Foundations of Semantic Communication Networks E-Book
116,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Wissenschaft und neue Technologien
- Sprache: Englisch
Comprehensive overview of the principles, theories, and techniques needed to build end-to-end semantic communication systems, with case studies included.
In this rapidly evolving landscape, the integration of connected intelligence applications highlights the pressing need for networks to gain intelligence in a non-siloed and ad hoc manner. The traditional incremental approach to network design is no longer sufficient to support the diverse and dynamic requirements of these emerging applications. This necessitates a paradigm shift towards more intelligent and adaptive network architectures.
From theory to application, Foundations of Semantic Communication Networks describes and provides a comprehensive understanding of everything needed to build end-to-end semantic communication systems. This book covers various interdisciplinary topics such as the mathematical foundations of semantic communications, information theoretical perspectives, joint-source channel coding, semantic-aware resource management strategies, interoperability under heterogeneous semantic communication users, advanced artificial intelligence (AI) and machine reasoning techniques for enabling connected intelligent applications, secure and privacy-preserving semantic communication systems, and the coexistence and interoperability of semantic, goal-oriented, and legacy systems.
The book examines unique features of end-to-end networking with semantic communications, including instilling reasoning behaviors in communication nodes, the role of the semantic plane in information filtering, control of communication and computing resources, transmit and receive signaling schemes, and connected intelligence device control. It emphasizes the importance of data semantics and age of information metrics. The book also discusses the profound impact of semantic communications on the telecom industry, highlighting changes in network performance, resource management, traffic, as well as spectral and energy efficiency.
Furthermore, the book provides insights into the mathematical constructs and AI theories for formulating semantic information, such as topology and category theory. It explores real-world applications, case studies, and future research directions as wireless technologies transition to 6G and beyond.
Written by four recognized experts in the field with a wealth of expertise from academia, industry, and research institutions, Foundations of Semantic Communication Networks addresses sample topics, including:
- Novel Semantic Information Formulations: Proposing new formulations using rigorous mathematical frameworks such as category theory and algebraic topology.
- Practical Applications and Networking Features: Focusing on real-world scenarios, addressing multiple access and networking challenges through collaborative frameworks for multi-modal transmissions, examining multiple access schemes to enhance transmission efficiency, and ensuring coexistence with legacy systems.
- AI-Native Air Interface and Semantic-Aware Resource Allocation: Enabling efficient large-scale systems for 6G and beyond wireless systems through AI-native air interfaces and semantic-aware resource allocation strategies.
- Advanced AI and Machine Reasoning: Utilizing causality and neuro-symbolic artificial intelligence for minimalistic transmissions, and achieving generalizability and transferability across contexts and data distributions to develop high-fidelity semantic communication systems.
- Multi-Domain Security Vulnerabilities: Examining security vulnerabilities associated with deep neural networks in semantic communications, and proposing encrypted, privacy-preserving semantic communication systems (ESCS) as a solution.
Foundations of Semantic Communication Networks is an excellent forward-thinking resource on the subject for readers with a strong background in the subject matter, including graduate-level students, academics, practitioners, and industry researchers.
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 596
Veröffentlichungsjahr: 2024
Ähnliche
Table of Contents
Cover
Table of Contents
Title Page
Copyright
Dedication
About the Editors
List of Contributors
Preface
Acknowledgment
Acronyms
1 Introduction to Semantic Communications
1.1 From Information Streams to Streams of Understanding: The Rise of Semantic Communication Networks
1.2 Reimagining Future G Applications with Semantic Communications
1.3 Structure and Path of the Book
Bibliography
Part I: Fundamentals of Semantic Communications
2 Semantic Compression and Communication: Fundamentals and Methodologies
2.1 Introduction
2.2 Semantic Index Assignment
2.3 The Rise of Machine Intelligence in Perception
2.4 Semantic Compression for Multimodal Sources
2.5 Conclusion
Bibliography
3 Toward a Theory of Semantic Information
3.1 Introduction
3.2 Cohomological Nature of Information
3.3 Axioms for Information Spaces
3.4 Comparison with Other Propositions of Semantic Information Measures
3.5 Carnap and Bar-Hillel Languages
3.6 Shepard’s Experiment
Bibliography
4 Deep Joint Source and Channel Coding
4.1 Introduction
4.2 DeepJSCC for MIMO Channels
4.3 DeepJSCC for Relay Channels
4.4 DeepJSCC for Feedback Channels
4.5 Concluding Remarks
Bibliography
Notes
5 When Information Is a Function of Data – Some Information Theoretic Perspectives on Semantic Communications
5.1 The Central Limit Theorem
5.2 Quantitative Bounds
5.3 General Polynomials
5.4 Examples and Applications
5.5 Further Generalizations
Bibliography
6 Interoperability and Coexistence of 6G Semantic, Goal-Oriented, and Legacy Systems
6.1 Introduction
6.2 Interoperability Issue in Goal-oriented and Semantic Systems
6.3 Coexistence of Semantic, Goal-Oriented, and Legacy Services in 6G
6.4 Conclusion
Acknowledgment
Bibliography
Note
Part II: Semantic Communications Networking
7 Optimization of Image Transmission in a Cooperative Semantic Communication Networks
7.1 Introduction
7.2 Representative Work
7.3 Value-Decomposition-based Entropy-Maximized Multi-Agent RL Method
7.4 Simulation Results and Analysis
7.5 Conclusion
Bibliography
8 Multiple Access Design for Joint Semantic and Classical Communications
8.1 Introduction
8.2 Heterogeneous Semantic and Bit Multiuser Network
8.3 NOMA-Enabled Heterogeneous Semantic and Bit Multiuser Communications
8.4 Semantic Communications-Enhanced NOMA
8.5 Concluding Remarks and Future Research
Bibliography
9 Contextual Reasoning-based Semantics-Native Communication
9.1 Semantics-Native Communication
9.2 Contextual Reasoning for Semantics-Native Communication
9.3 Context Synchronization for Semantics-Native Communication
9.4 Information Bottleneck Contextual Reasoning
9.5 Conclusion
Bibliography
10 Interoperable Semantic Communication
10.1 Pitfalls of Federated Learning for Semantic Alignment
10.2 Split Learning for Semantic Alignment
10.3 In-Context Learning for Semantic Alignment
10.4 Conclusion and Future Directions
Bibliography
Part III: Machine Reasoning for AI-Native Semantic Communication Networks
11 Causal Reasoning Foundations of Semantic Communication Systems
11.1 Introduction
11.2 Causality Primer
11.3 Causal Semantic Communications
11.4 Numerical Results
11.5 Conclusion
Bibliography
12 Reinforcement Learning-Based Unicast and Broadcast Wireless Semantic Communications
12.1 Introduction
12.2 System Model And Problem Formulation
12.3 SemanticBC-SCAL Schemes with Alternating Learning Mechanism
12.4 Performance Evaluation
12.5 Conclusions
Bibliography
Notes
13 Imitation Learning-based Implicit Semantic-aware Communication Networks
13.1 Introduction
13.2 System Model and Problem Formulation
13.3 iSAC Architecture
13.4 Extension to Collaborative Reasoning
13.5 Conclusion
Bibliography
Note
14 Semantic and Goal-Oriented Communication: A Data Valuation Perspective
14.1 Introduction
14.2 Data Valuation Principles
14.3 Semantic Communication For Earth Observation with LEO Satellites
14.4 Goal-Oriented Communications In FL
14.5 Conclusion
Bibliography
Part IV: Security of Semantic Networks
15 Securing Semantic Communications Against Adversarial Attacks
15.1 Introduction
15.2 Semantic Communications
15.3 Multitask Learning For Semantic Communications
15.4 Adversarial Attacks
15.5 Adversarial Attacks On Semantic Communications
15.6 Defense Against Adversarial Attacks
15.7 Adversarial Training as Defense Against Adversarial Attacks on Semantic Communications
15.8 Future Research Directions
15.9 Conclusion
Bibliography
16 Encrypted Semantic Communications for Privacy Preserving
16.1 Introduction
16.2 Basics Of Semantic Communication Systems
16.3 Security Issues Of Semantic Communication
16.4 Encrypted Semantic Communications
16.5 Adversarial Encryption Training
16.6 Conclusion
Bibliography
Appendix A
A.1 Proof of Lemma 11.3
A.2 Proof of Theorem 11.1
A.3 Proof of Proposition 11.5
A.4 Proof of Theorem 11.3
A.5 Proof of Theorem 11.2
Bibliography
Index
End User License Agreement
List of Tables
Chapter 2
Table 2.1 Optimal codeword assignment for semantic distortion problem.
