面向任务导向语义通信的信道自适应条件扩散生成方法

付玮琦; 赵霄宇; 金磊; 王莉; 费爱国

doi:10.11999/JEIT260410

面向任务导向语义通信的信道自适应条件扩散生成方法

doi: 10.11999/JEIT260410 cstr: 32379.14.JEIT260410

付玮琦^1,,
赵霄宇^2,,
金磊^3,,
王莉^1, ,,
费爱国^4,

1.
北京邮电大学计算机学院(国家示范性软件学院) 北京 100876
2.
东南大学网络空间安全学院南京 211189
3.
北京邮电大学电子工程学院北京 100876
4.
多域数据协同处理与控制全国重点实验室

基金项目: 国家自然科学基金(62525106)，北京邮电大学-中国联通联合创新中心联合研发项目(2025-STHZ-BJYDDX-004)

详细信息

作者简介:
付玮琦：男，博士生，研究方向为语义通信，邮箱：　fuweiqi@bupt.edu.cn

赵霄宇：男，副研究员，研究方向为低延时无线通信

金磊：男，副教授，研究方向为人工智能

王莉：女，教授，研究方向为应急指挥通信、能源互联网、边缘计算等，邮箱：　liwang@bupt.edu.cn

费爱国：男，研究员，研究方向为指挥信息系统和数据链技术等

通讯作者:
王莉　liwang@bupt.edu.cn

中图分类号: TN929.5; TP391
计量
- 文章访问数: 27
- HTML全文浏览量: 3
- PDF下载量: 5
- 被引次数: 0
出版历程
- 修回日期: 2026-10-28
- 录用日期: 2026-10-28
- 网络出版日期: 2026-10-30

Channel-Adaptive Conditional Diffusion Generation for Task-Oriented Semantic Communication

FU Weiqi^1
,,
ZHAO Xiaoyu^2
,,
JIN Lei^3
,,
WANG Li^{1
, ,},
FEI Aiguo^4
,

1.
School of Computer Science(National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing 100876, China
2.
School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China
3.
School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
4.
State Key Laboratory of Multi-domain Data Collaborative Processing and Control, China

Funds: The National Natural Science Foundation of China under Grant (62525106), BUPT-China Unicom Joint Innovation Center under Grant (2025-STHZ-BJYDDX-004)

