Future 6G networks should not only react after connections degrade, they should anticipate risk before service is interrupted.
Have you ever been on an important video call or playing an online game while traveling, only for the connection to suddenly freeze or drop? As we move toward ultra-fast 6G wireless technology, keeping mobile applications running smoothly while users move across changing network zones remains a major challenge.
One reason for these interruptions is that many current network decisions rely on a single estimate of future performance. When the network decides whether to keep a user connected to the current local server or move them to another one, that decision may be based on one predicted value. If this prediction is too optimistic, or if a short burst of interference occurs, the system may react too early, too late, or trigger unnecessary server changes. This can lead to avoidable signaling, degraded performance, or temporary service disruption.
Researchers from Technische Universität Braunschweig propose a new risk-aware approach to this problem. Instead of relying only on a single prediction, the framework estimates a range of possible future network speeds and calculates a calibrated probability that the connection may fall below the required service level. This allows the network to make decisions with a clearer understanding of uncertainty.
A simple way to think about it is a weather forecast. A basic forecast might say, “Tomorrow will be sunny.” A more useful forecast says, “The temperature will likely be between 20 and 24 degrees, with a 30% chance of rain.” In the same way, the proposed system does not only ask, “How fast will the connection be?” It also asks, “How confident are we, and how likely is a slowdown?”
Using this information, the network’s optimizer can decide whether to keep the current connection, switch to a better local server, or avoid risky choices that may soon degrade. The goal is to maintain smooth service while avoiding unnecessary server switching.
A key strength of the work is that it is designed to fit within existing 3GPP and ETSI MEC standards. The controller can run as an edge application and use existing mechanisms for network information, policy control, traffic steering, and application mobility. This makes the approach more practical for future deployment.
The evaluation shows that the risk-aware controller achieves higher expected throughput than a controller based only on single-value predictions. It also gives network operators a tuning knob to balance speed and reliability.
As 6G networks become more dynamic, this work shows that service continuity should not depend only on reacting after problems occur. By considering uncertainty and risk before making decisions, future networks can become more stable, adaptive, and reliable for users.
Publication Details
Conformal Prediction and Risk-Based Optimization of Service Continuity in 6G Edge Networks
Authors: Zied Ennaceur, Admela Jukan
MARE Partner: Technische Universität Braunschweig (TUBS)
Publication: IEEE International Conference on Communications (ICC) , Glasgow, Scotland, UK, 24–28 May 2026. Conference website.
Abstract: We study the problem of service continuity in 6G networks and propose a solution aligned with current mobile network standards that integrates conformal prediction and risk-aware throughput estimation to anticipate short-term degradations and, in addition, to optimize application and session continuity during network topology changes. Our pipeline combines throughput prediction over short horizons with calibration based on split conformal prediction to produce sharp intervals with coverage guarantees, and a throughput degradations estimator that provides the probability that throughput falls below the SLA threshold on each link. A risk-aware optimizer then selects associations that balance expected throughput, reliability, and continuity. Experiments on emulated mobile network traces demonstrate that the performance improves by incorporating uncertainty and calibrated risk estimation. Experiments on emulated mobile network traces demonstrate that the degradation risk model is well calibrated with a Brier score of 0.095, an Expected Calibration Error of 0.068, and a PR–AUC of 0.810, and that incorporating uncertainty and calibrated risk improves control decisions. The risk-aware controller consistently attains higher expected throughput than the commonly used point-estimate baseline at matched reassociation rates, and it can identify network operating regimes with low SLA degradation, e.g., 7.9% violations at 999 Mb expected throughput for ε = 0.40.
