The band pass filter is used to attenuating breathing originated signal from the heart originated BCG signal. The bandwidth of the both signals slightly overlap, hereby the complete attenuation of the breathing is not possible without also altering the heart originated BCG waveforms and the parameters which are obtained from the BCG. In our study we investigated the optimal lower cut-off frequency, and 1.3 Hz was found as the reasonable compromise between the attenuation of the breathing and the altering of the heart originated BCG.

The emerging 5G New Radio (NR) networks are expected to enable huge improvements, e.g., in terms of capacity, number of connected devices, peak data rates and latency, compared to existing networks. At the same time, a new trend referred to as the RF convergence is aiming to jointly integrate communications and sensing functionalities into the same systems and hardware platforms. In this paper, we investigate the sensing prospects of 5G NR systems, with particular emphasis on the user equipment side and their potential for joint communications and environment mapping. To this end, a radio-based sensing approach utilizing the 5G NR uplink transmit signal and an efficient receiver processing and mapping scheme are proposed. An indoor scenario is then studied and evaluated through real-world RF measurements at 28 GHz mm-wave band, showing that impressive mapping performance can be achieved by the proposed system. The measurement data is available at a permanent open repository.

Positioning is considered one of the most important features and enabler of various novel industry verticals in future radio systems. Since path loss or received signal strength-based measurements are widely available and accessible in the majority of wireless standards, path loss-based positioning has an important role among other positioning technologies. Conventionally path loss-based positioning has two phases; i) fitting a path loss model to training data, if such training data is available, and ii) determining link distance estimates based on the path loss model and calculating the position estimate. However, in both phases, the maximum measurable path loss is limited by measurement noise. Such immeasurable samples are called censored path loss data and such noisy data is commonly neglected in both the model fitting and in the positioning phase. In the case of censored path loss, the loss is known to be above a known threshold level and that information can be used in model fitting as well as in the positioning phase. In this paper, we examine and propose how to use censored path loss data in path loss model-based positioning and demonstrate with simulations the potential of the proposed approach for considerable improvements (over 30%) in positioning accuracy.

Future 5G networks will serve both terrestrial and aerial users, thanks to their network slicing and flexible numerology capabilities. The probability of Line-of-Sight (LoS) propagation will be intuitively higher for aerial users than for terrestrial users and this will provide a trade-off between increased capacity and increased interference. Our paper analyzes theoretically this trade-off and proposes solutions based on downlink multiantenna beamorming and joint optimization of the signal-to- interference ratio of multiple aerial users. It is shown that Multiple-Input-Single-Output solutions offer the most convenient tradeoff between complexity and capacity/interference performance. Simulation results are provided for mm Wave bands and low-altitude aerial vehicles.

Powerful in-band interference can saturate a receiver's front-end and limit the usefulness of digital interference suppression methods that are bounded by the receiver's limited dynamic range. This is especially true for the self-interference (SI) encountered in full-duplex (FD) radios, but also in the case of strong interference between co-located radios. However, unlike in FD radios, receivers co-located with interference sources do not typically have direct access to the transmitted interference. This work analyzes the performance of a digitally-assisted analog interference mitigation method and its implementation for the suppression of frequency-modulated (FM) interference before quantization in global navigation satellite system (GNSS) receivers that are co-located with interference sources. Over-the-air measurement results are presented that illustrate the effects of interference mitigation on GPS L1 and Galileo E1 reception in a commercial off-the-shelf GNSS receiver and a software-defined GNSS receiver. The analysis covers the effects of the interference mitigation on the radio frequency (RF) front-end, acquisition, tracking, and positioning stages.

Deep learning models are capable of achieving state-of-the-art performance on a wide range of time series analysis tasks. However, their performance crucially depends on the employed normalization scheme, while they are usually unable to efficiently handle non-stationary features without first appropriately pre-processing them. These limitations impact the performance of deep learning models, especially when used for forecasting financial time series, due to their non-stationary and multimodal nature. In this paper we propose a data-driven adaptive normalization layer which is capable of learning the most appropriate normalization scheme that should be applied on the data. To this end, the proposed method first identifies the distribution from which the data were generated and then it dynamically shifts and scales them in order to facilitate the task at hand. The proposed nor-malization scheme is fully differentiable and it is trained in an end-to-end fashion along with the rest of the parameters of the model. The proposed method leads to significant performance improvements over several competitive normalization approaches, as demonstrated using a large-scale limit order book dataset.

Fog computing brings computation and services to the edge of networks enabling real time applications. In order to provide satisfactory quality of experience, the latency of fog networks needs to be minimized. In this paper, we consider a peer computation offloading problem for a fog network with unknown dynamics. Peer competition occurs when different fog nodes offload tasks to the same peer FN. In this paper, the computation offloading problem is modeled as a sequential FN selection problem with delayed feedback. We construct an online learning policy based on the adversary multi-arm bandit framework to deal with peer competition and delayed feedback. Simulation results validate the effectiveness of the proposed policy.

