Beam Prediction Data Collection for Low-Overhead CSI Reporting

The patent applies preliminary action by collecting measurement data and training AI/ML models in advance during idle periods or when channel conditions are stable. The network node accumulates CSI measurements from multiple wireless devices and trains prediction models before they are needed for actual beam prediction, so that when prediction is required, the model is already ready and no additional real-time processing energy is consumed.

The system implements self-service by having the network node automatically collect measurement data, train models, and update prediction algorithms without requiring manual intervention. The AI/ML model continuously learns from incoming measurement data and adapts to changing channel conditions autonomously, reducing the need for external configuration and maintenance energy overhead.

The patent extracts only the essential measurement data needed for effective beam prediction. Instead of collecting and transmitting all possible CSI measurements for every DL RS resource, the system identifies and collects only the most relevant measurements that contribute to prediction accuracy, filtering out redundant information to reduce overhead while maintaining precision.

The system performs partial measurements by selecting a subset of DL RS resources for measurement based on prediction requirements. The AI/ML model is trained to make accurate predictions using measurements from fewer resources than would be needed for exhaustive measurement, achieving sufficient precision with reduced measurement and reporting overhead.

The patent performs preliminary measurements and model training in advance during periods when full measurements can be taken without time pressure. The network node accumulates measurement data from multiple sources and trains prediction models during idle periods or when channel conditions are stable, so that when actual beam prediction is needed, the model is already trained and can provide rapid predictions without requiring time-consuming comprehensive measurements.

The measurement strategy is made dynamic by adapting the measurement scope and frequency based on current channel conditions, mobility patterns, and prediction requirements. When channel conditions are stable, the system uses previously trained models with minimal measurements. When conditions change rapidly, the system increases measurement activity to update the model, optimizing the balance between precision and delay in real-time.