Table 2.2 Training parameters used for semantic VQ AE and classifier NNs.
Table 2.3 A message
s
randomly selected from the AG’s News test set and the ...
Table 2.4 Source compression results presented for 2000 text and 2000 image ...
Chapter 4
Table 4.1 Comparisons of DeepJSCC-MIMO-universal and BPG-LDPC schemes with s...
Chapter 7
Table 7.1 System parameters.
Chapter 12
Table 12.1 Network structure and hyperparameters used in SemanitcBC-SCAL.
Table 12.2 The default parameters in SemanitcBC-SCAL.
Table 12.3 Comparisons on the varying number of RXs of SemanticBC-SCAL under...
Chapter 15
Table 15.1 Encoder–decoder architectures for semantic communications.
List of Illustrations
Chapter 1
Figure 1.1 Hypothesis for Taming traffic: Telecom brain and AI lock horns.
Figure 1.2 The evolution of wireless networks from data-driven ones to reaso...
Figure 1.3 What is (is not) semantic communications [Chaccour et al., 2024]?...
Chapter 2
Figure 2.1 Characteristic graph with vertices and edges constructed with
Figure 2.2 The model of introduced semantic quantization.
Figure 2.3 The model of introduced semantic compression.
Figure 2.4 The model of introduced semantic VQ AE.
Figure 2.5 Projecting SBERT embeddings of text messages into and visualizi...
Figure 2.6 System time efficiency results over AWGN and Rayleigh Fading chan...
Chapter 3
Figure 3.1 One representative (atom) for each of the four orbits (types) of
Figure 3.2 Groupoid representing the orbit of type .
Figure 3.3 Changes of frames for the space of information , for of type
Figure 3.4 Types in Shepard’s experiment.
Chapter 4
Figure 4.1 Block diagram of the MIMO image transmission system: (a) conventi...
Figure 4.2 The pipeline of the DeepJSCC-MIMO scheme, where the source image
Figure 4.3 Performance comparisons between DeepJSCC-MIMO and BPG-Capacity ov...
Figure 4.4 Performance of DeepJSCC-MIMO compared with the BPG-Capacity bench...
Figure 4.5 (a) Performance of DeepJSCC-MIMO and conventional separation-base...
Figure 4.6 Performance of DeepJSCC-MIMO with different antennas number.
Figure 4.7 The half-duplex relay channel.
Figure 4.8 The processing of DeepJSCC-AF, -DF and -PF at the relay and des...
Figure 4.9 Comparison between DeepJSCC-AF and DeepJSCC-DF with dB. We also...
Figure 4.10 The PSNR and SSIM performance of the DeepJSCC-PF scheme compared...
Figure 4.11 Basic structure of the multihop communication network where the ...
Figure 4.12 The hybrid JSCC framework, where denotes the DeepJSCC encoder,...
Figure 4.13 The compression and decompression modules, and , respectively...
Figure 4.14 (a) The comparison of PSNR performance between the hybrid JSCC f...
Figure 4.15 Alternative JSCC schemes for channels with feedback, where the s...
Figure 4.16 The pipeline of our JSCCformer-f scheme. In the th transmission...
Figure 4.17 Performance of different schemes at various SNR values and bandw...
Figure 4.18 (a) Performance comparison of different models versus bandwidth ...
Figure 4.19 (a) Performance of JSCCformer-f over noisy feedback link in AWGN...
Figure 4.20 Visual comparisons of images transmitted by DeepJSCC-f, JSCCform...
Figure 4.21 Visualization of the reconstructed image with JSCCformer-f after...
Figure 4.22 Target PSNR analysis of different schemes for variable rate tran...
Chapter 5
Figure 5.1 Evaluation of average replacement error, obtained empirically and...
Figure 5.2 The computational versus the classic wiretap channel. Under certa...
Chapter 6
Figure 6.1 Coexistence and interoperability between heterogeneous agents, sh...
Figure 6.2 Latent space representations of the controller and the scout lang...
Figure 6.3 Performance comparison of different equalization techniques.
Figure 6.4 Latent space representations of the sender and the receiver langu...
Figure 6.5 Performance comparison of different equalization techniques.
Figure 6.6 Trade-off between legacy user performance and goal-effectiveness....
Chapter 7
Figure 7.1 The image semantic transmission framework of each server.
Figure 7.2 The cooperative multi-server image semantic communication wireles...
Figure 7.3 An example of semantic information extraction. (a) The extracted ...
Figure 7.4 The training process of the introduced VD-ERL algorithm.
Figure 7.5 The semantic scores distribution of semantic triples.
Figure 7.6 Correlation between transmitted semantic information and RB alloc...
Chapter 8
Figure 8.1 Conventional bit-based communications and new semantic communicat...
Figure 8.2 The considered heterogeneous semantic and bit multiuser network e...
Figure 8.3 Semi-NOMA for heterogeneous semantic and bit communications.
Figure 8.4 The achieved semantic-versus-bit rate region achieved by differen...
Figure 8.5 The proposed opportunistic semantic and bit communication approac...
Figure 8.6 The ergodic (equivalent) semantic rate achieved by the secondary ...
Chapter 9
Figure 9.1 A schematic illustration of semantics-native communication system...
Figure 9.2 A schematic illustration of Ogden and Richards’ triangle of meani...
Figure 9.3 A schematic illustration of semantic encoder/decoder triangles.
Figure 9.4 A schematic illustration of a contextual reasoning example in a r...
Figure 9.5 Reliability with respect to parameters and ranging to i...
Figure 9.6 Reliability versus communication rounds in SNC with different p...
Figure 9.7 A schematic illustration of context synchronization and SNC betwe...
Chapter 10
Figure 10.1 A schematic illustration of the federated codebook.
Figure 10.2 A brief illustration of the Federated Codebook: (a) Original, (b...
Figure 10.3 Reconstruction loss over the course of training with and without...
Figure 10.4 Reconstructed image results of encoder–decoder pairs trained in ...
