Dynamic Interference Prediction for In-X 6G Sub-networks

The sixth generation (6G) industrial Sub-networks (SNs) face several challenges in meeting extreme latency and reliability requirements in the order of 0.1-1 ms and 99.999 -to-99.99999 percentile, respectively. Interference management (IM) plays an integral role in addressing these requirements, especially in ultra-dense SN environments with rapidly varying interference induced by channel characteristics, mobility, and resource limitations. In general, IM can be achieved using resource allocation and \textit{accurate} Link adaptation (LA). In this work, we focus on the latter, where we first model interference at SN devices using the spatially consistent 3GPP channel model. Following this, we present a discrete-time dynamic state space model (DSSM) at a SN access point (AP), where interference power values (IPVs) are modeled as latent variables incorporating underlying modeling errors as well as transmission/protocol delays. Necessary approximations are then presented to simplify the DSSM and to efficiently employ the extended Kalman filter (EKF) for interference prediction. Unlike baseline methods, our proposed approach predicts IPVs solely based on the channel quality indicator (CQI) reports available at the SN AP at every transmission time interval (TTI). Numerical results demonstrate that our proposed approach clearly outperforms the conventional baseline. Furthermore, we also show that despite predicting with limited information, our proposed approach consistently achieves a comparable performance w.r.t the off-the-shelf supervised learning based baseline.

💡 Research Summary

The paper addresses the stringent latency (0.1–1 ms) and reliability (99.999 %–99.99999 %) requirements of sixth‑generation (6G) industrial sub‑networks (In‑X SNs). While interference management (IM) can be achieved through resource allocation or accurate link adaptation (LA), the authors focus on improving LA by predicting interference power values (IPVs) at the access point (AP) using only the channel quality indicator (CQI) reports that are already available in each transmission time interval (TTI).

First, the authors model inter‑SN interference using the spatially consistent 3GPP channel model. The model incorporates line‑of‑sight (LOS) and non‑LOS components, log‑normal shadowing, Rician/ Rayleigh small‑scale fading, and an ON‑OFF traffic process. This yields a realistic expression for the interference power seen by a UE, capturing the rapid variations caused by mobility, dense deployment, and limited resource allocation.

Next, they formulate a discrete‑time Gaussian dynamic state‑space model (DSSM) in which the IPV is treated as a latent state and the CQI‑derived SINR estimate is the observation. The state transition function F(·) and observation function G(·) are inherently nonlinear, which prevents direct application of a standard Kalman filter. To enable an extended Kalman filter (EKF), two key approximations are introduced.

1. Transition Approximation: The temporal correlation of the small‑scale fading is approximated by a zeroth‑order Bessel function B₀(ωτ), where ω is the Doppler spread and τ the TTI duration. This yields a correlation coefficient α_F ∈ (0,1) that is used to form a convex combination of the two most recent IPV estimates. The resulting linearized transition model ˜F(·) captures the inertia of interference while remaining computationally simple.

2. Observation Approximation: The observation function is linearized by taking the partial derivative of the SINR expression with respect to the IPV. Since the UE’s signal power S is not directly known at the AP, an average or pre‑defined mapping is substituted. Measurement noise is modeled as the sum of two independent Gaussian components: one representing the effective‑SNR‑mapping (ESM) compression loss and the other the CQI quantization error. This composite noise ˜u(t) is characterized by a known variance.

With these approximations, the EKF proceeds in the usual predict‑update cycle: the predicted IPV is obtained from the linearized transition, and the predicted observation is compared with the actual CQI‑derived SINR to update the state estimate and its covariance. The algorithm runs online at every TTI, requiring only the CQI feedback already present in the protocol stack, thus incurring no extra signaling overhead.

The authors evaluate the approach through extensive simulations that follow the 6G‑ANNA and Open6GHub grant parameters, using the 3GPP TR 38.901 channel specifications. Baselines include a simple moving‑average filter, a Wiener filter, and a supervised LSTM model trained on the same CQI inputs (the latter representing a state‑of‑the‑art data‑driven method). Results show that the EKF‑based predictor achieves significantly lower mean‑square error (MSE) and faster convergence than the baselines. Even without ground‑truth labels, its performance is within about 5 % of the LSTM, while offering far lower computational complexity and no need for label acquisition.

The paper’s contributions are threefold:

1. Accurate Interference Modeling: By integrating the spatially consistent 3GPP channel model with an ON‑OFF traffic model, the authors provide a realistic representation of inter‑SN interference dynamics.

2. Latent‑State DSSM with EKF Approximation: They formulate a Gaussian DSSM where IPV is the hidden state and develop practical linearization techniques that enable efficient EKF implementation.

3. CQI‑Only, Label‑Free Prediction: The proposed method predicts interference solely from CQI reports, eliminating extra signaling and training‑data requirements, and thereby supporting the ultra‑low‑latency, ultra‑reliable communication (HRLLC) goals of 6G industrial networks.

Overall, the work demonstrates that an AP‑centric, EKF‑based interference predictor can substantially improve link adaptation decisions in dense industrial 6G deployments, offering a viable path toward meeting the most demanding latency and reliability targets without incurring the overhead of more complex machine‑learning solutions.