Digital signal processing systems demand higher computational performance and more operations per second than ever before, and this trend is not expected to end any time soon. Processing architectures must adapt in order to meet these demands. The two techniques most prevalent for achieving throughput constraints are parallel processing and stream processing. By combining these techniques, significant throughput improvements have been achieved. These preliminary results apply to specific applications, and general tools for automation are in their infancy. In this dissertation techniques are developed to automatically generate efficient parallel streaming hardware architectures. / Ph. D. / The algorithms that process data have been getting more complicated requiring more operations in less time. This trend has been going on for many years with no end in sight. Techniques must be developed to allow the processing system to meet these requirements. Assembly line techniques, or stream processing allows multiple stages in which each stage is working on a different piece of data. Increasing the number of assembly lines can further increase the number of operations, but results in large overheads. This dissertation develops automation techniques to reduce these overheads resulting in efficient hardware.

Die P- und S-Wellen-Geschwindigkeitsstruktur der D” Schicht unter dem südwestlichen Pazifik wurde mittels kurzperiodischer Daten von 12 Tiefbeben in der Tonga-Fiji-Region untersucht, die vom J-Array und Hi-net-Array in Japan registriert wurden. Es wurde für Punktstreuer und ebene Schichten migriert, um schwache Signale zu extrahieren, die an relativ kleinräumigen Heterogenitäten des unteren Mantels entstehen. Um eine höhere Auflösung zu erzielen, wurde die Double Array-Methode (DAM) verwendet, die Empfängerarray und Quellarray gleichzeitig nutzt. Hierbei ist auch das Phase-Weighted Stack-Verfahren angewendet worden, um inkohärentes Rauschen zu reduzieren und somit schwache kohärente Signale aus dem unteren Mantel aufzulösen. Das Ergebnis der Ebenen-Schichten-Methode (RWB) zeigt, dass sich in der D”-Schicht negative Geschwindigkeitsdiskontinuitäten mit P-Wellen Geschwindigkeitskontrasten von höchstens –1 % in den Tiefen von 2520 km und 2650 km befinden. Zusätzlich befindet sich eine positive Geschwindigkeitsdiskontinuitäten in der Tiefe von 2800 km. Bei den S Wellen treten Geschwindigkeitsdiskontinuitäten in einer Tiefe von etwa 2550 km und 2850 km auf. Die scheinbare Verschiebung (50 km) der S-Wellen-Geschwindigkeitsdiskontinuität in der Tiefe von 2850 km deutet darauf hin, daß die S-Wellen-Geschwindigkeitsreduktion im unteren Mantel 2-3 mal stärker ist als die P- Wellen-Geschwindigkeitsreduktion. Ein zweidimensionaler Querschnitt, der mittels der RWB Methode und der Aufspaltung des Gesamtempfängerarrays in Subarrays gewonnen wurde, deutet darauf hin, dass die beobachteten Diskontinuitäten als intermittierende laterale Heterogenitäten mit einer Wellenlänge von einigen hundert km charakterisiert werden können. Die Kern-Mantel-Grenze (KMG) weist möglicherweise Undulationen mit einer Amplitude von 10 km auf. Die Migration weist nur schwache Hinweise für räumliche Streukörper auf. Die in der Migration abgebildeten heterogenen Regionen korrespondieren mit den mittels der RWB Methode gefundenen seismischen Diskontinuitäten. Bei den gefundenen Heterogenitäten könnte es sich um einen Teil eines aufsteigenden heißen Stroms unter dem südwestlichen Pazifik handeln. / The P- and S-wave velocity structure of the D” layer beneath the southwestern Pacific was investigated by using short-period data from 12 deep events in the Tonga-Fiji region recorded by the J-Array and the Hi-net in Japan. A migration method and reflected wave beamforming (RWB) were used in order to extract weak signals originating from small-scale heterogeneities in the lowermost mantle. In order to acquire high resolution, a double array method (DAM) which integrates source array beamforming with receiver array beamforming was applied to the data. A phase-weighted stacking technique, which reduces incoherent noise by employing complex trace analysis, was also applied to the data, amplifying the weak coherent signals from the lowermost mantle. This combination greatly enhances small phases common to the source and receiver beams. The results of the RWB method indicate that seismic energy is reflected at discontinuities near 2520 km and 2650 km, which have a negative P-wave velocity contrast of 1 % at the most. In addition, there is a positive seismic discontinuity at a depth of 2800 km. In the case of the S-wave, reflected energy is produced almost at the same depth (2550 km depth). The different depth (50 km) between the P-wave velocity discontinuity at the depth of 2800 and a further S-wave velocity discontinuity at the depth of 2850 km may indicate that the S-wave velocity reduction in the lowermost mantle is about 2-3 times stronger that that of P wave. A look at a 2D cross section, constructed with the RWB method, suggests that the observed discontinuities can be characterized as intermittent lateral heterogeneities whose lateral extent is a few hundred km, and that the CMB might have undulations on a scale of less than 10 km in amplitude. The migration shows only weak evidence for the existence of scattering objects. Heterogeneous regions in the migration belong to the detected seismic discontinuities. These anomalous structures may represent a part of hot plume generated beneath the southwestern Pacific in the lowermost mantle.