In this paper, the potential of extending 5G New Radio physical layer solutions to support communications in sub-THz frequencies is studied. More specifically, we introduce the status of third generation partnership project studies related to operation on frequencies beyond 52.6 GHz and note also the recent proposal on spectrum horizons provided by federal communications commission (FCC) related to experimental licenses on 95 GHz-3 THz frequency band. Then, we review the power amplifier (PA) efficiency and output power challenge together with the increased phase noise (PN) distortion effect in terms of the supported waveforms. As a practical example on the waveform and numerology design from the perspective of the PN robustness, link performance results using 90 GHz carrier frequency are provided. The numerical results demonstrate that new, higher subcarrier spacings are required to support high throughput, which requires larger changes in the physical layer design. It is also observed that new phase-tracking reference signal designs are required to make the system robust against PN. The results illustrate that single-carrier frequency division multiple access is significantly more robust against PN and can provide clearly larger PA output power than cyclic-prefix orthogonal frequency division multiplexing, and is therefore a highly potential waveform for sub-THz communications.

In this study, machine washing durability of working glove-integrated passive RFID tags is evaluated. These glove-tags are embedded inside 3D-printed thermoplastic polyurethane platforms. The results are compared to platforms embedded inside brush-painted encapsulant platforms. For a preliminary washing reliability evaluation, both types of glove- integrated platforms are washed in a washing machine for 5 times. Although both platforms can protect glove-tags from the effects of water, the main reliability challenge is found to be the fragile antenna-IC attachments. This paper introduces the two platform materials and the achieved washing test results. These preliminary results determine the future direction of this research: The next step is to study suitable methods to strengthen the interconnections, so that these glove-tags can survive the harsh environment inside a washing machine.

We present a headband loop antenna for wireless power transfer to multiple IMDs located in the cranial cavity at the depth of 10 mm from the skin. We characterize the wireless power transfer link in terms of the power gain and the power delivered to the IMD, when maximum SAR compliant transmission power is fed to the headband antenna at frequency of 5 MHz. We also consider two types of the misalignments i.e. lateral and angular, between the IMD antenna and the headband antenna and discuss their impact on the transducer gain, impedance matching and on the power delivered to the IMD.

We present a compact circularly polarized (CP) antenna for wearable passive UHF RFID tags. The antenna is a square-shaped microstrip patch antenna where we have applied corner truncation and slotting techniques in the top layer conductor for achieving the CP property and a shorting pin and loop structure for impedance matching. Despite using a lowpermittivity textile as antenna substrate, the antenna's footprint size is only 5-by-5 cm, which is approximately 15% of the operating wavelength. At the same time, the on-body measurements, the antenna's axial ratio is 0.9 dB and the measured attainable read range (reader's EIRP =3.28W) of the tag reaches 4.2 meters with a CP reader antenna and ranges from 2.9 meters to 3.4 meters for a linear reader antenna, depending on the rotation angle between the antennas.

The beyond-5G vehicular communications are expected not only to utilize the already explored millimeter-wave band but also to start harnessing the higher frequencies above 100 GHz ultimately targeting the so-called low terahertz band, 300 GHz-1 THz. In this paper, we perform a set of propagation measurements at 300 GHz band in representative vehicular environments. Particularly, we report on the reflection losses from the front, rear, and side of a regular vehicle. In addition, the penetration losses when propagating through, over, and under the vehicle are presented. Our study reveals that the vehicle body is extremely heterogeneous in terms of the propagation losses: the attenuation heavily depends on the trajectory of the 300 GHz signal through the vehicle. The reported measurement data may be used as a reference when developing the vehiclespecific channel and interference models for future wireless communications in the low terahertz band.

The interior structures of the comets and asteroids, still poorly known, might hold a unique key to understand the early Solar System. Considering the interaction of an illuminated electromagnetic wave with this kind of targets, these 'objects' are very large compared to the applicable wavelength. Consequently, tomographic imaging of such targets, i.e., reconstructing their interior structure via multiple measurements, constitutes a challenging inverse problem. To reach this objective and to develop and test inverse algorithms, we need to investigate electromagnetic fields that have interacted with structures analogous to real asteroids and comets. In this study, we focus on the acquisition of these fields considering three methods: calculated fields obtained with (1) time and (2) frequency domain methods and (3) microwave measurements performed for an analogue model, i.e., a small-scale asteroid model.

