Task Load Modeling for LTE Baseband Signal Processing with Regression Analysis Approach

Abstract: In telecommunication baseband signal processing systems, thousands of tasks are executed every millisecond. These tasks take in dierent parameters and cause heavy load to the system. The aim of the thesis is to build proper mathematical models for these tasks, enabling the prediction of their load given the corresponding parameters. For each task, data samples of task load measure and corresponding parameters are provided. No prior knowledge on the task load and its parameters is available. By studying the data samples, an explicit, accurate and simple model is expected. Graphical skills like scatter plots are used as a preliminary analysis of the data. Then rst-order and second-order linear models, piecewise-linear models and tree-based models are taken as prototypes for the task modeling. Methods like stepwise linear regression and partial correlation analysis are applied to select proper parameters from many available parameters to simplify the models. An automatic tool is further developed to automate the whole modeling process. There are 17 tasks in total. For 15 tasks, acceptable models are built with a RMSE lower than 2 times of the estimated noise standard deviation with the assumption of a Gaussian noise, while for the other 2, no adequate models are given. Reasons for not getting acceptable models are discussed and suggestions on future work are proposed.