摘要

摘要: 应急救援现场通常面临传输与计算资源受限的挑战，既不具备现场完成各种复杂智能感知任务所需的算力，也难以低时延地将感知数据回传至后方边缘服务器上处理。语义通信系统由于仅传输支持任务执行的必要的语义特征，能够在恶劣信道环境下实现对特定任务的高效传输，其关键在于增强通信系统发送端和解码端对语义特征的共性理解能力。随着基础模型的快速发展，凭借经海量数据训练获得的通用语义理解与生成能力，为构建高效的任务导向语义通信架构提供了新的思路。本文提出一种基于生成式基础模型Stable Diffusion (SD)的任务导向语义通信框架，支持在恶劣信道条件下高效完成图像分类任务。该框架在发送端设计任务-信道联合感知编码器，根据信噪比动态提取任务所需的语义特征；在接收端以预训练SD模型为核心生成器，结合ControlNet实现带噪条件的可控生成，并通过任务解码器输出分类结果。为降低端到端联合训练中扩散模型多步采样带来的高计算开销，本文提出一种两阶段训练框架：第一阶段在无噪声环境下利用信息瓶颈学习任务相关特征并训练任务解码器；第二阶段冻结扩散模型主干，仅训练任务编码器与ControlNet实现信道噪声适配，避免了多步采样带来的高昂的梯度回传成本。针对信道噪声在扩散生成过程中累积导致生成特征语义不一致的问题，进一步设计一种自适应条件引导策略，通过可学习权重网络动态调节噪声条件注入强度，使得模型在去噪各步中根据当前状态自适应地增强或减弱条件引导，以应对信道噪声的动态影响。实验结果表明，在不同信噪比条件下所提方法均取得了较优的分类性能。
- 任务导向语义通信 /
- 条件扩散模型 /
- 信道自适应 /
- Stable Diffusion
Abstract: Objective In emergency rescue scenarios, on-site unmanned aerial vehicles and other intelligent equipment often lack sufficient computing power to execute complex perception tasks such as disaster identification and personnel detection, while transmitting raw image data back to edge servers suffers from severe bandwidth limitations and channel degradation. Task-oriented semantic communication systems address this challenge by transmitting only task-relevant semantic features, thereby achieving high compression ratios and robust performance under poor channel conditions. However, existing methods require end-to-end training, which becomes computationally prohibitive when integrating large generative models like Stable Diffusion. Moreover, channel noise accumulated during the reverse diffusion process can cause semantic drift, leading to degraded task accuracy. To overcome these limitations, this paper proposes a channel-adaptive conditional diffusion generation method for task-oriented semantic communication, specifically targeting image classification under low-signal-to-noise-ratio (SNR) conditions. The objective is to achieve robust semantic feature transmission and accurate classification by leveraging a pre-trained diffusion model at the receiver while maintaining low training complexity and adapting dynamically to varying channel qualities. Methods The proposed framework comprises three key components: a task-channel joint sensing encoder at the transmitter, a conditional diffusion generator based on Stable Diffusion (SD) with ControlNet at the receiver, and an adaptive condition guidance strategy. At the transmitter, the encoder extracts task-relevant semantic features, taking the input image and the current SNR as inputs. The encoder adaptively adjusts its feature extraction according to the channel condition via an attention-based modulation module. The semantic features are transmitted over an additive white Gaussian noise channel, yielding noisy features. At the receiver, a pre-trained SD model serves as the core generator, and a trainable ControlNet takes the noisy features as input to produce residual corrections that guide the reverse diffusion process. After a fixed number of DDIM sampling steps, the generated features are decoded by a task decoder to output classification results.To avoid the high computational cost of end-to-end backpropagation through the multi-step diffusion chain, a two-stage training framework is proposed. In the first stage, under a noise-free environment, an information bottleneck is used to learn task-relevant features and train the task decoder. The loss function combines cross-entropy classification loss and KL divergence for variational information bottleneck. In the second stage, the SD backbone is frozen, and only the task encoder and ControlNet are trained to adapt to channel noise. The training loss includes a standard diffusion denoising loss and a distillation loss that aligns the transmitted features with the first-stage features. To mitigate semantic drift caused by noise accumulation during reverse diffusion, an adaptive condition guidance strategy is introduced. A learnable weight network takes the current timestep and SNR as inputs and outputs a scaling factor between zero and two. When the scaling factor equals one, the conditional residual is injected as is; when it is less than one, the conditional guidance is weakened; when greater than one, the conditional guidance is enhanced. Results and Discussions Experiments are conducted on the ImageNet-100 dataset, comprising 130,000 training images and 5,000 validation images across 100 classes. All images are resized to 256×256 pixels and compressed to 4×32×32 latent features (48× compression). The Stable Diffusion v1.5 backbone is frozen during the second stage, with only the task encoder and ControlNet trained for 15 epochs at a learning rate of 1e-4. The adaptive weight network is trained during the DDIM sampling process for 5 epochs with a batch size of 96 and a learning rate of 1e-4, using 20 sampling steps.The proposed method is compared with two baselines: T-DeepSC, a single-task semantic communication system without diffusion, and DDSC+VGG16, a diffusion-driven semantic communication method that maps channel noise to forward diffusion steps and cascades VGG16 for classification. Under SNR values from -3 dB to 7 dB, the proposed method achieves the highest classification accuracy across all conditions. Specifically, at -3 dB, the proposed method attains 44.0% accuracy, outperforming T-DeepSC at 39.8% by 4.2 percentage points. At -1 dB, the proposed method reaches 53.2%, surpassing T-DeepSC at 40.0% by 13.2 percentage points. At 7 dB, the proposed method achieves 66.8%, exceeding DDSC+VGG16 at 58.4% by 8.4 percentage points. These results demonstrate that the proposed two-stage training framework with adaptive conditional guidance effectively recovers task-relevant semantic features from noisy transmissions, delivering superior performance even under severe channel degradation.Ablation studies further validate the contribution of each component. Removing the adaptive weight control, denoted as configuration B, reduces accuracy by 3.5 percentage points at -3 dB and 3.6 percentage points at -1 dB compared to the full method, configuration C, which achieves 44.0% and 53.2% respectively. Compared to the DDSC+VGG16 baseline, configuration A, introducing the task encoder and ControlNet, configuration B, yields dramatic improvements, for example, from 58.4% to 65.0% at 7 dB. This confirms that the two-stage framework successfully learns task-relevant semantic features and adapts to channel noise without end-to-end backpropagation through the diffusion chain. Conclusions This paper presents a channel-adaptive conditional diffusion generation method for task-oriented semantic communication, targeting image classification under bandwidth and resource constraints. The proposed framework integrates a task-channel joint sensing encoder, a ControlNet-guided Stable Diffusion generator, and a two-stage training strategy that decouples task feature learning from channel noise adaptation, significantly reducing training complexity compared to end-to-end diffusion-based alternatives. The adaptive condition guidance strategy dynamically adjusts the injection strength of noisy conditional information during the reverse diffusion process, suppressing semantic drift and improving classification robustness. Experimental results on ImageNet-100 demonstrate that the proposed method outperforms both task-oriented semantic communication without diffusion (T-DeepSC) and diffusion-driven reconstruction-then-classification (DDSC+VGG16) across most SNR conditions. In addition, the framework can also generate reconstructed images at the receiver, which may serve as visual auxiliary evidence for human verification in emergency rescue scenarios. Future work will extend the framework to multi-modal semantic fusion, model lightweighting for edge deployment, and more complex tasks such as object detection and visual question answering, facilitating practical adoption in emergency rescue and other mission-critical scenarios.
- task-oriented semantic communication /
- conditional diffusion model /
- channel adaptation /
- Stable Diffusion