This paper presents the first prototype of a passive RFID-based textile touchpad. Our unique solution takes advantage of ICs from passive UHF RFID technology. These components are combined into a textile-integrated IC array, which can be used for handwritten character recognition. As the solution is fully passive and gets all the needed energy from the RFID reader, it enables a maintenance-free and cost-effective user interface that can be integrated into clothing and into textiles around us.

Deploying sub-THz frequencies for mobile communications is one timely research area, due to the availability of very wide and contiguous chunks of the radio spectrum. However, at such extremely high frequencies, there are large challenges related to, e.g., phase noise, propagation losses as well as to energy-efficiency, since generating and radiating power with reasonable efficiency is known to be far more difficult than at lower frequencies. To address the energy-efficiency and power amplifier (PA) nonlinear distortion related challenges, modulation methods and waveforms with low peak-to-average-power ratio (PAPR) are needed. To this end, a new modulation approach is formulated and proposed in this paper, referred to as constrained phase-shift keying (CPSK). The CPSK concept builds on the traditional PSK constellations, while additional constraints are applied to the time domain symbol transitions in order to control and reduce the PAPR of the resulting waveform. This new modulation is then compared with pulse-shaped π/2-BPSK and ordinary QPSK, in the discrete Fourier transform (DFT) spread orthogonal frequency division multiplexing (DFT-s-OFDM) context, in terms of the resulting PAPR distributions and the achievable maximum PA output power, subject to constraints in the passband waveform quality and out-of-band emissions. The obtained results show that the proposed CPSK approach allows for reducing the PAPR and thereon for achieving higher PA output powers, compared to QPSK, while still offering the same spectral efficiency. Overall, the CPSK concept offers a flexible modulation solution with controlled PAPR for the future sub-THz networks.

There is a strong interest in utilizing commercial cellular networks to support unmanned aerial vehicles (UAVs) to send control commands and communicate heavy traffic. Cellular networks are well suited for offering reliable and secure connections to the UAVs as well as facilitating traffic management systems to enhance safe operation. However, for the full-scale integration of UAVs that perform critical and high-risk tasks, more advanced solutions are required to improve wireless connectivity in mobile networks. In this context, integrated access and backhaul (IAB) is an attractive approach for the UAVs to enhance connectivity and traffic forwarding. In this paper, we study a novel approach to dynamic associations based on reinforcement learning at the edge of the network and compare it to alternative association algorithms. Considering the average data rate, our results indicate that the reinforcement learning methods improve the achievable data rate. The optimal parameters of the introduced algorithm are highly sensitive to the donor next generation node base (DgNB) and UAV IAB node densities, and need to be identified beforehand or estimated via a stateful search. However, its performance nearly converges to that of the ideal scheme with a full knowledge of the data rates in dense deployments of DgNBs.

The prospective millimeter-wave (mmWave) networks are envisioned to heavily utilize relay nodes to improve their performance in certain scenarios. In addition to the stationary mmWave relays already considered by 3GPP as one of the main focuses, the community recently started to explore the use of unmanned aerial vehicle (UAV)-based mmWave relays. These aerial nodes provide greater flexibility in terms of the relay placement in different environments as well as the ability to optimize the deployment height thus maximizing the cell performance. At the same time, the use of UAV-based relays leads to additional deployment complexity and expenditures for the network operators. In this paper, taking into account 3GPP-standardized mmWave-specific propagation, blockage, and resource allocation we compare the capacity gains brought by the static and the UAV-based mmWave relays in different scenarios. For each of the relay types, we investigate both uniform and clustered distribution of human users. The developed mathematical framework and a numerical study reveal that the highest capacity gains when utilizing the UAV-based relays instead of the static ones are observed in clustered deployments (up to 31%), while the performance difference between the UAV-based and the static mmWave relays under a uniform distribution of users is just 3%.

Cortical spreading depression (CSD) is a slowly propagating wave of depolarization of brain cells, followed by temporary silenced electrical brain activity. Major structural changes during CSD are linked to neuronal and possibly glial swelling. However, basic questions still remain unanswered. In particular, there are open questions regarding whether neurons or glial cells swell more, and how the cellular swelling affects the CSD wave propagation.In this study, we computationally explore how different parameters affect the swelling of neurons and astrocytes (starshaped glial cells) during CSD and how the cell swelling alters the CSD wave spatial distribution. We apply a homogenized mathematical model that describes electrodiffusion in the intraand extracellular space, and discretize the equations using a finite element method. The simulations are run with a twocompartment (extracellular space and neurons) and a threecompartment version of the model with astrocytes added. We consider cell swelling during CSD in four scenarios: (A) incorporating aquaporin-4 channels in the astrocytic membrane, (B) increasing the neuron/astrocyte ratio to 2:1, (C) blocking and increasing the Na⁺/K⁺-ATPase rate in the astrocytic compartment, and (D) blocking the Cl^- channels in astrocytes. Our results show that increasing the water permeability in the astrocytes results in a higher astrocytic swelling and a lower neuronal swelling than in the default case. Further, elevated neuronal density increases the swelling in both neurons and astrocytes. Blocking the Na⁺/K⁺-ATPase in the astrocytes leads to an increased wave width and swelling in both compartments, which instead decreases when the pump rate is raised. Blocking the Cl^- channels in the astrocytes results in neuronal swelling, and a shrinkage in the astrocytes. Our results suggest a supporting role of astrocytes in preventing cellular swelling and CSD, as well as highlighting how dysfunctions in astrocytes might elicit CSD.

