Foundations of Semantic Communication Networks

Walid Saad is a Professor with the Department of Electrical and Computer Engineering, Virginia Tech, USA, where he leads the Network Science, Wireless, and Security (NEWS) Laboratory. Christina Chaccour is a Network Solutions Manager at Ericsson Inc., USA, where she spearheads product solutions for 5G-Advanced, 6G, and AI integration across North America. Christo Kurisummoottil Thomas is a Post-Doctoral Fellow with the Department of Electrical and Computer Engineering, Virginia Tech, USA. Merouane Debbah is a Professor at Khalifa University of Science and Technology, UAE, and founding Director of the KU 6G Research Center.

• About the Editors xviiList of Contributors xxiPreface xxviiAcknowledgment xxixAcronyms xxxi1 Introduction to Semantic Communications 1Christina Chaccour, Christo Kurisummoottil Thomas, Walid Saad, and Merouane Debbah1.1 From Information Streams to Streams of Understanding: The Rise of Semantic Communication Networks 11.1.1 How Does It Work? 31.1.2 Why Now? What Factors Contribute to Our Ongoing Reliance on Traditional Communications? 61.1.3 What Is NOT Semantic Communications? 71.1.3.1 Semantic Communications Is Not Data Compression 71.1.3.2 Semantic Communications Is Not Only an “AI for Wireless” Concept 91.1.3.3 Semantic Communications Is Not Only Goal-Oriented Communications 91.1.3.4 Semantic Communications Is Not Only Application-Aware Communications 101.2 Reimagining Future G Applications with Semantic Communications 111.2.1 Semantic Communication for Next-Generation XR 111.2.2 Digital Reality and Massive Twinning: Speaking the Same Language 131.2.3 Semantic Communication and Sustainable Networks: A Convergence for Efficiency 141.3 Structure and Path of the Book 16Bibliography 17Part I Fundamentals of Semantic Communications 192 Semantic Compression and Communication: Fundamentals and Methodologies 21Emrecan Kutay and Aylin Yener2.1 Introduction 212.1.1 Notation 232.2 Semantic Index Assignment 232.2.1 System Model 242.2.1.1 Minimization of Semantic Distortion 252.2.1.2 Graph Coloring Problem 262.2.1.3 Joint Graph Coloring and Index Assignment 272.3 The Rise of Machine Intelligence in Perception 292.4 Semantic Compression for Multimodal Sources 322.4.1 System Model 322.4.1.1 Semantic Quantization 332.4.1.2 Semantic Compression 342.4.1.3 Semantic Vector Quantized Autoencoder 362.4.2 Results 392.5 Conclusion 44Bibliography 443 Toward a Theory of Semantic Information 49Jean-Claude Belfiore and Daniel Bennequin3.1 Introduction 493.2 Cohomological Nature of Information 503.3 Axioms for Information Spaces 523.4 Comparison with Other Propositions of Semantic Information Measures 543.5 Carnap and Bar-Hillel Languages 543.6 Shepard’s Experiment 57Bibliography 594 Deep Joint Source and Channel Coding 61Haotian Wu, Chenghong Bian, Yulin, Shao, and Deniz Gündüz4.1 Introduction 614.2 DeepJSCC for MIMO Channels 624.2.1 System Model 624.2.1.1 Open-loop MIMO with CSIR 634.2.1.2 Closed-loop MIMO with CSIT 644.2.2 A DeepJSCC-MIMO Solution 654.2.2.1 Image-to-Sequence Transformation 664.2.2.2 Channel Heatmap Construction 664.2.2.3 ViT Encoding 674.2.2.4 ViT Decoding 674.2.2.5 Loss Function 684.2.3 Training and Evaluation 684.2.3.1 Open-loop MIMO