面向图像恢复任务的语义通信网络能耗优化

陈阳; 马欢; 姬智; 李英奇; 梁佳宇; 郭兰

doi:10.11999/JEIT250915

面向图像恢复任务的语义通信网络能耗优化

doi: 10.11999/JEIT250915 cstr: 32379.14.JEIT250915

陈阳^1, ,,
马欢²,
姬智¹,
李英奇¹,
梁佳宇¹,
郭兰¹

1.
中国人民解放军陆军工程大学南京 210007
2.
同方电子科技有限公司九江 332005

基金项目: 国家优秀青年科学基金项目(编号：62122094)，江苏省科技计划项目(编号：BK20253031)

详细信息

作者简介:
陈阳：男，在读博士生，研究方向为面向任务的多模态语义通信资源管理

马欢：男，高级工程师，研究方向为通信信号分析与处理

通讯作者:
陈阳　chenyangbabm@126.com

中图分类号: TP393.1
计量
- 文章访问数: 6
- HTML全文浏览量: 3
- PDF下载量: 0
- 被引次数: 0
出版历程
- 修回日期: 2025-12-08
- 录用日期: 2025-12-08
- 网络出版日期: 2025-12-13

Optimization of Energy Consumption in Semantic Communication Networks for Image Recovery Tasks

1.
Army Engineering University of PLA Nanjing 210007 China
2.
Tong Fang Electronic Science & Technology Co., Ltd Jiujiang 332005 China

摘要

摘要: 针对语义通信网络在图像恢复任务中计算和传输能耗过高的问题，该文提出一种改进型多智能体近端策略优化算法驱动的网络能耗优化策略，以在保障任务性能的同时最小化网络总能耗。首先，量化分析了语义提取率、发射功率、计算资源与网络能耗间的耦合关系。随后，构建以小区总能耗最小化为目标，同时满足时延、图像恢复质量等多维约束的优化模型。最后，设计改进型多智能体近端策略优化算法对该模型进行求解。仿真结果表明，与基准算法相比，该文所提算法在维持相当能耗水平的同时，训练收敛速度提升66.7%-80%，网络能耗和用户时延稳定性显著提升，并能有效降低平均误符号率。
- 语义通信 /
- 资源管理 /
- 能耗 /
- 多智能体深度强化学习
Abstract: Objective With the rapid development of semantic communication and the increasing demand for high-fidelity image recovery, high computation and transmission energy consumption have become critical issues limiting network deployment. However, existing resource management strategies are mostly static and have limitations in adapting to dynamic wireless environments and user mobility. To address these challenges, a robust energy optimization strategy driven by a modified Multi-Agent Proximal Policy Optimization (MAPPO) algorithm has emerged as a promising approach. By jointly optimizing communication and computing resources, it is possible to minimize the total network energy consumption while strictly satisfying multi-dimensional constraints such as latency and image recovery quality. Methods First, a theoretical model for the semantic communication network is constructed , and a closed-form expression for the user Symbol Error Rate (SER) is derived via asymptotic analysis of the uplink Signal-to-Interference-plus-Noise Ratio (SINR). Subsequently, the coupling relationship among semantic extraction rate, transmit power, computing resources, and network energy consumption is quantified. Based on this, a joint optimization model is established to minimize the total energy under constraints of delay, accuracy, and reliability. To solve this complex mixed-integer nonlinear programming problem, a modified MAPPO algorithm is designed. This algorithm incorporates Long Short-Term Memory (LSTM) networks to capture temporal dynamics of user positions and channel states, and introduces a noise mechanism into the global state and advantage function to enhance policy exploration and robustness. Results and Discussions Simulation results demonstrate that the proposed algorithm significantly outperforms baseline methods (including standard MAPPO, NOISE-MAPPO, LSTM-MAPPO, MADDPG, and Greedy algorithms). Specifically, the proposed strategy accelerates the training convergence speed by 66.7%–80% compared to benchmarks. Furthermore, the algorithm exhibits superior stability in dynamic environments, improving the stability of network energy consumption by approximately 50% and user latency stability by over 96%. Additionally, the average SER is effectively reduced by 4%–16.33% without compromising the ultimate image recovery performance, verifying the algorithm's capability to balance energy efficiency and task reliability. Conclusions This paper addresses the challenge of energy optimization in semantic communication networks by integrating theoretical modeling with a modified deep reinforcement learning framework. The proposed decision-making method enhances the standard MAPPO algorithm by leveraging LSTM for temporal feature extraction and noise mechanisms for robust exploration. The method is evaluated through simulations in dynamic single-cell and multi-cell scenarios, and the results show that: (1) The proposed method significantly improves convergence efficiency and system stability over the baselines; (2) A better trade-off between energy consumption and service quality is achieved, providing a theoretical foundation and an efficient resource management framework for future energy-constrained semantic communication systems.
- Semantic communication /
- Resource management /
- Energy consumption /
- Multi-agent deep reinforcement learning

