Browsing by Author "Vorobyov, Sergiy"

Now showing 1 - 12 of 12

Algebraic and Adaptive MIMO Radar
(2015-12-14) Morency, Matthew
Sähkötekniikan korkeakoulu | Master's thesis
Breaking causality is the main distinction of the multiple-input multiple-output (MIMO) paradigm as used for active sensing/radar. This is because the transmitting side can be optimized in many ways to manipulate the capabilities of the system. Adaptive beamforming is a fundamental problem in array-processing, communications, and radar among other fields which has once again garnered significant research interest in recent years within the MIMO paradigm. In this work, transmit adaptive beamforming algorithms are developed. One class of algorithms allows search free DOA estimation in 1D and 2D MIMO radar with an arbitrary receive array geometry while allowing transmit power gain. The other uses polynomial ideals in order to recompose the rank-constrained beamforming problem from non-convex problem to a convex one. In the first case, modern algebra is used to analyze target identifiability in the radar system. In the second, algebra reshapes the problem formulation. In both cases, performance improvements are demonstrated compared to previous methods.
Computationally Efficient Algorithms for Radar Signal Design in Spectrally Busy Environment
(2018-12-10) Yli-Niemi, Markus
Sähkötekniikan korkeakoulu | Master's thesis
In this thesis the problem of designing radar transmitter waveforms in a spectrally busy environment is considered. Spectrally busy environment is an environment where radar operates in a congested frequency spectrum with other radiators. Both single waveform design and multiple waveforms design problems are considered. The solution algorithms for the design problems are based on Alternating Direction Method of Multipliers (ADMM) algorithm alongside computationally efficient projection techniques. Solution algorithms are verified to run on quadratic (i.e., O(N2), where N is the problem dimension) or cubic (i.e., O(N3)) time complexities by time complexity graphs. Solution algorithms are also tested with example environment simulations. The quality of designed transmitter waveforms are assessed with signal-tointerference-plus-noise ratio (SINR) and ambiguity function figures. According to these figures designed waveforms have adequate SINR while ambiguity functions have small Doppler leakages and sharp autocorrelation functions.
Deep RL for Radar Applications in an Overcrowded EM Spectrum
(2024-04-24) Kovanen, Saku
Perustieteiden korkeakoulu | Bachelor's thesis
Radar systems are employed in domains ranging from maritime navigation to weather forecasting. Recently, radar systems have started to evolve with the infusion of artificial intelligence (AI), specifically through the application of reinforcement learning (RL) to combat the overcrowding of the electromagnetic (EM) spectrum. The EM spectrum is being overcrowded due to multiple factors, but the most apparent cause is the increase in spectral usage by wireless communications, such as 5G and WiFi technologies. Thus, the accuracy of radar operations is at risk in environments where plenty of wireless communication is taking place. On the flip side of the problem, radars are also affecting wireless communication by causing interference, which could be reduced with a smaller EM spectrum footprint of a radar system. This bachelor’s thesis attempts to assess the research conducted on RL for radar applications in an overcrowded ME spectrum, by reflecting on the research question: "Are RL methods applicable for spectral resource management in radar applications?" The research involved analyzing 20 papers, primarily identified through forward snowballing from key papers suggested by the thesis advisor and searching IEEE Xplorer and Scopus AI using relevant keywords, of which 11 were used as references after assessing their relevance, recentness, and trustworthiness. The key findings from the research are that RL methods, such as Q-learning and deep Q-learning, have been shown to be effective in simulations and even in a real-world experiment. This indicates an opening for more real-world experiments to further demonstrate the capabilities of RL in spectral resource management for radar applications, which would be a good opportunity to experiment with the ways sense-and-avoid (SAA) compares in real-world scenarios to RL. Interestingly, the research shows that in some simulated scenarios, a simple SAA method might be more appropriate than a complex RL method. The drawback of RL methods in some specific scenarios is mainly due to the complexity of running and training required for RL, which necessitates either more powerful computers or larger latency in the radar operations.
