VTC 2025 Spring Conference’s Shorts

CGAN-based CSI Fusion for Frequency Division Duplex (FDD) Massive MIMO Systems

Massive Multiple Input Multiple Output (MIMO) is a cornerstone technology for achieving high capacity and spectral efficiency. A key challenge in frequency division duplex (FDD) massive MIMO systems lies in obtaining accurate downlink (DL) channel state information (CSI), as the absence of uplink (UL)-DL reciprocity hinders conventional estimation methods, creating a bottleneck in system performance. To address this issue, various approaches have been developed, broadly categorized into Conversion and Feedback approaches. However, partial reciprocity and quantization loss in feedback codebook design make accurate DL CSI acquisition a persistent challenge. In this paper, we propose a novel CGAN (Conditional Generative Adversarial Network)-based CSI-fusion framework that integrates both the UL channel statistics obtained from sounding reference signals (SRS) and the DL feedback from user equipments (UEs). These two sources of DL CSI-related information are fused and utilized as conditional inputs to CGAN to map the UL channel statistics to DL CSI. The proposed CGAN-based CSI-Fusion framework significantly enhances DL CSI acquisition accuracy, offering a practical solution to overcoming DL CSI acquisition challenges.CGAN-based CSI Fusion for Frequency Division Duplex (FDD) Massive MIMO SystemsTong Yi, Shengsong Luo, Bingnan Xiao, Chongbin Xu, Fudan University; Xin Wang, Fudan University, China

A Recursive Discretization Compression Framework Combined with Selective State Space Model...

The quality of channel state information (CSI) feedback is critical for maximizing the spectral efficiency of massive multiple-input multiple-output systems. With multiple antenna arrays, the overhead of direct CSI feedback in frequency division duplex mode is usually large, and many CSI compression techniques have been proposed to alleviate this problem. Deep learning (DL) has achieved tremendous strides in CSI feedback. However, most current DL-based CSI compression methods utilize fully connected layers to achieve dimensionality reduction, which may be suboptimal for network optimization and result in noteworthy information loss and reduced CSI reconstruction accuracy. In this paper, we propose a novel recursive discretization compression framework with a selective state space model for CSI feedback, namely CsiMamba-RDC. The framework employs improved residual vector quantization to recursively refine CSI representation, reducing information loss and storage overhead. Additionally, we present an encoder-decoder model leveraging a selective state space model to extract diverse channel features.A Recursive Discretization Compression Framework Combined with Selective State Space Model for Massive MIMO CSI FeedbackXinran Sun, Southeast university; Zhengming Zhang, Southeast University; wenzhe fu, southeast university; Chunguo Li, Southeast University, Nanjing, China; Yongming Huang, Luxi Yang, Southeast University

A GNN-Based Approach to AP Cooperation Cluster Formation in Cell-Free Massive MIMO

Forming effective access point (AP) cooperation clusters is a key challenge in user-centric cell-free massive MIMO (CF-mMIMO). Existing approaches to this task are either computationally prohibitive or overlook the complex interrelationships within communication networks. In this context, we introduce an innovative approach based on graph neural networks (GNNs). By leveraging the inherent graph structure of CF-mMIMO networks, we transform the rate maximization problem into a node classification task, enabling a competitive and robust solution. Simulation results show that the proposed method significantly outperforms conventional baselines in terms of spectral efficiency, computational complexity, and scalability.A GNN-Based Approach to AP Cooperation Cluster Formation in Cell-Free Massive MIMODariel Pereira-Ruisanchez, University of A Coruña; Michael Joham, Technical University of Munich; Óscar Fresnedo, Darian Pérez-Adán, Luis Castedo, University of A Coruña; Wolfgang Utschick, Technical University of Munich

Joint Transmission and Deblurring: A Semantic Communication Approach Using Events