System with CSIR 684.2.3.2 Closed-loop MIMO System with CSIT 714.3 DeepJSCC for Relay Channels 754.3.1 Cooperative Relay 754.3.1.1 System Model 754.3.1.2 DeepJSCC for Cooperative Relay 764.3.1.3 Numerical Experiments 794.3.2 Multihop Relay 814.3.2.1 System Model 824.3.2.2 Existing Methods 834.3.2.3 A Hybrid JSCC Solution 844.3.2.4 Numerical Experiments 884.4 DeepJSCC for Feedback Channels 914.4.1 System Model 914.4.2 A JSCCFormer-f Solution 934.4.2.1 ViT Encoder 934.4.2.2 ViT Decoder 954.4.3 Training and Evaluation 964.4.3.1 Transmission Performance 964.4.3.2 Impacts of Bandwidth Ratio and Block Number 974.4.3.3 Noisy Feedback Channel 1004.4.3.4 Adaptability 1004.4.3.5 High Resolution Dataset and Visualization 1024.4.3.6 Variable Rate Transmission 1024.5 Concluding Remarks 106Bibliography 1075 When Information Is a Function of Data – Some Information Theoretic Perspectives on Semantic Communications 111Alexander Mariona, Homa Esfahanizadeh, Rafael Gregorio Lucas D’Oliveira, and Muriel Médard5.1 The Central Limit Theorem 1125.2 Quantitative Bounds 1135.3 General Polynomials 1155.4 Examples and Applications 1195.4.1 The Computational Wiretap Channel 1205.5 Further Generalizations 122Bibliography 1236 Interoperability and Coexistence of 6G Semantic, Goal-Oriented, and Legacy Systems 125Emilio Calvanese Strinati, Mohamed Sana, Mattia Merluzzi, and Tomás Huttebraucker6.1 Introduction 1256.2 Interoperability Issue in Goal-oriented and Semantic Systems 1266.2.1 Language in Multiuser Communication 1286.2.2 A Measure of Semantic Mismatch 1296.2.3 Semantic Channel Equalization 1306.3 Coexistence of Semantic, Goal-Oriented, and Legacy Services in 6G 1346.3.1 Goal-Oriented Resource Allocation 1356.3.2 Goal-Driven Measures for Edge Inference 1356.4 Conclusion 137Acknowledgment 138Bibliography 138Part II Semantic Communications Networking 1417 Optimization of Image Transmission in a Cooperative Semantic Communication Networks 143Ye Hu and Mingzhe Chen7.1 Introduction 1437.1.1 Related Works 1447.2 Representative Work 1457.2.1 System Model 1457.2.2 Semantic Information Extraction 1467.2.3 Transmission Model 1497.2.4 Image Semantic Similarity Model 1507.2.5 Problem Formulation 1517.3 Value-Decomposition-based Entropy-Maximized Multi-Agent RL Method 1527.3.1 Components of VD-ERL Method 1527.3.2 VD-ERL Algorithm for Semantic-Oriented Resource Allocation 1557.3.3 Complexity and Convergence of the Introduced Algorithm 1577.4 Simulation Results and Analysis 1587.5 Conclusion 161Bibliography 1628 Multiple Access Design for Joint Semantic and Classical Communications 165Xidong Mu and Yuanwei Liu8.1 Introduction 1658.2 Heterogeneous Semantic and Bit Multiuser Network 1678.2.1 Multiple Access for the Heterogeneous Semantic and Bit Multiuser Network 1698.2.2 Interplay Between Semantic Communications and NOMA 1698.3 NOMA-Enabled Heterogeneous Semantic and Bit Multiuser Communications 1708.3.1 Semantic Rate: A New Performance Metric 1708.3.2 Semi-NOMA: A Unified Multiple Access Scheme 1718.3.3 Fundamental Limit: Semantic-Versus-Bit Rate Region 1738.4 Semantic Communications-Enhanced NOMA 1758.4.1 Early-Late Rate Disparity Issue in NOMA 1758.4.2 An Opportunistic Semantic and Bit Communication Approach for Noma 1778.4.3 Numerical Case