Contextual Multi-Armed Bandit-Based Link Adaptation for URLLC

We propose a contextual multi-armed bandit (MAB) link adaptation to ensure reliability in bursty traffic environments for ultra-reliable and low-latency communication (URLLC). The proposed link adaptation utilizes the estimated decoding success probability which is calculated with the log-likelihood ratio (LLR) as context to select the appropriate modulation and coding scheme (MCS) index for the channel, which is used in the soft demodulator for transport block decoding. Additionally, we introduce an algorithm update method that leverages the sequential relationship between MCS indices to expedite the exploration process. To evaluate the performance of the proposed link adaptation, simulations have been conducted in an new radio (NR) standard downlink transmission environment under file transfer protocol (FTP) model 3-based traffic scenarios. By comparing the proposed link adaptation with the conventional outer loop link adaptation and various MAB algorithm-based link adaptations, we demonstrate that the proposed link adaptation achieves the most reliable transmission in various environments while allocating resource efficiently.