Machine Learning in 5G URLLC Link Adaptation

Abstract

Ultra-reliable low latency communication (URLLC) is an emerging use-case of the 5G New Radio (5G-NR) standards. URLLC is an enabler of a multitude of mission-critical applications that demand very low latencies and a very reliable service. An important lement of meeting the strict URLLC requirements is a suitable link adaptation algorithm that must be able to select the appropriate modulation and coding schemes (MCS) according to the wireless link quality. Traditional link adaptation approaches, such as outer loop link adaptation (OLLA), which rely on the positive and negative acknowledgements (ACKs and NACKs) reported from the user equipment (UE), are impractical when the target block error rates (BLER) are very low. To eliminate this problem, calculating an estimate of the BLER at the UE and reporting it to the base station has been suggested. This work investigates the potential of utilizing the BLER feedback in designing machine learning algorithms that solve the problem of link adaptation in URLLC. The performance of the proposed algorithms is compared to the state-of-the-art link adaptation algorithms and it is shown that machine learning approaches perform significantly better than the state-of-the-art methods and that they provide a general framework to further the progress of meeting the URLLC requirements.