AI-Based UE Speed Estimation Using Uplink SRS Measurements

The patent replaces traditional spectral analysis methods (mechanical signal processing) with machine learning-based classification. The ML model learns patterns from channel state information and directly classifies speed categories, substituting complex mathematical transformations with trained classifiers that achieve similar or better accuracy with reduced computational burden during operation.

The patent performs preliminary training of machine learning models offline using labeled speed data and channel measurements. This preliminary action creates pre-trained classifiers that can be deployed in the network, transferring the computational burden from real-time operation to offline model development, thereby reducing online computational complexity.

The machine learning model performs self-service by automatically learning the relationship between channel state information and UE speed from training data. The model adapts to different network conditions and system parameters during training without requiring manual configuration or prior knowledge of specific system parameters, enabling it to serve various 5G-NR scenarios universally.

The patent changes the approach from fixed parameter-based estimation to adaptive parameter learning. The ML model learns optimal parameters and patterns from training data across different scenarios, allowing the system to adapt to varying network conditions, frequency bands, and deployment environments without requiring reconfiguration of system parameters.

Instead of performing complete beam scanning across all possible beams, the patent uses partial action by leveraging UE speed classification to predict which beams are likely to be optimal. The system performs beam management on a subset of candidate beams identified through speed-based prediction, reducing the time required while maintaining sufficient coverage for mobility management.

The patent implements feedback by using UE speed classification results to inform beam selection and prediction. The speed information feeds into beam prediction algorithms that anticipate future channel conditions and select appropriate beams in advance, reducing the need for exhaustive beam sweeping and enabling faster beam alignment during mobility events.