type pilots for compressed channel estimation and data transmission, by exploiting the sparsity of RIS-associated channels. To solve a bivariate pilot optimization problem with the objective of minimizing the mutual coherence of the measurement matrix influenced by pilot positions and pilot powers simultaneously, we propose two enhanced whale optimization algorithm (EWOA)-based approaches by leveraging the second order cone programming (SOCP) and compressed coding scheme (CCS). Simulation results validate the effectiveness of the proposed methods over benchmark schemes in various aspects. It corroborates that considering the coupling relationship between pilot positions and powers can help improve the channel estimation performance of the RIS-assisted MU-MISO-OFDM system with limited pilot overhead.

Speech Emotion Recognition (SER) plays a significant role in the field of Human–Computer Interaction (HCI) with a wide range of applications. However, there are still some issues in practical application. One of the issues is the difference between emotional expression amongst various individuals, and another is that some indistinguishable emotions may reduce the stability of the SER system. In this paper, we propose a multi-layer hybrid fuzzy support vector machine (MLHF-SVM) model, which includes three layers: feature extraction layer, pre-classification layer, and classification layer. The MLHF-SVM model solves the above-mentioned issues by fuzzy c-means (FCM) based on identification information of human and multi-layer SVM classifiers, respectively. In addition, to overcome the weakness that FCM tends to fall into local minima, an improved natural exponential inertia weight particle swarm optimization (IEPSO) algorithm is proposed and integrated with fuzzy c-means for optimization. Moreover, in the feature extraction layer, non-personalized features and personalized features are combined to improve accuracy. In order to verify the effectiveness of the proposed model, all emotions in three popular datasets are used for simulation. The results show that this model can effectively improve the success rate of classification and the maximum value of a single emotion recognition rate is 97.67% on the EmoDB dataset.