Doctoral School

Reading date: 06/10/2026

• ARBOLEDA CARVAJAL, ALEJANDRO: Nuevos índices para la predicción del momento óptimo de la extubación de pacientes asistidos mediante ventilación mecánica a partir de señales electromiográficas y cardíacas
  
  Author: ARBOLEDA CARVAJAL, ALEJANDRO
  Programme: DOCTORAL DEGREE IN BIOMEDICAL ENGINEERING
  Department: Department of Automatic Control (ESAII)
  Mode: Normal
  Deposit date: 03/09/2026
  Reading date: 06/10/2026
  Reading time: 11:00
  Reading place: Sala Polivalent, Edifici A , Escola d'Enginyeria de Barcelona Est (EEBE), Av. d'Eduard Maristany, 16, 08019 Barcelona
  Thesis director: GIRALDO GIRALDO, BEATRIZ FABIOLA
  Thesis abstract: Invasive mechanical ventilation constitutes one of the most critical life-support interventions in Intensive Care Units for patients with acute respiratory failure. Premature discontinuation may lead to respiratory failure, the need for reintubation, and increased morbidity and mortality. Conversely, delayed extubation increases the risk of ventilator-associated pneumonia and diaphragmatic atrophy due to disuse. Since traditional indices often fail to capture the systemic complexity of the patient, this thesis addresses the problem under the hypothesis that neuromuscular dynamics and cardiorespiratory coupling, measured noninvasively during the spontaneous breathing trial (SBT), contain predictive patterns related to the patient’s functional reserve to sustain autonomous ventilation.An experimental design based on a prospective observational study was structured, recruiting a cohort of 39 adult patients. Inclusion criteria required a period of mechanical ventilatory support of at least 48 hours, ensuring that patients had overcome the acute phase of the disease and met the clinical requirements to initiate extubation. A novel protocol was implemented to record respiratory activity from these patients using electromyographic (EMG) and electrocardiographic (ECG) signals. The cohort was monitored longitudinally for 48 hours following extubation; this follow-up enabled patient classification into the success group—those who maintained spontaneous breathing—or the failure group—those who exhibited intolerance to the SBT or required reintubation within 48 hours.According to the proposed protocol, multichannel surface diaphragmatic EMG (sEMGdi) and ECG signals were recorded simultaneously, together with clinical and ventilatory variables reflecting the patient’s condition at the time of the extubation trial. The signals underwent filtering and artifact reduction procedures. Subsequently, new derived signals were obtained: the EMG envelope to quantify muscle recruitment amplitude, the interpolated sEMG signal to standardize the analysis, ECG R-R interval fluctuations to assess heart rate variability (HRV) as an indicator of sympathetic-parasympathetic autonomic tone, and ECG-derived respiration (EDR), which reconstructs respiratory activity from variations in the heart’s electrical axis induced by thoracic volume changes.The derived signals were characterized using a multidimensional approach. In the time domain, activation cycle amplitudes and durations were evaluated; in the frequency domain, shifts in mean and central frequencies were analyzed to detect muscle fatigue, together with power distribution within the low- and high-frequency HRV bands. In addition, nonlinear analysis metrics, such as entropy measures, were applied to quantify the loss of system complexity. Cardiorespiratory coupling was assessed through cross-correlation and spectral coherence analyses, evaluating how the mechanical and neurological effort of breathing modulates the hemodynamic and cardiac response.Finally, statistical analysis identified a set of parameters showing statistically significant differences between the extubation success and failure groups. These properties reflected alterations in diaphragmatic effort efficiency, reduced cardiac variability, and loss of synchronization between cardiac and respiratory rhythms in patients who failed extubation. Based on these selected biomarkers, supervised classification models were implemented, including Naïve Bayes, Support Vector Machines, and k-Nearest Neighbors. The optimized model achieved an accuracy of 92% and a specificity greater than 90%.
  

