This is an archive of my graduation thesis for the Computer Engineering course at UFPE presented in 2018 as well as the related research that resulted in the paper presented at IEEE WCNC 2018. This archive contains final artifacts produced for each work as well as source code for the latex documents and experiments.

Comparativo de Técnicas de Aprendizado de Máquina para o Treinamento de Grids Regulares na Técnica de Localização por Fingerprinting

Outdoor location systems for mobile devices already has many different applications in literature. Among these techniques is the Fingerprinting method, which, although not as precise as the Global Positioning System (GPS), it can achieve tens of meters precision and be implemented with few or no modifications to the current cellular network infraestructure as a critical advantage. The fingerprinting based technique using divides the geographic region in a GRID of cells with Received Signal Strengh Indicator (RSSI) information to the radio base station providing service to the region. This information can be generated from RSSI measurements and machine learning techniques. This work makes a comparative and performance analysis of different machine learning algorithms applied in training of regular GRIDs, which have regular space between cells, used for outdoor location with fingerprinting.

A Proposal of a RF Fingerprint-based Outdoor Localization Technique using Irregular Grid Maps

Location techniques have an increasing demand nowadays. As opposed to the global positioning system (GPS), radio frequency (RF) fingerprint-based techniques are low-energy solutions with reasonable precision and easy implementation. In general, it works by dividing a location area into a grid map and defining an RF pattern to each grid cell in an attempt to uniquely identify it. The most common grid layout is the square grid model (i.e., regular grid) where all cells are equally spaced. In this paper, we propose an RF fingerprint-based location method using an irregular grid layout whose shape resembles the road network of an urban region. The proposed model aims to diminish computational complexity while maintaining or even improving the location estimation precision.