RIS-Enhanced Semantic Communication Systems Oriented towards Semantic Importance and Robustness
-
摘要: 为了在语义通信中高效传递语义信息，需要减少语义级损失并进一步降低物理噪声对语义特征的影响。对此，该文提出了一种面向语义重要性和鲁棒性并联合智能反射面(RIS)进行跨层设计的语义通信系统。在语义层，基于语义特征重要性和鲁棒性更新背景知识库并生成用于评估数据流传输优先级的特征优先级，发送端再根据特征优先级将数据流分割为高优先级数据流和低优先级数据流。在物理层，通过交替优化主动预编码波束成形矢量和RIS被动相移对抗信道衰落，然后由RIS辅助传输高优先级数据流，由衰落信道传输低优先级数据流，并在接收端依据特征优先级列表恢复原始文本。仿真结果表明，相较于基准方案，所提方案不仅有着更好的鲁棒性而且提高了双语评价替补(BLEU)分数和语义相似度，同时对不同长度的句子具有良好的适配性。Abstract:
Objective The deep integration of Deep Learning (DL) and Semantic Communication (SC) has become a key trend in next-generation communication systems. Current SC systems primarily adopt DL-based Joint Source-Channel Coding (JSCC) with end-to-end training to enable efficient semantic transmission. However, several limitations remain. Existing systems often optimize physical-layer channel characteristics or semantic-layer feature extraction in isolation, without establishing cross-layer mapping mechanisms. In addition, protection strategies for critical semantic features in fading channel environments are insufficient, limiting semantic recovery performance. To address these challenges, this study integrates Reconfigurable Intelligent Surfaces (RIS) into SC systems and proposes an intelligent transmission scheme based on dual-dimensional semantic feature metrics. The proposed approach effectively enhances semantic recovery capability under adverse channel conditions. This work provides a new intelligent solution for protecting semantic features in fading channels and establishes theoretical support for collaborative mechanisms between physical and semantic layers in SC systems. Methods This study develops a joint semantic importance-robustness metric model. Semantic importance is quantified using Bidirectional Encoder Representations from Transformers (BERT) combined with cosine similarity, while semantic robustness is assessed by measuring the loss increments of high-dimensional feature vectors during transmission. A dynamically updated background knowledge base is constructed to support a priority evaluation framework for semantic features ( Fig. 2 ). During transmission, the system partitions the original text into high- and low-priority data streams based on feature priority. High-priority streams are transmitted through RIS-assisted channels, whereas low-priority streams are transmitted over conventional fading channels. At the physical layer, an alternating optimization algorithm jointly designs active precoding beamforming vectors and RIS passive phase matrices. At the receiver, semantic reconstruction is performed under the guidance of feature priority index lists (Fig. 1 ).Results and Discussions The proposed SISR-RIS system effectively reduces the distortion effects of channel fading on critical semantic features by establishing cross-layer mapping between semantic features and physical channels. Simulation results show that, in medium-to-low Signal-to-Noise Ratio (SNR) environments, the SISR-RIS system maintains high low-order BLEU scores and approaches the theoretical performance boundary near the 10 dB SNR threshold, achieving approximately 95% recovery accuracy for BLEU-1 and 92% for BLEU-2 ( Fig.3 (a)). As the n-gram order increases, the system outperforms the baseline Deep-SC system by approximately 10% in BLEU-4, confirming its improved capability for contextual semantic reconstruction (Fig.3 (b)). Owing to the dual-dimensional metric mechanism, the system demonstrates stable performance with less than 1% variance in recovery accuracy across short and long sentences (Fig. 4 ). Case analysis indicates that when the original statements cannot be fully restored, the system maintains