Reading date: 07/10/2026

• NIU, YUYAO: Hardware-Aware Algorithms for Efficient Graph Traversal on High-End Parallel Architectures
  
  Author: NIU, YUYAO
  Programme: DOCTORAL DEGREE IN COMPUTER ARCHITECTURE
  Department: Department of Computer Architecture (DAC)
  Mode: Normal
  Deposit date: 09/09/2026
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: CASAS GUIX, MARC
  Thesis abstract: Graph traversal is a fundamental building block in scientific computing, data analytics, and artificial intelligence. Among traversal methods, Breadth First Search (BFS) and Depth First Search (DFS) are two core primitives for higher-level graph algorithms. As graph-structured data continues to grow, efficient traversal increasingly depends on the ability to exploit high-end parallel architectures. However, modern architectures rely on structured and throughput-oriented execution engines, such as long vector processors, dense matrix accelerators, and massively parallel GPUs, whereas graph traversal exhibits irregular memory accesses, data-dependent control flow, and dynamic workloads.This dissertation argues that the performance gap between graph traversal and modern hardware arises from three major mismatches: a structural mismatch between regular execution units and irregular graph structure, a computational mismatch between arithmetic-oriented acceleration and relation-driven traversal, and an architectural mismatch between throughput-oriented parallelism and dependency-driven traversal behavior. To address these challenges, the dissertation develops a hardware-aware design paradigm in which graph traversal algorithms are reformulated according to the characteristics and constraints of the target architecture.The first contribution is ALBBA, an algebraic BFS algorithm for long vector architectures. ALBBA reformulates direction-optimized BFS using a vector-friendly sparse representation, vectorized search, and a two-level bypass mechanism. These techniques allow the key algebraic BFS operations, sparse matrix-vector multiplication (SpMV) and sparse matrix-sparse vector multiplication (SpMSpV), to better exploit long vector execution while reducing unnecessary memory accesses and redundant computation. This shows that structural irregularity can be mitigated through hardware-aware data layout and execution design.The second contribution is BerryBees, a BFS framework for Tensor Core-based GPUs. BerryBees introduces the Binarized Row Slice format, which uses bit-level representations to support both SpMV and SpMSpV within a unified data layout while reducing memory overhead. It further employs a warp-level algorithm that leverages Tensor Core matrix-multiply-accumulate instructions to accelerate traversal. This shows that relation-driven graph computation can be reorganized to exploit hardware originally designed for dense matrix operations.The third contribution is DiggerBees, a parallel DFS algorithm for massively parallel GPUs. DiggerBees introduces a two-level stack structure aligned with the GPU memory hierarchy, warp-level DFS execution within thread blocks, and hierarchical work stealing across warps and blocks. These techniques expose scalable parallelism for dependency-driven traversal and improve load balance across the GPU. This shows that DFS, despite its inherent dependencies, can be effectively adapted to throughput-oriented parallel hardware.Together, these contributions demonstrate that BFS and DFS can be effectively accelerated when their data representations, computational formulations, and execution structures are designed with architectural constraints as first-class considerations. More broadly, this dissertation establishes hardware-aware reformulation as a general methodology for rethinking irregular graph algorithms on increasingly specialized parallel hardware.
  

• VILÀ INSA, MARC: Noncoherent detection and spatial cyclostationarity with massive arrays
  