Covering arrays with row limit, CARLs, are a new family of combinatorial objects which we introduce as a generalization of group divisible designs and covering arrays. In the same manner as their predecessors, CARLs have a natural application as combinatorial models for interaction test suites. A CARL(N;t,k,v:w), is an N×k array with some empty cells. A component, which is represented by a column, takes values from a v-set called the alphabet. In each row, there are exactly w non-empty cells, that is the corresponding components have an assigned value from the alphabet. The parameter w is called the row limit. Moreover, any N×t subarray contains every of v^t distinct t-tuples of alphabet symbols at least once. This thesis is concerned with the bounds on the size and with the construction of CARLs when the row limit w(k) is a positive integer valued function of the number of columns, k. Here we give a lower bound, and probabilistic and algorithmic upper bounds for any CARL. Further, we find improvements on the upper bounds when w(k)ln(w(k)) = o(k) and when w(k) is a constant function. We also determine the asymptotic size of CARLs when w(k) = Θ(k) and when w(k) is constant. Next, we study constructions of CARLs. We provide two combinatorial constructions of CARLs, which we apply to construct families of CARLs with w(k)=ck, where c<1. Also, we construct optimal CARLs when t=2 and w=4, and prove that there exists a constant δ, such that for any v and k≥4, an optimal CARL(2,k,v:4) differs from the lower bound by at most δ rows, with some possible exceptions. Finally, we define a packing array with row limit, PARL(N;t,k,v:w), in the same way as a CARL(N;t,k,v:w) with the difference that any t-tuple is contained at most once in any N×t subarray. We find that when w(k) is a constant function, the results on the asymptotic size of CARLs imply the results on the asymptotic size of PARLs. Also, when t=2, we consider a transformation of optimal CARLs with row limit w=3 to optimal PARLs with w=3.

Covering arrays with row limit, CARLs, are a new family of combinatorial objects which we introduce as a generalization of group divisible designs and covering arrays. In the same manner as their predecessors, CARLs have a natural application as combinatorial models for interaction test suites. A CARL(N;t,k,v:w), is an N×k array with some empty cells. A component, which is represented by a column, takes values from a v-set called the alphabet. In each row, there are exactly w non-empty cells, that is the corresponding components have an assigned value from the alphabet. The parameter w is called the row limit. Moreover, any N×t subarray contains every of v^t distinct t-tuples of alphabet symbols at least once. This thesis is concerned with the bounds on the size and with the construction of CARLs when the row limit w(k) is a positive integer valued function of the number of columns, k. Here we give a lower bound, and probabilistic and algorithmic upper bounds for any CARL. Further, we find improvements on the upper bounds when w(k)ln(w(k)) = o(k) and when w(k) is a constant function. We also determine the asymptotic size of CARLs when w(k) = Θ(k) and when w(k) is constant. Next, we study constructions of CARLs. We provide two combinatorial constructions of CARLs, which we apply to construct families of CARLs with w(k)=ck, where c<1. Also, we construct optimal CARLs when t=2 and w=4, and prove that there exists a constant δ, such that for any v and k≥4, an optimal CARL(2,k,v:4) differs from the lower bound by at most δ rows, with some possible exceptions. Finally, we define a packing array with row limit, PARL(N;t,k,v:w), in the same way as a CARL(N;t,k,v:w) with the difference that any t-tuple is contained at most once in any N×t subarray. We find that when w(k) is a constant function, the results on the asymptotic size of CARLs imply the results on the asymptotic size of PARLs. Also, when t=2, we consider a transformation of optimal CARLs with row limit w=3 to optimal PARLs with w=3.