Figure 10.5 Reconstructed image results of encoder–decoder pairs trained in ...
Figure 10.6 SLF result of MSE for reconstruction under semantic misalignment...
Figure 10.7 Graph of the results of applying SLF when a semantic misalignmen...
Figure 10.8 Semantic Knowledge Distillation (SKD) in language-oriented seman...
Figure 10.9 Semantic Knowledge Distillation (SKD) performance in various cha...
Figure 10.10 Samples of original prompts from Alice and customized prompts t...
Chapter 11
Figure 11.1 Hierarchical levels of causality.
Figure 11.2 Illustrative example showcasing the concept of semantic language...
Figure 11.3 Proposed ESC system time split between emergent language and dat...
Figure 11.4 Illustration of emergent language.
Figure 11.5 Example on how logical reasoning (task with two objects present)...
Figure 11.6 (a) The testbed for the illustrative example is the state descri...
Chapter 12
Figure 12.1 System model and comparisons of varied communication schemes [Lu...
Figure 12.2 The system model for semantic BC system.
Figure 12.3 Illustration of the self-critic optimization for decoder side at...
Figure 12.4 Optimization process under the alternate learning mechanism. (a)...
Figure 12.5 BLEU score and BERT-SIM versus SNR in AWGN channel for the point...
Figure 12.6 BLEU score and BERT-SIM versus SNR in Rayleigh fading channel fo...
Figure 12.7 An example on image-transmission extension of SemanticRL [Lu et ...
Figure 12.8 Comparison of semantic BC schemes in terms of BLEU-1 score, BLEU...
Figure 12.9 Convergence of SemanticBC-SCAL, wherein the reward curves are th...
Figure 12.10 Convergence performance of SemanticRL. From (a) to (b), we prov...
Figure 12.11 The impact of the number of parallel samples in the SemanticB...
Chapter 13
Figure 13.1 The proposed iSAC architecture: (a) A systematic review of the p...
Figure 13.2 A graphical representation of the semantic meaning of 64 concept...
Figure 13.3 Semantic constellation of 64 symbols in two selected dimensions ...
Figure 13.4 Semantic interpreter (communication phrase).
Figure 13.5 G-RML algorithm (training phrase).
Figure 13.6 Multi-user computing network.
Figure 13.7 G-RML multi-user algorithm.
Chapter 14
Figure 14.1 Semantic and goal-oriented transmission with strategic pull/push...
Figure 14.2 Proposed end-to-end scoring architecture for strategic push-base...
Figure 14.3 (a) ROC curve over change dataset (the AUC scores are shown in t...
Figure 14.4 Impact of: (a) on training performance, (b) risk factor () on...
Chapter 15
Figure 15.1 (a) Conventional communications (bottom) versus (b) semantic com...
Figure 15.2 Semantic task accuracy as a function of the SNR.
Figure 15.3 MSE of reconstruction as a function of the SNR.
Figure 15.4 Semantic task accuracy as a function of encoder output size.
Figure 15.5 MSE of reconstruction as a function of the encoder output size....
Figure 15.6 Attack success probability for untargeted FGSM attack versus Gau...
Figure 15.7 MSE for untargeted FGSM attack versus Gaussian perturbation unde...
Figure 15.8 Attack success probability for different untargeted attack metho...
Figure 15.9 Attack success probability for different untargeted attack metho...
Figure 15.10 Attack success probability for untargeted PGD attack as a funct...
Figure 15.11 Attack success probability for targeted FGSM attack versus Gaus...
Figure 15.12 MSE for targeted FGSM attack versus Gaussian perturbation under...
Figure 15.13 Attack success probability for different targeted attack method...
Figure 15.14 Attack success probability for different targeted attack method...
Figure 15.15 Classifier accuracy under the AWGN channel as a function of the...
Figure 15.16 Classifier accuracy under the Rayleigh channel as a function of...
Chapter 16
Figure 16.1 A typical diagram of learning-based semantic communication syste...
Figure 16.2 Semantic communication system under malicious attacks.
Figure 16.3 The structure of encrypted semantic communication system.
Figure 16.4 Physical layer encryptor and decryptor network structure.
Figure 16.5 The values of under adversarial encryption training and nonadv...
Figure 16.6 The values of under adversarial encryption training and nonadv...
Figure 16.7 The values of under adversarial encryption training and nonadv...
Figure 16.8 BLEU score versus SNR for Bob and Eve in EnSC with different tra...
Figure 16.9 From left to right, the original confidential image, the encrypt...
Guide
Cover
Table of Contents
Title Page
Copyright
Dedication
About the Editors
List of Contributors
Preface
Acknowledgment
Acronyms
Begin Reading
Appendix A
Index
End User License Agreement
Pages
ii
iii
iv
v
xvii
xviii
xix
xxi
xxii
xxiii
xxiv
xxv
xxvi
xxvii
xxviii
xxix
xxxi
xxxii
xxxiii
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
49
50
51
52
53
54
55
56
57
58
59
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
IEEE Press445 Hoes LanePiscataway, NJ 08854
IEEE Press Editorial BoardSarah Spurgeon, Editor in Chief
Moeness Amin
Jón Atli Benediktsson
Adam Drobot
James Duncan
Ekram Hossain
Brian Johnson
Hai Li
James Lyke
Joydeep Mitra
Desineni Subbaram Naidu
Tony Q. S. Quek
Behzad Razavi
Thomas Robertazzi
Diomidis Spinellis
Foundations of Semantic Communication Networks
Edited by
Walid Saad
Virginia Tech, Arlington, VA, USA
Christina Chaccour
Ericsson, Inc., Plano, TX, USA
Christo Kurisummoottil Thomas
Virginia Tech, Arlington, VA, USA
Merouane Debbah
Khalifa University of Science and Technology
Abu Dhabi, United Arab Emirates
Copyright © 2025 by The Institute of Electrical and Electronics Engineers, Inc. All rights reserved.
Published by John Wiley & Sons, Inc., Hoboken, New Jersey.Published simultaneously in Canada.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission.
Trademarks: Wiley and the Wiley logo are trademarks or registered trademarks of John Wiley & Sons, Inc. and/or its affiliates in the United States and other countries and may not be used without written permission. All other trademarks are the property of their respective owners. John Wiley & Sons, Inc. is not associated with any product or vendor mentioned in this book.
Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.
For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002.
Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com.
