In this thesis electromagnetic wave propagation is investigated in an indoor environment. The indoor environment is a furnished office building with corridors, corners and rooms. Particularly, there is an atrium through the building in the center. For the study there were measurements available from real building in the 2.1 GHz frequency band. One objective is to design a propagation model that should be simple but reflect the trend of the propagation measurements. Furthermore, a system performance evaluation is carried out based on the implemented model. The proposed 3D model is a combination of the Free Space Path Loss model, the Keenan-Motley model and the recursive diffraction model. The channel predictions from the 2D Keenan-Motley algorithm are quite different from the measurements. Therefore, the 3D Keenan-Motley algorithm is designed to depict the atrium effect and speed up the simulation at the same time. Besides a buttery radiation diagram is created to mimic Kathrein 80010709 antenna installed in the building. Finally, a diffracted path is added to improve the received signal strength for the users around the atrium areas. With all the above procedures, the final results from the model are in good quantitative agreement with the measurement data. With the implemented propagation model, a further analysis of the system performance on the Distributed Antenna System (DAS) is performed. A comparison for the system capacity between the closed building and the atrium building is conducted, showing that the former one benefits more when the number of the cells increases. The reason is the atrium cells suffer severe interference from neighbor cells during high traffic demand scenarios. Then some further cell configurations show that the number of the cells, the geometry performance and the balance of the user fraction should be considered to improve the system capacity.

[pt] A óptica de espaço livre (FSO), inicialmente estudada na década de 60, está atraindo atenção no contexto dos sistemas 5G+, que exige tecnologias de back-hauling sem fio com taxas de dados extremamente altas em distâncias de poucos quilômetros. A atual falta de um modelo global de perda de percurso para FSO motiva o estudo do comportamento do enlace óptico sob diferentes condições atmosféricas. Esta tese trata do desenvolvimento de modelos para prever a atenuação devido a neblina e chuva para projetar enlaces FSO em tecnologias 5G e além. Estudamos os bancos de dados de medições de superfície disponíveis em todo o mundo como referência para analisar as respostas do FSO considerando os dados disponíveis de longo prazo e descobrimos que o maior impacto na visibilidade em todo o mundo é causado por neblina, neve e chuva. Assim, selecionamos neblina e chuva como o foco deste estudo. Vários modelos de perda de percurso, baseados na visibilidade, são avaliados através de uma abordagem analítica. Propomos modelos de limite inferior e superior para estimar a atenuação específica devido ao nevoeiro considerando incertezas quanto à sensibilidade da atenuação específica, gama, em função da visibilidade em 1.550 micrômetros e o melhor ajuste aos resultados do modelo microfísico. Com relação à atenuação causada pela chuva, estudamos modelos disponíveis na literatura para enlaces de micro-ondas e os modificamos para aplicação em enlaces FSO. Quando os efeitos de espalhamento múltiplo são considerados, foi encontrada uma redução significativa na atenuação da chuva. Construímos a partir dos modelos, um método para combinar estatisticamente a atenuação em enlaces FSO causada por diferentes condições climáticas. O desempenho dos enlaces ópticos é avaliado em termos do máximo comprimento do enlace em relação à disponibilidade sob quatro condições climáticas: (i) somente chuva, (ii) somente neblina, (iii) chuva e neblina, e (iv) e todas as condições. Finalmente, eventos experimentais para os enlaces FSO e mmWave implementados no campus Leonardo do Politecnico di Milano são classificados e analisados dependendo do efeito atmosférico estudado: chuva estratiforme, chuva convectiva, chuva leve, vento, neblina pesada, neve e outros. / [en] Free-space optics (FSO), initially studied in the 60s, is attracting attention in the frame of 5G+ systems, which demands wireless back-hauling technologies with extremely high data rates over distances up to a few kilometers. The current lack of a global path loss model for FSO motivates studying the optical link s behavior under different atmospheric conditions. This thesis deals with the development of models for predicting the attenuation due to fog and rain for designing FSO links in 5G and beyond technologies. We study extensive surface measurement databases worldwide as a benchmark for analyzing FSO responses considering available long-term data. We find that the highest impact on visibility worldwide is caused by fog, snow, and rain. Thus, we select fog and rain as the focus of this study. Several path loss models based on visibility are assessed. We propose lower and upper-bound models to estimate the specific attenuation, gamma, due to fog which considers uncertainties as the sensitivity of gamma to the visibility at 1.550 µm and a micro-physical model of fog. Regarding attenuation caused by rain, we study models available in the literature for microwave links and modify them for application to FSO. When the multiple scattering effects are considered, a significant reduction in the rain attenuation has been found. From the findings, we build a method to statistically combine the attenuation effects on FSO caused by different weather conditions. The performance of FSO links is assessed in terms of the maximum link range against availability under four weather conditions: (i) rain only, (ii) fog only, (iii) rain and fog, and (iv) all conditions. Finally, experimental events for FSO and mmWave links implemented at Leonardo Campus of Politecnico di Milano are classified and analyzed depending on the studied atmospheric effects: stratiform rain, convective rain, light rain, wind, heavy fog, snow, and others.