This thesis deals with dual-polarized multiple input multiple output (MIMO) channels, an important issue for the practical deployment of multiple antenna systems. The MIMO architecture has the potential to dramatically improve the performance of wireless systems. Much of the focus of research has been on uni-polarized spatial MIMO configurations, the performance of which, is a strong function of the inter-element spacing. Thus the current trend of miniaturization, seems to be at odds with the implementation of spatial configurations in portable handheld devices. In this regard, dual-polarized antennas present an attractive alternative for realizing higher order MIMO architectures in compact devices. Unlike spatial channels, in the presence of polarization diversity, the subchannels of the MIMO channel matrix are not identically distributed. They differ in terms of average received power, envelope distributions, and correlation properties. In this thesis, we report on an indoor channel measurement campaign conducted at 2.4 GHz, to measure the copolarized and cross-polarized subchannels, under line-of-sight (LOS) and non-line-of-sight (NLOS) channel conditions. The measured data is then analyzed, to draw a fair comparison between spatial and dual-polarized MIMO systems, in terms of channel characteristics and achievable capacity. The main drawback of the MIMO architecture is that the gain in capacity comes at a cost of increased hardware complexity. Antenna selection is a technique using which we can alleviate this cost. We emphasize that this strategy is all the more relevant for compact devices, which are often constrained by complexity, power and cost. Using theoretical analysis and measurement results, this thesis investigates the performance of antenna selection in dual-polarized MIMO channels. Our results indicate that, antenna selection when combined with dual-polarized antennas, is an effective, low-complexity solution to the problem of realizing higher order MIMO architectures in compact devices.

Covering arrays with row limit, CARLs, are a new family of combinatorial objects which we introduce as a generalization of group divisible designs and covering arrays. In the same manner as their predecessors, CARLs have a natural application as combinatorial models for interaction test suites. A CARL(N;t,k,v:w), is an N×k array with some empty cells. A component, which is represented by a column, takes values from a v-set called the alphabet. In each row, there are exactly w non-empty cells, that is the corresponding components have an assigned value from the alphabet. The parameter w is called the row limit. Moreover, any N×t subarray contains every of v^t distinct t-tuples of alphabet symbols at least once. This thesis is concerned with the bounds on the size and with the construction of CARLs when the row limit w(k) is a positive integer valued function of the number of columns, k. Here we give a lower bound, and probabilistic and algorithmic upper bounds for any CARL. Further, we find improvements on the upper bounds when w(k)ln(w(k)) = o(k) and when w(k) is a constant function. We also determine the asymptotic size of CARLs when w(k) = Θ(k) and when w(k) is constant. Next, we study constructions of CARLs. We provide two combinatorial constructions of CARLs, which we apply to construct families of CARLs with w(k)=ck, where c<1. Also, we construct optimal CARLs when t=2 and w=4, and prove that there exists a constant δ, such that for any v and k≥4, an optimal CARL(2,k,v:4) differs from the lower bound by at most δ rows, with some possible exceptions. Finally, we define a packing array with row limit, PARL(N;t,k,v:w), in the same way as a CARL(N;t,k,v:w) with the difference that any t-tuple is contained at most once in any N×t subarray. We find that when w(k) is a constant function, the results on the asymptotic size of CARLs imply the results on the asymptotic size of PARLs. Also, when t=2, we consider a transformation of optimal CARLs with row limit w=3 to optimal PARLs with w=3.