Efficient ADMM-Based Algorithms for Convolutional Sparse Coding
(2022) Veshki, Farshad; Vorobyov, Sergiy
A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä
Convolutional sparse coding improves on the standard sparse approximation by incorporating a global shift-invariant model. The most efficient convolutional sparse coding methods are based on the alternating direction method of multipliers and the convolution theorem. The only major difference between these methods is how they approach a convolutional least-squares fitting subproblem. In this letter, we present a novel solution for this subproblem, which improves the computational efficiency of the existing algorithms. The same approach is also used to develop an efficient dictionary learning method. In addition, we propose a novel algorithm for convolutional sparse coding with a constraint on the approximation error. Source codes for the proposed algorithms are available online.
Efficient approximate online convolutional dictionary learning
(2023-12-15) Ghorbani Veshki, Farshad; Vorobyov, Sergiy
A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä
Most existing convolutional dictionary learning (CDL) algorithms are based on batch learning, where the dictionary filters and the convolutional sparse representations are optimized in an alternating manner using a training dataset. When large training datasets are used, batch CDL algorithms become prohibitively memory-intensive. An online-learning technique is used to reduce the memory requirements of CDL by optimizing the dictionary incrementally after finding the sparse representations of each training sample. Nevertheless, learning large dictionaries using the existing online CDL (OCDL) algorithms remains highly computationally expensive. In this paper, we present a novel approximate OCDL method that incorporates sparse decomposition of the training samples. The resulting optimization problems are addressed using the alternating direction method of multipliers. Extensive experimental evaluations using several image datasets and based on an image fusion task show that the proposed method substantially reduces computational costs while preserving the effectiveness of the state-of-the-art OCDL algorithms.
Generative multi-task learning for the air channel via hierarchical GANs
(2024-08-19) Kuikka, Juho
Perustieteiden korkeakoulu | Master's thesis
In wireless communication, channel model refers to an abstraction that aims to explain how a transmitted signal is altered in the process of wireless communication. Currently, most of the channel models are a compromise between accuracy and computational ex- penses. In order to achieve higher accuracy with lower computational costs, deep learning based generative modelling has been suggested for the channel modelling problem, with promising results. However, a major drawback within the framework of deep learning is the amount of training data required for success. Since channel measurements are expensive to obtain, methods for enhancing the data efficiency of generative modelling must be investigated. Specifically, as channel models for different locations share inherent similarities, multi-task learning from different, yet related datasets could reduce required data volume for an individual channel model. This thesis investigates the deep generative modelling via generative adversarial net- works (GANs), their Bayesian generalisation, and finally proposes a novel generative modelling scheme for multi-task generation, motivated by Bayesian hierarchical mod- elling. Our simulations show that our proposed scheme does not only greatly enhance the data efficiency of the channel modelling, but it also decreases instabilities usually present in GAN training. Furthermore, as our proposed modelling scheme is of great generality, it may be utilised in any modelling problem where multiple related, but limited datasets are present.
Image Fusion using Joint Sparse Representations and Coupled Dictionary Learning
(2020-05) Ghorbani Veshki, Farshad; Ouzir, Nora; Vorobyov, Sergiy
A4 Artikkeli konferenssijulkaisussa
The image fusion problem consists in combining complementary parts of multiple images captured, for example, with different focal settings into one image of higher quality. This requires the identification of the sharpest areas in sets of input images. Recently, it was shown that coupled dictionary learning can successfully capture the relationships between high- and low-resolution patches in the context of single image super-resolution. In this work, to identify the sharp image patches, we propose an improved discriminative coupled dictionary learning approach using joint sparse representations in blurred and focused dictionaries. In addition, a pixel-wise processing of the boundaries (i.e., patches containing blurred and focused pixels) is proposed. The experimental results using two natural image datasets, as well as a sequence of in vivo microscopy images, show the competitiveness of the proposed method compared to state-of-the-art algorithms in terms of accuracy and computational time.