At present, the current 4G systems provide a universal platform for broadband mobile services; however, mobile traffic is still growing at an unprecedented rate and the need for more sophisticated broadband services is pushing the limits on current standards to provide even tighter integration between wireless technologies and higher speeds. This has led to the need for a new generation of mobile communications: the so-called 5G. Although 5G systems are not expected to penetrate the market until 2020, the evolution towards 5G is widely accepted to be the logical convergence of internet services with existing mobile networking standards leading to the commonly used term “mobile internet” over heterogeneous networks, with several Gbits/s data rate and very high connectivity speeds. Therefore, to support highly increasing traffic capacity and high data rates, the next generation mobile network (5G) should extend the range of frequency spectrum for mobile communication that is yet to be identified by the ITU-R. The mm-wave spectrum is the key enabling feature of the next-generation cellular system, for which the propagation channel models need to be predicted to enhance the design guidance and the practicality of the whole design transceiver system. The present work addresses the main concepts of the propagation channel behaviour using ray tracing software package for simulation and then results were tested and compared against practical analysis in a real-time environment. The characteristics of Indoor-Indoor (LOS and NLOS), and indoor-outdoor (NLOS) propagations channels are intensively investigated at four different frequencies; 5.8 GHz, 26GHz, 28GHz and 60GHz for vertical polarized directional, omnidirectional and isotropic antennas patterns. The computed data achieved from the 3-D Shooting and Bouncing Ray (SBR) Wireless Insite based on the effect of frequency dependent electrical properties of building materials. Ray tracing technique has been utilized to predict multipath propagation characteristics in mm-wave bands at different propagation environments. Finally, the received signal power and delay spread were computed for outdoor-outdoor complex propagation channel model at 26 GHz, 28 GHz and 60GHz frequencies and results were compared to the theoretical models.

This thesis aims at proposing all the possible enhancements for the Multi-Resolution Frequency-Domain ParFlow (MR-FDPF) model. As a deterministic radio propagation model, the MR-FDPF model possesses the property of a high level of accuracy, but it also suffers from some common limitations of deterministic models. For instance, realistic radio channels are not deterministic but a kind of random processes due to, e.g. moving people or moving objects, thus they can not be completely described by a purely deterministic model. In this thesis, a semi-deterministic model is proposed based on the deterministic MR-FDPF model which introduces a stochastic part to take into account the randomness of realistic radio channels. The deterministic part of the semi-deterministic model is the mean path loss, and the stochastic part comes from the shadow fading and the small scale fading. Besides, many radio propagation simulators provide only the mean power predictions. However, only mean power is not enough to fully describe the behavior of radio channels. It has been shown that fading has also an important impact on the radio system performance. Thus, a fine radio propagation simulator should also be able to provide the fading information, and then an accurate Bit Error Rate (BER) prediction can be achieved. In this thesis, the fading information is extracted based on the MR-FDPF model and then a realistic BER is predicted. Finally, the realistic prediction of the BER allows the implementation of the adaptive modulation scheme. This has been done in the thesis for three systems, the Single-Input Single-Output (SISO) systems, the Maximum Ratio Combining (MRC) diversity systems and the wideband Orthogonal Frequency-Division Multiplexing (OFDM) systems.