Covering arrays with row limit, CARLs, are a new family of combinatorial objects which we introduce as a generalization of group divisible designs and covering arrays. In the same manner as their predecessors, CARLs have a natural application as combinatorial models for interaction test suites. A CARL(N;t,k,v:w), is an N×k array with some empty cells. A component, which is represented by a column, takes values from a v-set called the alphabet. In each row, there are exactly w non-empty cells, that is the corresponding components have an assigned value from the alphabet. The parameter w is called the row limit. Moreover, any N×t subarray contains every of v^t distinct t-tuples of alphabet symbols at least once. This thesis is concerned with the bounds on the size and with the construction of CARLs when the row limit w(k) is a positive integer valued function of the number of columns, k. Here we give a lower bound, and probabilistic and algorithmic upper bounds for any CARL. Further, we find improvements on the upper bounds when w(k)ln(w(k)) = o(k) and when w(k) is a constant function. We also determine the asymptotic size of CARLs when w(k) = Θ(k) and when w(k) is constant. Next, we study constructions of CARLs. We provide two combinatorial constructions of CARLs, which we apply to construct families of CARLs with w(k)=ck, where c<1. Also, we construct optimal CARLs when t=2 and w=4, and prove that there exists a constant δ, such that for any v and k≥4, an optimal CARL(2,k,v:4) differs from the lower bound by at most δ rows, with some possible exceptions. Finally, we define a packing array with row limit, PARL(N;t,k,v:w), in the same way as a CARL(N;t,k,v:w) with the difference that any t-tuple is contained at most once in any N×t subarray. We find that when w(k) is a constant function, the results on the asymptotic size of CARLs imply the results on the asymptotic size of PARLs. Also, when t=2, we consider a transformation of optimal CARLs with row limit w=3 to optimal PARLs with w=3.

Aside from performance, energy efficiency is an increasing challenge in database systems. To tackle both aspects in an integrated fashion, we pursue a hardware/software co-design approach. To fulfill the energy requirement from the hardware perspective, we utilize a low-energy processor design offering the possibility to us to place hundreds to millions of chips on a single board without any thermal restrictions. Furthermore, we address the performance requirement by the development of several database-specific instruction set extensions to customize each core, whereas each core does not have all extensions. Therefore, our hardware foundation is a low-energy processor consisting of a high number of heterogeneous cores. In this paper, we introduce our hardware setup on a system level and present several challenges for query processing. Based on these challenges, we describe two implementation concepts and a comparison between these concepts. Finally, we conclude the paper with some lessons learned and an outlook on our upcoming research directions.

The objective of this dissertation is to better understand the impact of the range extension and interference effects of opportunistic large arrays (OLAs), in the context of cooperative routing in multi-hop ad hoc networks. OLAs are a type of concurrent cooperative transmission (CCT), in which the number and location of nodes that will participate in a particular CCT cannot be known a priori. The motivation of this research is that the previous CCT research simplifies or neglects significant issues that impact the CCT-based network performance. Therefore, to develop and design more efficient and realistic OLA-based protocols, we clarify and examine through experimentation and analysis the simplified or neglected characteristics of CCT, which should be considered in the network-level system design. The main contributions of this research are (i) intra-flow interference analysis and throughput optimization in both disk- and strip-shaped networks, for multi-packet OLA transmission, (ii) CCT link modeling focusing on path-loss disparity and link asymmetry, (iii) demonstration of CCT range-extension and OLA-based routing using a software-defined radio (SDR) test-bed, (iv) a new OLA-based routing protocol with practical error control algorithm. In the throughput optimization in presence of the intra-channel interference, we analyze the feasibility condition of spatially pipelined OLA transmissions using the same channel and present numerical results with various system parameters. In the CCT link model, we provide the impact of path-loss disparity that are inherent in a virtual multiple-input-single-output (VMISO) link and propose an approximate model to calculate outage rates in high signal-to-noise-ratio (SNR) regime. Moreover, we present why link asymmetry is relatively more severe in CCT compared to single-input-single-output (SISO) links. The experimental studies show actual measurement values of the CCT range extension and realistic performance evaluation of OLA-based routing. Lastly, OLA with primary route set-up (OLA-PRISE) is proposed with a practical route recovery technique.