Providing sufficient mobile coverage during mass public events or critical situations is a highly challenging task for the network operators. To fulfill the extreme capacity and coverage demands within a limited area, several augmenting solutions might be used. Among them, novel technologies like a fleet of compact base stations mounted on Unmanned Aerial Vehicles (UAVs) are gaining momentum because of their time- and cost- efficient deployment. Despite the fact that the concept of aerial wireless access networks has been investigated recently in many research studies, there are still numerous practical aspects that require further understanding and extensive evaluation. Taking this as a motivation, in this paper, we develop the concept of continuous wireless coverage provisioning by the means of UAVs and assess its usability in mass scenarios with thousands of users. With our system-level simulations as well as a measurement campaign, we take into account a set of important parameters including weather conditions, UAV speed, weight, power consumption, and millimeter- wave (mmWave) antenna configuration. As a result, we provide more realistic data about the performance of the access and backhaul links together with the practical lessons learned about the design and real-world applicability of the UAV-enabled wireless access networks.

The limitations of state-of-the-art cellular modems prevent achieving low-power and low-latency Machine Type Communications (MTC) based on current power saving mechanisms alone. Recently, the concept of wake-up scheme has been proposed to enhance battery lifetime of 5G devices, while reducing the buffering delay. The existing wake-up algorithms use static operational parameters that are determined by the radio access network at the start of the userâ™s session. In this paper, the average power consumption of the wake-up enabled MTC UE is modeled by using a semi-Markov process and then optimized through a delay-constrained optimization problem, by which the optimal wake-up cycle is obtained in closed form. Numerical results show that the proposed solution reduces the power consumption of an optimized Discontinuous Reception (DRX) scheme by up to 40% for a given delay requirement.

This paper explores the activity of coding with smart toy robots Dash and Botley as a part of playful learning in the Finnish early education context. The findings of our study demonstrate how coding with the two toy robots was approached, conducted and played by Finnish preschoolers aged 5-6 years. The main conclusion of the study is that preschoolers used the toy robots with affordances related to coding mainly in developing gamified play around them by designing tracks for the toys, programming the toys to solve obstacle paths, and competing in player-generated contests of dexterity, speed and physically mobile play.

To meet the prospective demands of intelligent transportation systems (ITS), the Release 14 (Rel-14) and Rel-15 of the Long Term Evolution (LTE) specifications include solutions for enhanced vehicle-to-everything (V2X) communications. While the technical enablers of Rel-14 are suitable for delivering basic safety messages, Rel-15 supports more demanding ITS services with stringent latency and reliability. Starting in Rel-15 and continuing in Rel-16, the 3GPP was developing a novel radio interface for 5G systems, termed the New Radio (NR), which will enable ultra reliable and low latency communications suitable even for the most demanding ITS applications. In this paper, we overview the new V2X-specific features in Rel-15 and Rel-16. Further, we argue that future V2X and automotive radar systems may reuse common equipment, such as millimeter-wave antenna arrays. We finally discuss the vision of joint vehicular communications and radar sensing as well as characterize unified channel access for millimeter-wave vehicular communications and radar sensing.

Due to the closely-spaced antenna elements in largearray or massive MIMO transmitters, antenna crosstalk is inevitable. This imposes additional challenges when seeking to linearize the power amplifiers at the transmitter through digital predistortion (DPD). In the commonly applied indirect learning architecture (ILA), the antenna crosstalk is known to result in a large amount of additional basis functions (BFs) in order to account for all the coupling signal terms and achieve good linearization. In this article, we propose a novel closed-loop DPD architecture and associated parameter learning algorithms that can provide efficient linearization of digital MIMO transmitters under antenna crosstalk. The proposed solution does not need extra basis functions, and is thus shown to provide large benefits in terms of computational complexity compared to existing state-of-the-art. Comprehensive numerical results are also provided, showing excellent linearization performance outperforming the existing reference methods.

We briefly introduce two submissions to the Illumination Estimation Challenge, in the Int'l Workshop on Color Vision, affiliated to the 11th Int'l Symposium on Image and Signal Processing and Analysis. The fourier-transform-based submission is ranked 3rd, and the statistical Gray-pixel-based one ranked 6th.

