Interference Prediction Using Gaussian Process Regression and Management Framework for Critical Services in Local 6G Networks
Syed Luqman Shah, Nurul Huda Mahmood, and Matti Latva-aho, Fellow, IEEE · 2025
Read Interference Prediction Using Gaussian Process Regression and Management Framework for Critical Services in Local 6G Networks by Syed Luqman Shah, Nurul Huda Mahmood, and Matti Latva-aho, Fellow, IEEE (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-13500.txt.
Introduction
Syed Shah and co-authors forecast radio interference to protect key services in future 6G networks in a 2025 paper. Local 6G cells must serve tasks with strict delay and trust demands. Knowing interference a step ahead lets the network shift resources before quality dips.
Their model treats interference as a smooth random process that also reports how unsure it is. A few fresh samples update its beliefs fast, unlike deep networks that need much data. A proactive scheme then hands out resources by the forecast. Tests beat a moving-average rival and nearly matched an ideal cheat that knows the future.
Catalog record
| CREATOR | Syed Luqman Shah, Nurul Huda Mahmood, and Matti Latva-aho, Fellow, IEEE |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-13500
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Interference Prediction Using Gaussian Process Regression and Management Framework for Critical Services in Local 6G Networks Thanks: The research leading to this paper was supported by the Research Council of Finland (former Academy of Finland) through the projects 6G Flagship (grant number: 369116), and 6G-ConCoRSe (grant number: 359850). Syed Luqman Shah, Nurul Huda Mahmood, and Matti Latva-aho, Fellow, IEEE Abstract Interference prediction and resource allocation are critical challenges in mission-critical applications where stringent latency and reliability constraints must be met. This paper proposes a novel Gaussian process regression (GPR)-based framework for predictive interference management and resource allocation in future 6G networks. Firstly, the received interference power is modeled as a Gaussian process, enabling both the prediction of future interference values and their corresponding estimation of uncertainty bounds. Differently from conventional machine learning methods that extract patterns from a given set of data without any prior belief, a Gaussian process assigns probability distributions to different functions that possibly represent the data set which can be further updates using Bayes’ rule as more data points are observed. For instance, unlike deep neural networks, the GPR model requires only a few sample points to update its prior beliefs in real-time. Furthermore, we propose a proactive resource allocation scheme that dynamically adjusts resources according to predicted interference. The performance of the proposed approach is evaluated against two benchmarks prediction schemes, a moving average-based estimator and the ideal genie-aided estimator. The GPR-based method outperforms the moving average-based estimator and achieves near-optimal performance, closely matching the genie-aided benchmark. I Introduction In sixth-generation (6G) wireless networks, critical services for applications such as the tactile Internet [ 1 ] , autonomous vehicles [ 2 ] , and industrial automation [ 3 ] demand stringent connectivity and reliability requirements. To address these requirements, IMT-2030 identifies the hyper-reliable low-latency communication (HRLLC) use case, which define stringent constraints on latency, specified within the range of [0.1\text{ms},1\text{ms}] , and reliability, targeted within the interval [10^{-5},10^{-7}] [ 4 ] . Meeting these stringent requirements is challenging, particularly in environments with high interference. Predictive interference management has been proposed as a key enabler for ensuring the quality of service (QoS) in 6G networks [ 5 ] . By forecasting interference reliably, networks can proactively allocate resources to maintain the required performance for mission-critical applications in an efficient manner [ 6 , 7 , 8 , 9 ] . Interference prediction and resource allocation for critical services are gaining attention, with two main approaches: stochastic methods and machine learning (ML)-based techniques. Stochastic methods, including conventional link adaptation, predict future interference using historical samples, often relying on mean values. Such an approach overlooks large fluctuations resulting in suboptimal predictions [ 10 ] . More sophisticated approaches, such as autoregressive moving-average (MA) models, improve accuracy by utilizing the entire interference distribution [ 8 , 11 ] . In [ 9 ] , a discrete-time Markovian interference model is developed to forecast interference state transitions, enabling proactive resource allocation for wireless critical applications. In vehicular networks, predictive models