Statistical Characterization and Prediction of E2E Latency over LEO Satellite Networks

TL;DR

The paper addresses end-to-end latency variability in LEO satellite networks by exploiting Starlink's deterministic $15\text{s}$ periodicity. It deploys a high-rate $500~Hz$ measurement setup to segment periods, isolate handover boundary spikes of about $140~ms$ at the start and $75~ms$ at the end, and model the intra-period core with lightweight parametric and non-parametric methods, achieving 99th-percentile prediction errors below $50~ms$ after short sampling. By combining period-level prediction and classification (Good vs Degraded) with a tailored discounted service availability metric, the work demonstrates practical adaptive strategies to maintain QoS or switch to alternative interfaces. The findings offer a statistically grounded middleware blueprint for latency-aware operation in NTN/LEO networks and hold promise for generalizing to other constellations and transport protocols.

Abstract

Low Earth Orbit (LEO) satellite networks are emerging as an essential communication infrastructure, with standardized 5G-based non-terrestrial networks and their integration with terrestrial systems envisioned as a key feature of 6G. However, current LEO systems still exhibit significant latency variations, limiting their suitability for latency-sensitive services. We present a detailed statistical analysis of end-to-end latency based on 500Hz experimental bidirectional one-way measurements and introduce a segmentation of the deterministic 15-second periodic behavior observed in Starlink. We characterize handover-induced boundary regions that produce latency spikes lasting approximately 140 ms at the beginning and 75 ms at the end of each cycle, followed by a stable intra-period regime, enabling accurate short-term prediction. This analysis shows that latency prediction based on long-term statistics leads to pessimistic estimates. In contrast, by exploiting the periodic structure, isolating boundary regions, and applying lightweight parametric and non-parametric models to intra-period latency distributions, we achieve 99th-percentile latency prediction errors below 50 ms. Furthermore, period-level latency prediction and classification enable adaptive transmission strategies by identifying upcoming periods where application latency requirements cannot be satisfied, necessitating the use of alternative systems.

Figures (8)

• Figure 1: Diagram of the latency testbed at Aalborg University.
• Figure 2: Latency measurements for UL transmission. The beginning of the first period is set as time $t=0$, and the beginning of each period is indicated by a dashed line. Latency measurements for different periods are indicated with different colors.
• Figure 3: Mean-centered instantaneous uplink latency averaged over multiple 15 second intervals. The red areas, first 140 ms and last 75ms, are the periods affected by a sharp latency increase.
• Figure 4: Violin plot showing the intra-period latency for , , and across three different periods.
• Figure 5: Intra-period latency 95th and 99th quantiles across multiple periods for , , and .