A self-organization machine learning technique for sustainable infocommunications within communications, navigation, and surveillance / air traffic management (CNS/ATM) systems is proposed in the paper. The proposed technique is based on the modification of the expectation-maximization algorithm with adding of components of Gaussian mixture model. The proposed technique allows for an unsupervised self-organization of system parameters into ranges (e.g., frequency bands and any other groups of homogenous parameters), which simplifies a general tuning of infocommunications for aeronautical purposes in dynamically changing conditions. The proposed technique uses a norm transformation filtering to restrict possible influence of outliers and anomalies in input system parameters. The feature that only observed input system parameters are required for all stages of data processing characterizes the proposed technique. Setting of initial parameters, stopping criteria for internal and external iterative machine learning processes, robustness and computational cost within the proposed technique are described and analyzed. An example of simulation of the proposed technique, which presents an unsupervised automatic clustering of the available radio spectrum recourse, is also shown in the paper.