HTML全文

图 1 基于SD模型的任务导向语义通信框架

下载: 全尺寸图片幻灯片

图 2 基于SD模型的任务导向语义通信系统模型

下载: 全尺寸图片幻灯片

图 3 第一阶段训练任务相关语义特征与任务解码器

下载: 全尺寸图片幻灯片

图 4 第二阶段训练任务编码器与ControlNet

下载: 全尺寸图片幻灯片

图 5 训练权重控制网络

下载: 全尺寸图片幻灯片

图 6 不同SNR条件下下权重 α随DDIM去噪步数的变化

下载: 全尺寸图片幻灯片

表 1 高斯信道条件下，本文与当前先进算法在分类任务对比结果

方法	分类精度(%)
方法	SNR=$ - $3 dB	SNR=$ - $1 dB	SNR=1 dB	SNR=3 dB	SNR=5 dB	SNR=7 dB
T-DeepSC	39.8	40	40.2	40.2	40.2	40.2
DDSC + VGG16	2.6	8.4	20.5	34.4	47.8	58.4
本文方案	44	53.2	60.3	64.4	65.3	66.8

下载: 导出CSV

表 2 瑞利衰落信道条件下，本文与当前先进算法在分类任务对比结果

方法	分类精度(%)
方法	SNR=$ - $3dB	SNR=$ - $1dB	SNR=1dB	SNR=3dB	SNR=5dB	SNR=7dB
T-DeepSC	36.5	37.3	38	38.6	39.1	39.2
本文方案	27.6	36.4	38.2	45.5	51.3	55.2

下载: 导出CSV

表 3 消融实验结果

方法	分类精度(%)
方法	SNR=$ - $3dB	SNR=$ - $1dB	SNR=1dB	SNR=3dB	SNR=5dB	SNR=7dB
A: DDSC + VGG16	2.6	8.4	20.5	34.4	47.8	58.4
B: 无权重控制	40.5	49.6	56.2	61	63.5	65
C: 完整方案	44	53.2	60.3	64.4	65.3	66.8

下载: 导出CSV

表 4 不同发送端编码器参数量下的分类性能

编码器参数量	分类精度(%)
编码器参数量	SNR=$ - $3dB	SNR=$ - $1dB	SNR=1dB	SNR=3dB	SNR=5dB	SNR=7dB
34.2M	44	53.2	60.3	64.4	65.3	66.8
4.3M	42.4	51.6	58.4	62.7	64.7	65.2
1.6M	41.2	50.4	58.6	62.1	64.6	65.5

下载: 导出CSV

参考文献(19)