L’amincissement des cellules solaires semi-conductrices est motivé par la réduction des coûts de production et l’augmentation des rendements de conversion. Mais en deçà de quelques centaines de nanomètres, il requiert de nouvelles stratégies de piégeage optique. Nous proposons d’utiliser des concepts de la nanophotonique et de la plasmonique pour absorber la lumière sur une large bande spectrale dans des couches ultrafines de GaAs. Nous concevons et fabriquons pour ce faire des structures multi-résonantes formées de réseaux de nanostructures métalliques. Dans un premier temps, nous montrons qu’il est possible de confiner la lumière dans une couche de 25 nm de GaAs à l’aide d’une nanogrille bidimensionnelle pouvant servir de contact électrique en face avant. Nous analysons numériquement les modes résonants qui conduisent à une absorption moyenne de 80% de la lumière incidente entre 450 nm et 850 nm. Ces résultats sont validés par la fabrication et la caractérisation de super-absorbeurs ultrafins multi-résonants. Dans un second temps, nous appliquons une approche similaire dans le but d’obtenir des cellules photovoltaïques dix fois plus fines que les cellules GaAs records, avec des absorbeurs de 120 nm et 220 nm seulement. Un miroir arrière nanostructuré en argent, associé à des contacts ohmiques localisés, permet d’améliorer l’absorption tout en garantissant une collecte optimale des porteurs photo-générés. Nos calculs montrent que les densités de courant de court-circuit (Jsc) dans ces structures optimisées peuvent atteindre 22.4 mA/cm2 et 26.0 mA/cm2 pour les absorbeurs d’épaisseurs respectives t=120 nm et t=220 nm. Ces performances sont obtenues grâce à l’excitation d’une grande variété de modes résonants (Fabry-Pérot, modes guidés,…). En parallèle, nous avons développé un procédé de fabrication complet de ces cellules utilisant la nano-impression et le transfert des couches actives. Les mesures montrent des Jsc records de 17.5 mA/cm2 (t=120 nm) et 22.8 mA/cm2 (t=220 nm). Ces résultats ouvrent la voie à l’obtention de rendements supérieurs à 20% avec des cellules solaires simple jonction d’épaisseur inférieure à 200 nm. / The thickness reduction of solar cells is motivated by the reduction of production costs and the enhancement of conversion efficiencies. However, for thicknesses below a few hundreds of nanometers, new light trapping strategies are required. We propose to introduce nanophotonics and plasmonics concepts to absorb light on a wide spectral range in ultrathin GaAs layers. We conceive and fabricate multi-resonant structures made of arrays of metal nanostructures. First, we design a super-absorber made of a 25 nm-thick GaAs slab transferred on a back metallic mirror with a top metal nanogrid that can serve as an alternative front electrode. We analyze numerically the resonance mechanisms that result in an average light absorption of 80% over the 450nm-850nm spectral range. The results are validated by the fabrication and characterization of these multi-resonant super-absorbers made of ultrathin GaAs. Second, we use a similar strategy for GaAs solar cells with thicknesses 10 times thinner than record single-junction photovoltaic devices. A silver nanostructured back mirror is used to enhance the absorption efficiency by the excitation of various resonant modes (Fabry-Perot, guided modes,…). It is combined with localized ohmic contacts in order to enhance the absorption efficiency and to optimize the collection of photogenerated carriers. According to numerical calculations, the short-circuit current densities (Jsc) can reach 22.4 mA/cm2 and 26.0 mA/cm2 for absorber thicknesses of t=120 nm and t=220 nm, respectively. We have developed a fabrication process based on nano-imprint lithography and on the transfer of the active layers. Measurements exhibit record short-circuit currents up to 17.5 mA/cm2 (t=120 nm) and 22.8 mA/cm2 (t=220 nm). These results pave the way toward conversion efficiencies above 20% with single junction solar cells made of absorbers thinner than 200 nm.