Source-linked AI summary
Extreme URLLC: Vision, Challenges, and Key Enablers
Jihong Park, Sumudu Samarakoon, Hamid Shiri, Mohamed K. Abdel-Aziz, Takayuki Nishio, Anis Elgabli, Mehdi Bennis
TL;DR
URLLC remains challenged by rare-event characterization and the demands of emerging mission-critical applications. The article presents xURLLC, combining prediction, multimodal fusion, and communication-control co-design. It concludes that this vision targets beyond-5G/6G mission-critical applications, while non-RF data adds energy costs.
Problem
URLLC fundamentals remain elusive as new immersive and high-stakes control applications demand stricter reliability, latency, and scalability than 5G URLLC.
Source: §1 · p. 1Method
The article develops an xURLLC vision combining machine-learning prediction, RF and non-RF modalities, and communication-control co-design.
Source: §5 · p. 7Results
The article concludes that xURLLC is predictive, non-RF aided, and integrates communication and control to address extreme-event prediction and scalability.
Source: §5 · p. 7Takeaways & Limitations
The xURLLC vision is intended to spearhead beyond-5G/6G mission-critical applications, including vision-based control, VR, and drone swarms.
Source: §5 · p. 7Non-RF data adds energy costs for acquisition, processing, and control, including collecting training images across cameras and processing extracted features.
Source: §3.1 · p. 4
Abstract
from arXiv · showhide
Notwithstanding the significant traction gained by ultra-reliable and low-latency communication (URLLC) in both academia and 3GPP standardization, fundamentals of URLLC remain elusive. Meanwhile, new immersive and high-stake control applications with much stricter reliability, latency and scalability requirements are posing unprecedented challenges in terms of system design and algorithmic solutions. This article aspires at providing a fresh and in-depth look into URLLC by first examining the limitations of 5G URLLC, and putting forward key research directions for the next generation of URLLC, coined eXtreme ultra-reliable and low-latency communication (xURLLC). xURLLC is underpinned by three core concepts: (1) it leverages recent advances in machine learning (ML) for faster and reliable data-driven predictions; (2) it fuses both radio frequency (RF) and non-RF modalities for modeling and combating rare events without sacrificing spectral efficiency; and (3) it underscores the much needed joint communication and control co-design, as opposed to the communication-centric 5G URLLC. The intent of this article is to spearhead beyond-5G/6G mission-critical applications by laying out a holistic vision of xURLLC, its research challenges and enabling technologies, while providing key insights grounded in selected use cases.
1 INTRODUCTION
URLLC targets rare-event reliability and latency, but 5G’s simulation-led design leaves key wireless and control dynamics insufficiently understood. The article proposes xURLLC around prediction, non-RF modalities, and communication-control co-design to address stricter requirements and scalability.
1 INTRODUCTION
- Mission-critical wireless design must characterize extreme events such as latency-distribution tails, unlike average-based design; 3GPP uses system-level simulations and techniques including short packets and diversity to target 5-nine reliability and 1 ms latency.The article says understanding channel dynamics, estimation, control-loop stability, and robustness to unmodeled phenomena remains incomplete.Source: §1 · p. 1
- Factory automation calls for 7-nine reliability and sub-1 ms latency, while 6G targets 9-nine reliability and 0.1 ms latency for intelligent systems and real-time human-machine interactions.The article notes that high-precision robot control and autonomous vehicles cannot afford 5-nine reliability and millisecond latency.Source: §1 · p. 1
- URLLC must scale to applications combining massive connections and high data rates, including drone-swarm control and vision-based agricultural monitoring.The article says the URLLC-eMBB-mMTC combination is no longer a zero-sum game.Source: §1 · p. 1
- xURLLC uses machine learning to predict channels, traffic, states, and other performance indicators, raising questions about prediction reliability and horizon.Its research agenda asks whether wireless environments can be reliably predicted from past samples and over what future horizon.Source: §1 · p. 2
- xURLLC uses non-RF modalities such as RGB-D images for channel prediction, seeking to identify events such as blockages that RF-based solutions cannot capture under limited radio resources.The article also identifies RISs and metasurfaces as ways to tune channel randomness through surface reflections while inducing higher energy efficiency.Source: §1 · p. 2
- Communication-control co-design evaluates whether consecutive packet errors disrupt control and asks whether requirements can be relaxed while maintaining control stability.The article frames stability and safety against perturbations, state fluctuations, and inter-agent collision as central concerns.Source: §1 · p. 2
2 PREDICTIVE URLLC
Predictive URLLC uses data-driven forecasts to address extreme events in non-stationary environments, while its research agenda includes learning, prediction reliability, and perception-aware horizons. V2V AoI prediction and VR/AR network slicing illustrate these ideas.