  Author: VILÀ INSA, MARC
  Programme: DOCTORAL DEGREE IN SIGNAL THEORY AND COMMUNICATIONS
  Department: Department of Signal Theory and Communications (TSC)
  Mode: Normal
  Deposit date: 07/09/2026
  Reading date: 07/10/2026
  Reading time: 12:00
  Reading place: Aula de Teleensenyament ETSETB, Edifici B3, Campus Nord, Barcelona
  Thesis director: RIBA SAGARRA, JAUME
  Thesis abstract: Since the advent of digital communication technologies, wireless systems have relied on precise and regularly updated knowledge about the physical medium, referred to as channel state information (CSI). This resource has played a central role in the deployment of modern wireless networks, enabling the monumental growth in transmission rates and data coverage demanded by the current information era. In recent times, however, emerging use cases are revealing practical scenarios in which the acquisition of reliable CSI becomes spectrally inefficient or even infeasible. Clear examples may be found in industrial internet of things (IIoT), where communication is performed through small data packets and latency requirements outweigh throughput needs. Such characteristics appear antagonistic to pilot-based coherent detection, a discrepancy which is also present in systems over high mobility environments, characterized by very short coherence blocks. This has led some authors to reassess whether coherent approaches are the most suitable solutions to implement these applications, envisioned for next generation wireless systems.This shift in state-of-the-art communication theory has rekindled an interest in alternative strategies for wireless data transmission, particularly noncoherent detection. Systems operating noncoherently are those that bypass the need for instantaneous CSI, and rely instead on a crude statistical description of fading and noise. It is precisely within this context that the present thesis is embedded. This work contributes to the noncoherent communication corpus by exploring fundamental limitations of wireless systems based on massive arrays. Most notably, we derive necessary and sufficient conditions to achieve asymptotically error-free communication by increasing transmitted power and the number of receiving antennas, and relate them to unavoidable error floors found throughout the literature. These results are obtained by leveraging first principles from detection theory, thus remaining interpretable and insightful in light of investigating codeword design.Afterwards, the theory developed is harnessed to devise energy-based noncoherent systems that employ massive antenna arrays to improve communication reliability. In particular, we study a one-shot detection problem and conceive a family of suboptimal receivers founded on quadratic statistics of data. These approaches are then used as the building blocks of an index modulation scheme over orthogonal frequency-division multiplexing (OFDM). The solution obtained shows promising results, with near-optimal performance in a variety of scenarios at a fraction of the computational cost of maximum likelihood detection.In parallel, we study the spectral theory of periodically correlated stochastic processes, which naturally appear in some space-time noncoherent problems under spatially stationary fading. We develop a novel treatment based on the Karhunen-Loève representation of cyclostationary processes, and relate it to classical time-frequency approaches under the same framework. As a use case of interest, these techniques are applied to draw ultimate performance limits on a variety of Wiener filtering problems from a unified perspective. Finally, we utilize this toolkit to provide insights on noncoherent codeword detection and implement low-complexity receivers that exploit spectral approximations of data.
  

Reading date: 08/10/2026

• AGUILERA VIDAL, MARIA: Compound Rainfall-Wave Extremes in the Spanish Mediterranean: dependence structure, spatio-temporal variability, and climate change attribution
  
  Author: AGUILERA VIDAL, MARIA
  Programme: DOCTORAL DEGREE IN MARINE SCIENCES
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 04/09/2026
  Reading date: 08/10/2026
  Reading time: 15:00
  Reading place: Place: ETSECCPBUPC, Campus NordBuilding C1. Classroom: 002C/Jordi Girona, 1-308034 Barcelona
  Thesis director: JIMENEZ QUINTANA, JOSE ANTONIO
  Thesis abstract: Compound coastal flooding along the Spanish Mediterranean coast is primarily driven by the interaction between intense rainfall and energetic wave storms, which constitute the dominant terrestrial and marine flood drivers in this microtidal environment. This thesis develops an integrated framework to characterize rainfall-wave compound hazards by combining observational datasets, wave reanalysis, copula-based dependence analysis, synoptic weather classification, and climate-model attribution techniques. The framework examines how compound hazards vary across space, evolve over time, respond to climate change, and generate spatially connected impacts.The results reveal a pronounced spatial organization of compound hazard. Northern Catalonia and central Valencia emerge as the main hotspots, where compound events occur more frequently, exhibit stronger statistical dependence, and systematically more intense than single-driver extremes. Temporal analyses further show that rainfall-wave dependence has strengthened over recent decades along the northern Catalan coast, increasing the synchronization between both hazards and the probability of joint extremes. This evolution is accompanied by a greater likelihood of extreme precipitation. These changes are linked to an increased frequency of the dominant synoptic configurations, particularly Mediterranean cyclones and northern-blocking situations, that favour the concurrence of heavy rainfall and energetic wave conditions.Climate-change attribution demonstrate that the most severe compound events already exhibit a detectable anthropogenic signal through the intensification of extreme precipitation, substantially increasing the likelihood of rare compound extremes under present-day climate conditions. Beyond local hazard interactions, the thesis also demonstrates that compound flooding is inherently spatially connected. The same atmospheric configurations responsible for the strongest local compound events also generate widespread impacts extending from Catalonia to central Valencia, with Storm Gloria (January 2020) representing a characteristic example of this synoptic regime.Overall, this thesis demonstrates that compound coastal flood hazard in the western Mediterranean cannot be adequately understood using a single-driver, stationary, or purely local approaches. Instead, robust hazard assessment requires explicitly accounting for multivariate and evolving hazard conditions partly influenced by anthropogenic climate change, as well as for the spatial connectivity among driving processes across the region and the role of large-scale atmospheric circulation patterns. The integrated framework developed here provides a transferable methodology for improving compound flood assessments, supporting climate-resilient coastal planning, and advancing the understanding of compound coastal hazards in Mediterranean and other microtidal, wave-dominated environments.
  

