Browsing by Author "Vorobyov, Sergiy A., Prof., Aalto University, Department of Signal Processing and Acoustics, Finland"

Filter results by typing the first few letters
Now showing 1 - 3 of 3
  • Advances and New Opportunities in MIMO Radar - Theoretical Analysis and Algorithms
    (2018) Li, Yongzhe
     School of Electrical Engineering | Doctoral dissertation (article-based)
    The thesis focuses on researching the multiple-input multiple-output (MIMO) radar. Particular topics are the study of ambiguity properties by exploiting transmit beamspace (TB) strategies, opportunities and challenges of clutter and jammer suppression in both the conventional and TB-based MIMO radar configurations, waveform designs ensuring good correlation properties, and joint multi-dimensional transmission and receive adaptive design guaranteeing best signal-to-interference-plus-noise ratio performance. A detailed overview of the MIMO radar research related to the above research aspects is presented, including subjects such as waveform(s)/code(s) design, clutter and jammer suppression, TB design, and parameter estimation and detection. The TB-based MIMO radar AF which deals with the case of far-field targets and narrowband waveforms is proposed. It incorporates the effects of transmit coherent processing gain, waveform diversity, and the array geometry in its definition, and can serve as a generalized AF form for which the phased-array (PA) and conventional MIMO radar AFs are important special cases. It shows interesting relationships with the existing AF results. Furthermore, the maximum achievable ''clear region'' of the TB-based MIMO radar AF, which is free of sidelobes, is derived, and two limiting cases that help obtain tight bounds for the ''clear region'' are identified. For addressing the clutter and jammer suppression problem, a series of reduced-dimensional (RD) spatial and/or temporal adaptive processing algorithms with reasonable complexity are developed, including the space-(fast) time adaptive processing algorithms which can maintain the cold clutter stationarity over the slow time domain, 3D space-time adaptive processing (STAP) algorithms for joint hot and cold clutter mitigation, and RD beamspace and robust beamforming techniques. Fast and efficient algorithms for generating aperiodic unimodular waveforms with good correlation properties are also proposed. The waveform designs are based on minimizing the integrated sidelobe level (ISL) or weighted ISL (WISL) of waveforms and are formulated as nonconvex quartic optimization problems in frequency domain. By means of the majorization-minimization technique, the quartic problems are then simplified into quadratic ones, where the inherent algebraic structures in the objective functions are exploited. For the WISL minimization based design, an alternative quartic form that allows to apply the quartic-quadratic transformation is additionally derived. The developed algorithms are applicable to large-scale design problems as they have faster convergence speed and lower complexity than the state-of-the-art algorithms. In addition, an efficient approach for jointly synthesizing the space-(slow) time transmission with unimodular waveforms and designing the receive STAP filter is proposed by means of the minorization-maximization technique. Two cases of known Doppler information and Doppler uncertainty on clutter bins are considered. The proposed algorithms demonstrate good performance with fast convergence speed and low complexity.
  • Channel Estimation in Large-Scale Multi-Antenna Systems for 5G and Beyond - Novel Pilot Structures and Algorithms
    (2018) Upadhya, Karthik
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Efficient use of the limited quantity of available spectrum to cater to the exponentially increasing demand for throughput has been the focus of communication and signal processing engineers for the past few decades. With the advent of technologies such as the Internet of things (IoT) or machine-type communications (MTC), devices and appliances around us which have predominantly been offline are being equipped with sensors that generate data and are now driving the demand for throughput. The forthcoming fifth generation (5G) standard is being developed to cater to these use cases and to also increase throughput for conventional mobile users. One of the enabling technologies of 5G is the use of antenna arrays with orders of magnitude more elements than in conventional fourth generation (4G) transceivers. Large-scale multi-antenna systems impose constraints on channel training and transceiver architecture. In this thesis, we consider the problem of channel estimation in large-scale multi-antenna systems at conventional sub-6 GHz and millimeter-wave (mmWave) frequencies. In coherent receivers, channel state information (CSI) is obtained using training, which involves sending known pilots from the transmitter. In multi-cell networks, these pilots will have to be reused in different cells in order to limit the channel estimation overhead, resulting in a detrimental phenomenon known as pilot contamination. Pilot contamination, which causes interference and decreases throughput, is a fundamental challenge in large-scale multi-antenna systems. In the first part of this thesis, we address the issue of pilot contamination and propose using superimposed pilots for avoiding/mitigating interference. We also consider variants of superimposed pilots such as the hybrid system and staggered pilots to improve throughput. Next, we address the problem of estimating spatial covariance matrices (SCMs) in massive MIMO systems in the presence of pilot contamination. SCMs are useful for mitigating the effects of pilot contamination, but have to be estimated from contaminated observations of the user channels, and consequently, are also contaminated. In the second part of this thesis, we propose a novel pilot structure for estimating contamination-free SCMs. The shift to mmWave frequencies opens up large swathes of spectrum for communication, enabling the large throughputs that 5G demands. However, the channel propagation characteristics at these frequencies are markedly different from sub-6 GHz channels and communicating at mmWave frequencies imposes significant constraints on the transceiver architecture. Both factors in turn influence the design of signal processing algorithms. In the third part of the thesis, we address the problem of channel tracking in mmWave transceivers and develop novel semi-blind algorithms to track the channel with a low overhead.
  • Constructing Accelerated Algorithms for Large-scale Optimization - Framework, Algorithms, and Applications
    (2018) Florea, Mihai Iulian
     School of Electrical Engineering | Doctoral dissertation (article-based)
    A wide range of inverse problems and various machine learning tasks can be expressed as large-scale optimization problems with a composite objective structure, where the gradient of the smooth part has a global Lipschitz constant that is either impractical to compute, or considerably larger than the local values. The smooth part may be strongly convex as well, especially in certain medical imaging applications. The only fast methods that are able to address this entire class of problems are similar to or based on the black-box algorithms developed and analyzed by Nesterov using the estimate sequence mathematical framework. In this work, we introduce the augmented estimate sequence framework, a relaxation of the estimate sequence. When the lower bounds incorporated in the augmented estimate functions are hyperplanes or parabolae, this framework generates a conceptually simple gap sequence. We use this gap sequence to construct the Accelerated Composite Gradient Method (ACGM), a versatile first-order scheme applicable to the entire composite problem class. ACGM is endowed with an efficient dynamic Lipschitz constant estimation (line-search) procedure and features monotonicity. Motivated by the absence of an accurate complexity measure applicable to all first-order methods, we also introduce the wall-clock time unit (WTU). The WTU accounts for variations in algorithmic per-iteration complexity and more consistently reflects the performance of first-order methods in practice. When analyzed using WTU, ACGM has the best provable convergence rate on the composite problem class, both in the strongly and non-strongly convex cases. We confirm the superiority of ACGM within its class using an extensive simulation benchmark. ACGM also excels in terms of robustness and usability. In particular, ACGM is guaranteed to converge without requiring any quantitative prior information. Additional information, if available, leads to an improvement in performance at least on par with competing methods. Moreover, ACGM actually encompasses several popular algorithms for large-scale optimization, including Nesterov's Fast Gradient Method (FGM) and the Fast Iterative Shrinkage Thresholding Algorithm (FISTA), along with their most common variants. The efficiency and generality of ACGM enables new applications, particularly in ultrasound image reconstruction. In contrast with the unrealistic models of existing approaches, we propose two ultrasound image formation models based on spatially varying kernel convolution that account for arbitrary boundary conditions. We provide these models and their adjoints with resource efficient matrix-free implementations. Using either of our models, a variant of ACGM optimized for this task is able to efficiently reconstruct large ultrasound images with accuracy vastly superior to the state-of-the-art.