Source-linked AI summary
Semantic Communications in Networked Systems: A Data Significance Perspective
Elif Uysal, Onur Kaya, Anthony Ephremides, James Gross, Marian Codreanu, Petar Popovski, Mohamad Assaad, Gianluigi Liva, Andrea Munari, Touraj Soleymani, Beatriz Soret, Karl Henrik Johansson
TL;DR
Communication networks are optimized mainly for delivering data, while real-time machine systems need significant information delivered to the right place at the right time. The paper proposes an end-to-end semantic communication architecture that redesigns information generation, transmission, and usage together. It presents semantic metrics and related sampling, coding, access, and control directions, including evidence that fewer transmissions can improve freshness.
Problem
Existing raw-data communication paradigms create bottlenecks and leave a knowledge gap relative to proliferating real-time machine systems.
Method
The paper proposes an end-to-end semantic communication architecture based on information significance relative to the purpose and timing of data exchange.
Results
A 30% reduced update rate can achieve 6% lower Age of Information while saving energy.
Takeaways & Limitations
Semantic-aware redesign can align data generation and network operation with real-time communication and control requirements.
Abstract
from arXiv · showhide
We present our vision for a departure from the established way of architecting and assessing communication networks, by incorporating the semantics of information for communications and control in networked systems. We define semantics of information, not as the meaning of the messages, but as their significance, possibly within a real time constraint, relative to the purpose of the data exchange. We argue that research efforts must focus on laying the theoretical foundations of a redesign of the entire process of information generation, transmission and usage in unison by developing: advanced semantic metrics for communications and control systems; an optimal sampling theory combining signal sparsity and semantics, for real-time prediction, reconstruction and control under communication constraints and delays; semantic compressed sensing techniques for decision making and inference directly in the compressed domain; semantic-aware data generation, channel coding, feedback, multiple and random access schemes that reduce the volume of data and the energy consumption, increasing the number of supportable devices.
INTRODUCTION
Networked systems increasingly require significant information to reach computation or actuation at the right time, but raw-data communication creates bottlenecks. The paper envisions semantic communication as an end-to-end redesign of information generation, transmission, and use.
- Motivation: Semantic communication provisions the right significant information to the right computation or actuation point at the right time.Its significance is defined relative to the purpose of the data exchange and may include real-time constraints.
- Examples: Process-aware sparse sampling can improve remote-estimation MSE by orders of magnitude over uniform sampling while reducing sampling rate, device energy use, and network consumption.The setting includes random communication delays and low-power or energy-harvesting sensors.
- Examples: Semantic-aware transmission can prioritize time-dependent information among autonomous vehicles while meeting safety demands for timely maneuver consensus.The motivating scenario includes unforeseen events such as the sudden appearance of pedestrians.
- Motivation: Raw-data streams create enormous data volumes and bottlenecks that limit real-time decision-making systems such as smart grids and networked control.The paper identifies a knowledge gap between existing communication paradigms and the needs of proliferating real-time systems.
- Architecture: Resolving these bottlenecks requires transforming communication-system design and reformulating data generation and transfer.The envisioned architecture redesigns the complete information process rather than only improving delivery.
A SEMANTIC-AWARE END-TO-END ARCHITECTURE FOR INFORMATION FLOW
The envisioned architecture requires new semantic metrics and coordinated principles across communication layers. It replaces assumptions of uncontrolled traffic and semantic-agnostic networking with end-to-end information-flow optimization.
- Semantic-aware architecture: The architecture must incorporate semantics-based metrics into the communication process through new theoretical results.These metrics serve as objectives for the envisioned semantics-aware design approach.
- Semantic-aware architecture: Semantic-aware sampling relaxes the assumption of exogenous data arrivals that constrains current communication protocols.The paper explicitly calls for non-uniform sampling within communication-system design.
- Semantic-aware architecture: Link, transport, and application-layer principles must be developed in concert to meet semantic targets under scarce network resources and energy.The proposal treats cross-layer coordination as necessary for semantic-related objectives.