Το πρόβλημα της μοντελοποίησης του ασύρματου διαύλου συνίσταται σε ένα ολοένα και αυξανόμενο πλήθος παραμέτρων και ιδιαίτερων χαρακτηριστικών που πρέπει να λαμβάνονται υπόψη και να ενσωματώνονται στα πλαίσια του RF σχεδιασμού (Radio-Frequency planning) ενός ασύρματου δικτύου εύρωστης παροχής υπηρεσιών. Ο χαρακτηρισμός του ασύρματου καναλιού προϋποθέτει κάτι περισσότερο από μία ντετερμινιστική εξίσωση που υπολογίζει τις απώλειες οδεύσεως ελεύθερου χώρου συναρτήσει της απόστασης πομπού-δέκτη και της συχνότητας. Ένα αξιόπιστο μοντέλο απωλειών σε μία τέτοια περίπτωση απαιτεί τον συνυπολογισμό των απωλειών λόγω των διάφορων μηχανισμών διάδοσης και εξασθένησης της Η/Μ ακτινοβολίας, συμπεριλαμβανομένων των απωλειών λόγω εμποδίων, αλλά και των φαινομένων σκίασης λόγω ανθρώπινης παρεμβολής. Η σημασία της συμβολής της διατριβής έγκειται στην συγκριτική αντιπαραβολή των βασικότερων μοντέλων απωλειών οδεύσεως και στην αριθμητική διόρθωση και επαναξιολόγηση του μοντέλου της ITU για ένα σύνολο τοπολογιών σύνθετης ράδιο-διάδοσης. Οι προτεινόμενες μας αλλαγές στο μοντέλο ITU αύξησαν σημαντικά την αξιοπιστία του όχι μόνο σε σχέση με την αρχική του απόδοση αλλά και συγκριτικά με τα υπόλοιπα μοντέλα. Επιπρόσθετα, υπολογίσαμε αναλυτικά για κάθε τοπολογία την εξασθένηση ανά απόσταση και προβήκαμε στην μοντελοποίηση της γεωγραφικής διασποράς της εξασθένησης ανά απόσταση με τρόπο που μπορεί να αποτελέσει το θεμέλιο ενός προγνωστικού εργαλείου για την παράμετρο αυτή. Επίσης, εξετάσαμε τις διαλείψεις μεγάλης κλίμακας και καταλήξαμε σε μία καινοτόμο μέθοδο για τον υπολογισμό του βάθους σκίασης απευθείας από τα εμπόδια της εκάστοτε τοπολογίας. Η μέθοδος αυτή προτάθηκε και ελέγχθηκε για την ακρίβειά της στα 2.4 GHz, αλλά μπορεί να εφαρμοστεί για κάθε συχνότητα ενδιαφέροντος εφόσον ακολουθηθούν οι διαδικασίες καταμέτρησης των απωλειών σκίασης των εμποδίων κάθε τύπου και πλήθους. Με τη μέθοδο αυτή μπορεί να υπολογιστεί και η εξασθένηση ανά απόσταση μέσω του βάθους σκίασης, συνεπώς είναι εφικτός ο χαρακτηρισμός της εξασθένησης εντός της τοπολογίας χωρίς την αναγκαιότητα πραγματοποίησης εκτεταμένων μετρήσεων. Επιπρόσθετα, εξετάσαμε πώς διαφορετικοί μηχανισμοί ράδιο-διάδοσης οδηγούν σε διαφορετική γεωγραφική διασπορά των τοπικών τιμών της λαμβανόμενης ισχύος. Τέλος, παρουσιάστηκε η λύση της Ασφάλειας Φυσικού Επιπέδου (Wireless Information-Theoretic Security, WITS) με διερεύνηση της επίδρασης των ρεαλιστικών απωλειών οδεύσεων ενός ασύρματου διαύλου με εμπόδια και διαλείψεις στα όρια των τιμών των παραμέτρων, ενώ επίσης εξετάστηκε για πρώτη φορά σε κλειστή μορφή η Αποκοπή Χωρητικότητας Ασφαλείας (Outage Secrecy Capacity). Μελετήσαμε την επίδραση της κινητικότητας των χρηστών στην αξιοπιστία της μεθόδου σε επίπεδο προσομοιώσεων αλλά και με πειραματικές μετρήσεις, αξιολογώντας ποσοτικά και ποιοτικά την επίδρασή τους στα όρια τιμών των παραμέτρων υπό την παρουσία ενός ωτακουστή, και εν συνεχεία σε καθεστώς πολλαπλών ωτακουστών, θεωρώντας μέθοδο συνεργασίας ωτακουστών SC και MRC. / Wireless Channel Modeling consists of an even more increasing number of factors and parameters that need to be accounted for, in the context of RF planning. Wireless Channel Characterization requires more than a deterministic formula for calculating free space path loss in relation to frequency and distance. A reliable path loss model needs to incorporate in its formula the various propagation mechanisms that influence signal attenuation, including losses due to obstacles and human body shadowing. The importance of our contribution lies in the comparative evaluation and validation of the most fundamental RF path loss models and in the numerical adjustment and re-evaluation of the ITU path loss model for a number of different complex propagation topologies. Our suggested correction to the ITU model has significantly increased its prediction precision compared not only to the initial ITU model but also to all other path loss models in question. In addition, we have calculated the attenuation over distance and modeled the distribution of the local mean values versus distance throughout each topology, in a way that can serve as the foundation