User Element Geolocation in Massive-MIMO Networks with Dynamic Spectrum Access Using Hybrid Super Machine Learning

An innovative three stage Hybrid Super Machine Learning (HSML) architecture for street block User Element (UE) localization in a Massive-Multi-Input-Multi-Output (M-MIMO) networks with Dynamic Spectrum Access (DSA) is developed and presented.The approach requires a large set of UE observation samples to be collected and split into three different datasets. Two of which are used for training and validation, and the last dataset is used to measure the performance of the offline trained and validated proposed HSML structure and provide the required online testing accuracy. Two-thirds of the overall UE observation samples are required for the training and the rest is equally divided for validation and testing phases.Pre-processing is also required to initiate optimal feature extraction and classification experience, including z-score normalization and non-linear transformation. Excessive experiments on several Supervised Machine and Deep Learning (SML and DML) algorithms are accomplished to obtain premium SML and DML feature extractors. These features are fed to SML classifiers and combined using super learning to achieve high-performance classification testing accuracy. The HSML architecture is examined in a synthetic urban environment generated by the Wireless Insite (WI) platform. This unique technique is required in critical scenarios where Conventional receiver localization techniques including Global Positioning System (GPS), range-free and range-based receiver localization techniques tend to fail in geolocating an intended UE.