- Semantic-aware architecture: New protocol principles should optimize information flow in control systems rather than support only data streaming or file transfer.Current network-control systems operate on classical networks optimized for those conventional traffic types.
EXAMPLES OF SEMANTIC MEASURES FOR THE ENVISIONED ARCHITECTURE
For machine-type communications, semantics concerns the significance of data attributes needed for effective real-time operation, and the paper illustrates this perspective with semantic measures.
- Semantic significance: In machine-type communications, semantic significance includes data timeliness, priority, and value for real-time decision scenarios.The paper distinguishes this perspective from emerging communication notions that define semantics more closely as message meaning.
- Semantic significance: The paper frames semantic measures as attributes that support the effective operation of the receiver.For human communication, earlier formulations instead emphasized conveying desired meaning and influencing desired conduct.
FRESHNESS
Freshness is a semantic concern distinct from packet latency because it depends jointly on sampling and network delay. The paper presents AoI and related measures for analyzing and optimizing this concern.
- Age of Information: Age of Information (AoI) measures freshness as the elapsed time since the newest destination-available sample was generated at the source.For a newest received timestamp u(t), the age is Δ(t) = t−u(t).
- Related measures: Query-AoI and Age of Incorrect Information extend age-based analysis beyond basic freshness.The paper identifies these as emerging metrics alongside age penalties g(Δ(t)).
- Age of Information: AoI combines source sampling patterns with network delays, unlike designs that handle sampling and delay separately.Standard TCP and UDP do not explicitly support freshness, which can produce extreme inefficiency.
- Freshness versus latency: Latency guarantees alone do not ensure freshness: with 1 ms random sampling, Peak AoI reaches 2 ms even under 0.5-1.0 ms one-way latency.Reducing Peak AoI to 1.1 ms requires a 0.1 ms sampling period tailored to the network latency.
RELEVANCE
Semantic relevance improves estimation and control by selecting information according to its age, process change, or value rather than relying on uniform transmission.
- Estimation: Process-aware non-uniform sampling can match uniform-sampling MSE while using fewer transmissions under random channel delays.The paper identifies process change as a relevance criterion beyond age.
- Estimation: MMSE-optimal and age-aware sampling can be arbitrarily better than uniform sampling as the maximum allowed sampling rate varies.The comparison assumes IID unit-mean exponentially distributed channel delays.
- Control: Value-of-Information policies transmit sensory information when its benefit can justify transmission cost in feedback control.The inverted-pendulum example shows VoI-driven scheduling events and control trajectories.
COMMUNICATION ARCHITECTURE
The proposed architecture represents information by freshness, relevance, value, and priority, using semantic metrics to align sampling and transmission with receiver needs.
- Semantic attributes: Freshness, relevance, and value correspond respectively to sending at the right time, generating the right information, and delivering it to the right computation point.Value additionally accounts for timeliness and destination within cyber-physical and hierarchical control systems.
- Semantic metrics: QAoI combines age and value in pull-based systems where information is useful only at particular query instants.QAoI-optimized scheduling can improve freshness at those query instants relative to AoI-optimized scheduling.
- Semantic metrics: AoII combines freshness and relevance, remaining flat when the sampled phenomenon has not changed significantly even as AoI rises.The metric is illustrated for video streaming and industrial machine-breakdown settings.
- Sensing: Sparsity, information aging, and path-delay statistics can support semantic-aware causal reconstruction from sub-Nyquist random samples.The proposed direction combines signal and network characteristics for reconstruction.
- Semantic attributes: Priority can be derived from data as a semantic property when comparing information across flows in real-time transportation and satellite-tracking scenarios.This differs from treating priority only as an external constraint or resource-allocation outcome.
CHALLENGE #1: SAMPLING, SOURCE CODING AND COMPRESSION
The paper challenges sample-then-compress pipelines and conventional throughput- or latency-centered design by proposing sensing and reconstruction that incorporate semantics and timing.