Joint Space-(Slow) Time Transmission With Unimodular Waveforms and Receive Adaptive Filter Design for Radar
(2018) Li, Yongzhe; Vorobyov, Sergiy
A4 Artikkeli konferenssijulkaisussa
A novel computationally efficient method for jointly designing the space-(slow) time (SST) transmission with unimodular waveforms and receive adaptive filter is developed for different radar configurations. The range sidelobe effect and Doppler characteristics are considered. In particular, we develop a novel approach for jointly synthesizing unimodular SST waveforms and minimum variance distortionless response receive adaptive filter for two cases of known Doppler information and presence of uncertainties on clutter bins. Corresponding non-convex optimization problems are formulated and efficient algorithms are derived. The main ideas of the algorithm developments are to decouple composite objective function of the formulated problems, generate minorizing surrogates, and then solve the joint design problem iteratively, but in closed form for each iteration by means of minorization-maximization technique. The proposed algorithms demonstrate good performance and have fast convergence speed and low complexity.
Radar Signal Design in Spectrally Dense Environment
(2017-08-28) Yli-Niemi, Markus
Sähkötekniikan korkeakoulu | Bachelor's thesis
Robust Least Mean Squares Estimation of Graph Signals
(2019-05-01) Miettinen, Jari; Vorobyov, Sergiy; Ollila, Esa
A4 Artikkeli konferenssijulkaisussa
Recovering a graph signal from samples is a central problem in graph signal processing. Least mean squares (LMS) method for graph signal estimation is computationally efficient adaptive method. In this paper, we introduce a technique to robustify LMS with respect to mismatches in the presumed graph topology. It builds on the fact that graph LMS converges faster when the graph topology is specified correctly. We consider two measures of convergence speed, based on which we develop randomized greedy algorithms for robust interpolation of graph signals. In simulation studies, we show that the randomized greedy robust least mean squares (RGRLMS) outperforms the regular LMS and has even more potential given a robust sampling design.
Soil Moisture Estimation with GNSS Interferometric Reflectometry and Multispectral Satellite Model
(2024-08-19) Padrón, Nicolás
Sähkötekniikan korkeakoulu | Master's thesis
During the last decade, GNSS Reflectometry (GNSS-R) together with its ground application, GNSS Interferometric Reflectometry (GNSS-IR), have been gaining momentum with satellite missions and ground campaigns. These passive remote sensing technique lies within the microwave remote sensing technology and makes use of satellite navigation signals for Earth Observation (EO) purposes. Another passive EO technology is multispectral satellite imagery from missions such as Landsat-8. Multispectral imagery is an optical remote sensing technology and covers various wavelengths of the optical spectrum, allowing to analyze the spectral response of the surface materials. This work focuses on GNSS Interferometric Reflectometry processing for soil moisture estimation, followed by a GNSS-IR aided multispectral model using Landsat-8 satellite data, with the aim of providing an accurate, cost-effective solution with wide-area coverage. In this research, a GNSS-IR processing chain is developed to estimate volumetric soil moisture surrounding static geodetic receivers. Results from this technique are used to fit a linear model with Landsat-8 data combining several optical indexes that have high correlation with soil moisture. Finally, the proof-of-concept of a multispectral imagery model, aided by local GNSS-IR results, is demonstrated and verified against data from the Soil Moisture Active Passive (SMAP) satellite mission for a wide-area coverage.
Tensor decompositions in wireless communications and mimo radar
(2021-04) Chen, Hongyang; Ahmad, Fauzia; Vorobyov, Sergiy; Porikli, Fatih
A2 Katsausartikkeli tieteellisessä aikakauslehdessä
The emergence of big data and the multidimensional nature of wireless communication signals present significant opportunities for exploiting the versatility of tensor decompositions in associated data analysis and signal processing. The uniqueness of tensor decompositions, unlike matrix-based methods, can be guaranteed under very mild and natural conditions. Harnessing the power of multilinear algebra through tensor analysis in wireless signal processing, channel modeling, and parametric channel estimation provides greater flexibility in the choice of constraints on data properties and permits extraction of more general latent data components than matrix-based methods.Tensor analysis has also found applications in Multiple-Input Multiple-Output (MIMO) radar because of its ability to exploit the inherent higher-dimensional signal structures therein. In this paper, we provide a broad overview of tensor analysis in wireless communications and MIMO radar. More specifically, we cover topics including basic tensor operations, common tensor decompositions via canonical polyadic and Tucker factorization models, wireless communications applications ranging from blind symbol recovery to channel parameter estimation, and transmit beamspace design and target parameter estimation in MIMO radar.