of a prediction method for this parameter. The large-scale fading of the received signal has also been characterized and a new empirical method for the calculation of the shadow depth has been introduced. This method has been validated for the 2.4 GHz frequency, but it can be applied to any frequency of interest as long as the respective obstacle-caused losses are measured. Extending this method can be applied in order to calculate the attenuation over distance as a function of shadow depth, therefore allowing us to predict this parameter without extensive measurements. The impact of different propagation mechanisms on the variation of distribution of local mean values of received signal power throughout the topology has also been studied. Finally, the concept of Wireless Information-Theoretic Security (WITS) was discussed, by investigating the impact of channel-dependent variation of path loss on the boundaries of secure communications as defined by WITS parameters, whereas a closed-form expression for Outage Secrecy Capacity was introduced. The impact of user mobility on the range of these parameters was investigated in terms of simulations and experimental measurements, in the presence of a single eavesdropper and also for a multiple eavesdroppers scenario, assuming SC and MRC schemes.

Submitted by Johnny Rodrigues (johnnyrodrigues@ufcg.edu.br) on 2018-05-01T13:36:45Z No. of bitstreams: 1 DALTON CÉZANE GOMES VALADARES - DISSERTAÇÃO PPGCC 2015..pdf: 6500440 bytes, checksum: f8b244a64bf81646d322b68023c204d9 (MD5) / Made available in DSpace on 2018-05-01T13:36:45Z (GMT). No. of bitstreams: 1 DALTON CÉZANE GOMES VALADARES - DISSERTAÇÃO PPGCC 2015..pdf: 6500440 bytes, checksum: f8b244a64bf81646d322b68023c204d9 (MD5) Previous issue date: 2015-07 / No âmbito industrial, é sabido que as redes sem fio são as mais indicadas, já que possuem baixo custo de implantação, maior flexibilidade e são menos invasivas ao ambiente. Na literatura, dentre as tecnologias de rede sem fio aplicadas em ambientes industriais, pouco destaque é dado às tradicionais 802.11a/b/g, conhecidas pelo uso doméstico e chamadas WiFi (acrônimo de Wireless Fidelity). A pesquisa, ora descrita, tem como objetivo avaliar o desempenho de uma rede 802.11g em um ambiente industrial de uma usina termoelétrica (UTE). O cenário de análise foi constituído de 4 enlaces de comunicação, com o principal deles possuindo uma distância de, aproximadamente, 150 metros. Para a avaliação do desempenho da rede, foram consideradas três métricas: taxa de perda de pacotes, Taxa de Transferência e tempo de resposta. Os testes foram realizados por meio do estabelecimento de comunicação entre pontos dentro da sala de motores (primeiro enlace) e o servidor localizado na sala de administração (último enlace). Os resultados obtidos para potência do sinal foram comparados aos valores estimados por meio do modelo Log Distance Path Loss. Observou-se que o desempenho da comunicação realizada pela rede não sofreu degradação significativa, mesmo em um ambiente sujeito à interferência eletromagnética e demais características intrínsecas a uma UTE. Após a análise concluída, verificou-se a viabilidade do uso da tecnologia 802.11g para comunicação de dados em ambientes com características similares às de uma usina termoelétrica. / In industry, it is known that wireless networks are the most suitable, since they have low implementation cost, greater flexibility and are less invasive to the environment. In the literature, among the wireless network technologies applied in industrial environments, little attention is given to traditional 802.11a/b/g, known by the domestic use and by the acronym WiFi (Wireless Fidelity). This work aims to evaluate the performance of an 802.11g network in an industrial environment of a thermal power plant. The scenario consists of four communication links, with the main one having a distance of approximately 150m. For the evaluation of network performance, we considered three metrics: packet loss rate, bandwidth and response time. Tests were carried out through the establishment of communication between points within the engine square (first link) and the server located in the boardroom (last link). The obtained results for the signal strength at each point were compared with the estimated values by the Log Distance Path Loss Model. It was verified that the performance of the communication performed by the network did not suffer significant degradation, even being in an environment subject to considerable electromagnetic interference. After the concluded analysis, it was found feasibility of using 802.11g technology for data communication in environments with characteristics similar to those of a thermal power plant.