Studies 1778.5 Concluding Remarks and Future Research 179Bibliography 1799 Contextual Reasoning-based Semantics-Native Communication 181Hyowoon Seo, Yoon Huh, Heekang Song, Wan Choi, and Mehdi Bennis9.1 Semantics-Native Communication 1819.1.1 System Model 1829.1.1.1 Information-Theoretic Model Description 1839.1.1.2 Motivation from Triangle of Meaning Model 1839.2 Contextual Reasoning for Semantics-Native Communication 1849.2.1 Motivation from Referential Game 1859.2.2 Single-Sided Contextual Reasoning 1859.2.3 Double-Sided Contextual Reasoning 1889.2.4 Multi-round Contextual Reasoning 1899.3 Context Synchronization for Semantics-Native Communication 1929.3.1 Bayesian Inverse Contextual Reasoning 1939.3.2 Inverse Linearized Contextual Reasoning 1949.3.2.1 Linearizing Contextual Reasoning 1959.3.2.2 Invertible Linearized Contextual Reasoning 1969.4 Information Bottleneck Contextual Reasoning 1979.4.1 Information Bottleneck Method 1979.4.2 Implementing Information Bottleneck with Contextual Reasoning 1989.5 Conclusion 198Bibliography 19910 Interoperable Semantic Communication 201Jinhyuk Choi, Hyelin Nam, Jihong Park, Seung-Woo Ko, Jinho Choi, Mehdi Bennis, and Seong-Lyun Kim10.1 Pitfalls of Federated Learning for Semantic Alignment 20110.2 Split Learning for Semantic Alignment 20310.3 In-Context Learning for Semantic Alignment 20710.4 Conclusion and Future Directions 211Bibliography 212Part III Machine Reasoning for Ai-native Semantic Communication Networks 21511 Causal Reasoning Foundations of Semantic Communication Systems 217Christo Kurisummoottil Thomas, Christina Chaccour, Walid Saad, and Merouane Debbah11.1 Introduction 21711.2 Causality Primer 21911.3 Causal Semantic Communications 22211.3.1 System Model 22211.3.1.1 How to Pose the Proper Interventions and Counterfactuals via Queries? 22411.3.2 Emergent Language Model 22611.3.3 Semantic Information Measure 22711.3.4 Signaling Game Model and Generalized Nash Equilibrium Problem 23011.3.5 Characterization of the Generalized Local NE 23211.3.6 Analysis of the Signaling Game Equilibria for Emergent Language 23311.3.7 Average Semantic Representation Length for Classical and Emergent Language Based ESC 23511.4 Numerical Results 23611.4.1 Illustrative Example for NeSy AI’s Potential in Wireless Versus Classical AI Based Wireless 23611.5 Conclusion 239Bibliography 24012 Reinforcement Learning-Based Unicast and Broadcast Wireless Semantic Communications 241Zhilin Lu, Rongpeng Li, Ekram Hossain, Zhifeng Zhao, and Honggang Zhang12.1 Introduction 24112.2 System Model And Problem Formulation 24512.2.1 Unicast Model 24512.2.2 Broadcast Model 24612.2.3 Problem Formulation 24812.3 SemanticBC-SCAL Schemes with Alternating Learning Mechanism 25012.3.1 The Markov Decision Process (MDP) Framework 25012.3.2 Self-Critical Optimization Under Alternate Learning Mechanism 25212.4 Performance Evaluation 25812.4.1 Simulation Settings 25812.4.2 Numerical Results And Analysis 26012.4.2.1 Performance In Point-to-Point SC 26012.4.2.2 Performance In Semantic BC 26312.4.2.3 Convergence Analyses 26412.4.2.4 Ablation Study 26512.5 Conclusions 267Bibliography 26713 Imitation Learning-based Implicit Semantic-aware Communication Networks 273Yiwei Liao, Zijian Sun, Yong Xiao, Guangming