Machine Learning Technique Based on Gaussian Mixture Model for Environment Friendly Communications.

A machine learning technique, which is based on the Gaussian mixture model and uses a developed parametric and criteria features modification of the expectation-maximization (EM) algorithm with removing components of the Gaussian mixture model for a deep statistical analysis of cross-correlations between code structures in low power environment friendly direct sequence spread spectrum (DSSS) non-orthogonal multiple access (NOMA) communications, is proposed in the paper. The features of the EM-algorithm for this purpose are described and analyzed. The proposed modification of the EM-algorithm contains the justification of the initial number of components of a mixture, the initial model parameters, and three additional clustering criteria for adjusting the procedures of EM-algorithm under conditions of mathematical singularities in the log-likelihood function. An example of working of the proposed technique for DSSS NOMA communications is presented and analyzed.