Learning Scheme for Adaptive Modulation and Coding in 5G New Radio

Adaptive modulation and coding (AMC) is a link adaptation technique based on the physical layer of fifth-generation (5G) new radio (NR) systems. Studies on AMC and AMC improvement techniques have primarily focused on adding a feedback and retransmission mechanism or on improving channel prediction algorithms, including pilot signal insertion algorithms. A common improvement solution for AMC in machine learning is the K nearest neighbor (KNN)-based modulation and coding scheme selection method. However, to use this method, channel information must be complete and accurate. We adopted a supervised learning-based AMC scheme that considers the effects of channel and noise bias on AMC performance and explored the effects of the K value, distance metrics, and eigenvectors on scheme performance. The K value associated with the highest average classification accuracy rate was determined and verified through ten-fold cross-validation, and the channel quality indicator estimation errors were analyzed. We compared the classification performance of the scheme by using training samples with channel and noise values and those with estimated channel and noise values. The results indicated that the classification accuracy was higher when training samples with estimated values were adopted. In addition, the KNN-based AMC scheme outperformed genetic algorithms regarding bit error rates and spectral efficiency.