Source-linked AI summary
6G Networks: Beyond Shannon Towards Semantic and Goal-Oriented Communications
Emilio Calvanese Strinati, Sergio Barbarossa
TL;DR
The paper asks how 6G can move beyond packet-level Shannon communication to support meaning, task effectiveness, and sustainability. It proposes semantic and goal-oriented network architectures that identify relevant information and combine knowledge-based reasoning with machine learning. The resulting vision targets greater effectiveness and reliability without necessarily increasing bandwidth or energy, while acknowledging computational, knowledge-base, and federated-learning constraints.
Problem
The paper addresses the gap between Shannon-style symbol reliability and communication systems that must convey meaning or accomplish goals sustainably.
Method
It proposes 6G architectures and communication strategies centered on semantic relevance, goal accomplishment, knowledge representation, reasoning, and machine learning.
Results
The vision targets increased effectiveness and reliability without necessarily increasing bandwidth or energy by transmitting only information needed for meaning extraction or goal execution.
Takeaways & Limitations
Semantic and goal-oriented communication are presented as foundations for more effective and sustainable 6G networks, including semantic learning at the edge.
Takeaways & Limitations
Semantic processing adds receive-side computational complexity and delay, while knowledge bases are incomplete and federated learning faces channel, device, and model heterogeneity.
Abstract
from arXiv · showhide
The goal of this paper is to promote the idea that including semantic and goal-oriented aspects in future 6G networks can produce a significant leap forward in terms of system effectiveness and sustainability. Semantic communication goes beyond the common Shannon paradigm of guaranteeing the correct reception of each single transmitted packet, irrespective of the meaning conveyed by the packet. The idea is that, whenever communication occurs to convey meaning or to accomplish a goal, what really matters is the impact that the correct reception/interpretation of a packet is going to have on the goal accomplishment. Focusing on semantic and goal-oriented aspects, and possibly combining them, helps to identify the relevant information, i.e. the information strictly necessary to recover the meaning intended by the transmitter or to accomplish a goal. Combining knowledge representation and reasoning tools with machine learning algorithms paves the way to build semantic learning strategies enabling current machine learning algorithms to achieve better interpretation capabilities and contrast adversarial attacks. 6G semantic networks can bring semantic learning mechanisms at the edge of the network and, at the same time, semantic learning can help 6G networks to improve their efficiency and sustainability.
1. Introduction
The paper argues that future 6G networks should shift beyond Shannon’s symbol-level reliability paradigm toward semantic and goal-oriented communications. This shift targets relevant-information extraction, task effectiveness, and more sustainable network operation.
- Motivation: 6G research must move beyond Shannon’s framework to address emerging services and future individual and societal needs.The paper presents this as a vision for Beyond-5G connect-compute networks.
- Motivation: Growing demands from services such as virtual reality and autonomous driving face eventual spectrum and energy scarcity.Higher carrier frequencies provide bandwidth but introduce blocking, atmospheric absorption, and power-amplifier efficiency problems.
- Three communication levels: Shannon and Weaver distinguish technical transmission, semantic exchange, and the effectiveness of exchanged information.Shannon’s theory deliberately focused on the technical problem, while the paper advocates incorporating the other levels.
- Proposed vision: The proposed 6G vision makes semantics and effectiveness central, identifying only information relevant to intended meaning or predefined goals.The authors connect this focus to reduced transmitted and recovered data and savings in bandwidth, delay, and energy.
- Paper scope: The paper develops this vision through use cases, KPIs, a new architecture, semantic and goal-oriented communications, and learning-based strategies.These topics are organized across Sections 2–6.
2. 6G Use Cases, KPIs and New Services
The paper frames 6G as a computation-oriented network supporting immersive, intelligent, and machine-centric services alongside conventional KPI evolution. Its proposed direction adds effectiveness, learning, and sustainability objectives to communication performance.
- KPIs: 6G retains 5G-style KPIs while introducing new measures for computation-oriented networks and distributed artificial intelligence.Candidate new KPIs include learning reliability and energy consumption associated with goal accomplishment.
- Use cases: 6G use cases span extended reality, holographic and multisensory communications, industrial automation, autonomous systems, and massive sensor networks.Holographic links may require several terabits per second and stringent end-to-end latency.
- KPIs: Not all KPIs need to be achieved simultaneously or everywhere; selected targets should adapt to application and service needs.The paper discusses a factor of 10 to 100 improvement between successive wireless generations and a possible 100-fold capacity increase.
- Network evolution: Uplink growth from many connected devices increases the importance of pervasive sensing, computation, and resource-efficient network design.The paper notes a reduction or inversion of traditional downlink-uplink traffic asymmetry beginning with 4G.
- New services: The proposed services include distributed-intelligence machine communications, globally enhanced mobile broadband, ultra-reliable low-latency computation-communication-control, and semantic services.Semantic services support shared knowledge across human-human, human-machine, and machine-machine interactions.