have been used to mitigate packet collisions by anticipating interference [ 12 ] . On the other hand, ML-based approaches particularly those using neural networks, improve prediction accuracy by leveraging large datasets to learn interference patterns. Interference-aware resource allocation for efficient deep learning (DL) on graphics processing units is explored in [ 13 ] , while models like long short-term memory have been applied to wireless interference prediction [ 7 ] . Further advancements include the use of deep neural networks (DNNs) with transformer architecture, as discussed in [ 6 ] . However, ML-based methods have three key drawbacks: they do not provide confidence levels for predictions, they cannot easily incorporate prior knowledge, and they typically require large training datasets. This work introduces a GPR-based framework for predicting interference power and enabling proactive resource allocation in local 6G deployments. Local 6G networks, which operate in confined environments such as industrial and healthcare settings, require HRLLC services due to their stringent application demands [ 6 ] . Compared to traditional stochastic models, which rely on oversimplified assumptions and DNN-based methods that require large datasets and lack interpretability, the GPR-based approach offers several key advantages [ 14 , 15 ] : (i) its non-parametric nature adapts to complex, non-linear interference dynamics without a predefined model structure; (ii) its data efficiency allows accurate predictions with minimal training data, ideal for data-constrained 6G environments [ 16 ] ; (iii) its real-time adaptability ensures dynamic updates as new data is available, responding to evolving interference conditions; (iv) its transparency and interpretability enable hyperparameter tuning and the integration of prior knowledge, overcoming limitations of black-box ML methods; and (v) its uncertainty quantification provides confidence intervals for predictions, facilitating informed resource allocation and meeting the rigorous service quality requirements of HRLLC applications [ 14 , 16 ] . The primary contributions of this work are threefold. First, we develop and implement a GPR-based model for predicting received interference power, specifically tailored for local 6G networks. Second, we integrate the predicted interference power into a proactive resource allocation framework, enabling efficient resource distribution to meet the stringent service requirements of mission-critical applications. Third, we conduct a performance evaluation of the proposed GPR-based model, comparing it against two benchmark methods MA-based estimation and genie-aided estimation, with results presented in terms of achieved outage versus target outage 1 1 1 To facilitate reproducibility, the simulation codes are publicly accessible at: https://github.com/Syed-Luqman-Shah-19/GPR-IEEEWCNC2025 . . Following the introduction, review of existing techniques, and outline of the contributions in Section I , the system model is detailed in Section II . Section III presents the non-parametric GPR model for interference power prediction. A proactive resource allocation strategy is introduced in Section IV . Section V presents the results, while Section VI discusses potential future directions and concludes the paper. Fig. 1 : Proposed indoor wireless system model depicting a desired signal and N interfering signals to a UE. II System Model We consider the downlink communication of a local 6G network deployed in an indoor mission-critical environment, such as a factory or industrial setting. The system comprises a serving access point (AP) communicating with user equipment (UE), surrounded by N interfering APs, all operating within the same frequency band, as illustrated in Fig. 1 . Due to the stationary nature of the devices in this setting, the system is assumed to be quasi-static during each transmission period. The communication channels are subject to multipath propagation, resulting in Rayleigh fading for all links due to reflections from surrounding structures. The N interfering APs introduce co-channel interference, which may degrade the signal quality received by the UE. The link between the serving AP and the UE is characterized by an average signal-to-noise ratio (SNR), ( \bar{\gamma} ). Each interfering AP contributes an interference-to-noise ratio (INR), ( \gamma_{i} ), for i=1,2,\dots,N , with maximum and minimum values \gamma_{\text{max}} and \gamma_{\text{min}} , respectively. We assume that the UE connects to the AP with the strongest signal, where \bar{\gamma}>\gamma_{\text{max}} , and that there is no cooperation among the interfering APs (e.g., power control or coordinated beamforming). The received signal at the UE, R_{r}(t) , is expressed as, R_{r}(t)=\sqrt{P_{d}}h_{d}(t)s_{t}(t)+\sum_{i=1}^{N}\sqrt{P_{i}}h_{i}(t)s_{i}(t)+n(t), (1) where P_{d} is the transmitted power from the serving