• GARCIA SALANOVA, ANTONIO: Arquitectura amb propòsit per a una gestió sostenible dels equipaments públics
  
  Author: GARCIA SALANOVA, ANTONIO
  Programme: DOCTORAL DEGREE IN ARCHITECTURAL, CIVIL AND URBAN HERITAGE AND REFURBISHMENT OF EXISTING BUILDINGS
  Department: Departamento de Representación Arquitectónica (RA)
  Mode: Normal
  Deposit date: 09/09/2026
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: REDONDO DOMINGUEZ, ERNEST
  Thesis abstract: PURPOSEFUL ARCHITECTURE FOR THE SUSTAINABLE MANAGEMENT OF PUBLIC FACILITIESThis research analyses, formulates and reasonably contrasts a comprehensivea model of sustainable management of public facilities throughout their life cycle, from planning and design to operation and dismantling, and incorporates data-driven governance aimed at facilitating collaborative decision-making between the different agents involved. It is based on the hypothesis that the adoption of a comprehensive sustainable management model, based on collaborative digital methodologies —especially Building Information Modeling (BIM)—, sustainability criteria and life cycle analysis (LCA), makes it possible to overcome the limitations of traditional models of management of public facilities, characterized by the fragmentation of processes, the lack of traceability of information and the low optimization of the use phase. The work adopts a mixed methodological approach that combines systematic review of the literature, comparative analysis between management models, interviews with experts from the public and private sectors, quantitative validation and the development of an original methodological instrument: a methodological framework of 100 sustainable technical criteria applicable to the entire life cycle, which operationalizes the proposed management model. Unlike predominantly theoretical approaches, the research is carried out in a real context of public management of educational facilities and incorporates recent data and professional experiences (2024–2025), especially in relation to the implementation of BIM in the operation phase.The research also develops a practical case, a hypothetical Audiovisual Vocational Training Institute of Catalonia in the Tres Xemeneies del Besòs site, which acts as a proof of concept of the proposed comprehensive management model. This case makes it possible to contrast the coherence and feasibility of the model, as well as its architectural translation in relation to the pedagogical requirements of vocational education.The results positively show the three hypotheses formulated. Firstly, the comprehensive management model based on BIM–LCA improves the efficiency, quality and sustainability of processes. Secondly, its application determines a new flexible, modular architectural typology aimed at efficient maintenance. Thirdly, the integral model is adapted in a coherent way to the functional and pedagogical requirements of Vocational Training, reinforcing the connection between architecture, management, educational innovation and participatory governance processes. Of particular relevance is the contribution in the field of digital operation management, still underdeveloped in current institutional practice,Finally, the research proposes future lines aimed at data-driven exploitation, the integration of intelligent systems and the development of new professional profiles linked to life cycle management.