Library of Congress Cataloging-in-Publication Data Applied for:
Hardback ISBN: 9781394247882
Cover Design: WileyCover Image: © Tuomas A. Lehtinen/Getty Images
“To those who believe that semantics matter,”
Walid Saad
“To those who believe ‘less is more’, finding substance, depth, and beyond,”
Christina Chaccour
“To my parents,”
Christo Kurisummoottil Thomas
“To my wife Radja and my three kids Lyne, Ines, and Noor,”
Merouane Debbah
About the Editors
Walid Saad received his PhD from the University of Oslo, Norway, in 2010. He is currently a Professor at the Department of Electrical and Computer Engineering at Virginia Tech, where he leads the Network intelligEnce, Wireless, and Security (NEWS) laboratory. His research interests include wireless networks (5G/6G/beyond), machine learning, game theory, quantum communications/learning, security, UAVs, semantic communications, cyber-physical systems, and network science. Dr. Saad is a Fellow of the IEEE. He is also the recipient of the NSF CAREER award in 2013, the AFOSR summer faculty fellowship in 2014, and the Young Investigator Award from the Office of Naval Research (ONR) in 2015. He was the (co-)author of twelve conference best paper awards at IEEE WiOpt in 2009, ICIMP in 2010, IEEE WCNC in 2012, IEEE PIMRC in 2015, IEEE SmartGridComm in 2015, EuCNC in 2017, IEEE GLOBECOM (2018 and 2020), IFIP NTMS in 2019, IEEE ICC (2020 and 2022), and IEEE QCE in 2023. He is the recipient of the 2015 and 2022 Fred W. Ellersick Prize from the IEEE Communications Society, of the IEEE Communications Society Marconi Prize Award in 2023, and of the IEEE Communications Society Award for Advances in Communication in 2023. He was also a co-author of the papers that received the IEEE Communications Society Young Author Best Paper award in 2019, 2021, and 2023. Other recognitions include the 2017 IEEE ComSoc Best Young Professional in Academia award, the 2018 IEEE ComSoc Radio Communications Committee Early Achievement Award, and the 2019 IEEE ComSoc Communication Theory Technical Committee Early Achievement Award. From 2015 to 2017, Dr. Saad was named the Stephen O. Lane Junior Faculty Fellow at Virginia Tech and in 2017, he was named College of Engineering Faculty Fellow. He received the Dean’s award for Research Excellence from Virginia Tech in 2019. He was also an IEEE Distinguished Lecturer in 2019-2020. He has been annually listed in the Clarivate Web of Science Highly Cited Researcher List since 2019. He currently serves as an Area Editor for the IEEE Transactions on Communications. He is the Editor-in-Chief for the IEEE Transactions on Machine Learning in Communications and Networking.
Christina Chaccour, a distinguished research scientist, engineer, and entrepreneur, currently serves as a Network Solutions Manager at Ericsson Inc. In her role, Dr. Chaccour seamlessly bridges strategy, product solutions, and research, spearheading developments in 5G Advanced, 6G networks, and AI-integrated solutions while ensuring the responsible and innovative use of AI, addressing both policy and technical dimensions. Dr. Chaccour actively represents Ericsson as a delegate in prominent industry bodies and councils including FCC CSRIC IX, 5G Americas, and Next G Alliance. With a PhD in Electrical Engineering from Virginia Tech, her academic journey yielded significant contributions, particularly in 6G systems at THz frequencies for next-gen XR and holographic systems, her work also laid the groundwork for pioneering research in AI-native networks and semantic communications. Her dissertation contributions and their subsequent impact were recognized with the Bill and LaRue Blackwell Award.
Dr. Chaccour has been recognized with several honors, including the Best Paper Award at the 10th IFIP Conference on New Technologies, Mobility, and Security (NTMS) in 2019 and the Exemplary Reviewer Award from IEEE Transactions on Communications in 2021, a distinction given to fewer than 2% of reviewers. Her paper in IEEE Communication Surveys and Tutorials was featured in the Top Access article listing from June to November 2022. In 2024, she was named among the “Top 100 Brilliant and Inspiring Women in 6G.”
Dr. Chaccour serves on the editorial board of IEEE Transactions on Machine Learning in Communications and Networking, IEEE Transactions on Cognitive Communications and Networking, Wireless Personal Communications (Springer), and the guest editorial board of IEEE Communications Standards Magazine within the series “AI for Wireless.” She is also on the advisory board of “The Data Science Conference.”
In her entrepreneurial endeavors, Dr. Chaccour co-founded the startup “Internet of Trees,” which has garnered numerous local and international awards.
Christo Kurisummoottil Thomas received his BS in Electronics and Communication Engineering from National Institute of Technology, Calicut, India, in 010, MS in Telecommunication Engineering from Indian Institute of Science, Bangalore, India, in 2012, and PhD from EURECOM, France, in 2020. He is currently a postdoctoral fellow at the Electrical and Computer Engineering Department at Virginia Tech. His research interests include semantic communications, statistical signal processing, and advanced machine learning using causal inference and neurosymbolic AI for wireless communications. From 2012 to 2014, he was a staff design engineer on 4G LTE with Broadcom communications, Bangalore, and from 2014 to 2017, he was a design engineer with Intel corporation, Bangalore. During November 2020 till June 2022, he was a staff engineer on 5G modems with wireless research and development division of Qualcomm Inc., Espoo, Finland. He was a recipient of the best student paper award at IEEE SPAWC 2018, Kalamata, Greece, and also received third prize for his team titled “Learned Chester” ML5G-PHY channel estimation challenge, as part of the ITU AI/ML in 5G challenge, conducted at NCSU, US, 2020. He has delivered multiple tutorials on variational Bayesian inference at IEEE conferences such as ICASSP and EUSIPCO.
Merouane Debbah is a professor at Khalifa University of Science and Technology in Abu Dhabi and founding Director of the KU 6G Research Center. He is a frequent keynote speaker at international events in the field of telecommunication and AI. His research has been lying at the interface of fundamental mathematics, algorithms, statistics, information, and communication sciences with a special focus on random matrix theory and learning algorithms. In the communication field, he has been at the heart of the development of small cells (4G, massive MIMO (5G), and large intelligent surfaces (6G) technologies. In the AI field, he is known for his work on large language models, distributed AI systems for networks and semantic communications. He received multiple prestigious distinctions, prizes, and best paper awards (more than 40 IEEE best paper awards) for his contributions to both fields. He is an IEEE Fellow, a WWRF Fellow, a Eurasip Fellow, an AAIA Fellow, an Institut Louis Bachelier Fellow, an AIIA Fellow, and a Membre émérite SEE.
List of Contributors
Jean-Claude Belfiore
Advanced Wireless Technology Lab.
Huawei Technologies
Boulogne Billancourt
France
Daniel Bennequin
Advanced Wireless Technology Lab.
Huawei Technologies
Boulogne Billancourt
France
and
UFR de Mathématiques
Universite Paris Cite
Paris
France
Mehdi Bennis
Centre for Wireless Communications
University of Oulu
Oulu
Finland
Chenghong Bian
Department of Electrical and Electronic Engineering
Imperial College London
London
UK
Van Phuc Bui
Department of Electronic Systems
Aalborg University
Aalborg Øst
Denmark
Christina Chaccour
Ericsson, Inc.