Deep learning-based joint source-channel coding (JSCC) is emerging as a promising technology for effective image transmission. However, most existing approaches focus on transmitting clear images, overlooking real-world challenges such as motion blur caused by camera shaking or fast-moving objects. Motion blur often degrades image quality, making transmission and reconstruction more challenging. Event cameras, which asynchronously record pixel intensity changes with extremely low latency, have shown great potential for motion deblurring tasks. However, the efficient transmission of the abundant data generated by event cameras remains a significant challenge. In this work, we propose a novel JSCC framework for the joint transmission of blurry images and events, aimed at achieving high-quality reconstructions under limited channel bandwidth. This approach is designed as a deblurring task-oriented JSCC system. Since RGB cameras and event cameras capture the same scene through different modalities, their outputs contain both shared and domain-specific information. To avoid repeatedly transmitting the shared information, we extract and transmit their shared information and domain-specific information, respectively. At the receiver, the received signals are processed by a deblurring decoder to generate clear images. Additionally, we introduce a multi-stage training strategy to train the proposed model. Simulation results demonstrate that our method significantly outperforms existing JSCC-based image transmission schemes, addressing motion blur effectively.Joint Transmission and Deblurring: A Semantic Communication Approach Using EventsPujing Yang, Guangyi Zhang, Yunlong Cai, Zhejiang University; Lei Yu, Wuhan University; Guanding Yu, Zhejiang University

Intelligent Scheduling with Volatile Wireless Path State Prediction for 5G+ Multi-access N...

Modern 5G+ network terminals are equipped with multiple access interfaces (e.g. cellular and WiFi), enabling the implementation of multipath protocols, providing means for better network resource utilisation and better Quality of Service (QoS). This paper addresses a critical challenge in multipath network environments: the degradation of application performance and increase in out-of-order packet delivery due to the volatility of wireless paths. The paper presents a new scheduling algorithm, Best Path First (BPF), which leverages a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) model. By analysing historical path data, the model predicts the path state, enabling selection of the optimal path in a multipath framework. BPF uses the path time-throughput characteristics (rolling standard variation, path bottleneck queue, path throughput) in real-time to prioritise the path with the highest expected quality, predicting with a 2% Mean Absolute Error (MAE). A Mininet testbed with real mobility traces is used to demonstrate the solution and evaluate its performance. BPF can reduce UDP end-to-end delays and jitter by 20% and 21% respectively when compared to the standard Cheapest-path-first (CPF) multipath scheduler. Tests on TCP traffic demonstrate that BPF improves by 7% and 6% TCP end-to-end delay and jitter and reduces FTP download times by 7% compared to CPF and up to 30% compared to Peekaboo.Intelligent Scheduling with Volatile Wireless Path State Prediction for 5G+ Multi-access NetworksGregorio Maglione, City St George’s University of London; Veselin Rakocevic, City, University of London; Markus Amend, Deutsche Telekom

Idempotent Semantic Communication Against Distortion Accumulation

Despite the remarkable success of semantic image transmission, existing approaches face the challenge of distortion accumulation. Specifically, as a received image is further forwarded to other devices, reconstruction distortion will accumulate, leading to decreased system stability. In this paper, we propose an idempotent semantic communication system for image transmission to enhance stability. We systematically analyze the factors contributing to this accumulation effect and propose several strategies to mitigate it. First, we design the system using a right-invertible neural network to achieve idempotence, ensuring the decoder functions as the right inverse of the encoder. Second, we introduce a feature discretization mechanism to further reduce distortion accumulation, leveraging the benefits of digitalization over analog transmission. Finally, we employ a recursive training strategy, which incorporates the reconstructed images into the training process to significantly improve overall stability. Empirical results demonstrate that our proposed strategies effectively enhance system stability, minimizing quality degradation and enhancing output consistency across multiple transmissions.Idempotent Semantic Communication Against Distortion AccumulationGuangyi Zhang, Pujing Yang, Yunlong Cai, Qiyu Hu, Guanding Yu, Zhejiang University

Beam Interference Identification and SINR Prediction in Millimeter Wave Systems

Millimeter wave (mmWave) has emerged as a key technology for next-generation communications to address spectrum scarcity in traditional low-frequency wireless systems. However, its dynamic channel, high propagation losses and blocking susceptibility impose higher demands on resource allocation in complex interference environments, especially in factory scenarios. Efficient resource allocation in mmWave systems relies on accurately identifying and modeling interference, but current prevalent research remains constrained by idealized channel state information (CSI) assumptions and statistical modeling approaches that focus on stable CSI and consistent interference and noise. These conventional methods are inadequate for transient systems with instantaneous channel variations. To overcome these challenges, this paper proposes an interference identification model based on directional beams, which integrates the cluster scattering characteristics of mmWave signals and formulates the received power as the sum of the directional beam gain-weighted multipath powers. Using intelligent algorithms, we train this model to precisely identify interference and match arbitrary input resource relationships to achieve accurate signal-to-interference-plus-noise ratio (SINR) prediction for supporting the subsequent resource allocation. Numerical results demonstrate that the proposed algorithm improves the prediction accuracy by 4.7% – 35.1% compared to the benchmark scheme.Beam Interference Identification and SINR Prediction in Millimeter Wave SystemsYantong Zhou, Chunjing Hu, Tao Peng, Yichen Guo, Yujie Zhao, Yijing Niu, Wenbo Wang, Beijing University of Posts and Telecommunications