HTML全文

图 1 单小区语义通信网络示意图

下载: 全尺寸图片幻灯片

图 2 用户与基站通过上行链路进行图像传输与恢复过程示意图

下载: 全尺寸图片幻灯片

图 3 不同算法下网络链路速率性能示意图

下载: 全尺寸图片幻灯片

图 4 多小区场景不同算法网络性能示意图

下载: 全尺寸图片幻灯片

表 1 仿真参数

仿真参数	参数值
小区数量	1
基站天线数量N_r	32
小区用户数量M	25
用户天线数量	1
用户最大发射功率P_k	100 mw
功率$ \left\{{p}_{level1},{p}_{level2},{p}_{level3},{p}_{level4}\right\} $	{25 mw, 50 mw, 75 mw, 100 mw}
语义提取率$ \{{\rho }_{level1},{\rho }_{level2}, $ ${\rho }_{level3},{\rho }_{level4},{\rho }_{level5},{\rho }_{level6}\} $	{1/6,2/6,3/6,4/6,5/6,1}
小区半径	900 m
用户上行信道带宽W	1 MHz
能耗系数$ \kappa $	10^–28
时延约束$ {t}_{th} $	100 ms
用户计算容量f	2 GHz
ECIW计算容量$ {F}_{\max } $	25 GHz
噪声功率	–174 dBm/Hz

下载: 导出CSV

表 2 所提算法与基准算法的网络误符号性能对比

算法	所有用户每回合最大平均误符号率均值	所有用户每回合最大平均误符号率方差
所提算法	$ 1.3796\times {10}^{-4} $	$ 6.5535\times {10}^{-7} $
MAPPO	$ 1.4420\times {10}^{-4} $	$ 6.8344\times {10}^{-7} $
LSTM-MAPPO	$ \text{1.5349}\times {\text{10}}^{\text{-4}} $	$ 7.8464\times {10}^{-7} $
NOISE-MAPPO	$ \text{1.6489}\times {\text{10}}^{\text{-4}} $	$ 7.2612\times {10}^{-7} $
MADDPG	$ \text{1.5251}\times {\text{10}}^{\text{-4}} $	$ 7.1457\times {10}^{-7} $
贪婪算法	$ \text{8.9753}\times {\text{10}}^{\text{-3}} $	$ 6.5866\times {10}^{-6} $

下载: 导出CSV

表 4 所提算法与基准算法的图像恢复性能对比(Imagenet数据集)

算法	所有用户每回合平均峰值信噪比均值(dB)	所有用户每回合平均峰值信噪比方差
所提算法	30.1345	0.0017
MAPPO	30.1005	0.0024
LSTM-MAPPO	29.5246	0.0084
NOISE-MAPPO	29.3471	0.0101
MADDPG	28.8562	0.0075
贪婪算法	11.1175	0.0159

下载: 导出CSV

表 3 所提算法与基准算法的图像恢复性能对比(MNIST数据集)

算法	所有用户每回合平均峰值信噪比均值(dB)	所有用户每回合平均峰值信噪比方差
所提算法	35.1475	0.0026
MAPPO	35.1005	0.0054
LSTM-MAPPO	34.5487	0.0097
NOISE-MAPPO	34.2211	0.0097
MADDPG	33.9256	0.0065
贪婪算法	15.6624	0.0219

下载: 导出CSV

参考文献(22)

