RAN Training Data Collection Through Periodic DCMF Control

The patent implements discontinuous collection of training data by introducing periodic data collection cycles controlled by a data collection management function. Instead of continuous data collection from multiple rollout workers, the system collects data in periodic intervals, allowing processing and deployment of model updates between collection periods. This periodic action reduces the latency in model updates while still accumulating sufficient training data over time.

The patent implements dynamic control of data collection and model update deployment through a data collection management function. The system dynamically adjusts the frequency and timing of data collection based on training progress, data quality metrics, and system conditions. This dynamic approach allows the system to optimize training efficiency while maintaining stability by adapting update frequencies rather than using fixed frequent updates.

The patent implements a universal data collection management function that handles data collection from multiple rollout workers through a single, standardized interface and process. The data collection management function serves multiple purposes: coordinating data collection across workers, managing data aggregation, controlling collection timing, and orchestrating model updates. This multi-functional approach enables diverse data collection from multiple sources without proportionally increasing system complexity.