5G • 5G-Advanced (Rel-18)
AI/ML Beam Prediction
Spatial-Temporal AI/ML Beam Tracking & Prediction
Predicts optimal future mmWave and Sub-THz narrow beams in time and frequency using recurrent neural networks, mitigating blockage without frequent CSI-RS sweeping.
Technical Explanation
High-frequency mmWave and sub-THz beams suffer from severe penetration loss and sudden blockage from obstacles and user body movements. Rel-18 AI/ML beam prediction processes historical beam measurements (SSB-RSRP and CSI-RSRP) using recurrent neural networks (RNNs) and transformers to anticipate channel degradation, predicting optimal target beams before link degradation occurs.
Key Functions
- Forecasts future beam signal quality based on historical trajectory and RSRP trends
- Eliminates beam sweeping overhead by proactively identifying candidate beam pairs
- Mitigates abrupt human body and structural blockage in FR2 and FR3 bands
- Decreases latency for beam failure recovery from tens of milliseconds to sub-frame intervals
Specifications
3GPP TR 38.843, TS 38.331 (Rel-18)
Interfaces
UuMAC CE
Related 5G Concepts
Want to memorize 5G concepts like this one?
Study it with SuperMemo SM-2 spaced repetition flashcards.