Uplink Link Adaptation Using Q-Learning for MCS Prediction

The patent transforms the static link adaptation system into a dynamic one by implementing a machine learning model that continuously learns from transmission outcomes (ACK/NACK feedback) and adapts MCS selections to specific cell environments and UE characteristics. The system evolves its decision-making process over time, moving from fixed look-up tables to adaptive predictions based on learned patterns.

The patent changes the fundamental parameter selection approach from static table-based indexing to dynamic machine learning predictions. The system learns optimal MCS values by analyzing relationships between channel conditions, UE parameters, and transmission outcomes, allowing flexible adaptation to varying conditions rather than relying on pre-defined static mappings.

The patent introduces a machine learning model as an intermediary between raw input features (SINR, UE parameters, transmission type) and MCS selection. This intermediary processes and integrates multiple information sources, learning complex relationships and patterns that static tables cannot capture, thereby reducing information loss while maintaining reasonable computational requirements.

The machine learning model serves multiple functions simultaneously: it processes SINR estimates, incorporates UE-specific parameters, accounts for transmission type (new vs. retransmission), and learns cell-specific patterns. This multi-functional approach consolidates what would otherwise require separate processing systems, achieving comprehensive information utilization with efficient resource consumption.

The patent implements a feedback mechanism where the machine learning model receives ACK/NACK outcomes from previous transmissions and uses this information to refine future MCS selections. The system learns from past performance, adjusting its predictions based on whether transmissions succeeded or failed, thereby adapting to changing channel conditions while maintaining stable operational goals.

The machine learning model performs preliminary learning during system operation, accumulating knowledge about cell-specific characteristics and UE behavior patterns. This preliminary action of learning and adaptation occurs continuously in the background, enabling the system to proactively adjust to changing conditions rather than reactively responding to performance degradation.