3. Beyond Shannon? A new architecture
The paper argues that 6G must evolve from reliable data delivery toward a communicate-and-compute system that also supports knowing, learning, decision-making, and sustainable operation. It therefore motivates resource-aware design beyond simply increasing data rates.
- Beyond the legacy: Wireless generations have traditionally pursued reliable, high-capacity, low-latency communication, while energy consumption increasingly limits practical operation.This legacy is associated with broader bandwidths and exploration of higher frequency bands.
- Communicate-and-compute: 5G begins a communicate-and-compute transition in which sensing, computing, controlling, and actuating form a continuous loop.The paper presents 6G as extending this trajectory toward networked knowledge and decision-making.
- Sustainability: The central design question is whether more intelligent mobile services can be delivered without proportionally increasing capacity, infrastructure, or energy.The paper argues that wider bandwidths and higher data rates alone are unlikely to handle this challenge efficiently.
Level A. How accurately can the symbols of communication be transmitted?
The paper extends communication design from symbol transmission to semantic interpretation and goal effectiveness. Its proposed architecture combines layered communication functions with knowledge representation, machine learning, and adaptive control.
- Three levels: The three communication levels ask how accurately symbols are transmitted, how precisely they convey meaning, and how effectively meaning affects conduct.These correspond to technical, semantic, and effectiveness problems.
- Motivation: The paper argues that semantic and effectiveness aspects become necessary as networks interconnect humans and machines with varying intelligence.6G is therefore expected to incorporate all three Shannon-Weaver levels.
- Architecture: The proposed architecture places semantic and effectiveness levels above a technical protocol stack and connects them with physical, digital, and application spaces.The architecture also includes artificial intelligence tools, knowledge representation, machine learning, and control-actuation loops.
- Architecture: Virtualization extends beyond network functions to application, semantic, and effectiveness functions, distributing computation and communication across virtual machines.This builds on 5G virtualization, NFV, and SDN.
- Communication paradigms: Semantic communication focuses on conveying meaning, while goal-oriented communication focuses on accomplishing a goal under time and resource constraints.Online learning can adapt traffic, coding, decoding, and scheduling based on network monitoring.
4. Semantic Communications
Semantic communications focus on recovering meaning and increasing knowledge rather than preserving symbol sequences, using knowledge representation and reasoning across multi-level communication architectures. They can also exploit language knowledge to correct some syntactic errors without retransmission.
- Semantics and knowledge representation systems: Semantic communication prioritizes message meaning and knowledge gain over the probabilistic properties of encoded symbols.A semantically correct exchange recovers equivalent content or increases the destination’s knowledge.
- Semantics and knowledge representation systems: Knowledge representation models entities and relations so reasoning can generate new knowledge from represented symbols.Graph-based representations use nodes for entities and edges for relations.
- Semantics and knowledge representation systems: Knowledge bases are inherently incomplete because domain relations are extensive and complete reasoning may exceed computational time constraints.Answers may therefore be correct only with a certain degree of reliability when time limits apply.
- Semantic communication architecture: The three-level architecture connects technical transmission, semantic exchange, and effectiveness among human or machine agents.Its protocol stack and AI machinery combine knowledge representation, machine learning, network data, control, and actuation.
- Semantic source and channel coding: Message-to-symbol mappings may be one-to-many or many-to-one, creating semantic equivalence and natural-language ambiguity.Context can resolve ambiguity, but disambiguation remains a key NLP challenge.
- Semantic source and channel coding: Semantic decoding can recover the intended content despite some syntactic errors by exploiting the rules of the communicating language.This can reduce retransmissions in ARQ when errors are detected at the semantic rather than syntactic level.
Example of application to text transmission
Semantic text transmission replaces bit-by-bit recovery with recovering sentence meaning, using learned semantic/channel encoding and decoding. The same framework extends to speech, video, and holographic communications, while trading reduced transmitted information for receiver computation and delay.
- Text transmission: DeepSC combines semantic and channel encoding in a deep neural network trained on variable-length sentences to minimize semantic error and transmitted symbols.Its performance is evaluated by similarity between emitted and reconstructed sentences over AWGN and Rayleigh fading channels.
- Feedback and cross-layer interaction: Semantic decoding can request retransmission only when message meaning is unclear, rather than whenever packets contain syntactic errors.Semantic feedback may also request a different message version or a lower data rate when the receiver can predict the content.
- Semantic recovery: Semantic prediction can reconstruct missing information or compensate for syntactic errors without retransmitting every affected packet.The framework prioritizes recovering source content and completing tasks over error-free symbol reception.
- Speech: Speech transmission can use recognition and language processing to correct words or sentences and reconstruct speech lost during channel fades.Context-based recovery can reduce pressure on forward error-correction codes.
- Video streaming: Video interpreters can predict missing frames and reproduce a semantically equivalent sequence when channel fades cause losses without major scene changes.The reconstructed video need not equal the transmitted video if the captured flow of events remains intact.
- Holographic communications: Receiver-side digital models can reconstruct holographic views, reducing the data rate needed to provide the desired quality of experience.This approach assumes source and destination share relevant background knowledge.