semantic equivalence through appropriate synonym substitutions. Additionally, core verbs and nouns are consistently assigned higher feature priority scores, which reduces the effect of channel fading on critical semantic features (Tables 2 and3 ;Figs. 5 and6 ).Conclusions This study proposes a RIS-enhanced SC system designed to account for semantic importance and robustness. By extracting semantic importance and robustness features to prioritise transmission and implementing a joint physical-semantic layer design enabled by RIS, the system provides enhanced protection for high-importance, low-robustness semantic features. Evaluations based on BLEU scores, BERT Semantic Similarity (BERT-SS) metrics, and case analyses demonstrate the following: (1) The proposed system achieves a 15% performance improvement over baseline systems in low SNR environments, with performance approaching theoretical limits near the 10 dB SNR threshold; (2) In high-SNR conditions, the system performs comparably to state-of-the-art methods across both BLEU and BERT-SS metrics; (3) The dual-dimensional semantic feature metric mechanism enhances contextual semantic relevance, reduces the recovery discrepancy between long and short sentences to below 1% in high-SNR scenarios, and demonstrates strong adaptability to varying text lengths. -
1 语义特征提取算法
1: 输入：背景知识库K，文本数据集$ s $，衰落信道系数$ {h_d} $，高斯噪声$ n $，已完成预训练的神经网络$ E_\beta ^{\rm C}\left( \cdot \right),E_\alpha ^{\rm S}\left( \cdot \right),D_\delta ^{\rm C}\left( \cdot \right),D_\chi ^{\rm S}\left( \cdot \right) $ 2: 加载预训练模型BERT 3: while未遍历完所有文本时，do 4: 利用BERT模型得到$ s $中第$ l $个句子的第$ i $个单词的向量表示$ {{\boldsymbol{B}}_\psi }({s_{l,i}}) $ 5: 经语义编码器$ E_\alpha ^{\rm S}\left( \cdot \right) $后$ s $被映射为高维特征向量$ {s^ * } $ 6: 语义重要性计算： 7: 通过式(8)和式(9)计算$ {{\boldsymbol{B}}_\psi }({s_{l,i}}) $的平均语义关联度$ \varPsi ( \cdot ) $ 8: 通过式(10)基于语义损失计算语义重要性，得到语义重要性分数列表$ {\boldsymbol{I}} $ 9: 处理$ s $中的特殊标记和停用词，将其在$ {\boldsymbol{I}} $中对应的重要性分数置为0 10: 检测分词器可能生成的子词，重新整合$ {\boldsymbol{I}} $ 11: 语义鲁棒性计算： 12: 随机产生扰动噪声$ \delta $ 13: 加入扰动噪声$ \delta $对高维特征向量$ {{\boldsymbol{s}}^ * } $进行干扰 14: 通过式(12)和式(13)计算扰动特征向量通过SC系统后的损失增量 15: 通过式(14)基于损失增量进行语义鲁棒性计算，语义鲁棒性分数列表$ {\boldsymbol{r}} $ 16: end while 17: 基于$ {\boldsymbol{I}} $和$ {\boldsymbol{r}} $加权计算特征优先级列表$ {\boldsymbol{\sigma}} $，并对K进行更新 18: 输出：包含特征优先级列表$ {\boldsymbol{\sigma}} $的背景知识库K 
下载: 导出CSV
2 SISR-RIS系统端到端训练步骤
1: 输入：背景知识库K，文本数据集$ s $，衰落信道系数$ {h_{\mathrm{d}}},{{\boldsymbol H}_{\rm{dr}}},{{\boldsymbol{H}}_{\rm T}},{{\boldsymbol{H}}_{\rm R}} $，高斯噪声$ n $，已完成预训练的神经网络$ E_\beta ^{\rm C}\left( \cdot \right),E_\alpha ^{\rm S}\left( \cdot \right),D_\delta ^{\rm C}\left( \cdot \right) $,
$D_\chi ^{\rm S}\left( \cdot \right) $2: 初始化参数集 3: while停止条件不满足时，do 4: 利用交替优化算法解决$ {{\mathrm{P}}_1} $，得到优化后的RIS反射相移系数矩阵 5: 发送机： 6: 从K中取出一批次$ s $ 7: $ E_\alpha ^{\rm S}(s) \to {\boldsymbol{m}} $，$ E_\beta ^{\rm C}({\boldsymbol{m}}) \to {\boldsymbol{x}} $ 8: 依据K中的$ {\boldsymbol{\sigma}} $进行切割$ {{\boldsymbol x}_{\rm L}},{{\boldsymbol x}_{\rm H}} = {F_{\rm T}}({\boldsymbol{x}};{\boldsymbol{\sigma}} ) $ 9: 通过无RIS辅助衰落信道和RIS辅助衰落信道分别对$ {{\boldsymbol x}_{\rm L}} $和$ {{\boldsymbol x}_{\rm H}} $进行传输 10: RIS：根据得到的RIS反射相移系数矩阵对输入信号$ {{\boldsymbol x}_{\rm H}} $进行反射 11: 接收机： 12: 依据K中的$ {\boldsymbol{\sigma}} $对$ {{\boldsymbol y}_{\rm H}} $和$ {{\boldsymbol y}_{\rm L}} $进行恢复，得到$ {\boldsymbol{y}} = {F_{\rm R}}({{\boldsymbol y}_{\rm H}},{{\boldsymbol y}_{\rm L}};{\boldsymbol{\sigma}} ) $ 13: $ D_\delta ^{\rm C}\left( {\boldsymbol{y}} \right) \to \hat {\boldsymbol{m}} $，$ D_\chi ^{\rm S}\left( {\hat {\boldsymbol{m}}} \right) \to \hat s $ 14: 通过式(18)计算损失$ {\varGamma _{{\mathrm{CE}}}}(s,\hat s;\beta ,\alpha ,\chi ,\delta ) $，并用Adam优化器训练$ \beta ,\alpha ,\chi ,\delta $ 15: end while 16: 输出：已完成训练的神经网络$ E_\beta ^{\rm C}( \cdot ),E_\alpha ^{\rm S}( \cdot ),D_\delta ^{\rm C}( \cdot ),D_\chi ^{\rm S}( \cdot ) $