Graphics Processing Units (GPU) have been widely used in various fields of scientific computing, such as in signal processing. GPUs have a hierarchical memory structure with memory layers that are shared between GPU processing elements. Partly due to the complex memory hierarchy, GPU programming is non-Trivial, and several aspects must be taken into account, one being memory access patterns. One of the fastest GPU memory layers, shared memory, is grouped into banks to enable fast, parallel access for processing elements. Unfortunately, it may happen that multiple threads of a GPU program may access the same shared memory bank simultaneously causing a bank conflict. If this happens, program execution slows down as memory accesses have to be rescheduled to determine which instruction to execute first. Bank conflicts are not taken into account automatically by the compiler, and hence the programmer must detect and deal with them prior to program execution. In this paper, we present an algebraic approach to detect bank conflicts and prove some theoretical results that can be used to predict when bank conflicts happen and how to avoid them. Also, our experimental results illustrate the savings in computation time.

The Internet of Things (IoT) enables long-range outdoor networks, such as smart grid and municipal lighting, as well as short-range indoor systems for smart homes, residential security, and energy management. Wireless connectivity and standardized communication protocols become an essential technology baseline for these diverse IoT applications. The focus of this work is wireless connectivity for smart metering systems. One of the recent protocols in this field is Wireless M-BUS, which is being widely utilized for remote metering applications across Europe. Therefore, in this paper, we detail a novel multi-platform framework designed to serve as a data generator for the protocol in question. The developed software allows to construct Wireless M-Bus telegrams with a high level of detail according to the EN 13757-4 specification and then schedule them for periodic transmission. The evaluation of the data generator is done in real scenario by using previously developed prototype equipped with IQRF TR72DA communication module acting as a smart meter with implemented software framework. As a result, the evaluation of communication distance between the developed Wireless MBus prototype and commercial gateway was tested in case of indoor scenario at Brno University of Technology, Faculty of Electrical Engineering and Communication.

Narrowband IoT (NB-IoT) stands for a radio access technology standardized by the 3GPP organization in Release 13 to enable a large set of use-cases for massive Machine-type Communications (mMTCs). Compared to legacy human-oriented 4G (LTE) communication systems, NB-IoT has game-changing features in terms of extended coverage, enhanced power saving modes, and a reduced set of available functionality. At the end of the day, these features allow for connectivity of devices in challenging positions, enabling long battery life and reducing device complexity. This article addresses the development of the universal testing device for delay-tolerant services allowing for in-depth verification of NB-IoT communication parameters. The presented outputs build upon our long-term cooperation with the Vodafone Czech Republic a.s. company.

In this paper we discuss how does the input magnitude data setting influence the behavior of error-reduction algorithm in the case of the one-dimensional discrete phase retrieval problem. We present experimental results related to the convergence or stagnation of the algorithm. We also discuss the issue of the zeros distribution of the solution, when the solution of the problem exists.

360° videos are increasingly used for media and entertainment, but the best practices for editing them are not yet well established. In this paper, we present a study in which we investigated the user experience of 360° music videos viewed on computer monitor and VR goggles. The research was conducted in the form of a laboratory experiment with 20 test participants. During the within-subject study, participants watched and evaluated four versions of the same 360° music video with a different cutting rate. Based on the results, an average cutting rate of 26 seconds delivered the highest-quality user experience both for computer monitor and VR goggles. The cutting rate matched with participants' mental models, and there was enough time to explore the environment without getting bored. Faster cutting rates made the users nervous, and a video consisting of a single shot was considered to be too static and boring.

This paper proposes a low algorithmic latency adaptation of the deep clustering approach to speaker-independent speech separation. It consists of three parts: a) the usage of long-short-term-memory (LSTM) networks instead of their bidirectional variant used in the original work, b) using a short synthesis window (here 8 ms) required for low-latency operation, and, c) using a buffer in the beginning of audio mixture to estimate cluster centres corresponding to constituent speakers which are then utilized to separate speakers within the rest of the signal. The buffer duration would serve as an initialization phase after which the system is capable of operating with 8 ms algorithmic latency. We evaluate our proposed approach on two-speaker mixtures from Wall Street Journal (WSJ0) corpus. We observe that the use of LSTM yields around one dB lower SDR as compared to the baseline bidirectional LSTM in terms of source to distortion ratio (SDR). Moreover, using an 8 ms synthesis window instead of 32 ms degrades the separation performance by around 2.1 dB as compared to the baseline. Finally, we also report separation performance with different buffer durations noting that separation can be achieved even for buffer duration as low as 300 ms.

