TROLL: Training of Outer Loop Link Adaptation in Wireless Networks via Back-propagation

The role of link adaptation (LA) is critical to allow wireless networks to achieve high spectral efficiency and meet latency and reliability requirements. However, due to wireless channel aging and unpredictable interference, predicting the signal to interference plus noise ratio (SINR) at transmission time has always been a major issue. For this reason, outer loop link adaptation (OLLA) schemes have been proposed and implemented in current wireless systems, allowing to correct the SINR estimates to guarantee a long-term match of the first transmission target block error rate (FTB). However, OLLA’s limitations emerge targeting low FTB for highly reliable services, and a crucial challenge arise if the OLLA parameters are not optimized on the scenario handled by the network. In the present paper, we propose training of outer loop link adaptation (TROLL), a method to train OLLA’s parameters to optimize its performance in any generic scenario. TROLL procedure is derived in close form and consists in minor modification and computation load increase with respect to baseline OLLA. We also enhance OLLA by smoothing the effect of SINR updates, allowing to better leverage previous experienced transmission successes/failures. The proposed algorithm is evaluated with data from 3GPP compliant system level simulation, demonstrating its gains in term of FTB matching at initialization and spectral efficiency when compared to current OLLA, with negligible complexity increase with respect to other solutions based on neural networks.