Made available in DSpace on 2014-12-17T15:48:12Z (GMT). No. of bitstreams: 1 LeonardoDO_DISSERT.pdf: 3357434 bytes, checksum: 75c2c837997e2220e8de164c844c9b5d (MD5) Previous issue date: 2014-06-05 / Due to the constantly increasing use of wireless networks in domestic, business and industrial environments, new challenges have emerged. The prototyping of new protocols in these environments is typically restricted to simulation environments, where there is the need of double implementation, one in the simulation environment where an initial proof of concept is performed and the other one in a real environment. Also, if real environments are used, it is not trivial to create a testbed for high density wireless networks given the need to use various real equipment as well as attenuators and power reducers to try to reduce the physical space required to create these laboratories. In this context, LVWNet (Linux Virtual Wireless Network) project was originally designed to create completely virtual testbeds for IEEE 802.11 networks on the Linux operating system. This paper aims to extend the current project LVWNet, adding to it the features like the ability to interact with real wireless hardware, provides a initial mobility ability using the positioning of the nodes in a space coordinates environment based on meters, with loss calculations due to attenuation in free space, enables some scalability increase by creating an own protocol that allows the communication between nodes without an intermediate host and dynamic registration of nodes, allowing new nodes to be inserted into in already in operation network / Com o crescimento constante da utiliza??o de redes sem fio em ambientes dom?sticos, empresariais e at? industriais, aparecem novos desafios. A prototipa??o de novos protocolos nesses ambientes tipicamente ? restrita a ambientes de simula??o, onde existe a necessidade de uma dupla implementa??o, uma no ambiente de simula??o, onde se realiza uma prova de conceito inicial e outra em um ambiente real. Al?m disso, uma vez que se parta para ambientes reais, n?o ? trivial a cria??o de um testbed para redes sem fio de alta densidade, dada a necessidade de uso de v?rios equipamentos reais, e uso de atenuadores, redutores de pot?ncia, para tentar reduzir o espa?o f?sico necess?rio para cria??o desses laborat?rios. Nessa lacuna, o projeto LVWNet (Linux Virtual Wireless Network) foi inicialmente concebido para cria??o de testbeds completamente virtuais para redes IEEE 802.11 sobre o sistema operacional Linux. Este trabalho tem como objetivo extender o atual projeto LVWNet adicionando a ele os recursos de possibilitar a intera??o com hardwares wireless reais, dar um suporte inicial ? mobilidade atrav?s do posicionamento dos n?s em um ambiente de coordenadas no espa?o baseado em metros, j? com c?lculos de perda decorrente da atenua??o em espa?o livre, aumentar a escalabilidade com a cria??o de um mecanismo que permita a comunica??o direta entre os n?s sem necessidade de um host intermedi?rio al?m do registro din?mico de n?s, de modo que novos n?s podem ser inseridos na rede com a mesma j? em opera??o

