https://scholar.dgist.ac.kr/handle/20.500.11750/16001 2026-07-30T11:03:17Z https://scholar.dgist.ac.kr/handle/20.500.11750/60545 Title: Feature Importance-Aware Deep Joint Source-Channel Coding for Computationally Efficient and Adjustable Image Transmission Author(s): Choi, Hansung; Seo, Daewon Abstract: Recent advances in deep learning-based joint source-channel coding (deepJSCC) have substantially improved communication performance, but their growing computational cost hinders practical deployment. Furthermore, certain applications require the ability to dynamically adapt computational complexity. To address these issues, we propose a feature importance-aware deepJSCC (FAJSCC) model for image transmission that is both computationally efficient and adjustable. FAJSCC employs axis-dimension specialized computation, which performs efficient operations individually for each spatial and channel axis, significantly reducing computational cost while representing features effectively. It further incorporates selective deformable self-attention, which applies self-attention only to selected and adaptively adjusted features, leveraging the importance and relations of input features to efficiently capture complex feature correlations. Another key feature of FAJSCC is that the number of selected important areas can be controlled separately by the encoder and the decoder, depending on the available computational budget. It makes FAJSCC the first deepJSCC architecture to allow independent adjustment of encoder and decoder complexity within a single trained model. The experimental results show that FAJSCC achieves superior image transmission performance under various channel conditions while requiring less computational complexity than recent state-of-the-art (SOTA) models. Furthermore, experiments independently varying the encoder and decoder's computational resources reveal, for the first time in the deepJSCC literature, that understanding the meaning of noisy features in the decoder demands the greatest computational cost. The code is publicly available at github.com/hansung-choi/FAJSCCv2 2026-05-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60200 Title: Activity recognition and user identification using mmWave radar with a shared-backbone graph network and task-specific heads Author(s): Eom, Jun Yong; Seo, Daewon Abstract: Identity-aware activity recognition is a key enabler for customized services. However, joint modeling of activity recognition and user identification from wireless signals remains underexplored. This work presents a dual-task graph model for millimeter-wave (mmWave) frequency-modulated continuous-wave (FMCW) radar point-cloud sequences. We construct directed graphs that capture a user’s spatial structure and motion over time. A shared graph neural backbone processes these graphs and produces node embeddings that encode local spatial features and short-term dynamics. Each task-specific head first aggregates node embeddings into a graph-level representation and then performs activity or identity classification. Experiments on two public datasets demonstrate that the proposed scheme achieves classification performance comparable to single-task baselines for both activity recognition and user identification while maintaining low-latency inference. Codes are available at https://github.com/junyongeom/mmActId/ . © 2026 The Authors. 2026-03-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60015 Title: On the Fundamental Tradeoff of Joint Communication and QCD: The Monostatic Case Author(s): Lim, Sung Hoon; Seo, Daewon Abstract: This paper investigates the fundamental tradeoff between communication and quickest change detection (QCD) in integrated sensing and communication (ISAC) systems under a monostatic setup. We introduce a novel Joint Communication and quickest Change subblock coding Strategy (JCCS) that leverages feedback to adapt coding dynamically based on real-time state estimation. The achievable rate-delay region is characterized using state-dependent mutual information and KL divergence, providing a comprehensive framework for analyzing the interplay between communication performance and detection delay. Moreover, we provide a partial converse demonstrating the asymptotic optimality of the proposed detection algorithm within the JCCS framework. To illustrate the practical implications, we analyze binary and MIMO Gaussian channels, revealing insights into achieving optimal tradeoffs in ISAC system design. © 2025 Elsevier B.V., All rights reserved. 2025-12-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60014 Title: Region-of-Interest-Guided Deep Joint Source-Channel Coding for Image Transmission Author(s): Choi, Hansung; Seo, Daewon Abstract: Deep joint source-channel coding (deepJSCC) methods have shown promising improvements in communication performance over wireless networks. However, existing approaches primarily focus on enhancing overall image reconstruction quality, which may not fully align with user experiences, often driven by the quality of regions of interest (ROI). Motivated by this, we propose ROI-guided joint source-channel coding (ROI-JSCC), a novel deepJSCC framework that prioritizes high-quality transmission of ROI. The ROI-JSCC consists of four key components: (1) Image ROI embedding, (2) ROI-guided split processing, (3) ROI-based loss function design, and (4) ROI-adaptive bandwidth allocation. Together, these components enable ROI-JSCC to selectively enhance the ROI reconstruction quality at varying ROI positions while maintaining overall image quality with minimal computational overhead. Experimental results under diverse communication environments demonstrate that ROI-JSCC significantly improves ROI reconstruction quality while maintaining competitive average image quality compared to recent state-of-the-art methods. 2025-11-30T15:00:00Z