The Time Difference of Arrival (TDoA) of a sound wavefront impinging on a microphone pair carries spatial information about the source. However, captured speech typically contains dynamic non-speech interference sources and noise. Therefore, the TDoA estimates fluctuate between speech and interference. Deep Neural Networks (DNNs) have been applied for Time-Frequency (TF) masking for Acoustic Source Localization (ASL) to filter out non-speech components from a speaker location likelihood function. However, the type of TF mask for this task is not obvious. Secondly, the DNN should estimate the TDoA values, but existing solutions estimate the TF mask instead. To overcome these issues, a direct formulation of the TF masking as a part of a DNN-based ASL structure is proposed. Furthermore, the proposed network operates in an online manner, i.e., producing estimates frame-by-frame. Combined with the use of recurrent layers it exploits the sequential progression of speaker related TDoAs. Training with different microphone spacings allows model re-use for different microphone pair geometries in inference. Real-data experiments with smartphone recordings of speech in interference demonstrate the network's generalization capability.

The energy efficiency of modern MPSoCs is enhanced by complex hardware features such as Dynamic Voltage and Frequency Scaling (DVFS) and Dynamic Power Management (DPM). This paper introduces a new method, based on convex problem solving, that determines the most energy efficient operating point in terms of frequency and number of active cores in an MPSoC. The solution can challenge the popular approaches based on never-idle (or As-Slow-As-Possible (ASAP)) and race-to-idle (or As-Fast-As-Possible (AFAP)) principles. Experimental data are reported using a Samsung Exynos 5410 MPSoC and show a reduction in energy of up to 27 % when compared to ASAP and AFAP.

Developing accurate financial analysis tools can be useful both for speculative trading, as well as for analyzing the behavior of markets and promptly responding to unstable conditions ensuring the smooth operation of the financial markets. This led to the development of various methods for analyzing and forecasting the behaviour of financial assets, ranging from traditional quantitative finance to more modern machine learning approaches. However, the volatile and unstable behavior of financial markets forbids the accurate prediction of future prices, reducing the performance of these approaches. In contrast, in this paper we propose a novel price trailing method that goes beyond traditional price forecasting by reformulating trading as a control problem, effectively overcoming the aforementioned limitations. The proposed method leads to developing robust agents that can withstand large amounts of noise, while still capturing the price trends and allowing for taking profitable decisions.

1D Convolutional Neural Networks (CNNs) have recently become the state-of-the-art technique for crucial signal processing applications such as patient-specific ECG classification, structural health monitoring, anomaly detection in power electronics circuitry and motor-fault detection. This is an expected outcome as there are numerous advantages of using an adaptive and compact 1D CNN instead of a conventional (2D) deep counterparts. First of all, compact 1D CNNs can be efficiently trained with a limited dataset of 1D signals while the 2D deep CNNs, besides requiring 1D to 2D data transformation, usually need datasets with massive size, e.g., in the »Big Data» scale in order to prevent the well-known »overfitting» problem. 1D CNNs can directly be applied to the raw signal (e.g., current, voltage, vibration, etc.) without requiring any pre- or post-processing such as feature extraction, selection, dimension reduction, denoising, etc. Furthermore, due to the simple and compact configuration of such adaptive 1D CNNs that perform only linear 1D convolutions (scalar multiplications and additions), a real-time and low-cost hardware implementation is feasible. This paper reviews the major signal processing applications of compact 1D CNNs with a brief theoretical background. We will present their state-of-the-art performances and conclude with focusing on some major properties. Keywords - 1-D CNNs, Biomedical Signal Processing, SHM.

Forecasting time series has several applications in various domains. The vast amount of data that are available nowadays provide the opportunity to use powerful deep learning approaches, but at the same time pose significant challenges of high-dimensionality, velocity and variety. In this paper, a novel logistic formulation of the well-known Bag-of-Features model is proposed to tackle these challenges. The proposed method is combined with deep convolutional feature extractors and is capable of accurately modeling the temporal behavior of time series, forming powerful forecasting models that can be trained in an end-to-end fashion. The proposed method was extensively evaluated using a large-scale financial time series dataset, that consists of more than 4 million limit orders, outperforming other competitive methods.

Block partition structure is a critical module in video coding scheme to achieve significant gap of compression performance. Under the exploration of future video coding standard by the Joint Video Exploration Team (JVET), named Versatile Video Coding (VVC), a new Quad Tree Binary Tree (QTBT) block partition structure has been introduced. In addition to the QT block partitioning defined by High Efficiency Video Coding (HEVC) standard, new horizontal and vertical BT partitions are enabled, which drastically increases the encoding time compared to HEVC. In this paper, we propose a fast QTBT partitioning scheme based on a Machine Learning approach. Complementary to techniques proposed in literature to reduce the complexity of HEVC Quad Tree (QT) partitioning, the propose solution uses Random Forest classifiers to determine for each block which partition modes between QT and BT is more likely to be selected. Using uncertainty zones of classifier decisions, the proposed complexity reduction technique is able to reduce in average by 30% the encoding time of JEM-v7.0 software in Random Access configuration with only 0.57% Bjontegaard Delta Rate (BD-BR) increase.