- Trade-offs: Semantic communication offers resource savings but adds receiver computational complexity and delay, which can become a serious bottleneck in some applications.The paper connects this trade-off to predictive processing inspired by human brains.
5. Goal-oriented communication
Goal-oriented communication focuses on transmitting only information relevant to accomplishing a specified task, rather than preserving all source information. The framework seeks efficient mappings that satisfy service constraints while reducing communication resources.
- 5. Goal-oriented communication: Goal-oriented communication defines performance by how effectively a specified goal is fulfilled under resource and application constraints.The goal may be parameter estimation or classification, with requirements such as delay, accuracy, and energy consumption.
- 5. Goal-oriented communication: The proposed framework uses feedback from the decision maker to the source encoder to adapt transmitted data to the communication goal.The system maps observations X into transmitted data Z for a fusion center that makes decisions.
- 5. Goal-oriented communication: A minimal sufficient statistic can replace the full dataset without losing information needed to estimate the target parameter.Because its entropy may be much smaller than that of the original data, it can require significantly fewer transmitted bits.
- 5. Goal-oriented communication: The information bottleneck formulation trades source compression against learning accuracy through a tunable parameter β.Small β favors compression, whereas large β favors learning accuracy.
Example of application: Edge online learning
The edge-learning example adapts transmission and computation to classification requirements. It combines classifier feedback with dynamic source encoding to balance accuracy, delay, and sensor energy.
- Example of application: Edge online learning: The proposed edge system minimizes a weighted sum of energy consumption and classification accuracy under average end-to-end delay constraints.It dynamically adjusts quantization bits, transmit power, and edge-server CPU scheduling on MNIST and HSM datasets.
- Example of application: Edge online learning: An ensemble combines parallel SVM and MLP learners, with the decision selected from the learner producing the most reliable output.Classifier-output entropy is used as a heuristic for posterior class uncertainty because correct classification cannot be directly evaluated in practice.
- Example of application: Edge online learning: Feedback based on classifier-output entropy dynamically adjusts image quantization bits to satisfy delay constraints and trade classification accuracy against sensor energy.Low output entropy corresponds to greater classifier reliability in the stated heuristic.
- Example of application: Edge online learning: Goal-oriented video analytics can reduce peripheral-camera data rates by filtering frames that are irrelevant to the ensuing analysis.The paper presents FFS-VA as an example of a pipelined multistage video-analytics system.
- Example of application: Edge online learning: Combining semantic and goal-oriented communication aims to transmit only semantic information strictly relevant to goal achievement without reducing accomplishment accuracy or reliability.
6. Online Learning-Based Communication and Control
The paper envisions AI-native 6G networks that use online learning, edge intelligence, and semantic knowledge to manage constrained services. It argues that combining machine learning with knowledge representation can improve learning robustness and network efficiency.
- 6. Online Learning-Based Communication and Control: AI-native 6G networks are intended to support learning tools that reshape network operation according to requirements and constraints.The vision includes machine-driven network design and machine learning at the edge for context-aware, delay-critical services.
- 6.5. Semantic machine learning: Merging machine learning with semantics combines data-driven inductive strategies with model-based deductive strategies.The paper attributes potential benefits to improved semantic interpretation, contextual disambiguation, robustness against adversarial attacks, and resource optimization.
- 6. Online Learning-Based Communication and Control: Edge computing supports intelligent services requiring smart decisions within tight delay and jitter constraints by moving computation closer to users.Service delay may include communication, computation, storage access, and control-actuation components.
- 6. Online Learning-Based Communication and Control: Online learning interleaves learning and testing so models can adapt as wireless channels vary over time.The paper distinguishes reinforcement learning among online algorithm families.
- 6.3. Federated learning: Federated learning sends local model estimates or gradients instead of raw device data and aggregates updates at a fusion center.This setting faces heterogeneity in channels, device behavior, and models, which can affect convergence and final accuracy.
- 6.5. Semantic machine learning: Knowledge representation and reasoning can guide learning toward solutions coherent with an underlying language of entities and logical rules.The paper presents this integration as a way to incorporate external world knowledge and context into decision-making.
7. Conclusions
The proposed 6G vision prioritizes semantic and goal-oriented communications to improve effectiveness and sustainability by transmitting and processing only goal-relevant information. Its central implementation challenge is distributed computing that can learn and extract meaning using knowledge representation.
- Semantic and goal-oriented communications identify relevant information needed to recover meaning or achieve predefined goals efficiently.This enables selective transmission, processing, inference, and memory rather than handling all information equally.
- The approach seeks increased effectiveness and reliability without necessarily increasing bandwidth or energy.
- Sustainability is positioned as a key property of future networks, replacing the usual drive for continually increasing energy and bandwidth resources.
- Implementing the vision requires distributed computing mechanisms that learn and extract meaning through knowledge representation systems.These mechanisms must also identify and exploit strictly relevant information in goal-oriented communications.