Accelerating Multi-Modal Radio Frequency Fingerprinting by Efficient Early Fusion

Deep learning has been applied to radio identification techniques for identifying individual radio frequency (RF) transmitters. While multi-modal neural networks can achieve high identification accuracy, the inference speed at edge devices needs to be accelerated because of the computational cost. In this paper, we propose a method for combining modalities to use different resolutions in early fusion. The method makes it possible to strategically reduce the input size of only the modalities that do not affect the inference accuracy. The evaluation results show that the proposed method increases the inference speed to 13.0 times on the Hailo–8 edge AI device without sacrificing accuracy.Accelerating Multi-Modal Radio Frequency Fingerprinting by Efficient Early FusionKazutoshi Hirose, Seiya Shibata, Taichi Ohtsuji, Takashi Takenaka, NEC Corporation

Maturing Federated Transfer Learning for Adaptive Beam Selection in mmWave MIMO Systems

Millimeter-wave (mmWave) beam selection in MIMO systems presents significant challenges in dynamic environments due to computational constraints, data heterogeneity, and privacy concerns. In this paper, we propose a novel Maturing Federated Transfer Learning (MFTL) framework that integrates radar and image data to enhance beam prediction accuracy while ensuring user data privacy. The proposed approach utilizes ResNet-50 as a pre-trained model, fine-tuned locally at distributed Antenna Units (AUs) to adapt to diverse scenarios. To mitigate the effects of data heterogeneity, we evaluate multiple aggregation strategies, with FedMedian demonstrating superior robustness compared to FedAvg and weighted averaging. Our optimized MFTL configuration, utilizing a learning rate of 0.005, a batch size of 32, and five aggregation rounds, significantly improves model performance. Experimental results on the DeepSense 6G dataset indicate that our approach achieves a Top-1 accuracy of 57.58% and a Top-5 accuracy of 94.7% , outperforming state-of-the-art methods. These findings highlight the effectiveness of federated learning, transfer learning, and robust aggregation techniques in improving beam selection accuracy for next-generation mmWave communication systems.Maturing Federated Transfer Learning for Adaptive Beam Selection in mmWave MIMO SystemsShaimaa Hassanein, Qatar University; Elias, Yaacoub; Tamer Khattab, Aiman Erbad, Qatar University

FedEXD: Self-Propelled Federated Learning with Extraction-Based Knowledge Distillation in ...

Federated learning (FL) is a pivotal paradigm for decentralized model training while preserving data privacy. However, data heterogeneity among clients significantly degrades model performance and convergence efficiency. In response, we introduce a federated knowledge distillation mechanism, FedEXD, that addresses robustness and convergence in diverse client environments through a self-propelled learning architecture. FedEXD employs a novel density ratio-based data extraction algorithm, leveraging KLIEP to select representative data, enhancing global knowledge synthesis and local model adaptability while preserving privacy. Extensive evaluations on benchmark datasets demonstrate FedEXD’s substantial improvements in efficiency and accuracy, demonstrating a substantial 1.51% accuracy improvement over state-of-the-art methods under firm heterogeneity while reducing communication rounds by over 46.3%. These findings underscore FedEXD’s potential to advance FL systems’ generalizability across complex, non-IID data distributions, offering a scalable solution for privacy-conscious, high-performance distributed learning.FedEXD: Self-Propelled Federated Learning with Extraction-Based Knowledge Distillation in Heterogeneous EnvironmentsYunfan Li, Xidian university; Jie Feng, Lei Liu, Xidian University; Bodong Shang, Eastern Institute for Advanced Study Eastern Institute of Technology; Jing Lei, Xidian university; Qingqi Pei, Xidian University