In recent years, the performance of phase-aware speech enhancement neural networks has steadily improved. However, dealing with complex-valued Short-Time Fourier Transform (STFT) spectrograms involves complex operations and phase estimation, which increases the complexity and parameter number of the model. To address this, we have built upon the foundation of DCTCRN and introduced real-valued Short-Time Discrete Cosine Transform (STDCT) spectrograms as input features, which avoids the complexities associated with phase estimation and modeling amplitude-phase relationships. To further enhance skip connections without increasing parameters, we have incorporated the SimAM attention mechanism. Additionally, we have added dual-path RNN modules between the encoder and decoder to capture long dependencies in both time and frequency dimensions. We have also introduced Hardtanh as the new scaling function. Through comparative experiments and ablation studies, we have confirmed the effectiveness of using STDCT spectrograms, attention mechanism and Hardtanh scaling function. Our approach demonstrates higher competitiveness in objective performance metrics compared to recent speech enhancement models. Notably, it achieves this while maintaining a relatively low parameter number, thus raising the performance ceiling of the DCTCRN series models.

Approximate computing (AC) has been increasingly used to meet the growing demand for energy-efficient and high-performance computing in error-tolerant applications, particularly in the fields of machine learning, signal processing, and multimedia applications. Approximate synthesis tools are an essential component in approximate circuit design and optimization as they enable designers to trade-off accuracy for reduced power consumption, area, and delay. However, these approximate synthesis tools are quite vulnerable to adversarial attacks as it is quite challenging to distinguish between a change in the functionality of the given circuit due to an approximation or a Hardware Trojan (HT). In this paper, we propose an approach for HT detection using Triple modular Redundancy in APproximate Synthesis (HT-TRAPS) tools. The approach also utilizes functional and side channel power and timing analysis to distinguish approximations from HTs and is applicable for either identifying a clean netlist or detecting the presence of a HT at the gate-level post-approximate synthesis stage. We implement the HT-TRAPS approach using a full stack of open-source and commercial tools, including BLASYS, ABACUS, SCOAP and Xilinx Vivado, and thoroughly evaluate its effectiveness by detecting various Trojan benchmarks, such as B19-T100, B19-T200, and B19-500 Trojans, inserted in several circuits, such as the approximated Absolute Difference calculator, Butterfly filter, a classifier, x2, a 16-bit Multiplier and a 32-bit Adder.

Large-scale interactive systems require efficient performance and scalability to accommodate user demands. The response time of individual requests can be a limiting factor for overall performance. One effective solution is to employ a consistent hashing algorithm for request distribution, which alleviates the load on each server and enhances response speed. However, existing consistent hash algorithms often fail to account for the heterogeneity of resources, which may degrade performance and result in underutilized resources. To solve this problem, this paper proposes a solution based on the study of cluster load balancing using a consistent hash algorithm to increase the resource utilization. Firstly, The scheme adopts a two-dimensional spatial coordinate positioning. Partitions are formed by dividing the ring created by the value range of the CRC32 or CRC64 hash function. The partition and relative position for a key are determined by its vertical and horizontal coordinate, respectively. Secondly, nodes with different performance are assigned with different weights, and each partition also has a corresponding weight which determined by the number and weight of nodes within it. Experiments have shown that this scheme can achieve a search rate of more than 30 million keys per second, with a resource utilization rate of over 80% within the cluster, and maintain stability when changing the number of nodes, thereby improving the overall performance of the system.

In online social media, various fake news highly impact experience quality (QoE) of consumers. Nowadays, researchers in this area concentrated on either semantic analysis or contextual awareness. However, social media is a spatiotemporal place, and much important information such as social activities, social connection and social attributes, can make sense. Hence, this work proposes an intelligent fake news detection method by leveraging semantic spatiotemporal fusion. It uses the RoBERTa model to process posts made by users to understand their semantic content. Then, the processed data is fed into a Long Short-Term Memory network (LSTM) to capture the temporal features of the posts. Additionally, the study constructs a spatial model using Graph Attention Networks (GATs) to simulate the spread of fake news. Performance of the proposed method is testified on a realworld dataset specifically collected by our research team. And the results reveal that this innovative approach not only improves the accuracy of fake news detection but also optimizes consumer service quality.

