Abstract
Ultra-reliable and low-latency communications (URLLC) are considered as one of three new application scenarios in the fifth generation cellular networks. In this work, we aim to reduce the user experienced delay through prediction and communication co-design, where each mobile device predicts its future states and sends them to a data center in advance. Since predictions are not error-free, we consider prediction errors and packet losses in communications when evaluating the reliability of the system. Then, we formulate an optimization problem that maximizes the number of URLLC services supported by the system by optimizing time and frequency resources and the prediction horizon. Simulation results verify the effectiveness of the proposed method, and show that the tradeoff between user experienced delay and reliability can be improved significantly via prediction and communication co-design. Furthermore, we carried out an experiment on the remote control in a virtual factory, and validated our concept on prediction and communication co-design with the practical mobility data generated by a real tactile device.
Introduction
The advent of 5G has introduced Ultra-Reliable Low-Latency Communication (URLLC) as a cornerstone for the "Tactile Internet," enabling applications like remote surgery, autonomous driving, and industrial automation. However, achieving near-instantaneous communication remains a significant challenge. Even with high-speed wireless interfaces, total end-to-end latency is often dominated by additive delays from data buffering, backhaul congestion, and core network processing.
In traditional systems, if the network takes 20ms to deliver a packet, the user experiences a 20ms lag. To overcome this, we propose a "co-design" strategy that utilizes state prediction to mask network delays. By predicting a device's future state and transmitting it early, the receiver can obtain the "current" data exactly when it is needed, effectively achieving a "zero-latency" experience.
Masking Delay via Prediction
The core of this framework is the joint optimization of state prediction and communication resources. A device uses a linear state-transition model to predict its future state (e.g., position or velocity) based on current observations. This predicted state is then sent through the network.
As shown in the timing diagram below, the prediction horizon ($T_n^p$) is designed to offset the communication delay ($D_n^c$). If $T_n^p$ is set equal to $D_n^c$, the user-experienced delay ($D_n^e$) can theoretically reach zero. However, this introduces a fundamental tradeoff: predicting further into the future increases the likelihood of prediction error, while shorter horizons increase the perceived latency.
Reliability and Optimization
To maintain the extreme reliability required for URLLC (error rates between $10^{-5}$ and $10^{-8}$), we define "overall reliability" as the sum of three distinct error probabilities:
- Prediction Error ($\epsilon^p$): The probability that the predicted state deviates beyond a tolerable threshold.
- Queueing Delay Violation ($\epsilon^q$): The probability that a packet remains in the buffer longer than its allocated time.
- Transmission Error ($\epsilon^t$): The probability of packet loss or corruption over the wireless link, modeled using finite blocklength theory for short packets.
We formulate an optimization problem to maximize the number of supported URLLC users by balancing bandwidth allocation and the prediction horizon. Since the problem is non-convex, we employ a near-optimal algorithm using binary search to find the ideal parameters for each device.
Experimental Validation and Results
The framework was validated using both simulated mobility models and real-world experimental data. For the latter, we used a 3D System Touch haptic device to record human-generated movement in a virtual factory setting. This confirmed that linear prediction models are effective for practical tactile data.
Key findings include:
- Significant Delay Reduction: The co-design method reduced user-experienced delay by 23 ms in simulations compared to traditional non-predictive methods while maintaining a $10^{-5}$ reliability target.
- Spectral Efficiency: Increasing the number of antennas at the access point proved highly effective. Doubling the antennas from 32 to 64 allowed the system to save 75% of the required bandwidth for the same number of users.
- Prediction as a Resource: The results demonstrate that the prediction horizon can be treated as a system resource, similar to power or bandwidth, to improve the overall delay-reliability tradeoff.
Conclusion
This work demonstrates that prediction and communication co-design is a viable path toward achieving zero-latency URLLC. While linear models proved effective for simple motion, future work will explore non-linear prediction methods, such as Deep Neural Networks, to handle more complex robotic and biological movements in dynamic environments.
Summary updated