Plano, TX
USA
Mingzhe Chen
Department of Electrical and Computer Engineering
University of Miami
Coral Gables, FL
USA
Zhiyong Chen
Cooperative Medianet Innovation Center
Shanghai Jiao Tong University
Shanghai
China
Jinho Choi
School of Electrical and Mechanical Engineering
the University of Adelaide
North Terrace, Adelaide, SA
Australia
Jinhyuk Choi
School of Electrical & Electronic Engineering
Yonsei University
Seodaemun-Gu, Seoul
Korea
Wan Choi
Department of Electrical and Computer Engineering
Seoul National University
Seoul
Republic of Korea
Merouane Debbah
6G Center
Khalifa University of Science and Technology
Abu Dhabi
United Arab Emirates
Homa Esfahanizadeh
Radio Systems Research Group
Nokia Bell Labs
Murray Hill, NJ
USA
Rafael Gregorio Lucas D’Oliveira
School of Mathematical and Statistical Sciences
Clemson University
Clemson, SC
USA
Deniz Gunduz
Department of Electrical and Electronic Engineering
Imperial College London
London
UK
Ekram Hossain
Department of Electrical and Computer Engineering
University of Manitoba
Winnipeg, Manitoba
Canada
Ye Hu
Department of Industrial and Systems Engineering
University of Miami
Coral Gables, FL
USA
Yoon Huh
Department of Electrical and Computer Engineering
Seoul National University
Seoul
Republic of Korea
Tomás Huttebraucker
CEA-Leti
Université Grenoble Alpes
Grenoble
France
Seong-Lyun Kim
School of Electrical & Electronic Engineering
Yonsei University
Seodaemun-Gu, Seoul
Korea
Seung-Woo Ko
Department of Smart Mobility Engineering
Inha University
Yeonsu-Gu, Incheon
Korea
Emrecan Kutay
Department of Electrical and Computer Engineering
INSPIRE@OhioState Research Center
The Ohio State University
Columbus, OH
USA
Rongpeng Li
College of Information Science and Electronic Engineering
Zhejiang University
Hangzhou
China
Yingyu Li
School of Mechanical Engineering and Electronic Information
China University of Geosciences
Wuhan, Hubei
China
Yiwei Liao
School of Electronic Information and Communications
Huazhong University of Science and Technology
Wuhan, Hubei
China
Yuanwei Liu
Department of Electrical and Electronic Engineering
The University of Hong Kong
Hong Kong, Island
Zhilin Lu
College of Information Science and Electronic Engineering
Zhejiang University
Hangzhou
China
Xinlai Luo
Cooperative Medianet Innovation Center
Shanghai Jiao Tong University
Shanghai
China
Alexander Mariona
Research Laboratory of Electronics
Massachusetts Institute of Technology
Cambridge, MA
USA
Muriel Médard
Research Laboratory of Electronics
Massachusetts Institute of Technology
Cambridge, MA
USA
Mattia Merluzzi
CEA-Leti
Université Grenoble Alpes
Grenoble
France
Xidong Mu
School of Electronic Engineering and Computer Science
Queen Mary University of London
London
UK
Hyelin Nam
School of Electrical & Electronic Engineering
Yonsei University
Seodaemun-Gu, Seoul
Korea
Shashi Raj Pandey
Department of Electronic Systems
Aalborg University
Aalborg Øst
Denmark
Jihong Park
Information Systems Technology and Design Pillar
Singapore University of Technology and Design
Singapore
H. Vincent Poor
Department of Electrical and Computer Engineering
Princeton University
Princeton, NJ
USA
Petar Popovski
Department of Electronic Systems
Aalborg University
Aalborg Øst
Denmark
Walid Saad
Department of Electrical and Computer Engineering
Virginia Tech
Arlington, VA
USA
Yalin Evren Sagduyu
Nexcepta
Gaithersburg, MD
USA
Mohamed Sana
CEA-Leti
Université Grenoble Alpes
Grenoble
France
Hyowoon Seo
Department of Electronics and Communications Engineering
Kwangwoon University
Seoul
Republic of Korea
Yulin Shao
State Key Laboratory of Internet of Things for Smart City and Department of Electrical and Computer Engineering
University of Macau
Macau
China
Guangming Shi
School of Electronic Engineering
Xi’an University of Electronic Science and Technology
Xi’an, Shaanxi
China
Heekang Song
School of Electrical Engineering
Korea Advanced Institute of Science and Technology
Daejeon
Republic of Korea
Emilio Calvanese Strinati
CEA-Leti
Université Grenoble Alpes
Grenoble
France
Zijian Sun
School of Electronic Information and Communications
Huazhong University of Science and Technology
Wuhan, Hubei
China
Meixia Tao
Department of Electronic Engineering and the Cooperative Medianet Innovation Center
Shanghai Jiao Tong University
Shanghai
China
Christo Kurisummoottil Thomas
Virginia Tech Research Center
Virginia Tech
Arlington, VA
USA
Sennur Ulukus
Department of Electrical and Computer Engineering
University of Maryland
College Park, MD
USA
Zhongwei Wang
Cooperative Medianet Innovation Center
Shanghai Jiao Tong University
Shanghai
China
Haotian Wu
Department of Electrical and Electronic Engineering
Imperial College London
London
UK
Yong Xiao
School of Electronic Information and Communications
Huazhong University of Science and Technology
Wuhan, Hubei
China
Aylin Yener
Department of Electrical and Computer Engineering
INSPIRE@OhioState Research Center
The Ohio State University
Columbus, OH
USA
Honggang Zhang
College of Information Science and Electronic Engineering
Zhejiang University
Hangzhou
China
and
Zhejiang Lab
Hangzhou
China
Zhifeng Zhao
College of Information Science and Electronic Engineering
Zhejiang University
Hangzhou
China
and
Zhejiang Lab
Hangzhou
China
Preface
The wireless industry is on the brink of a revolution with the emergence of connected intelligence applications such as the metaverse, connected autonomy, holographic societies, haptics, telehealth, smart manufacturing, and smart cities. This revolution has brought forth many concerted efforts that focus on various design challenges of wireless networks, in general, and 6G systems and beyond, in particular. However, those efforts remain focused primarily on incremental, but important advances to conventional communication system technologies such as massive MIMO, reconfigurable intelligent surfaces, network slicing, and the likes.
In this rapidly evolving landscape, the integration of connected intelligence applications highlights the pressing need for networks to gain intelligence in a non-siloed and ad hoc manner. The traditional incremental approach to network design is no longer sufficient to support the diverse and dynamic requirements of these emerging applications. This necessitates a paradigm shift toward more intelligent and adaptive network architectures.
The genesis of this book stems from our recognition of semantic communication as an emerging cornerstone in the telecommunication industry. We believe that exploring this concept through diverse viewpoints – encompassing information theoretic, physical layer, networking, and security perspectives – is paramount in unraveling its complexities and harnessing its potential. While our exploration may traverse differing schools of thought and encounter opposing views on granular concepts, we view this diversity as essential in shaping a comprehensive and consensus-driven understanding of semantic communication systems.
This book represents our endeavor to establish a staple resource in the realm of semantic communication. We envision it as a guiding beacon for researchers, practitioners, and enthusiasts alike, navigating the intricate pathways of this burgeoning field.
To achieve our objectives, this book is structured into four parts, each delving into fundamental aspects of semantic communications and networking. The introductory chapter sets the stage, providing an overarching framework for subsequent discussions. Each part is meticulously crafted with specific goals in mind, offering readers a comprehensive exploration of the subject matter.
As editors and authors, we are immensely grateful to the contributors whose expertise and dedication have enriched this volume. We extend our heartfelt thanks to the reviewers whose insights and feedback have helped refine the content.
Ultimately, we believe that this book will serve as a cornerstone in the ongoing discourse surrounding semantic communication. It is our sincere hope that it will inspire further exploration, innovation, and collaboration in this dynamic field.
March, 2024United States of America
Sincerely,
Walid Saad
Christina Chaccour
Christo Kurisummoottil Thomas
Merouane Debbah
Acknowledgment
This effort was supported by the Office of Naval Research (ONR) under MURI Grant N00014-19-1-2621 and by the U.S. National Science Foundation under Grant CNS-2225511.
