Wireless channel modelling is becoming a decisive enabler for the successful realization of 6G networks. Unlike previous generations, 6G envisions ultra-dense deployments, operation across sub-6 GHz, mmWave, and sub-THz bands, integrated sensing and communication, and intelligent, environment-aware services. These ambitious targets significantly increase channel complexity due to higher frequencies, wider bandwidths, extreme mobility scenarios, and heterogeneous deployment environments spanning terrestrial, aerial, and satellite platforms. Conventional models struggle to scale under such diversity and lack predictive capability for unseen environments and future states, which is critical for proactive resource allocation and adaptive network control in 6G. In this context, AI/ML techniques are employed to learn latent representations of the channel environment directly from high-dimensional channel state information (CSI), capturing both large-scale effects and small-scale fading characteristics. Channel charting embeds these measurements into a low-dimensional space that preserves spatial relationships, enabling relative user localization without explicit positioning. Uncertainty estimation and Deep Reinforcement Learning (DRL) enhance robustness and adaptively refine channel and location predictions in dynamic conditions.
6G-Leader approach
In the 6G-Leader project, DRL is explored to predict wireless channel availability without relying on explicit occupancy estimation. This approach aligns with Europe’s strategic vision for intelligent, energy-efficient, and self-optimizing 6G infrastructures.
The proposed methodology enables a DRL agent to learn optimal transmission behavior directly from interaction with the wireless environment. Instead of depending on traditional channel-state models or supervised predictors, the agent infers availability implicitly through its learned policy. The system integrates reduced Channel State Information (CSI), Signal-to-Noise Ratio (SNR)/Interference-to-Noise Ratio (INR) measurements, and collision history into a Markov Decision Process (MDP), enabling robust decision-making in dynamic radio conditions.
For validation, the model leverages the DeepMIMO dataset, which provides realistic propagation, user mobility, and multi-user interference. This innovation contributes directly to Europe’s 6G priorities, including autonomous spectrum access, energy-aware operation, and scalable multi-agent intelligence.
Future developments include multi-agent reinforcement learning extensions, RIS integration, beamforming-aware policies, and experimental validation on real testbeds. These advancements will support the creation of resilient, adaptive, and sustainable 6G networks across Europe.