A system of zero-knowledge proofs on graph signatures has been proposed, where a graph can be signed, and the owner of the graph signature can prove a graph relation such as the connectivity and isolation of any two vertexes on the graph without disclosing all information about the graph. The correctness of the graph information is guaranteed by the signature. One of the applications is a virtualized infrastructure, where an infrastructure provider manages a distributed system, and each tenant is allocated a specific portion of this infrastructure for use. Tenants need to check with the provider that their resources are properly connected (connectivity) and that their resources are properly separated from the resources of other tenants (isolation). On the other hand, the provider cannot simply disclose the entire infrastructure topology to each tenant. Using the zero-knowledge proof system on graph signatures, both requirements can be addressed. Previously, an efficient zero-knowledge proof system on graph signatures using a bilinear-map accumulator has been proposed, where the verification time and the size of the proof data do not depend on the number of graph vertexes and edges. However, this system has two problems. First, since the proof does not include labels, it is not possible to prove the connectivity considering network bandwidth and cost. Second, since it assumes undirected graphs, it cannot handle applications on directed graphs such as network flows. In this paper, we extend the previous system and propose a zero-knowledge proof system of the connectivity for directed graphs where each edge has labels. We implemented our system on a PC using a pairing library and evaluate it by measuring the processing times. Compared to the conference version of this paper, we show the formal definitions and the security proofs of our proposed system, and add implementation-based evaluations reflecting the application to the virtualized infrastructure.

Partial transmit sequence (PTS) is an effective technique to mitigate the peak-to-average power ratio (PAPR) problem in orthogonal frequency-division multiplexing (OFDM) systems. Some past research proposed a Q-section coded PTS scheme with the progressive edge-growth (PEG) search algorithm for construction of the employed quasi-cyclic low-density parity-check (QC-LDPC) codes. In this paper, a new approach of the Q-section PTS scheme for PAPR reduction and error correction is devised based on algebraically constructed QC-LDPC codes with masking. The corresponding Q-section constraint is then turned into one only on the masking matrix which is easy to satisfy, based on which QC-LDPC codes can be flexibly constructed and designed without extra search complexity. Our method allows for a single algebraically constructed base matrix to generate a family of Q-section QC-LDPC codes with different lengths and rates. The performance of the constructed codes can be guaranteed by employing analysis such as density evolution. The corresponding efficient ways of encoding and decoding applicable to general QC-LDPC codes are also proposed. Our scheme need not transmit extra side information and has no extra cost compared with the traditional PTS scheme. The PAPR control patterns are embedded in specific bit positions of the codeword so that the information bits can be recovered without attempting all possible control patterns during decoding. Finally, simulation results show that our design can provide better error performance than previous codes and have similar PAPR reduction performance.

Automatic Micro-Expression (ME) spotting is a fundamental yet essential task in Micro-Expression analysis from videos. Due to the brief and subtle of ME, the ME spotting task is challenging and the spotting performance need to be further improved. However, current works generally neglect the correlation between expression proposals in a video. In this work, we propose a two-stage relation-aware graph convolutional network (MES-RANet) to locate the temporal positions of macro- and micro-expression. First, we adopt a temporal evaluation module (TEM) to predict the frame-level probabilities from the spatial-temporal feature sequences and generate candidate proposals for the subsequent module. Specifically, in relation-aware module (RAM), we formulate video proposals as graph nodes, and proposal-proposal correlations as edges to construct graphs. Then apply the relation-aware network to model the relations among proposals and learn powerful representations for the boundary regression. Comprehensive experiment results show that MES-RANet is effective and achieves competitive performance compared with state-of-the-art methods on two public benchmark datasets CAS(ME)² and SAMM-LV. Codes are available at https://github.com/hahaluluyo/MES-RANet.