In recent years, several studies have established a relationship between mammographic parenchymal patterns and breast cancer risk. However, there is a lack of publicly available data and software for objective comparison and clinical validation. This paper presents an open and adaptable implementation (OpenBreast v1.0) of a fully-Automatic computerized framework for mammographic image analysis for breast cancer risk assessment. OpenBreast implements mammographic image analysis in four stages: breast segmentation, detection of region-of-interests, feature extraction and risk scoring. For each stage, we provide implementations of several state-of-The-Art methods. The pipeline is tested on a set of 305 full-field digital mammography images corresponding to 84 patients (51 cases and 49 controls) from the breast cancer digital repository (BCDR). OpenBreast achieves a competitive AUC of 0.846 in breast cancer risk assessment. In addition, used jointly with widely accepted risk factors such as patient age and breast density, mammographic image analysis using OpenBreast shows a statistically significant improvement in performance with an AUC of 0.876 (\mathrm{p}<0.001). Our framework will be made publicly available and it is easy to incorporate new methods.

Photonic neuromorphic hardware can provide significant performance benefits for Deep Learning (DL) applications by accelerating and reducing the energy requirements of DL models. However, photonic neuromorphic architectures employ different activation elements than those traditionally used in DL, slowing down the convergence of the training process for such architectures. An initialization scheme that can be used to efficiently train deep photonic networks that employ quadratic sinusoidal activation functions is proposed in this paper. The proposed initialization scheme can overcome these limitations, leading to faster and more stable training of deep photonic neural networks. The ability of the proposed method to improve the convergence of the training process is experimentally demonstrated using two different DL architectures and two datasets.

Ground Penetrating Radar (GPR) is generally used as a non-destructive method of inspection for structures and for finding defects in concrete slabs. In this paper, GPR is used in the detection of water inside the cavities of concrete hollow core slabs. We propose an algorithm that determines the water level inside the concrete slab by analyzing the time delays of the reflections originating from inside the cavity. The algorithm is based on utilizing prior knowledge about the geometry of the hollow core slab. The presence of water was successfully detected and an estimate for the height of the water surface was obtained with a GPR system operating with a central frequency of 2.7 GHz. Based on the experiments, the proposed method holds promise in providing a robust and accurate method for the detection of water inside the concrete slabs. Results, possible future research and analysis of the feasibility of GPR systems in water detection are presented and discussed.

Terahertz (THz) band communications, capable of achieving the theoretical capacity of up to several terabits-per-second, are one of the attractive enablers for beyond 5G wireless networks. THz systems will use extremely directional narrow beams, allowing not only to extend the communication range but also to partially secure the data already at the physical layer. The reason is that, in most cases, the Attacker has to be located within the transmitter beam in order to eavesdrop the message. However, even the use of very narrow beams results in the considerably large area around the receiver, where the Attacker can capture all the data. In this paper, we study how to decrease the message eavesdropping probability by leveraging the inherent multi-path nature of the THz communications. We particularly propose sharing the data transmission over multiple THz propagation paths currently available between the communicating entities. We show that, at a cost of the slightly reduced link capacity, the message eavesdropping probability in the described scheme decreases significantly even when several Attackers operate in a cooperative manner. The proposed solution can be utilized for the transmission of the sensitive data, as well as to secure the key exchange in THz band networks beyond 5G.

The intermittent millimeter-wave radio links as a result of human-body blockage are an inherent feature of the 5G New Radio (NR) technology by 3GPP. To improve session continuity in these emerging systems, two mechanisms have recently been proposed, namely, multi-connectivity and guard bandwidth. The former allows to establish multiple spatially-diverse connections and switch between them dynamically, while the latter reserves a fraction of system bandwidth for sessions changing their state from non-blocked to blocked, which ensures that the ongoing sessions have priority over the new ones. In this paper, we assess the joint performance of these two schemes for the user- and system-centric metrics of interest. Our numerical results reveal that the multi-connectivity operation alone may not suffice to increase the ongoing session drop probability considerably. On the other hand, the use of guard bandwidth significantly improves session continuity by somewhat compromising new session drop probability and system resource utilization. Surprisingly, the 5G NR system implementing both these techniques inherits their drawbacks. However, complementing it with an initial AP selection procedure effectively alleviates these limitations by maximizing the system resource utilization, while still providing sufficient flexibility to enable the desired trade-off between new and ongoing session drop probabilities.