The diploma thesis describes mobile communication systems GSM and UMTS including the operating principle of these systems. It also deals with the GPS issues and the determination of the location by this system. The signal path losses issues between the transmitter and receiver are also described. The focus is also put on Telit UC864-G model and the measurement system, operated by the created programme with the graphic user interface, which measures parameters of the cellular network according to entered input parameters and saves the measured values into the file. The part of the diploma thesis describes another created programme with the graphic user interface displaying the measured network parameters in map data with the possibility to compare the distribution of the measured signal value in both outdoor and indoor environment and the theoretical value set by the propagation models.

[pt] A perda de percurso (PL) é um parâmetro essencial em modelos de propagação e crucial na determinação da área de cobertura de sistemas móveis. Os métodos de aprendizado de máquina (ML) tornaram-se ferramentas promissoras para a previsão de propagação de rádio. No entanto, ainda existem alguns desafios para sua implantação completa, relacionados à seleção das entradas mais significativas do modelo, à compreensão de suas contribuições para as previsões do modelo e à avaliação adicional da capacidade de generalização para amostras desconhecidas. Esta tese tem como objetivo projetar modelos de PL baseados em ML otimizados para diferentes aplicações das tecnologias 5G e além. Essas aplicações abrangem links de ondas milimétricas (mmWave) para ambientes indoor e outdoor na faixa de frequência de 26,5 a 40 GHz, cobertura de macrocélulas no espectro sub-6 GHz e comunicações veiculares usando campanhas de medições desenvolvidas em CETUC, Rio de Janeiro, Brazil. Vários algoritmos de ML são explorados, como redes neurais artificiais (ANN), regressão de vetor de suporte (SVR), floresta aleatória (RF) e aumento de árvore de gradiente (GTB). Além disso, estendemos dois modelos empíricos para mmWave com previsão de PL melhorada. Propomos uma metodologia para seleção robusta de modelos de ML e uma metodologia para selecionar os preditores mais adequados para as máquinas consideradas com base na melhoria de desempenho e na interpretabilidade do modelo. Além disso, para o canal veículo-veículo (V2V), uma técnica de rede neural convolucional (CNN) também é proposta usando uma abordagem de aprendizado por transferência para lidar com conjuntos de dados pequenos. Os testes de generalização propostos mostram a capacidade dos modelos de ML de aprender o padrão entre as entradas do modelo e a PL, mesmo em ambientes e cenários mais desafiadores de amostras desconhecidas. / [en] Path loss (PL) is an essential parameter in propagation models and critical in determining mobile systems’ coverage area. Machine learning (ML) methods have become promising tools for radio propagation prediction. However, there are still some challenges for its full deployment, concerning to selection of the most significant model s inputs, understanding their contributions to the model s predictions, and a further evaluation of the generalization capacity for unknown samples. This thesis aims to design optimized ML-based PL models for different applications for the 5G and beyond technologies. These applications encompass millimeter wave (mmWave) links for indoor and outdoor environments in the frequency band from 26.5 to 40 GHz, macrocell coverage in the sub-6 GHz spectrum, and vehicular communications using measurements campaign carried out by the Laboratory of Radio-propagation, CETUC, in Rio de Janeiro, Brazil. Several ML algorithms are exploited, such as artificial neural network (ANN), support vector regression (SVR), random forest (RF), and gradient tree boosting (GTB). Furthermore, we have extended two empirical models for mmWave with improved PL prediction. We proposes a methodology for robust ML model selection and a methodology to select the most suitable predictors for the machines considered based on performance improvement and the model’s interpretability. In adittion, for the vehicle-to-vehicle (V2V) channel, a convolutional neural network (CNN) technique is also proposed using a transfer learning approach to deal with small datasets. The generalization tests proposed shows the ability of the ML models to learn the pattern between the model’s inputs and PL, even in more challenging environments and scenarios of unknown samples.