[1]	王博文, 郑建, 孙彦景, 等. 应急场景无人机自组网部分重叠信道动态分配方法[J]. 电子与信息学报, 2024, 46(12): 4373–4382. doi: 10.11999/JEIT240377. WANG Bowen, ZHENG Jian, SUN Yanjing, et al. Partially overlapping channels dynamic allocation method for UAV Ad-hoc networks in emergency scenario[J]. Journal of Electronics & Information Technology, 2024, 46(12): 4373–4382. doi: 10.11999/JEIT240377.
[2]	王文远, 周明宇, 王朝炜, 等. 面向空天地网络的弹性语义通信[J]. 电子与信息学报, 2025, 47(10): 3646–3657. doi: 10.11999/JEIT250077. WANG Wenyuan, ZHOU Mingyu, WANG Chaowei, et al. Resilient semantic communication for space-air-ground networks[J]. Journal of Electronics & Information Technology, 2025, 47(10): 3646–3657. doi: 10.11999/JEIT250077.
[3]	ZHANG Guangyi, HU Qiyu, QIN Zhijin, et al. A unified multi-task semantic communication system for multimodal data[J]. IEEE Transactions on Communications, 2024, 72(7): 4101–4116. doi: 10.1109/TCOMM.2024.3364990.
[4]	BOURTSOULATZE E, KURKA D B, and GÜNDÜZ D. Deep joint source-channel coding for wireless image transmission[J]. IEEE Transactions on Cognitive Communications and Networking, 2019, 5(3): 567–579. doi: 10.1109/TCCN.2019.2919300.
[5]	LOKUMARAMBAGE M U, GOWRISETTY V S S, REZAEI H, et al. Wireless end-to-end image transmission system using semantic communications[J]. IEEE Access, 2023, 11: 37149–37163. doi: 10.1109/ACCESS.2023.3266656.
[6]	GETU T M, KADDOUM G, and BENNIS M. A survey on goal-oriented semantic communication: Techniques, challenges, and future directions[J]. IEEE Access, 2024, 12: 51223–51274. doi: 10.1109/ACCESS.2024.3381967.
[7]	邓梓焌, 何相腾, 彭宇新. 文本到视频生成: 研究现状、进展和挑战[J]. 电子与信息学报, 2024, 46(5): 1632–1644. doi: 10.11999/JEIT240074. DENG Zijun, HE Xiangteng, and PENG Yuxin. Text-to-video generation: Research status, progress and challenges[J]. Journal of Electronics & Information Technology, 2024, 46(5): 1632–1644. doi: 10.11999/JEIT240074.
[8]	HU Jiangjing, WU Haotian, ZHANG Wenjing, et al. Zero-shot semantic communication with multimodal foundation models[J]. IEEE Transactions on Vehicular Technology, 2026, 75(5): 8759–8764. doi: 10.1109/TVT.2025.3632893.
[9]	JIANG Feibo, PENG Yubo, DONG Li, et al. Large AI model-based semantic communications[J]. IEEE Wireless Communications, 2024, 31(3): 68–75. doi: 10.1109/MWC.001.2300346.
[10]	ZHANG Lvmin, RAO Anyi, and AGRAWALA M. Adding conditional control to text-to-image diffusion models[C]. 2023 IEEE/CVF International Conference on Computer Vision, Paris, France, 2023: 3813–3824. doi: 10.1109/ICCV51070.2023.00355.
[11]	GRASSUCCI E, BARBAROSSA S, and COMMINIELLO D. Generative semantic communication: Diffusion models beyond bit recovery[J]. IEEE Transactions on Cognitive Communications and Networking, 2026, 12: 8171–8185. doi: 10.1109/TCCN.2026.3689849.
[12]	YUAN Weiwen, REN Jinke, WANG Chongjie, et al. Generative semantic communication for joint image transmission and segmentation[C]. 2025 IEEE International Conference on Communications Workshops, Montreal, QC, Canada, 2025: 1110–1115. doi: 10.1109/ICCWorkshops67674.2025.11162317.
[13]	LI Ming, YANG Taojiannan, KUANG Huafeng, et al. ControlNet++: Improving conditional controls with efficient consistency feedback[C]. 18th European Conference on Computer Vision, Milan, Italy, 2024: 129–147. doi: 10.1007/978-3-031-72667-5_8.
[14]	CHEN Shoufa, SUN Peize, SONG Yibing, et al. Diffusiondet: Diffusion model for object detection[C]. 2023 IEEE/CVF International Conference on Computer Vision, Paris, France, 2023: 19773–19786. doi: 10.1109/ICCV51070.2023.01816.
[15]	GUO Lei, CHEN Wei, SUN Yuxuan, et al. Diffusion-driven semantic communication for generative models with bandwidth constraints[J]. IEEE Transactions on Wireless Communications, 2025, 24(8): 6490–6503. doi: 10.1109/TWC.2025.3553851.
[16]	HO J and SALIMANS T. Classifier-free diffusion guidance[Z]. arXiv: 2207.12598, 2022. doi: 10.48550/arXiv.2207.12598.
[17]	MEI Kangfu, NAI N G, and PATEL V M. Improving conditional diffusion models through re-noising from unconditional diffusion priors[C]. 2025 IEEE/CVF Winter Conference on Applications of Computer Vision, Tucson, AZ, USA, 2025: 3792–3801. doi: 10.1109/WACV61041.2025.00373.
[18]	ALEMI A A, FISCHER I, DILLON J V, et al. Deep variational information bottleneck[C]. 5th International Conference on Learning Representations, Toulon, France, 2017.
[19]	XU Jialong, AI Bo, CHEN Wei, et al. Wireless image transmission using deep source channel coding with attention modules[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(4): 2315–2328. doi: 10.1109/TCSVT.2021.3082521.