Acronyms
AI
artificial intelligence
AoI
age-of-information
AP
affinity propagation
APC
average power constraint
AR
augmented reality
AR
augmented reality
AUC
area under the curve
AWGN
additive white Gaussian noise
BER
bit error rate
BERT
bidirectional encoder representations from transformers
BIM
basic iterative method
BIP
basic invariance principle
BLER
block error rate
BLEU
bilingual evaluation understudy
BPG
better portable graphics
BPSK
binary phase-shift keying
B-RNNs
bi-directional recurrent networks
BS
base station
BSC
binary symmetric channel
CA
channel attention
CDF
cumulative distribution function
CLT
central limit theorem
CNN
convolutional neural network
CVaR
conditional value at risk
DeepJSCC
deep joint source-channel coding
DL
deep learning
DMC
discrete memory channel
E2E
end-to-end
EnSC
encrypted semantic communication
FC
fully connected
FGSM
fast GradientSign method
FL
federated learning
GAN
generative adversarial network
GenAI
generative artificial intelligence
GO
goal oriented
IB
information bottleneck
iCR
inverse contextual reasoning
IID
independent and identically distributed
IoT
Internet of Things
IoU
intersection over union
ISS
image-to-graph semantic similarity
JSCC
joint source-channel coding
KL
Kullback–Liebler
LCR
linearized contextual reasoning
LDPC
low-density parity check
LLM
large language model
LPIPS
learned perceptual image patch similarity
LSC
language-oriented SC
MAB
multi-arm bandits
MAC
multiple access channels
MCMC
Markov chain Monte Carlo
MIM
momentum iterative method
MIMO
multiple-input and multiple-output
ML
machine learning
MSE
mean squared error
NE
nash equilibrium
NLP
natural language processing
NOMA
non-orthogonal multiple access
OMA
orthogonal multiple access
O-RAN
open-RAN
PGD
projected gradient descent
PPC
peak power constraint
PSNR
peak signal-to-noise ratio
PSR
perturbation-to-signal ratio
QPSK
quadrature phase shift keying
RHS
right-hand side
RL
reinforcement learning
ROC
receiver operating characteristic
RS
reed Solomon
RX
receiver
SBERT
sentence-BERT
SC
semantic communication
SCM
structural causal model
SER
symbol error rate
SGD
stochastic gradient descent
SKD
semantic knowledge distillation
SL
split learning
SNC
semantics-native communication
SNR
signal-to-noise ratio
SSCC
separate source and channel coding scheme
SV
Shapley value
SVD
singular-value decomposition
T2I
text-to-image
T2T
text to text interpreter
TV
total variation
TX
transmitter
UAV
unmanned aerial vehicle
UCB
upper confidence bound
UE
user equipment
UMAP
uniform manifold approximation and projection for dimension Reduction
V2X
vehicle-to-everything
VD-ERL
value decomposition entropy-maximized multi-agent RL
VG
visual genome
ViT
vision transformer
VMAF
video multimethod assessment fusion
VQ AE
vector quantized autoencoder model
VQ-VAE
vector-quantized variational autoencoder
VR
virtual reality
VR
virtual reality
WSN
wireless sensor networks
XR
extended reality
1Introduction to Semantic Communications
Christina Chaccour1, Christo Kurisummoottil Thomas2, Walid Saad3, and Merouane Debbah4
1Ericsson, Inc., Plano, TX, USA
2Virginia Tech Research Center, Virginia Tech, Arlington, VA, USA
3Department of Electrical and Computer Engineering, Virginia Tech, Arlington, VA, USA
46G Center, Khalifa University of Science and Technology, Abu Dhabi, United Arab Emirates
1.1 From Information Streams to Streams of Understanding: The Rise of Semantic Communication Networks
Throughout history, networks have mirrored the evolution of our relationship with information. Initially, they facilitated the transmission of human voices, carrying the ephemeral nature of conversation across vast distances. Subsequently, they evolved to transport the richness of media, bringing the world into our living rooms. As technology progressed, networks connected a multitude of smaller devices, forming the nascent internet of things. However, with the dawn of 6G and beyond, and the promise of a hyper-connected future, the focus of our network infrastructure shifts from mere information transport to the interconnection of intelligence.
This fundamental shift necessitates a radical transformation in how we communicate. As we envision a world of holographic teleportation, the immersive metaverse, and the sophisticated automation of Industry 5.0, the limitations of traditional networks become increasingly apparent. These networks, acting as mere “bit pipes,” rely on brute force, pushing ever-increasing volumes of data through ever-widening bandwidths, simply cannot sustain the demands of this intelligent future. Complexity would spiral out of control, and the very infrastructure that underpins this progress would become its own Achilles’ heel.
To address this challenge effectively, we need to move away from makeshift fixes and fundamentally rethink the structure of our network. This means giving greater importance to the connections between network nodes. These connections shouldn’t just pass on data blindly; they should possess the capacity to comprehend the significance of the information they carry. This transformation, often termed as the “humanization” of the network, forms the core of semantic communication – a paradigm shift that transcends mere data transmission to facilitate the exchange of meaningful information.
It is worth emphasizing that semantic communication does not aim to replicate the full complexity of human interaction in transmitting data. Rather, its focus lies in conveying the essence of the data along with pertinent metadata, irrespective of specific data formats, and functions at the network’s foundational layers. This approach starkly contrasts with traditional wireless networks, where communication relies solely on raw, unprocessed bits. As we navigate the increasingly intricate landscape of future applications, our reliance on spectrum and coverage alone becomes unsustainable. Semantic communication represents a shift in networking philosophy toward a deeper understanding and intelligent handling of data traffic.
With the advancement of applications and the proliferation of consumer AI, the intelligence embedded within AI systems is evolving rapidly. Tasks as simple as a Google search now translate into dynamic generative pre-trained transformer (GPT) conversations, while a single recorded video can spawn numerous generated videos through AI algorithms. Moreover, the landscape is expanding with complex applications like emerging services such as next-generation extended reality (XR), full and others. In this context, it becomes imperative for networks to adapt their approach to data and traffic. Rather than adhering to the traditional “bit pipe” model, networks must mirror the intelligence embedded within applications. We envision a future where traffic continues to escalate, but instead of exponentially increasing to the point where new spectrum allocation becomes unfeasible, the concept of semantic communications intervenes to transform networks into truly AI-native entities (as shown in Figure 1.1). In this future, traffic is managed and controlled through semantic understanding, alleviating the relentless demand for spectrum that is becoming increasingly scarce.
However, by embracing semantic communication, we unlock a powerful new paradigm. We can leverage our computing resources as “in-memory” networks, reducing the strain on spectrum and enabling efficient handling of complex data. Additionally, the growing ubiquity of AI, both in consumer devices and across diverse applications, necessitates a network that can keep pace with the rising tide of data volume and intelligence. Semantic communication networks represent the crucial leap forward, enabling us to navigate this exciting future where machines not only share information but also share understanding.
Figure 1.1 Hypothesis for Taming traffic: Telecom brain and AI lock horns.
