Introduction
In the evolving landscape of 5G and 6G, Ultra-Reliable Low-Latency Communications (URLLC) are critical for applications such as autonomous driving, industrial automation, and remote surgery. Traditional communication models often assume that packets are processed sequentially. However, in high-speed control systems, a new message (Message 2) may arrive at the transmitter while a previous message (Message 1) is still being transmitted. This creates a "heterogeneous" scenario where messages have staggered arrival times and overlapping decoding deadlines.
Standard information-theoretic limits, which rely on the assumption of infinitely long blocklengths, are insufficient for URLLC because these applications utilize very short packets. This paper addresses the challenge of determining the "second-order" achievable rates—a more precise performance metric for finite blocklengths—for broadcast channels under these complex timing constraints.
System Model and Proposed Methodology
The authors consider a scenario where two messages, $m_1$ and $m_2$, arrive at times $a_1$ and $a_2$ and must be decoded by $d_1$ and $d_2$, respectively. To handle the overlap, they propose a three-phase encoding strategy based on an adaptation of Marton’s coding strategy for the finite-blocklength regime.
As shown in the figure above, the transmission is partitioned into three distinct intervals:
- Phase 1 ($t \in [a_1 : a_2 - 1]$): Only the first message $m_1$ is transmitted.
- Phase 2 ($t \in [a_2 : d_1]$): Both messages are available. The transmitter uses joint encoding (Marton's strategy) to send information for both receivers simultaneously.
- Phase 3 ($t \in [d_1 + 1 : d_2]$): After the deadline for $m_1$ has passed, the transmitter sends only the remaining information for $m_2$.
Key technical innovations include the use of power-shell codebooks, where codewords are distributed uniformly on a multi-dimensional sphere to provide tighter reliability bounds than standard Gaussian codebooks. The authors derive second-order rate bounds that follow the normal approximation format: $R \approx C - \sqrt{V/n}Q^{-1}(\epsilon)$, where $V$ is the channel dispersion and $\epsilon$ is the error probability.
Key Insights and Results
The research demonstrates that joint encoding during the overlap period significantly outperforms traditional time-sharing methods. By exploiting the transmitter's knowledge of both messages, the scheme effectively manages interference, similar to the principles of Dirty Paper Coding (DPC).
Numerical simulations in additive white Gaussian noise (AWGN) channels validate the theoretical findings. The proposed scheme's performance is compared against the Meta-Converse bound (the absolute theoretical limit) and the standard normal approximation for point-to-point channels.
As illustrated in the performance plot, the proposed coding scheme (red dashed line) closely tracks the converse bound and significantly improves upon simpler approximations as the blocklength $n_{1,1}$ increases. Furthermore, the authors show that as the number of channel uses grows, their finite-blocklength bounds converge to the asymptotic Marton’s inner bound, confirming the optimality of the approach in the long-blocklength limit while maintaining robustness for short-packet URLLC applications.
Conclusion and Future Work
This work provides a rigorous framework for analyzing broadcast channels with staggered deadlines, offering a concrete way to guarantee reliability (e.g., $10^{-5}$ error rates) in latency-sensitive environments. While the current study focuses on Gaussian channels and two-user scenarios, future research will explore the impact of channel fading and the scalability of these joint encoding strategies to $K$-user systems with multiple overlapping arrivals.
Summary updated