Cardiovascular disease (CVD) remains the leading cause of mortality worldwide, with coronary artery disease (CAD) being the most prevalent form. To improve screening efficiency, there is a critical need for accurate, non-invasive, and cost-effective CAD detection methods. This study presents Co-Attention Dual-Modal ViT (CAD-ViT), a novel classification framework based on the Vision Transformer that integrates both electrocardiogram (ECG) and phonocardiogram (PCG) signals. Unlike prior approaches that process ECG and PCG features independently or fuse them through simple concatenation, the proposed model introduces two key modules: a Co-Attention mechanism that enables bidirectional cross-modal interaction to effectively capture complementary features between ECG and PCG signals, and a Dynamic Weighted Fusion (DWF) module that adaptively adjusts the contribution of each modality for robust feature fusion. CAD-ViT is evaluated on a private clinical dataset comprising 132 CAD and 101 non-CAD subjects, achieving an accuracy of 97.08%, precision of 97.18%, specificity of 98.52%, F1-score of 97.04, and recall of 96.94%. Additional validation on two public datasets confirms the model's robustness and generalization capability. These results demonstrate the effectiveness of the proposed approach and its potential for practical deployment in CAD screening using multimodal biosignals.

Soccer is a globally popular sporting event, typically characterized by long matches and distinctive highlight moments. Recent advances in Multimodal Large Language Models (MLLMs) show promising capabilities in temporal grounding and video understanding. However, generating soccer commentary requires both precise temporal localization and semantically rich descriptions over long-form videos. Existing soccer MLLMs often rely on temporal priors for caption generation, which limits their ability to process the entire video in an end-to-end manner. Traditional approaches, on the other hand, follow a complex two-step paradigm that fails to capture the global context, leading to suboptimal performance. To solve the above issues, we present TimeSoccer, the first end-to-end soccer MLLM for Single-anchor Dense Video Captioning (SDVC) in full-match soccer videos. TimeSoccer jointly predicts timestamps and generates captions in a single pass, enabling global context modeling across 45-minute matches. To support long video understanding of soccer matches, we introduce MoFA-Select, a training-free, motion-aware frame compression module that adaptively selects representative frames via a coarse-to-fine strategy, and incorporates complementary training paradigms to strengthen the model's ability to handle long temporal sequences. Extensive experiments demonstrate that our TimeSoccer achieves State-of-The-Art (SoTA) performance on the SDVC task in an end-to-end form, generating high-quality commentary with accurate temporal alignment and strong semantic relevance. For more information, please visit: https://vpx-ecnu.github.io/TimeSoccer-Website/.

Recently, neural radiance fields have made significant advancements in rendering new views. However, limited research has focused on dynamically loading implicit radiance fields with efficient memory utilization and extended scene representation. This paper introduces HeRF, a novel framework with a hierarchical scene representation based on layered sparse voxels. With such an adaptive design, our method is able to partition scenes into different levels for faster modeling and reduced memory cost. Furthermore, these partitioned scenes can be dynamically loaded and joined for a better immersive experience. Quantitative and qualitative analysis using objectlevel, indoor, and outdoor datasets demonstrates the effectiveness of HeRF. Remarkably, our proposed method requires only about 38% of the training rays and 45% of the GPU memory cost, yet achieves a 9% improvement in PSNR compared to NeRFusion on the ScanNet dataset. The code is available at https://github.com/Minisal/HeRF.

Automatic modulation recognition (AMR) is a widely used technique in various communication systems. In this work, we propose a complex-valued transformer (CV-TRN) network for AMR. Considering the in-phase (I) and quadrature (Q) components of the signal are two consistent data with only a phase difference of π/2, they can teach the network independently which in disguise augment the training data, but the I/Q components are collectively needed to measure similarity in the multi-head self-attention (MHSA). We input the I/Q data individually into the network with shared parameters, and they are transmitted independently in the network except in the MHSA, where a complex-valued MHSA (CMHSA) is proposed to let the information from I/Q components integrate. Moreover, CV-TRN adopts the relative position embedding, with a mathematical analysis of its advantages for AMR. A data augmentation method of random phase offset is introduced to further improve the robustness. Experimental results on RML2016.10a and RML2018.01a datasets demonstrate that the proposed CV-TRN outperforms state-of-the-art AMR methods and is parameter-efficient.

Detection of high frequency (HF) signal in the wideband is challenging since the HF environment is chaotic. Recent works adopt deep learning-based object detectors to capture signals in wideband spectrogram, but the task of signal detection exhibits different characteristics from that of generic object detection, which causes the classical deep learning-based detectors to have defects such as limited receptive field and prior anchor mismatch. Based on the task analysis, this letter proposes a deep learning framework which extracts features along the time axis at each frequency bin, and predicts multiple characteristics of the signal, including the center frequency and the shape attributes. The unique advantages of the framework are the utilization of all time features and no prior anchor which adapt to the slender shape of signal with a strong generalization ability. The numerical studies on simulation signal + real-world background prove the superiority of the proposed framework both in accuracy and speed.

Protocol format reverse engineering aims to extract the protocol fields automatically without access to the protocol specification. Existing works focus on the methodology of deriving the protocol format efficiently, but neglect the relationship between the statistical characteristics of protocol data and the intrinsic properties of the protocol format. In this paper, we study two problems to see how the protocol specification affects the statistical properties, and how the latter affect the difficulty of format reverse analysis. Through empirical analysis of known protocols, we first verify the stationarity of protocol features, which is the stand for developing trace-based reverse methods. We study the position arrangement and value distribution of protocol fields, and investigate their influence on the statistical properties of the protocol format. Then we propose an HMP-based model of protocol data. Using this model, we define two quantitative indicators by protocol fields' structure and content to reflect the reversibility of protocol format: the field non-interlacing ratio and the field information variation. We apply the analysis of format reversibility to a number of typical realistic protocols. The results suggest that the fields of most protocols can be partially revealed, but there are also certain fields difficult for reverse analysis. The quantitative results can provide hints for improving protocol reverse engineering approaches.