1.1.1 How Does It Work?
Semantic communication necessitates a fundamental reconsideration of the communication quandary, as delineated by Weaver’s three-tier framework [Weaver, 1953]. Following Shannon’s inception of information theory, Weaver delineated communication hurdles across three levels [Weaver, 1953]: (i) Level A, pertaining to the technical precision of communication symbols to be transmitted; (ii) Level B, addressing the semantic accuracy of transmitted symbols vis-à-vis intended meaning; and (iii) Level C, evaluating the impact of received meaning on overall system conduct. Traditional communication systems primarily tackled Level A challenges. Nonetheless, a judiciously crafted semantic communication system can leverage AI advancements and computational prowess to potentially transcend Weaver’s framework, incorporating a reasoning plane [Chaccour et al., 2024], achieving more with less. This entails a paradigm shift from conventional transmitter–receiver pairs to what we propose as teacher and apprentice nodes, endowed with the following capabilities:
Transitioning from a bit-driven transmitter to a knowledge-driven teacher:
The conventional transmitter model needs to evolve beyond merely serving as a conduit for bits of data. Instead, it should metamorphose into a
teacher
endowed with the ability to decipher multiple
semantic content elements
embedded within the source data. This entails the extraction of distinct meanings, or semantics, from the message. Subsequently, for each identified semantic content element, the teacher must generate a
semantic representation
possessing desirable attributes. Fundamentally, the semantic content represents the essence of the data, while the semantic representation encapsulates this essence in a minimal form – akin to how individuals meticulously select words to articulate their thoughts. Moreover, various semantic content elements may correspond to different modalities within the data. For instance, in an audio recording, the tone of voice and the spoken words constitute distinct semantic content elements. Human cognition effortlessly disentangles and comprehends these elements, a capability lacking in current communication system transmitters. Hence, there is a pressing need to reconfigure transmitters to emulate human reasoning capabilities to the best of their ability. This necessitates
reasoning
at the transmission end – the capacity enabling the transmitting agent to identify, differentiate, and efficiently represent each semantic content element within the data. This stands in stark contrast to the conventional approach in today’s networks, where transmitters treat input as a random, uncertain string of information transmitted through a bit-pipeline, devoid of semantic understanding. Furthermore, the shift from a data-driven to a reason-driven approach is showcased in
Figure 1.2
.
Shifting from a bit-driven receiver to a knowledge-driven apprentice:
Similarly, the receiver’s role should evolve from merely processing bits to embodying an apprentice endowed with
reasoning
abilities. This transformation enables the receiver to
comprehend the minimal semantic representation
utilized by the teacher, thereby mapping it back to its corresponding semantic content element. Additionally, the apprentice must harness its computational resources to accurately recreate the semantic content element derived from the transmitted semantic representation with utmost fidelity. For instance, if a holographic element is transmitted, the apprentice must possess the capability to reproduce it with identical resolution as transmitted by the teacher. Furthermore, the developed reasoning capabilities empower the apprentice to utilize both causal and associational (statistical) logic across the networking stack. These logic frameworks, derived from an evolving knowledge base, enable the apprentice to undertake diverse projections and decisions concerning the received semantic representation.
Figure 1.2 The evolution of wireless networks from data-driven ones to reasoning-driven ones [Chaccour et al., 2024].
Transitioning from a bit-pipeline to a semantic language:
In the realm of semantic communications, the fundamental unit of meaning is encapsulated within a semantic representation. These representations, when organized sequentially, form what we term a
semantic language
. Unlike natural languages, semantic languages prioritize automation by minimizing emphasis on syntax and pragmatics. Moreover, the semantic representations within a semantic communication language must adhere to three crucial properties:
Minimalism
: This property entails characterizing information structure using the fewest possible language elements (and their corresponding bits). The aim is to reduce the number of exchanged messages over time while maintaining a comprehensive understanding of the underlying data structure.
Generalizability
: A critical aspect of semantic representations is their ability to represent or comprehend underlying structures invariant to changes in distribution, domain, and context. Context, in this context, refers to the thematic umbrella under which various semantics coexist (e.g., steering actions on a tennis court for a robot). Generalizability enables a node to apply learned representations across diverse domains, distributions, and contexts, akin to how words in natural language generalize their meaning to describe various events.
Efficiency
: The apprentice’s proficiency in regenerating information with high fidelity in minimal time is crucial. This means that the generated data must match or exceed the resolution achievable by a traditional receiver. For instance, if reconstructing an audio clip, the apprentice must replicate the intended resolution precisely. It is worth noting that while efficiency can be quantified through concrete metrics (as defined in
Chapter 11
), evaluating the efficacy of the system is more intricate. Thus, for brevity, we consider “efficiency” in this tutorial to encompass both efficacy and efficiency.
1.1.2 Why Now? What Factors Contribute to Our Ongoing Reliance on Traditional Communications?
Since the inception of digital communication, Shannon’s formulation of the “reconstruction problem” has been at the core of its challenges. This problem is primarily rooted in the limitations of computing power, which consequently restricts the intelligence of AI-guided tasks within communication systems. Traditionally, the focus of communication has revolved around replicating transmitted messages at another point, relying heavily on compression, transmission, and decompression techniques. However, these techniques predominantly address stochasticity arising from the source, channel, and destination, often overlooking meaningful information about the message’s significance or context.
Within this traditional framework, various shortcomings emerge in efficiently conveying the intended meaning of transmitted messages. These shortcomings manifest in multiple issues within end-to-end (E2E) communication system design. Firstly, there’s a prevalent need for repetitive transmissions in many cases to adequately transfer application information. This is primarily because transmitters lack automation in message generation at the receiver and fail to leverage memory or historical patterns in the data. Secondly, the asymmetrical nature of communication, characterized by passive receivers and active transmitters, renders receivers vulnerable to channel conditions and hardware errors.
Introducing symmetry into communication, where receivers can learn message structures and leverage historical context, holds the potential to enhance robustness against irregularities. Semantic communication, emerging as a transformative approach, facilitates radio nodes in extracting meaning from datastreams, thereby fostering shared understanding and efficient communication, particularly in scenarios with unreliable links [Lan et al., 2021].
Addressing intermittent behavior through receiver generative capabilities, semantic communication minimizes back-and-forth communication for a reliable link. By delving into low–level data analysis, including causal roots of events, semantic systems empower communication networks to efficiently achieve specific goals.
The transformation of current systems into reasoning-driven semantic communication platforms presents a promising avenue for enhancing efficiency and intelligence in wireless networks. This paradigm shift, leveraging computing resources and emphasizing languages, reasoning, and causality, offers significant potential to achieve more with less. Subsequent sections will delve deeper into exploring and understanding this transformative journey.
1.1.3 What Is NOT Semantic Communications?
Semantic communications might appear as a mere evolution of existing approaches and methods. However, in this section, we aim to clarify this misconception by emphasizing the core distinctions between conventional techniques and semantic communications. This is further illustrated in Figure 1.3.
1.1.3.1 Semantic Communications Is Not Data Compression
In a traditional communication framework, as described in information theory by Shannon 1948, data compression, also called source coding, involves encoding information using fewer bits than the original data stream. This compression exploits statistical redundancies within the data bits, allowing for data representation without sacrificing any information, thus ensuring reversibility at the receiver end. While both data compression and semantic communications share the objective of minimizing the size of transmitted data, they are fundamentally distinct concepts:
Data compression achieves minimalism, i.e., shrinking the size of a particular data stream, by identifying and eliminating statistical redundancy. For instance, the most prominent lossless compressors employ probabilistic models such as prediction by partial matching [Zhang and Adjeroh,
2008
]. Notably, there is a close connection between data compression and machine learning (ML), in that they both specifically attempt to predict the posterior probabilities of a sequence given its history. However, data compression does not bear any learning ability that contributes to a particular training memory or a trained model. Put simply, from an ML perspective, data compression techniques tend to overfit as they solely aim to reduce the current data stream without considering futuristic ones. They lack learning abilities and reasoning. From a minimalist viewpoint, data compression might offer short-term benefits over semantic communications. However, it fails to provide contextual information and knowledge-driven memory to the receiver.
