A simulation study reports that a new control framework supported up to 50 simultaneous public-safety MCX users while meeting requirements for delivered rate, delay, jitter, information freshness and terminal non-delivery. The same tests found that ordinary mobile traffic stayed within a configured protection limit at the framework’s largest supported MCX load.
The work is an arXiv preprint, version v1, dated 26 Aug 2026. The evaluation uses simulated service and packet-delivery trajectories for MCX and ordinary mobile services in a shared O-RAN network, where both compete for the same resources.
A short-horizon view of network capacity
The framework estimates how much service the network can execute over a finite period, rather than relying only on a long-run average. It combines network states observed through the O-RAN E2 interface with delays between observation and control action, and it accounts for connectivity diversity while considering correlation between network paths.
The calculation and the real-time decision are separated. Capability estimates and assurance certificates are produced offline. At runtime, the Near-RT RIC, the network’s near-real-time control component, solves a finite profile-selection problem to choose among available operating profiles.
The paper also gives a formal proposition saying that the proposed finite-horizon effective-capacity framework becomes consistent with conventional effective capacity over long horizons. In practical terms, the model is designed to retain information about the network’s starting condition and control delay when the decision window is short, then narrow the gap with long-run estimates as the window grows.
State information mattered most when time was short
In one short-horizon comparison, the favorable and degraded capability curves differed by about 0.88 Mbit/s. As the horizon lengthened, the difference between the finite-horizon and conventional effective-capacity estimates decreased, with the proposed measure gradually converging toward conventional effective capacity.
The model also showed a clear sensitivity to stale information. Capability loss increased as the age of the E2 observation increased, while the configuration with rich E2 exposure showed the smallest loss among the exposure settings tested.
For assurance checks, running several profiles concurrently without constraints pushed the observed risk above the stored bound as more profiles became active. Execution that preserved the calibrated contract kept the diagnostic ratio at or below one, the condition reported by the paper as indicating that the stored bound remained valid.
Protection for ordinary traffic came with trade-offs
The proposed method’s largest supportable MCX load kept non-MCX utility loss below the configured protection threshold. The Static-QPP comparison faced a different problem: it either breached the protection threshold at high load or held back more resources than necessary at low load.
Connectivity choices changed the balance between resilience and capacity. Always-DP, which sends duplicate copies, was reported to provide greater delivery diversity under severe independent degradation, but it admitted fewer flows during normal operation because every duplicate uses radio resources.
Under O-DU degradation, the abstract reports greater MCX supportability when connectivity was selected adaptively, while the configured multi-QoS and non-MCX protection requirements were satisfied. No numerical effect size is given in the supplied analysis for that comparison.
What the simulations do not establish
The evaluation used a finite-state Markov-additive wireless service model and packet-level simulations. MATLAB was used for capability and profile computation, while ns-3 with 5G-LENA handled packet-level delivery; the comparison schemes used the same radio resources and packet-level simulator.
The results therefore depend on the assumed traffic, channel, failure and execution-class conditions. The paper does not report exact trajectory counts, repetitions, random seeds or per-condition sample sizes, and most plotted values and confidence budgets are not tabulated. The post-selection guarantee is conditional on exchangeability between calibration and runtime execution classes and on those execution conditions remaining within the stored envelope.
The theorem states that, with calibration confidence at least one minus the two stated error budgets, every selected nonzero profile simultaneously satisfies the listed calibrated guarantees. That is a statement about the model and its calibration conditions, not an estimate of real-world probabilities for every public-safety network or every raw E2 reading.
The framework is presented as a unified way to turn capability estimates into executable O-RAN profiles that can respond to changing conditions while protecting ordinary traffic. Its evidence comes from analytical derivations, formal results and shared-O-RAN simulations.
The authors report support from the National Natural Science Foundation of China under Grant 62341132 and the Natural Science Basic Research Program of Shaanxi under Grant 2024JC-YBQN-0642.
Paper data and sources
Original title: E2-Conditioned Finite-Horizon Effective Capacity for Public-Safety MCX over Shared O-RAN
Authors: Jingqing Wang, Wenchi Cheng
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text