While COVID-19 poses significant global challenges, this study proposes leveraging deep learning models to predict adverse events post-vaccination. Using the Vaccine Adverse Event Reporting System (VAERS) dataset, which includes potential side effects of Pfizer, Janssen, and Moderna vaccines, the study partitions the data to predict vaccination adverse events (VAE) such as “died,” “hospitalized,” and “recovered.” The proposed DeepCNNBDLSTM model combines deep neural network layers with convolutional and bidirectional long short-term memory layers. Baseline models include deep neural networks, bidirectional long short-term memory, convolutional neural networks, and long short-term memory. Performance metrics such as confusion matrix, AUCROC, accuracy, F1-score, precision, and recall are evaluated. Experimental results show the proposed model outperforms by achieving an accuracy of 89.05% for predicting ‘recovered’ VAE (trained on Pfizer dataset), 99.03% accuracy for predicting ‘hospitalized’ VAE, and 99.86% accuracy for predicting ‘died’ VAE (both trained on Moderna dataset). These insights may aid doctors in selecting the most effective COVID-19 vaccine for patient protection.

Quantization is an effective way to reduce memory and computational costs in the inference of convolutional neural networks. However, it remains unclear which model can achieve higher recognition accuracy while minimizing memory and computational costs: a large model (with a large number of parameters) quantized to an extremely low bit width (1 or 2 bits) or a small model (with a small number of parameters) quantized to a moderately low bit width (3, 4, or 5 bits). In this paper, we define a metric that combines the numbers of parameters and computations with the bit widths of quantized weight parameters. By utilizing this metric, we demonstrate that Pareto-optimal performance, where the best accuracy is attained at a given memory or computational cost, is achieved when a small model is moderately quantized, not when a large model is extremely quantized. Based on this finding, we empirically show that the Pareto frontier is improved by 4.3 × in a post-training quantization scenario for a quantized ResNet-50 model using the ImageNet dataset.

Accurate prediction of transportation carbon emissions is crucial for the government to formulate transportation strategies. However, due to the characteristics of the traffic carbon emission data, such as nonlinearity, uncertainty, and abrupt changes, accurately predicting traffic carbon emissions faces significant challenges. In this paper, we propose the LNSSA-BiLSTM-MHA model, which combines the improved Sparrow Search Algorithm based on the logistic-tent-cosine chaotic mapping and the Narrowed Opposition-Based Learning Strategy (LNSSA) with the BiLSTM-MHA model to improve the accuracy of transportation carbon emission prediction. Firstly, the LNSSA algorithm is proposed for model hyperparameter search, which combines Logistic-Tent-Cosine chaotic mapping with Narrowed Opposition-Based Learning strategy to improve the SSA algorithm. Among them, the logistic-tent-cosine chaotic mapping is used to uniformly initialize the population, and the Narrowed Opposition-Based Learning strategy is used to reduce the solution space. Secondly, the BiLSTM model is used to analyze the temporal data features in both forward and reverse directions. Finally, we combine the multihead attention mechanism to learn data features from different subspaces. The private data set of six intersections of roads in Xiangyang City from 2023 to 2024 is collected. We perform model training and experimental analysis on this private dataset. Using the evaluation metrics RMSE, MSE, and MAE, the results showed that the LNSSA-BiLSTM-MHA model outperformed the comparison models LSTM, GRU, LSTM-Attention, GRU-Attention, CNN-LSTM-Attention, CNN-GRU-Attention and ResNet-GRU-Attention by 32%, 54% and 35%, respectively, in six datasets.

In this study, blood vessel shape was accurately measured to evaluate the viscoelastic properties of the radial artery. For a robust measurement, the shape parameters of the elliptical blood vessel were determined by integrating the brightness over the entire vessel instead of using the brightness gradient at the vessel boundary. The usefulness of the proposed method is demonstrated through simulations and in vivo experiments. The results obtained exhibit considerable potential for estimating the viscoelastic properties of elliptically deformed blood vessels.

Symmetric 2-adic complexity stands out as a better measure for appraising the strength of a periodic sequence against the rational approximation attack. In this letter, we determine the symmetric 2-adic complexity of a class of interleaving cascade binary sequences with period 4N (where N ≡ 3 (mod 4)), which was introduced recently. Our results indicate that the symmetric 2-adic complexity of these sequences is not less than 2N.

