1 of 20

Towards Efficient Semantic Communication Systems via Deep Learning Techniques

Shiyao Ma

2 of 20

2

  • Conventional Communication and Semantic Communication
  • Transmission-oriented Semantic Communication
  • Goal-oriented Semantic Communication
  • Future Work

Outline

3 of 20

3

The exponential growth of communication data requires new solutions

Data will consume 20% of world’s energy

› 9 Billion of connected people on Earth

› 1 Trillion of connected devices

› Wireless communications and AI assistance will be a commodity

› The Cyber and Physical space fusion turn humans, things and events into information

​

The system capacity is gradually approaching to the Shannon limit

​

Background

4 of 20

4

Three key issues:

  • Level A (The technical problem)
  • Level B (The semantic problem.)
  • Level C (The effectiveness problem.)

Three Levels in Communication System

[1] G. Shi, Y. Xiao, Y. Li and X. Xie, "From Semantic Communication to Semantic-Aware Networking: Model, Architecture, and Open Problems,"

5 of 20

5

Conventional Communication System

I have an automobile

Conventional Communications

I have an automobile

I have an automobile

Semantic Communications

I own a car

Semantic Communication System

6 of 20

6

General Communication System

7 of 20

7

Definition of Semantic Communication

 

 

8 of 20

8

Difference of semantic communication and conventional system

Conventional Communications:

  • Only consider the data recovery accurately
  • Information redundancy are removed in entropy-domain
  • All information (including useless and irrelevant) is transmitted to the receiver

Semantic Communications:

  • Consider the transmitted symbols convey the desired meaning
  • Information redundancy are removed in semantic domain
  • Only useful and relevant information transmitted to the receiver

9 of 20

Why deep learning

9

  • End-to-end loss optimization for global optimality
  • More efficient semantic feature extraction
  • End-to-end and semantic communications

Challenges in conventional communication systems

  • Complex mathematical models and optimality
  • Effective compression methods
  • Spectrum efficiency limited by Shannon capacity

Advantages of deep learning

10 of 20

10

Semantic Communication Systems for Image Transmission

 

[2] E. Bourtsoulatze, D. Burth Kurka and D. Gündüz, "Deep Joint Source-Channel Coding for Wireless Image Transmission,"

11 of 20

11

Semantic Communication Systems for Speech Transmission

[3] Z. Weng and Z. Qin, "Semantic Communication Systems for Speech Transmission,"

 

12 of 20

12

[4] H. Xie, Z. Qin, G. Y. Li and B. -H. Juang, "Deep Learning Enabled Semantic Communication Systems,"

Semantic Encoder

Channel

Semantic Decoder

Semantic Communication Systems for Text transmission

13 of 20

13

Loss Function

 

[4] H. Xie, Z. Qin, G. Y. Li and B. -H. Juang, "Deep Learning Enabled Semantic Communication Systems,"

14 of 20

The goal of the system is to learn a set of parameters θ from a set of observations X

Goal: Solution:

Goal-oriented Semantic Communication Systems

[5] Strinati E C, Barbarossa S. “6G networks: Beyond Shannon towards semantic and goal-oriented communications”

14

15 of 20

Future Work: Goal-oriented Multimodal Semantic Communications

15

  • Is it necessary to transmit all the data?
  • Current semantic communication systems are mostly based on single-modal
  • Most tasks or goals in the real world are multimodal
  • We need a new paradigm to solve the above problem

16 of 20

16

Multimodal

17 of 20

17

Previous Multimodal Semantic Communications

Semantic Encoder

Semantic Decoder

VQA Semantic Decoder

[6] H. Xie, Z. Qin and G. Y. Li, "Task-Oriented Multi-User Semantic Communications for VQA,"

18 of 20

18

Future Work: Goal-oriented Multimodal Semantic Communications

Semantic Encoder

Semantic Decoder

cat

Semantic Encoder

Semantic Decoder

. . .

cat

19 of 20

Conclusions

  • Conventional communications almost to the limit
  • Semantic communications can significantly improve transmission efficiency
  • It is necessary to design an efficient semantic coding via deep learning
  • Semantic communication is the future of wireless communication

19

20 of 20

Q&A

Thanks for listening!