5G • 5G-Advanced (Rel-18)
Sub-Band AI Beamforming
Frequency-Selective Neural Precoding Across Sub-Bands
Applies deep learning models to predict optimal frequency-dependent analog and digital precoding vectors across wideband channels with high beam squint.
Technical Explanation
In wideband mmWave channels, phase shifters introduce 'beam squint' where beams point in slightly different directions at carrier edges compared to the center frequency. Rel-18 sub-band AI beamforming uses neural networks to calculate true-time-delay and digital precoding matrices per sub-band, preserving maximum antenna gain across the full 400 MHz carrier width.
Key Functions
- Compensates for wideband beam squint phenomena in mmWave antenna arrays
- Generates customized frequency-selective beamforming weights per resource block group
- Maximizes antenna array directivity across ultra-wideband carrier channels
- Improves cell-edge throughput in high-bandwidth FR2 and FR3 bands
Specifications
3GPP TR 38.843, TS 38.214 (Rel-18)
Interfaces
Uu
Want to memorize 5G concepts like this one?
Study it with SuperMemo SM-2 spaced repetition flashcards.