In this paper, we present a new classification method for the defining sets of η-constacyclic codes of length n = $\frac{q^4+1}{2}$. This classification can simplify the calculation of the dimensions of the Hermitian hulls for these codes, while deriving the numbers of sharing pre-shared maximally entangled states of the corresponding entanglement-assisted quantum error-correcting codes under various design distances.

In the constant dimension codes (CDCs) case, to explore the largest possible value 𝒜_q(n, d, k) is one of the most fundamental topics in subspace coding. Echelon-Ferrers is an approach proposed in 2009 for CDCs’ construction, and has received extensive attentions. In this letter, in order to further improve the constructions, we study the method which combines linkage construction and echelon-Ferrers construction. This allows us to produce some improvements to lower bounds for constant dimension codes, including 𝒜_q(13,4,4), 𝒜_q(17, 4, 4), 𝒜_q(19, 6, 6) etc.

This letter studies the robust transceiver designs for multiple-input multiple-output (MIMO) interference channels with imperfect channel state information (CSI). The sum power minimization problem is considered subject to signal-to-interference-plus-noise ratio (SINR) constraint under CSI mismatch with statistical error model. Since the problem is a joint design of transmitter and receiver beamformers, an alternating approach is developed by exploiting the primal decomposition. Specifically, closed-form expressions are derived as a solution for the robust transceiver design. Numerical results are provided to demonstrate the robustness of the proposed algorithm and the performance efficiency in combating channel errors.

Automatic modulation recognition in low-computation edge devices under noisy environments has become increasingly critical. To address the dual challenges of lightweight design and noise resistance, this paper proposes WCTFormer, a novel lightweight network that integrates discrete wavelet transform for frequency-domain noise suppression and a Transformer for global feature extraction, enabling robust performance in low SNR conditions. Experiments on open-source datasets demonstrate that WCTFormer achieves superior recognition accuracy, with 92.40% accuracy at 0 dB SNR, requiring only 60K parameters. WCTFormer combines high recognition performance and computational efficiency, making it suitable for deployment in resource-constrained edge devices.

In this letter, we extend the analysis of outage probability of decode-and-forward dual-hop system over Nakagami-m fading to include non-integer fading parameters, which addresses a limitation of previous studies that were restricted to integer values. By deriving a generalized expression, we can provide a unified framework applicable to more general and practical scenarios. Extensive numerical results are presented to validate that our proposed closed-form formula can embrace more general Nakagami-m fading environments, obviously showing its compactness and accuracy.

Flying Ad-hoc Networks (FANETs) frequently suffer from widespread disconnections due to unpredictable external disruptions, consequently leading to network partitioning into segments in which UAVs become isolated from the base station. To address this issue, this letter proposes the Infection-based Leader Election (ILE) algorithm to autonomously restore network connectivity after disruptions. The ILE identifies and elects a leader within each disconnected segment, which subsequently guides the restoration process. By limiting information exchange to neighboring nodes, the ILE achieves significant reductions in both computational complexity and communication overhead. After the leadership election process is completed, the elected leader coordinates nodes within the partition to execute cascading movements, enabling rapid restoration of network connectivity. The simulation results validate the effectiveness of our algorithm, demonstrating notable improvements in communication overhead and restoration time.

Precise and automatic segmentation of pulmonary lesions is crucial for assisting pulmonologists in accurate diagnosis and decision-making. Despite advances in deep learning, segmenting pulmonary nodules remains challenging due to factors like small lesions, irregular boundaries, and data imbalance. We propose an edge detail enhancement method (EDC-UNet) for pulmonary nodule segmentation, which integrates deformable convolutional layers to improve flexibility for various lesion morphologies and dilated convolution-based residual blocks to enhance feature extraction. Additionally, a Sobel-based detail supervision module in the decoder helps capture low-level spatial details, improving segmentation of blurred edges. Extensive experiments on the LIDC-IDRI and LUNA16 datasets demonstrate that EDC-UNet outperforms other models, highlighting its potential for medical image analysis.