下载: 导出CSV
表 1 语义层网络参数设置表
各部分名称 具体层名 维度 相关细节 语义编码器 嵌入层 22234 ×12822234 为字典大小语义编码器 Transformer编码器×3 128(8个头) Linear激活 信道编码器 全连接层 128×256 ReLU激活 信道编码器 全连接层 256×16 ReLU激活 信道 RIS或瑞利信道 无 无 信道解码器 全连接层 16×256 ReLU激活 信道解码器 全连接层 256×128 ReLU激活 语义解码器 Transformer解码器×3 128(8个头) Linear激活 语义解码器 预测层 22234 Softmax激活
下载: 导出CSV
表 2 验证集中示例短句比较
示例句子 the vote will be taken at noon on thursday Proposed-SISR-RIS this vote will be taken at noon on thursday Deep-SC-Rayleigh the vote will be taken at noon on tuesday RLJSCC-Rayleigh the vote will be taken at afternoon on thursday
下载: 导出CSV
表 3 验证集中示例长句比较
示例句子 this report is significant for small airports and their income could be dramatically reduced as a result Proposed-SISR-RIS this report is significant for small airports and their income could be dramatically reduced as a result Deep-SC-Rayleigh this report is considerable for small businesses and their income could be treated reduced as a result RLJSCC-Rayleigh this report is significant for small airports and their income could be significantly reduced as a result
下载: 导出CSV
-
[1] FARSHBAFAN M K, SAAD W, and DEBBAH M. Curriculum learning for goal-oriented semantic communications with a common language[J]. IEEE Transactions on Communications, 2023, 71(3): 1430–1446. doi: 10.1109/TCOMM.2023.3236671. [2] SHI Guangming, XIAO Yong, LI Yingyu, et al. From semantic communication to semantic-aware networking: Model, architecture, and open problems[J]. IEEE Communications Magazine, 2021, 59(8): 44–50. doi: 10.1109/MCOM.001.2001239. [3] XIE Huiqiang, QIN Zhijin, LI G Y, et al. Deep learning enabled semantic communication systems[J]. IEEE Transactions on Signal Processing, 2021, 69: 2663–2675. doi: 10.1109/TSP.2021.3071210. [4] XIE Huiqiang and QIN Zhijin. A lite distributed semantic communication system for Internet of Things[J]. IEEE Journal on Selected Areas in Communications, 2021, 39(1): 142–153. doi: 10.1109/JSAC.2020.3036968. [5] WANG Sixian, DAI Jincheng, LIANG Zijian, et al. Wireless deep video semantic transmission[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(1): 214–229. doi: 10.1109/JSAC.2022.3221977. [6] TONG Wanjie, LIU Fangfang, SUN Zhengfen, et al. Image semantic communications: An extended rate-distortion theory based scheme[C]. Proceedings of 2022 IEEE Globecom Workshops, Rio de Janeiro, Brazil, 2022: 1723–1728. doi: 10.1109/GCWkshps56602.2022.10008733. [7] HAN Yu, TANG Wankai, JIN Shi, et al. Large intelligent surface-assisted wireless communication exploiting statistical CSI[J]. IEEE Transactions on Vehicular Technology, 2019, 68(8): 8238–8242. doi: 10.1109/TVT.2019.2923997. [8] HUANG Chongwen, ZAPPONE A, ALEXANDROPOULOS G C, et al. Reconfigurable intelligent surfaces for energy efficiency in wireless communication[J]. IEEE Transactions on Wireless Communications, 2019, 18(8): 4157–4170. doi: 10.1109/TWC.2019.2922609. [9] LIU Yuanwei, LIU Xiao, MU Xidong, et al. Reconfigurable intelligent surfaces: Principles and opportunities[J]. IEEE Communications Surveys & Tutorials, 2021, 23(3): 1546–1577. doi: 10.1109/COMST.2021.3077737. [10] WU Qingqing, ZHANG Shuowen, ZHENG Beixiong, et al. Intelligent reflecting surface-aided wireless communications: A tutorial[J]. IEEE Transactions on Communications, 2021, 69(5): 3313–3351. doi: 10.1109/TCOMM.2021.3051897. [11] DU Hongyang, WANG Jiacheng, NIYATO D, et al. Semantic communications for wireless sensing: RIS-aided encoding and self-supervised decoding[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(8): 2547–2562. doi: 10.1109/JSAC.2023.3288231. [12] JIANG Peiwen, WEN Chaokai, JIN Shi, et al. RIS-enhanced semantic communications adaptive to user requirements[J]. IEEE Transactions on Communications, 2024, 72(7): 4134–4148. doi: 10.1109/TCOMM.2024.3369697. [13] MA Junyu, LI Qiang, LIU Ronghua, et al. Enhanced semantic information transfer on RIS-assisted communication systems[J]. IEEE Wireless Communications Letters, 2024, 13(8): 2225–2229. doi: 10.1109/LWC.2024.3408420. [14] 郭帅帅, 李树静. 