A Gaussian Mixture Model with Firm Expectation-Maximization Algorithm for Effective Signal Power Coverage Estimation

Abstract

Generally, the strength of propagated signal power over wireless cellular communication systems is known to be very stochastic due to their susceptibility to various impacting fading conditions of the radio frequency propagation channels. In previous works, one popular method of analyzing and characterizing such signal dataset in high-dimensional space is by means of single Gaussian density function modelling. However, under many circumstances, it is difficult to give an accurate description of such stochastic signal dataset using a single Gaussian density function model, especially when it does not follow approximately ellipsoidal distribution. In this research paper, a machine learning technique based on Gaussian Mixture Model (GMM) is employed for the analysis and modeling of highly dimensional signal power with mix stochastic shadow fading and long term fading components. The signal power was acquired with the aid of telephone mobile system software investigation tools over the air transmission interface of operational Long Term Evolution cellular network, belonging to a commercial network operator in Port Harcourt City. First, in our contribution, we model the probability density of signal power coverage dataset acquired in the high-dimensional space around six different deployed cell sites by the network operator. In our second contribution, we propose a GMM, which employs a firm iterative maximum likelihood estimation technique combined with the expectation-maximization algorithm to effectively model and characterized signal power dataset. By means of Akaike information criterion, the robustness of the proposed GMM with firm iterative EM algorithm over the commonly used single Gaussian density function modelling method is shown. This technique can serve as a valuable step toward effective monitoring and analyzing operational cellular radio network performance.