AI-Native PHY-Layer in 6G Orchestrated Spectrum-Aware Networks

Area of Science:

• Telecommunications Engineering
• Artificial Intelligence
• Signal Processing

Background:

• The transition to 6G wireless networks requires a move towards AI-native orchestration within the radio access network (RAN).
• Current systems often rely on higher-layer signaling for network parameter identification, which can be inefficient.

Purpose of the Study:

• To develop AI-based enablers for physical (PHY)-layer awareness in 6G networks.
• To enable intrinsic intelligence within the RAN for improved spectrum awareness and system adaptability.

Main Methods:

• Development of an AI-based waveform classifier to differentiate between Orthogonal Frequency-Division Multiplexing (OFDM) and Orthogonal Time Frequency Space (OTFS) signals using in-phase/quadrature (IQ) samples.
• Implementation of an AI-based numerology detector to ascertain parameters like subcarrier spacing, FFT size, slot duration, and cyclic prefix type without higher-layer information.

Main Results:

• Waveform classification achieved 99.5% accuracy.
• Numerology detection exceeded 99% accuracy for most parameters.
• Demonstrated robust joint inference of waveform and numerology features, confirming AI-native spectrum awareness feasibility.

Conclusions:

• AI-native PHY-layer awareness is feasible for 6G wireless systems.
• The developed enablers pave the way for self-optimizing, context-aware, and adaptive 6G networks.
• Intrinsic intelligence in the RAN enhances spectrum utilization and system performance.