基于语义重要度的不等错误保护数据传输机制[J]. 中兴通讯技术, 2023, 29(2): 62–66. doi: 10.12142/ZTETJ.202302012.GUO Shuaishuai and LI Shujing. Semantic importance-aware data transmission with unequal error protection[J]. ZTE Technology Journal, 2023, 29(2): 62–66. doi: 10.12142/ZTETJ.202302012. [15] GUO Shuaishuai, WANG Yanhu, LI Shujing, et al. Semantic importance-aware communications using pre-trained language models[J]. IEEE Communications Letters, 2023, 27(9): 2328–2332. doi: 10.1109/LCOMM.2023.3293805. [16] LIU Chuanhong, GUO Caili, YANG Yang, et al. OFDM-based digital semantic communication with importance awareness[J]. IEEE Transactions on Communications, 2024, 72(10): 6301–6315. doi: 10.1109/TCOMM.2024.3397862. [17] GUO Shuaishuai, WANG Yanhu, YE Jia, et al. Semantic importance-aware communications with semantic correction using large language models[J]. IEEE Transactions on Machine Learning in Communications and Networking, 2025, 3: 232–245. doi: 10.1109/TMLCN.2025.3530875. [18] HU Qiyu, ZHANG Guangyi, QIN Zhijin, et al. Robust semantic communications with masked VQ-VAE enabled codebook[J]. IEEE Transactions on Wireless Communications, 2023, 22(12): 8707–8722. doi: 10.1109/TWC.2023.3265201. [19] CHU Hongyun, PAN Xue, JIANG Jing, et al. Adaptive and robust channel estimation for IRS-aided millimeter-wave communications[J]. IEEE Transactions on Vehicular Technology, 2024, 73(7): 9411–9423. doi: 10.1109/TVT.2024.3385776. [20] GUO Shuaishuai, WANG Yanhu, and ZHANG Peng. Signal Shaping for Semantic Communication Systems with A Few Message Candidates[C]. Proceedings of 2022 IEEE 96th Vehicular Technology Conference, London, United Kingdom, 2022: 1–5. doi: 10.1109/VTC2022-Fall57202.2022.10012981. [21] DEVLIN J, CHANG Mingwei, LEE K, et al. BERT: Pre-training of deep bidirectional transformers for language understanding[C]. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Minneapolis, Minnesota, 2019: 4171–4186. [22] ZHOU Kequan, ZHANG Guangyi, CAI Yunlong, et al. FAST: Feature arrangement for semantic transmission[C]. Proceedings of 2024 IEEE Wireless Communications and Networking Conference, Dubai, United Arab Emirates, 2024: 1–6. doi: 10.1109/WCNC57260.2024.10570995. [23] KOEHN P. Europarl: A parallel corpus for statistical machine translation[C]. Proceedings of Machine Translation Summit X: Papers, Phuket, Thailand, 2005: 79–86. [24] WU Qingqing and ZHANG Rui. Intelligent reflecting surface enhanced wireless network via joint active and passive beamforming[J]. IEEE Transactions on Wireless Communications, 2019, 18(11): 5394–5409. doi: 10.1109/TWC.2019.2936025. [25] HUFFMAN D. A method for the construction of minimum-redundancy codes[J]. Proceedings of the IRE, 1952, 40(9): 1098–1101. doi: 10.1109/JRPROC.1952.273898. [26] BERROU C, GLAVIEUX A, and THITIMAJSHIMA P. Near shannon limit error-correcting coding and decoding: Turbo-codes. 1[C]. Proceedings of ICC’93 - IEEE International Conference on Communications, Geneva, Switzerland, 1993: 1064–1070. doi: 10.1109/ICC.1993.397441. [27] LU Kun, LI Rongpeng, CHEN Xianfu, et al. Reinforcement learning-powered semantic communication via semantic similarity[EB/OL]. http://arxiv.org/abs/2108.12121, 2021. -
图(6) / 表(5)
计量
- 文章访问数: 160
- HTML全文浏览量: 63
- PDF下载量: 48
- 被引次数: 0
下载:
