Search

The largest part of this thesis is concerned with the study of a fragmentation process in which rectangles break up into progressively smaller pieces at rates that depend on their shape. Long, thin rectangles are more likely to break quickly, and are also more likely to split along their longest side. We are interested in the evolution of the system at large times: how many fragments are there of different shapes and sizes, and how did they reach that state? We give an almost sure growth rate along paths by studying an equivalent branching random walk. Our analysis is highly technical due to the spatial dependence of the rates and the fact that we work under weaker assumptions than the usual large deviations regime for random walks. In the second part of the thesis we focus on a different, but related problem: estimating the probability that the paths of a random walk stay close to a given function. We prove a small deviation result about the unscaled paths of either a compound Poisson process, or a random walk in discrete time. Our proof strategy involves a Brownian motion approximation on smaller time intervals, which allows us to take advantage of the sharpest estimates currently available on the probability that a Brownian motion lies in a tube about a given function.

The operations of Mobile Autonomous Systems (MAS) rely on real-time data analysis. For instance, autonomous vehicles’ navigation requires the low-latency analysis of high resolution images to detect, and avoid, objects. Unfortunately, many Mobile Autonomous Systems (MASs) have constrained computing and energy resources, and the continuous execution of state-of-the-art algorithms is out of their reach. By offloading the processing load to compute-capable device located at the network edge, the edge computing paradigm can mitigate this issue. However in practical real-world settings, the wireless channel connecting the mobile devices to the edge server often presents erratic capacity patterns due to mobility.As a result, the overall delay perceived by the mobile application may be affected by large variations, which in turn harm control.This thesis explores solutions to the problem described above. To this aim, in addition to new concepts and edge offloading strategies, a complete real-world platform – the HyDRA platform, was developed to support design and evaluation, as well as dataset collection.HyDRA is a fully open source software and hardware platform realizing flexible machine learning-empowered computing for MAS. From a hardware perspective, HyDRA is composed of several Unmanned Aerial Vehicles (UAVs) and ground devices collaboratively performing data analysis to accomplish system-wide goals. From a software perspective, the HyDRA middleware enables real-time control of data and task routing within the system, organized as a distributed set of modules transforming the data captured by the MAS into actuable control.Taking HyDRA as a starting point, this thesis makes the following conceptual contributions:‚the end-to-end delay in remote computing settings for MAS – and specifically autonomous quad-copters – were characterized by means of real-world experiments that produced a comprehensive dataset focused on object detection from images. The study considered both Wi-Fi and Long-Term Evolution connectivity, and several embedded computing platforms. The results demonstrates the instability of application level delay even in line-of-sight settings and relatively slow vehicle motion. A framework for the dynamic control of task offloading in MASs with extreme temporal variations is developed. The frameworks is based on a preliminary experimental analysis, which indicates that there is no dominant feature, including obvious features such as channel quality, and that prediction necessitates an ensemble of weaker features. We first mathematically formulate a Redundant Task Offloading Problem. Then, we create predictors that can help manage the resource usage/performance trade-off. Specifically, we propose a myopic predictor as baseline and a DRL-based approach, which operates on a set of features from application, network and device-level components. To the best of our knowledge, this is the first framework addressing the problem of redundant task offloading in MAS with a data-driven approach which efficacy is verified in a real-world testbed and with replicable dataset-based experiments.‚A modeling and optimization framework based on Markov Decision Processes (MDP) was developed to analyze the structural properties of dynamic control strategies determining where information is processed in collaborative computing scenarios for MAS. In this section of the thesis, the focus of control is primarily between local and remote analysis. Using recent split Deep Neural Networks deep neural network (DNN) techniques, the framework also controls at a fine-grain how the analysis task (the DNN in this case) is divided between the mobile device and the edge server based on current system parameters.

Photovoltaic (PV) energy generation is one of the most prominent technologies for the clean production of electricity. In PV systems of any scale, power electronic converters are the essential interface between the renewable energy generators and the load. Traditionally, multiple PV modules are interfaced to one converter with maximum power point tracking (MPPT) capability. Although this is a well-known and simple architecture, research has highlighted the loss of energy attributed to partial shading and electrical characteristic mismatching among PV modules. During the last two decades, substantial research and industry efforts converged towards distributed maximum power point tracking (DMPPT) architectures adopting interfacing converters, referred to as module integrated converters (MICs), on a per-PV module basis. These architectures maximise energy harvest, regardless of partial shading or electrical characteristic mismatching among PV modules. Other advantages include the possibility of operating PV modules of distinct technology together, or of varying power rating, or installed along different orientations. Moreover, MICs provide opportunities for granular monitoring and diagnostic of PV systems, since the health status of individual PV modules may be observed through the converters. This thesis is focussed on the analysis of a PV interfacing dc-dc converter (dc-MIC), involving modelling, modulation and advanced control techniques. In the studied application, each PV module feeds a dc-MIC, and the dc-MIC outputs are connected in series. The non-inverting buck-boost is the selected converter topology, as it is one of the most promising candidates for DMPPT. Before describing the contributions on dc-MIC control, a chapter of this thesis is dedicated to PV modelling, as the understanding of this topic is necessary to simulate any PV interfacing converter. Firstly, a pragmatic review of PV modelling techniques is undertaken, describing the PV equivalent circuit parameter estimation. Secondly, a contribution to the practical simulation of the PV module electrical characteristic is outlined. The operation of the non-inverting buck-boost dc-MIC is then analysed. One research contribution presented, gives evidence of the improvement of the converter performance during the changeover between buck and boost operating modes. Attention is then devoted to modelling and controller design, according to linear control techniques. The small-signal model of the non-inverting buck-boost dc-MIC is derived, and a cascaded controller is designed. Its performance is compared against a traditional single-loop voltage controller. It is verified that the cascaded controller yields a considerable improvement in the regulation of the PV module voltage, as it greatly reduces the disturbance caused by the operation of the gate driving circuit. The final contribution regards the utilisation of the feedback linearisation control (FLC) technique to the dc-MIC application. This emerging non-linear control technique provides the benefit of analysing the converter as a linear system, whose dynamic behaviour is independent of the operating point. The dynamic performance of the dc-MIC under FLC and cascaded control is experimentally compared. It is proven that, with FLC, the converter transient response is indeed insensitive to the system operating point, further enhancing the PV module voltage regulation. All converter control techniques are digitally implemented in a floating point microcontroller, driving a 250 W dc-MIC prototype, having a switching frequency of 200 kHz. The outlined converter control techniques can be applied in existing dc-MICs, adding value to state of the art control for PV interfacing dc-dc converters.

A seam effect occurs in panel studies when within-wave changes are less frequent than between-wave changes (comparing data gathered from two different interviews). This study explores the changes in the magnitude of seam effects among labor force states (employment, unemployment, not in labor force) using the last seven waves of the Panel Study of Income Dynamics collected between 1995 and 2005. The panel underwent several changes: data were collected with conventional questionnaires (CQ) until 2001. The interval between waves was changed from one year to two years in 1997. The data regarding labor force transitions were collected with Event History Calendar (EHC) instruments starting in 2003. The questionnaire was also changed: one modification took place when implementing the two year recall period and the second when starting data collection with EHC. Data collected with EHCs show a decrease of seam effects in comparison to the previous waves collected with CQ on a two year recall period. A new undocumented phenomenon was also found in the data. When implementing the two year recall period in the CQ waves, a within-wave seam effect appeared, that is a seam effect between the first year and the second year of the two-year reference period reported in the same interview. This effect disappeared in EHC interviews and is most likely due to a change in the questionnaire design. The estimates of labor status changes most affected by seam bias were the transitions from “employed” to “not in the labor force”, and from “not in the labor force” to “employed”, regardless of the data collection methodology or the length of the recall period. Lastly, the magnitude of one year recall period seam bias was lower than any two year recall period seam points. The results are discussed in light of theories of seam effects, questionnaire design, and with references to the literature on EHC and CQ interviewing methods.