Machine Learning Algorithms for Uplink Link Adaptation for LTE CAT M1 Users

In this paper, an Uplink Link Adaptation in LTE for CAT M1 users is designed using machine learning algorithms in varying channel conditions. Modulation Coding Scheme (MCS) and Repetition Level (RL) values are estimated based on past data using K-Nearest Neighbor (K-NN) and Support Vector Machine Algorithms (SVM). Accuracy test for K-NN algorithm is performed to find the optimal value of k (number of neighbors), used to classify test data. As the accuracy of SVM with RBF kernel depends upon regularization parameter (C) and spread of kernel (gamma), cross-validation over different values of C and gamma is performed to find the value, which gives the best accuracy. To validate the effectiveness of machine learning algorithms, we performed throughput simulation analysis in varying channel conditions using live and real-time eNodeB (eNB) data. This data is then fed to our python-based simulator. Simulation results show that machine learning algorithms with dynamic RL adaptation can improve performance up to 15% in the case of LTE CAT M1 users as compared to conventional link adaptation algorithms.