MACHINE LEARNING BASED ADAPTIVE MODULATION AND CODING IN 5G NETWORKS

Abstract

In wireless communication, resources like bandwidth and energy are scarce and extremely valuable, any system should serve as many users as possible while preserving high Quality of Service (QoS) for the best user experience. Accordingly, the Base Station (BS) has the responsibility to optimally schedule its resources to the users based on the available information. Consequently, the whole process of scheduling is truly demanding and requires high complex calculations from the overall system. Hence, the request of more sophisticated and effective methods is substantial in order to minimize the challenges of scheduling. This bachelor’s thesis focuses on the Modulation and Coding Scheme (MCS) selection in a Time Division Duplex (TDD) based mobile network. The main objective is the simplification and optimization of the downlink process at the base station by predicting the MCS index for a single User Equipment (UE), using Machine Learning (ML). The developed machine learning algorithms is in accordance with the LTE-Advanced Pro (release 12, 13,14) lookup tables and is based on similar parameters. For a given frame, this thesis targets predicting the MCS index of future subframes. Thus, the resource allocation process for independent users is becoming quicker and easier for the BS. The results are based on laboratory measurements at Ericsson, where the collection of data logs for several stationary UEs, occurred on a network testing environment and their different cell characteristics investigated thoroughly. Concluding, the accuracy level which the ML classification algorithm achieved was approximately 75 percent. Therefore, the prediction accuracy can be described as sufficient for the BS to decrease the computation complexity and energy consumption during the downlink process. The data logs that the project took into account cannot be generalized for real-time scenarios as it is explained in detail finally