2 PREDICTIVE URLLC
- xURLLC uses machine learning for proactive decisions in non-stationary and unpredictable environments, with examples including latency, AoI, and traffic-demand prediction from visuo-haptic perceptions.The article contrasts this with 5G URLLC’s static channel assumptions and its spectral-efficiency cost when characterizing extreme events.Source: §2 · p. 2
2.1 Challenges and Opportunities
- Reliable prediction requires models to be retrained when data distributions change; federated learning exchanges locally trained parameters to reduce the cost of predictive URLLC.Prediction reliability can be assessed through generalization error or Bayesian methods such as GPR, whose posterior variance provides prediction confidence; adversarial training addresses non-stationary distributions.Source: §2.1 · p. 2
- Prediction horizons trade available resources against accuracy, so perception-aware forecasting should minimize the horizon; human drivers’ perception-reaction time is around 2.5 s, versus several milliseconds for driverless cars.For multiple modalities, the article says perceptual relationships and integrated resolutions measured using JND should be considered.Source: §2.1 · pp. 2–3
2.2.1 Predictive AoI for Ultra-Reliable V2V Communication
- More samples sharpen the tail distribution of V2V AoI estimation error, while the GPR posterior variance that characterizes prediction reliability decreases.AoI prediction uses past AoI and resource-block selections to choose the next transmission power and resource block for a receiver’s bounded-AoI target.Source: §2.2.1 · p. 3, Figure 2 · p. 3
2.2.2 VR/AR Perception-Aware Proactive Network Slicing
- Proactive URLLC deactivation improves eMBB data rate under orthogonal and non-orthogonal slicing, particularly at high URLLC target data rates.The scenario deactivates the URLLC link when the predicted future field of view contains no touchable object; the prediction horizon is five frames, or 41.7 ms at 120 Hz.Source: §2.2.2 · pp. 3–4, Figure 3 · p. 3
3 NON-RF AIDED URLLC
xURLLC uses non-RF sensing alongside RF to predict rare channel events and support scalable communication. The section presents multimodal fusion and its energy overhead as key design challenges.
3 NON-RF AIDED URLLC
- Non-RF modalities such as RGB-D and cameras can provide features for predicting extreme events and blockage, with negligible signaling overhead and without sacrificing spectral efficiency.Proposed use cases include vision-based channel prediction and mobility management, and vision-aided coordination and control of robotic swarms.Source: §3 · p. 4
3.1 Challenges and Opportunities
- Split learning fuses modalities through data-specific lower NN segments and a shared upper segment; accelerometers and RIS-endowed walls are additional non-RF options.Non-RF acquisition, processing, and fusion consume energy that may negate modality gains, so energy efficiency should be examined.Source: §3.1 · p. 4
3.2.1 RGB-D Aided mmWave Received Power Prediction
- RF+Img achieves the highest accuracy and detects LoS/NLoS transitions, while Img minimizes extreme-error probability and RF+Img minimizes mean prediction error.RF-only and image-only baselines fail to accurately predict transitions or short-term channel fluctuations.Source: Figure 4 · p. 4, §3.2.1 · p. 5
- A split NN fuses image features and past received powers to predict 60 GHz mmWave received power 120 ms ahead from 30 Hz RGB-D video.Two convolutional layers extract image features, a single-layer NN processes received powers, and a recurrent layer predicts from their sequence.Source: §3.2.1 · p. 4
3.2.2 ML Based Energy-Efficient RIS
- The RIS scenario divides 64 elements among K three-layer NN controllers that infer binary phase shifts from user location using supervised training.The model is trained on 650 samples against exhaustive-search ground truth.Source: §3.2.2 · p. 5
- K = 2 maximizes energy efficiency by balancing the lower energy use of smaller RIS controllers against their reduced model capacity.The proposed method achieves higher energy efficiency than random phase shifting and nearly the same spectral efficiency as exhaustive search, without its energy dissipation.Source: §3.2.2 · p. 5
4 CONTROL CO-DESIGNED URLLC
Co-designed URLLC treats control stability and the value of individual packets as central to communication design. The section discusses challenges and illustrates ML-aided control for a remote UAV and an autonomous UAV swarm.
4 CONTROL CO-DESIGNED URLLC
- CoCoCo prioritizes whether packet errors disrupt control and uses packet importance and stability, rather than treating all transmission errors equally.Control dynamics offer an opportunity to relax stringent communication requirements; stability matters because communication can become useless after a device drifts beyond controllable states.Source: §4 · p. 5
4.1 Challenges and Opportunities
- Wireless links between sensors, controllers, and actuators can lose or distort state and command receptions; predicted poor channels may motivate relocating control to the actuator.The section also proposes ramping up controller transmission power when preceding sensor delay is too long, and predicting future states and commands from past ones.Source: §4.1 · p. 5
- Control stability can be underestimated or overestimated by communication reliability alone, and must be reconciled with reliability requirements across single-agent, swarm, and string control.The section describes BIBO stability for linear systems, Lyapunov stability for nonlinear systems, swarm stability through relative velocities, and string stability through bounded inter-agent spacing under finite disturbance.Source: §4.1 · pp. 5–6
- High-dimensional control, unknown dynamics, interacting agents, and human intervention challenge scalability; ML control, mean-field games, and transferring human knowledge are proposed directions.Mean-field games let agents interact with a locally estimated population distribution rather than exchanging all states.Source: §4.1 · p. 6
4.2.1 ML-Aided Single UAV Remote Control
- Switch UL maintains Lyapunov stability more frequently throughout travel and achieves shorter travel time than action UL or model UL.Action UL provides faster control cycles and more state observations, while model UL can autonomously control a faraway UAV after action UL loses control.Source: §4.2.1 · p. 7
4.2.2 ML Aided Massive Autonomous UAV Control
- Action+MF guarantees convergence given initial state exchange and avoids collisions while swarming faster than action-only control, even with small transmission power.Each UAV locally runs action and mean-field neural networks; the action network is penalized when maximum relative velocity exceeds a threshold.Source: §4.2.2 · p. 7
5 CONCLUSIONS
The conclusion presents xURLLC as a next-generation URLLC vision that moves beyond reactive, RF-based, communication-centric design. It argues that future designs should draw on ML, non-RF modalities, and control while accounting for their costs.
5 CONCLUSIONS
- xURLLC is a next-generation URLLC vision that moves beyond reactive, RF-based, communication-centric 5G design.The vision targets extreme-event prediction and scalability, adds diversity enhancements with minimal overhead, and uses information value to relax latency and reliability requirements.Source: §5 · p. 7
- Future xURLLC design should build on ML, non-RF, and control while factoring in the costs of those domains.The conclusion frames this within data-driven decision-making and predictions.Source: §5 · p. 7