[pt] A perda de percurso (PL) é um parâmetro essencial em modelos de propagação e crucial na determinação da área de cobertura de sistemas móveis. Osmétodos de aprendizado de máquina (ML) tornaram-se ferramentas promissoras para a previsão de propagação de rádio. No entanto, ainda existem algunsdesafios para sua implantação completa, relacionados à seleção das entradasmais significativas do modelo, à compreensão de suas contribuições para asprevisões do modelo e à avaliação adicional da capacidade de generalizaçãopara amostras desconhecidas. Esta tese tem como objetivo projetar modelosde PL baseados em ML otimizados para diferentes aplicações das tecnologias5G e além. Essas aplicações abrangem links de ondas milimétricas (mmWave)para ambientes indoor e outdoor na faixa de frequência de 26,5 a 40 GHz,cobertura de macrocélulas no espectro sub-6 GHz e comunicações veicularesusando campanhas de medições desenvolvidas em CETUC, Rio de Janeiro,Brazil. Vários algoritmos de ML são explorados, como redes neurais artificiais(ANN), regressão de vetor de suporte (SVR), floresta aleatória (RF) e aumentode árvore de gradiente (GTB). Além disso, estendemos dois modelos empíricospara mmWave com previsão de PL melhorada. Propomos uma metodologiapara seleção robusta de modelos de ML e uma metodologia para selecionar ospreditores mais adequados para as máquinas consideradas com base na melhoria de desempenho e na interpretabilidade do modelo. Além disso, para o canalveículo-veículo (V2V), uma técnica de rede neural convolucional (CNN) também é proposta usando uma abordagem de aprendizado por transferência paralidar com conjuntos de dados pequenos. Os testes de generalização propostosmostram a capacidade dos modelos de ML de aprender o padrão entre as entradas do modelo e a PL, mesmo em ambientes e cenários mais desafiadoresde amostras desconhecidas. / [en] Path loss (PL) is an essential parameter in propagation models and critical in determining mobile systems coverage area. Machine learning (ML) methods have become promising tools for radio propagation prediction. However, there are still some challenges for its full deployment, concerning to selection of the most significant model s inputs, understanding their contributions to the model s predictions, and a further evaluation of the generalization capacity for unknown samples. This thesis aims to design optimized ML-based PL models for different applications for the 5G and beyond technologies. These applications encompass millimeter wave (mmWave) links for indoor and outdoor environments in the frequency band from 26.5 to 40 GHz, macrocell coverage in the sub-6 GHz spectrum, and vehicular communications using measurements campaign carried out by the Laboratory of Radio-propagation, CETUC, in Rio de Janeiro, Brazil. Several ML algorithms are exploited, such as artificial neural network (ANN), support vector regression (SVR), random forest (RF), and gradient tree boosting (GTB). Furthermore, we have extended two empirical models for mmWave with improved PL prediction. We proposes a methodology for robust ML model selection and a methodology to select the most suitable predictors for the machines considered based on performance improvement and the model s interpretability. In adittion, for the vehicle-to-vehicle (V2V) channel, a convolutional neural network (CNN) technique is also proposed using a transfer learning approach to deal with small datasets. The generalization tests proposed shows the ability of the ML models to learn the pattern between the model’s inputs and PL, even in more challenging environments and scenarios of unknown samples.