- Sampling and compression: Sample-then-compress systems collect high-resolution raw data before computation-intensive compression, creating discarded data and unnecessary network traffic.The paper connects this structure to poor scalability.
- Sampling and compression: Semantic compressed sensing performs detection, classification, or learning directly in the compressed domain without complete signal reconstruction.The vision targets real-time prediction and reconstruction under communication constraints and delays.
- Sampling and compression: The proposed sensing framework augments measurements with reliability or accuracy attributes and jointly uses sparsity and information aging for causal reconstruction.Noisy, outdated, or unreliable samples may be coarsely quantized or discarded, with average sampling below Nyquist.
- Design objectives: Maximum throughput and minimum delay are neither necessary nor sufficient for optimal operation in rapidly evolving applications based on status updates and remote estimation.The paper uses this limitation to motivate semantic performance criteria beyond classical communication KPIs.
CHALLENGE #2: PACKETIZATION, SCHEDULING, RESOURCE ALLOCATION
Semantic communication requires redesigning feedback, packetization, scheduling, and resource allocation around information significance, energy availability, and realistic packet constraints.
- System design: Optimal trade-offs between semantic metrics and energy require revising feedback, packetization, scheduling, and resource allocation while relaxing fresh-information and unit-energy assumptions.Machine learning may support these redesigned policies.
- Feedback and scheduling: Semantic-aware feedback outperforms equal-opportunity packet feedback when a constrained feedback frame allows updates for only selected users.The comparison concerns packet scheduling at a base station in a multiuser downlink.
- Packetization: Short packets devote significant space to metadata and auxiliary operations, motivating packet structures optimized for semantic objectives.Feedback and retransmissions must be studied under realistic channels, bit-error rates, and non-asymptotic block lengths.
- Energy-aware allocation: 30% fewer updates achieved 6% lower AoI in one reported example, showing that reduced transmission rate can coincide with improved freshness.The paper presents this as evidence that scheduling should factor in energy availability rather than assume fresher data requires more sampling.
CHALLENGE #3: SEMANTIC-AWARE MULTIPLE AND RANDOM ACCESS
Current access techniques such as NOMA optimize users or throughput, but semantic goals like freshness require new scheduled and random-access principles. Age-aware random-access protocols have already improved spectral efficiency, freshness, and energy consumption relative to slotted ALOHA.
- NOMA targets user count or throughput, while its spectral-efficiency gains do not directly ensure better freshness or other semantic attributes.
- Semantic principles for scheduled and random access remain a fundamental problem for status updates in mMTC, IoT, and related applications.
- Age-aware random-access protocols improve spectral efficiency, freshness, and energy consumption relative to slotted ALOHA.
CHALLENGE #4: OPERATION AT THE TRANSPORT LAYER AND ABOVE
Operation at the transport layer and above should jointly account for sampling, flow control, communication, and control objectives rather than treating network performance through conventional layer boundaries. This semantic-aware perspective targets resource-efficient, timely information exchange across large-scale networked systems.
- Flow control is coupled to sample generation, while sampling, flow control, and retransmissions jointly determine the system’s input flow and resource use.
- Separately optimizing communication networks and control systems can be highly suboptimal because control objectives are not reflected in network design.
- Control-cost and information-cost trade-off regions can guide packet scheduling while respecting processing and communication delays.
- Semantic-aware ICTs support connections among machines, communities, physical things, processes, and content in data- and information-centric ecosystems.
- Semantic measures can organize traffic over scarce wireless spectrum to support safety-critical, timely consensus in smart mobility.
- Connected-device growth and narrow channels make spectrum scarcity a central constraint for IoT services, including remote applications using satellite connectivity.
- VDES illustrates semantic-driven IoT design by adjusting update rates to vessel location or speed because route changes are more critical in busy harbours.
- The vision requires coordinated advances across signal processing, communication, networking, and control, which traditionally address different aspects of information.