semantic communication
- Explanation 1: Semantic communication can understand the business requirements and environment in advance, carry out the semantic characteristics of the information source
理解、提取及传输, and at the same time ensure that the sink can understand the semantic characteristics of the received information source, so as to successfully restore the information source information based on semantics. Through the accurate extraction and efficient transmission of semantic features, semantic communication will greatly reduce the transmission bandwidth requirements of large-bandwidth services such as video and pictures in new 6G application scenarios, thereby greatly improving communication efficiency and improving user experience. - Explanation 2: Semantic communication refers to the communication method in which semantic information is extracted from the source and encoded, and transmitted in a noisy channel. Traditional grammatical communication requires that the decoding information at the receiving end is strictly consistent with the encoding information at the sending end, that is, to achieve
比特级的无差错传输. On the contrary, semantic communication does not require a strict match between the decoding sequence and the encoding sequence.只要求接收端恢复的语义信息与发送语义信息匹配即可
Syntax, Semantics and Pragmatics
- Grammar:
通信符号How to transmit exactly - Semantics: how precisely the transmitted symbols express the intended meaning
- Pragmatics: How Effectively Conveyed Meaning Affects Desired Behavior
The difference between syntactic communication and semantic communication
Reference: Beijing University of Posts and Telecommunications-Zhang Ping Semantic communication technology development trend and countermeasures 38 minutes
Semantic communication system for 6G mobile communication
- At the sending end, the information generated by the information source is first sent to the semantic extraction module to generate a semantic representation sequence, and then sent to the semantic source encoder to compress and encode the semantic features, and then sent to the channel encoder to generate a channel encoding sequence and sent to the transmission channel.
- At the receiving end, the channel output signal is first sent to the channel decoding module, and the output decoding sequence is sent to the semantic source decoder, and the obtained semantic representation sequence is sent to the semantic recovery and reconstruction module, and finally the source data is sent to the sink.
- The channel codec belongs to the classical communication system, while the semantic extraction and encoding module belongs to the semantic communication system; the
classical communication channel is统计转移概率modeled by , while the semantic channel is语义标签之间的逻辑转移概率modeled by .
Differences between Semantic Communication and Classical Communication
- The most important difference between semantic communication and classical communication is that the semantic encoding and decoding modules are based on the knowledge base trained with massive data, and extract and reconstruct semantic information through the deep learning network. This process provides strong prior knowledge for classical signal transmission, effectively improving Transmission effectiveness and reliability. At the sending end, the semantic extraction module extracts semantic features from the source message based on the knowledge base and deep learning network. Among them, the semantic extraction module adopts deep learning network models of different structures according to the redundancy characteristics of information sources. For example, the timing and text sources use the recurrent neural network (RNN) network model, the image source uses the convolutional neural network (CNN) model, and the graph data source uses the graph convolutional network (GCN) model.
- At the receiving end, the semantic synthesis module reconstructs the received semantic information based on the knowledge base and deep learning network. If the information sources are multi-modal or heterogeneous, semantic synthesis of multi-source data is also required during semantic extraction and encoding. The two ends of the transceiver endow the neural network with prior knowledge in specific scenarios
共享云端知识库through the method.数据驱动
Problems to be Solved in Semantic Communication
- The transmission of semantics is also limited by different industry application scenarios
传输带宽,收发端计算能力and存储能力the following problems need to be solved:
①According to different industry application scenarios, analyze the impact of different important semantic feature selection on semantic errors, and clarify the relationship between semantic transmission errors and channel bandwidth, intelligent The relationship between multiple resource parameters such as algorithm models, computing power requirements, and storage requirements ②Construct the theory and mechanism of optimal allocation of joint resources
under the guidance of semantics ③Evaluate the performance indicators of semantic errors, important semantic features, and semantic transmission efficiency
Taxonomy of Semantic Communication
Existing semantic encoding and decoding modules are mainly implemented based on model-free machine learning methods. These machine learning-based solutions can be roughly divided into two categories: modular design and all-in-one design
模块化设计Introduce semantic encoding and decoding into the existing communication system as an independent module. The semantic encoding and decoding module realizes the mutual conversion between syntax and semantic information, and improves the efficiency of text, voice, and image transmission.一体化设计Based on信源-信道联合编码the idea of semantic enhancement, the semantic coding/decoding module and the source-channel coding module are jointly designed to achieve end-to-end transmission optimization.
Reference Towards 6G Intent-Driven Network - A New Paradigm of Network Based on Semantic Communication
semantic knowledge base
-
Definition Semantic knowledge base is a kind of knowledge network model that can provide relevant semantic
knowledge description for data information结构化具备记忆能力 -
Classification
The semantic knowledge base for semantic communication can be divided into source , channel , and task knowledge bases , which provide multi-level semantic knowledge representation for source data , channel transmission environment , and task requirements respectively.①Information source data can be text, picture, video
②Channel transmission environment can be the location and shape information of obstacles and scatterers during transmission, intelligent reflective surface location information and configuration matrix, etc. ③Task requirements
can be image classification, 3D reconstruction, semantic segmentation
Different Implementations of Semantic Communication
- Solution 1: End-to-end network, source-channel joint coding
- Solution 2: Modularization, using the semantic encoder as an advanced source encoding, followed by channel encoding
- Solution 3: Modularization, add a "semantic extractor" before the traditional source code
Reference: Tao Meixia, Shanghai Jiaotong University - Design of Semantic Communication System Empowered by AI 10 minutes
references
[1] Niu Kai, Dai Jinsheng, Zhang Ping, Yao Shengshi, Wang Sixian. Semantic Communication for 6G [J]. Mobile Communications, 2021,45(04):85-90.