[1]	张平. 语义通信: 未来通信系统的智简之道[J]. 中兴通讯技术, 2024, 30(1): 1. doi: 10.12142/ZTETJ.202401001. ZHANG Ping. Semantic communication: The intelligent and concise way to the future communication system[J]. ZTE Technology Journal, 2024, 30(1): 1. doi: 10.12142/ZTETJ.202401001.
[2]	朱政宇, 梁馨月, 孙钢灿, 等. 智能超表面赋能语义通信系统研究综述[J]. 电子与信息学报, 2025, 47(2): 287–295. doi: 10.11999/JEIT240984. ZHU Zhengyu, LIANG Xinyue, SUN Gangcan, et al. Research overview of reconfigurable intelligent surface enabled semantic communication systems[J]. Journal of Electronics & Information Technology, 2025, 47(2): 287–295. doi: 10.11999/JEIT240984.
[3]	ZHANG Ping, LIU Yiming, SONG Yile, et al. Advances and challenges in semantic communications: A systematic review[J]. National Science Open, 2024, 3(4): 20230029. doi: 10.1360/nso/20230029.
[4]	SHAO Yulin, CAO Qi, and GÜNDÜZ D. A theory of semantic communication[J]. IEEE Transactions on Mobile Computing, 2024, 23(12): 12211–12228. doi: 10.1109/TMC.2024.3406375.
[5]	PENG Xiang, QIN Zhijin, TAO Xiaoming, et al. A robust image semantic communication system with multi-scale vision transformer[J]. IEEE Journal on Selected Areas in Communications, 2025, 43(4): 1278–1291. doi: 10.1109/JSAC.2025.3531413.
[6]	VAN CHIEN T, PHONG L H, PHUC D X, et al. Image restoration under semantic communications[C]. 2022 International Conference on Advanced Technologies for Communications, Ha Noi, Vietnam, 2022: 332–337. DOI: 10.1109/ATC55345.2022.9943000.
[7]	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.
[8]	YAN Zhigang and LI Dong. Semantic communications for digital signals via carrier images[J]. IEEE Wireless Communications Letters, 2025, 14(6): 1816–1820. doi: 10.1109/LWC.2025.3557843.
[9]	LIANG Chengyang and LI Dong. Image generation with supervised selection based on multimodal features for semantic communications[J]. IEEE Transactions on Communications. doi: 10.1109/TCOMM.2025.3615798.
[10]	DAI Jincheng, WANG Sixian, TAN Kailin, et al. Nonlinear transform source-channel coding for semantic communications[J]. IEEE Journal on Selected Areas in Communications, 2022, 40(8): 2300–2316. doi: 10.1109/JSAC.2022.3180802.
[11]	YANG Pujing, ZHANG Guangyi, and CAI Yunlong. Rate-adaptive generative semantic communication using conditional diffusion models[J]. IEEE Wireless Communications Letters, 2025, 14(2): 539–543. doi: 10.1109/LWC.2024.3515656.
[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, Canada, 2025: 1110–1115. doi: 10.1109/ICCWorkshops67674.2025.11162317.
[13]	ZHANG Guangyi, YANG Pujing, CAI Yunlong, et al. From analog to digital: Multi-order digital joint coding-modulation for semantic communication[J]. IEEE Transactions on Communications, 2025, 73(6): 4257–4271. doi: 10.1109/TCOMM.2024.3511949.
[14]	TUNG T Y, KURKA D B, JANKOWSKI M, et al. DeepJSCC-Q: Constellation constrained deep joint source-channel coding[J]. IEEE Journal on Selected Areas in Information Theory, 2022, 3(4): 720–731. doi: 10.1109/JSAIT.2022.3231042.
[15]	YANG Zhaohui, CHEN Mingzhe, ZHANG Zhaoyang, et al. Energy efficient semantic communication over wireless networks with rate splitting[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(5): 1484–1495. doi: 10.1109/JSAC.2023.3240713.
[16]	ZHANG Hongwei, SHAO Shuo, TAO Meixia, et al. Deep learning-enabled semantic communication systems with task-unaware transmitter and dynamic data[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(1): 170–185. doi: 10.1109/JSAC.2022.3221991.
[17]	翟凯. 基于MMSE检测的MIMO及大规模MIMO系统性能精确分析[D]. [博士论文], 西南交通大学, 2021. doi: 10.27414/d.cnki.gxnju.2021.000015. ZHAI Kai. Exact performance analysis of MIMO and massive MIMO systems with MMSE receiver[D]. [Ph. D. dissertation], Southwest Jiaotong University, 2021. doi: 10.27414/d.cnki.gxnju.2021.000015.
[18]	QIN Langtian, LU Hancheng, CHEN Yuang, et al. Toward decentralized task offloading and resource allocation in user-centric MEC[J]. IEEE Transactions on Mobile Computing, 2024, 23(12): 11807–11823. doi: 10.1109/TMC.2024.3399766.
[19]	张峻伟, 吕帅, 张正昊, 等. 基于样本效率优化的深度强化学习方法综述[J]. 软件学报, 2022, 33(11): 4217–4238. doi: 10.13328/j.cnki.jos.006391. ZHANG Junwei, LÜ Shuai, ZHANG Zhenghao, et al. Survey on deep reinforcement learning methods based on sample efficiency optimization[J]. Journal of Software, 2022, 33(11): 4217–4238. doi: 10.13328/j.cnki.jos.006391.
[20]	HU Siyue and HU Jian. Noisy-MAPPO: Noisy advantage values for cooperative multi-agent actor-critic methods[J]. arXiv: 2106.14334v1, 2021. doi: 10.48550/arXiv.2106.14334. (查阅网上资料,未能确认文献类型,请确认).
[21]	SCHULMAN J, WOLSKI F, DHARIWAL P, et al. Proximal policy optimization algorithms[J]. arXiv: 1707.06347, 2017. doi: 10.48550/arXiv.1707.06347.(查阅网上资料,未能确认文献类型,请确认).
[22]	LILLICRAP T P, HUNT J J, Pritzel A, et al. Continuous control with deep reinforcement learning[C]. Proceedings of the 4th International Conference on Learning Representations, San Juan, Puerto Rico, 2016.