Shi, Yingyu Li, H. Vincent Poor, Walid Saad, Merouane Debbah, and Mehdi Bennis13.1 Introduction 27313.1.1 Framework of Implicit Semantic Communications 27513.1.1.1 Representation of Semantics 27513.1.1.2 Explicit Semantics 27513.1.1.3 Implicit Semantics U V 27513.1.1.4 Reasoning Mechanism Π 27613.1.2 Knowledge Base 27613.1.3 Reasoning Mechanism Modeling and Learning 27713.2 System Model and Problem Formulation 27713.2.1 System Model 27713.3 iSAC Architecture 28013.3.1 Source User Side 28013.3.1.1 Semantic Encoding 28013.3.1.2 Semantic Distance 28313.3.2 Destination User Side 28413.3.2.1 Semantic Interpreter 28413.3.3 Algorithm and Theoretical Analysis 28413.4 Extension to Collaborative Reasoning 28613.4.1 Collaborative Reasoning Computing Network 28613.4.2 Algorithm for Collaborative Reasoning 28813.5 Conclusion 288Bibliography 28914 Semantic and Goal-Oriented Communication: A Data Valuation Perspective 291Shashi Raj Pandey, Van Phuc Bui, and Petar Popovski14.1 Introduction 29114.2 Data Valuation Principles 29214.2.1 Semantic Approaches In Satellite Communications For Earth Observation 29314.2.2 Goal-Oriented Problems in FL 29314.3 Semantic Communication For Earth Observation with LEO Satellites 29414.3.1 Energy Model 29614.3.2 Energy Efficient Data Downloading With Change Detection Constraint 29614.3.3 Preprocessing and Semantic Encoding 29714.3.3.1 Preprocessing 29714.3.3.2 Cloud Removing 29714.3.3.3 Change Scoring and Semantic Encoding 29814.3.4 Numerical Results 29814.4 Goal-Oriented Communications In FL 30014.4.1 Contribution-Based Aggregation 30114.4.2 Contribution-Based Participation 30114.4.3 Performance Evaluations 30214.5 Conclusion 303Bibliography 304Part IV Security of Semantic Networks 30715 Securing Semantic Communications Against Adversarial Attacks 309Yalin Evren Sagduyu, Aylin Yener, and Sennur Ulukus15.1 Introduction 30915.2 Semantic Communications 31115.3 Multitask Learning For Semantic Communications 31315.4 Adversarial Attacks 31815.4.1 Untargeted Adversarial Attack 31915.4.2 Targeted Adversarial Attack 32015.4.3 Fast Gradient Sign Method (FGSM) Attack 32015.4.4 Projected Gradient Descent (PGD) Attack 32115.4.5 Basic Iterative Method (BIM) Attack 32215.4.6 Momentum Iterative Method (MIM) Attack 32215.4.7 DeepFool Attack 32315.5 Adversarial Attacks On Semantic Communications 32415.6 Defense Against Adversarial Attacks 33015.7 Adversarial Training as Defense Against Adversarial Attacks on Semantic Communications 33115.8 Future Research Directions 33415.9 Conclusion 335Bibliography 33516 Encrypted Semantic Communications for Privacy Preserving 339Zhiyong Chen, Meixia Tao, Zhongwei Wang, and Xinlai Luo16.1 Introduction 33916.2 Basics Of Semantic Communication Systems 34016.3 Security Issues Of Semantic Communication 34316.3.1 Security Risk Of Effective Transmission 34316.3.2 Security Risk Of Privacy Protection 34516.4 Encrypted Semantic Communications 34616.4.1 Overall System Architecture 34616.4.2 Physical-Layer Encryptor And Decryptor Structure In EnSC 34816.4.3 Semantic Encoder And Decoder Structure In EnSC 34916.5 Adversarial Encryption Training 34916.5.1 Loss Functions 34916.5.2 Training 35116.5.3 Performance Evaluation 35216.6 Conclusion 356Bibliography 356Appendix A 361Index 367