UE DRX and ML Life Cycle Coordination for Power-Reliability Tradeoffs

The patent implements dynamic adjustment of DRX cycle configurations based on ML life cycle stage requirements. During critical ML phases such as data collection and model training, the system dynamically modifies DRX parameters to ensure continuous operation, while allowing more aggressive DRX settings during stable operational phases. This dynamic adaptation resolves the contradiction by making the DRX behavior flexible rather than static, allowing power saving when safe and reliability when needed.

The system changes physical and operational parameters of the DRX cycle based on ML model requirements. Specifically, it adjusts DRX cycle lengths, on-duration timers, and offset values according to the current ML life cycle stage. During data collection phases, parameters are modified to ensure continuous monitoring, while during inference phases, more aggressive power-saving parameters can be applied. This parameter adaptation directly addresses the contradiction between power saving and ML reliability.

The system performs preliminary actions by proactively adjusting DRX configurations before critical ML operations begin. When the ML model enters data collection or training phases, the system pre-configures the DRX cycle to maintain continuous or near-continuous operation, ensuring that data collection is not disrupted. This preliminary adaptation prevents the degradation of predictive accuracy while still allowing energy-efficient operation during non-critical phases.

The system implements feedback mechanisms where the ML model's performance metrics and data collection status are continuously monitored. Based on this feedback, the DRX cycle configuration is dynamically adjusted. If predictive accuracy begins to deteriorate due to DRX-induced data collection delays, the system receives feedback and modifies DRX parameters to restore accurate measurements. This closed-loop control resolves the contradiction by using feedback to maintain predictive accuracy while maximizing energy efficiency.

The patent segments the ML life cycle into distinct phases (data collection, model training, validation, inference) and applies different DRX cycle configurations to each phase. This segmentation allows the UE to autonomously manage DRX settings based on the current phase, improving operational flexibility. The complexity is managed by providing clear phase-based rules rather than requiring complex real-time optimization algorithms.

The UE is empowered to autonomously manage its own DRX and ML cycle coordination without requiring constant network intervention. The device uses self-service mechanisms where it automatically detects ML life cycle stages and adjusts DRX parameters accordingly, reducing the need for complex network-UE coordination protocols. This self-management capability improves operational flexibility while keeping the implementation complexity manageable through rule-based autonomous decision-making.