In contrast to data compression, semantic communications focus on identifying, learning, and representing patterns that map to structure and semantic content instead of merely compressing data redundancies. While data compression aims to overfit statistical characteristics of data streams, semantic communications prioritize characterizing structure rather than pure randomness exhibited in data. Random data points are often better transmitted conventionally. Semantic communication achieves “minimalism” through semantic representations that not only comprise fewer total bits but also teach the receiver to learn, generate, and automate tasks or messages. This is enabled by the characteristics of the adopted representation and the acquired reasoning capabilities, which minimize the number of bits per transmission and the total number of transmissions required to convey a message or accomplish a task. Semantic communication networks achieve minimalism through mechanisms that surpass simple bit compression. Moreover, when radio nodes operate based on organized knowledge, they can make more informed logical conclusions across the networking stack, a capability unique to reasoning-driven semantic communication networks.
Figure 1.3 What is (is not) semantic communications [Chaccour et al., 2024]?
1.1.3.2 Semantic Communications Is Not Only an “AI for Wireless” Concept
AI has played a significant role in addressing various wireless-related challenges, including channel estimation, beamforming, network management, and receiver design. While AI has indeed enhanced the accuracy, precision, or key performance indicators (KPIs) relevant to these tasks, the fundamental functionality and dynamics of the corresponding wireless tasks have largely remained unchanged. For instance, AI-enabled channel estimation represents an advanced method for conducting classical channel estimation, albeit with the same core task. In contrast, semantic communication systems do not simply add an AI layer atop existing tasks; they represent a fundamental shift in communication paradigms. Semantic communications do not merely improve the air interface with AI; instead, they introduce a novel approach to communication tasks. With semantic communications, user equipment (UEs) and base stations (BSs) no longer rely solely on continuous channel probing. In scenarios where radio nodes have amassed substantial knowledge, the necessity for frequent channel estimation diminishes. Furthermore, rather than relying on bit-driven signaling messages to gauge the channel, semantic communications enable a continual understanding of contextual information from previous messages. This contextual awareness can facilitate learning channel characteristics while utilizing computational resources efficiently. Essentially, semantic communications fundamentally alter the mechanism of communication tasks, as exemplified by the transformation in channel estimation.
1.1.3.3 Semantic Communications Is Not Only Goal-Oriented Communications
A goal-oriented communication system involves a number of agents that interact and exchange messages to achieve a joint goal or separate goals that include the same environment. For example, two robots can interact with each other to execute a common mission. Here, in contrast to sending the information gathered by sensors bit-by-bit, the robots can exchange multiple feedback messages of their current semantic action, their next expected outcome, all while achieving a unique joint goal. In a goal-oriented framework, the nodes, e.g., the teacher and apprentice, can also be achieving two separate goals. Much of the early-on work on semantic communication has equated it with such goal-oriented communication systems [Xie et al., 2022; Zhang et al., 2022; Farshbafan et al., 2023]. However, there are fundamental differences between the two concepts. In some sense, goal-oriented communications falls under the umbrella of semantic communications. For instance, in every goal-oriented communication system, the nodes will have to embed semantic representations to ultimately achieve a particular goal. In contrast, under the broader auspices of a semantic communication system, the generation and communication of semantic representations is not necessarily done for the purpose of serving a system-wide goal. In this regard, limiting the concept of semantic communications to the confines of goal-oriented systems will therefore unnecessarily limit its use to a subset of use cases that have a competitive or cooperative nature. Meanwhile, there are many instances in which the teacher and the apprentice do not necessarily share any joint goals nor interact with a common environment. For instance, the teacher can be a server that is transmitting highly data-intensive content (e.g., XR content) to a particular user. Here, every standalone content transmitted can have an entirely different goal, and there are no cooperative or competing goals between the teacher and the apprentice. Yet, in this case, semantic communications can still be used to: (i) rely less on the channel to transfer massive information content and (ii) empower radio nodes with reasoning to make versatile decisions, which can enhance network’s capability in meeting the stringent requirements of future applications.
1.1.3.4 Semantic Communications Is Not Only Application-Aware Communications
Implementing the context of information within the transmission of messages may seem at first glance similar to the traditional concept of application-aware communication. In fact, there are many prior works (e.g., Shajaiah et al. 2019, Chen et al. 2019, Temdee and Prasad 2018, and Chen et al. 2020) that have fine-tuned the network optimization process to address application-level requirements. For example, in XR applications, the XR content transmitted by users may exhibit a particular correlation. Here, some works such as Chen et al. 2019 exploit this correlation to ultimately improve the management of uplink and downlink wireless transmissions. Notably, it is important to distinguish between the “context-awareness” concept defined by such frameworks and the one granted with semantic communications systems, as the former is a mere application and use case-specific awareness. In contrast, in semantic communication systems, “context” is a concept defined with respect to the low-level structure of exchanged data streams between the transmitter and the receiver. Such low-level intelligence opens the door for an inter-application, intra-application, and out-of-domain generelizability. In other words, a radio node can leverage the meaning attributed to low-level data corresponding to service A by using it to improve the E2E performance for service B.
1.2 Reimagining Future G Applications with Semantic Communications
In this section, we will explore how the adoption and deployment of semantic communication will fundamentally transform emerging applications like XR, digital reality, and concepts such as sustainable networks.
1.2.1 Semantic Communication for Next-Generation XR
The realm of XR, encompassing augmented reality (AR), virtual reality (VR), and evolutionary merged reality concepts like holographic teleportation, promises to revolutionize the way we interact with the world around us. However, current XR experiences face significant challenges, particularly in the realm of data transmission. The vast quantities of data required to render realistic environments create bottlenecks, hindering immersion and user experience [Chaccour et al., 2022]. Essentially, XR experiences often require extremely high data rates as well as high reliability, availability, resilience, and low latency. This combination is often difficult to achieve due to physical layer limitations. Strategies include advanced compression, edge computing, network slicing, adaptive streaming, and foveated rendering. These strategies could gain added flexibility with semantic communication as the backbone instead of classical methods.
From raw data to semantic meta-data:
Current XR systems transmit vast amounts of data to represent complex environments [Chaccour et al.,
2022
]. Semantic communication breaks this paradigm by prioritizing the transmission of critical metadata, such as object type, location, and attributes as well as their inherent structure. This metadata carries the semantic meaning, significance, or structure of the scene, allowing for efficient reconstruction at the receiving end. For instance, an AR system utilizing semantic communication might transmit the “control” semantic information “fire exit” associated with a specific door, rather than the entire door image. This significantly reduces data volume and bandwidth requirements, leading to a more lightweight and efficient XR experience.
Reliability and error correction in semantic XR:
While advancements like massive multiple-input and multiple-output (MIMO) and higher carrier frequencies promise high