This paper investigates the performance of energy detection-based spectrum sensing over Fisher-Snedecor F fading channels. To this end, an analytical expression for the corre- sponding average detection probability is firstly derived and then this is extended to account for collaborative spectrum sensing. The complementary receiver operating characteristics (ROC) are analyzed for different conditions of the average signal-to- noise ratio (SNR), time-bandwidth product, multipath fading, shadowing and number of collaborating users. It is shown that the energy detection performance is strongly linked to the severity of the multipath fading and amount of shadowing, whereby even small variations in either of these physical phenomena significantly impact the detection probability. Also, the versatile modeling capability of the Fisher-Snedecor F distribution is veridfied in the context of energy detection based spectrum sensing as it provides considerably more accurate characterization than the conventional Rayleigh fading model. To confirm the validity of the analytical results presented in this paper, we compare them with the results of some simulations.

Markov Decision Processes (MDPs) provide important capabilities for facilitating the dynamic adaptation of hardware and software configurations to the environments in which they operate. However, the use of MDPs in embedded signal processing systems is limited because of the large computational demands for solving this class of system models. This paper presents Sparse Parallel Value Iteration (SPVI), a new algorithm for solving large MDPs on resource-constrained embedded systems that are equipped with mobile GPUs. SPVI leverages recent advances in parallel solving of MDPs and adds sparse linear algebra techniques to significantly outperform the state-of-the-art. The method and its application are described in detail, and demonstrated with case studies that are implemented on an NVIDIA Tegra K1 System On Chip (SoC). The experimental results show execution time improvements in the range of 65 % -78% for several applications. SPVI also lifts restrictions required by other MDP solver approaches, making it more widely compatible with large classes of optimization problems.

This paper proposes an active learning method to control a labeling process for efficient annotation of acoustic training material, which is used for training sound event classifiers. The proposed method performs K-medoids clustering over an initially unlabeled dataset, and medoids as local representatives, are presented to an annotator for manual annotation. The annotated label on a medoid propagates to other samples in its cluster for label prediction. After annotating the medoids, the annotation continues to the unexamined sounds with mismatched prediction results from two classifiers, a nearest-neighbor classifier and a model-based classifier, both trained with annotated data. The annotation on the segments with mismatched predictions are ordered by the distance to the nearest annotated sample, farthest first. The evaluation is made on a public environmental sound dataset. The labels obtained through a labeling process controlled by the proposed method are used to train a classifier, using supervised learning. Only 20% of the data needs to be manually annotated with the proposed method, to achieve the accuracy with all the data annotated. In addition, the proposed method clearly outperforms other active learning algorithms proposed for sound event classification through all the experiments, simulating varying fraction of data that is manually labeled.

To detect the class, and start and end times of sound events in real world recordings is a challenging task. Current computer systems often show relatively high frame-wise accuracy but low event-wise accuracy. In this paper, we attempted to merge the gap by explicitly including sequential information to improve the performance of a state-of-the-art polyphonic sound event detection system. We propose to 1) use delayed predictions of event activities as additional input features that are fed back to the neural network; 2) build N-grams to model the co-occurrence probabilities of different events; 3) use se-quentialloss to train neural networks. Our experiments on a corpus of real world recordings show that the N-grams could smooth the spiky output of a state-of-the-art neural network system, and improve both the frame-wise and the event-wise metrics.

We present an overview of the challenge entries for the Acoustic Scene Classification task of DCASE 2017 Challenge. Being the most popular task of the challenge, acoustic scene classification entries provide a wide variety of approaches for comparison, with a wide performance gap from top to bottom. Analysis of the submissions confirms once more the popularity of deep-learning approaches and mel frequency representations. Statistical analysis indicates that the top ranked system performed significantly better than the others, and that combinations of top systems are capable of reaching close to perfect performance on the given data.

This paper presents a low-latency neural network based speech enhancement system. Low-latency operation is critical for speech communication applications. The system uses the time-frequency (TF) masking approach to retain speech and remove the non-speech content from the observed signal. The ideal TF mask are obtained by supervised training of neural networks. As the main contribution different neural network models are experimentally compared to investigate computational complexity and speech enhancement performance. The proposed system is trained and tested on noisy speech data where signal-to-noise ratio (SNR) ranges from -5 dB to +5 dB and the results show significant reduction of non-speech content in the resulting signal while still meeting a low-latency operation criterion, which is here considered to be less than 20 ms.

Mean square error (MSE) has been the preferred choice as loss function in the current deep neural network (DNN) based speech separation techniques. In this paper, we propose a new cost function with the aim of optimizing the extended short time objective intelligibility (ESTOI) measure. We focus on applications where low algorithmic latency (≤ 10 ms) is important. We use long short-term memory networks (LSTM) and evaluate our proposed approach on four sets of two-speaker mixtures from extended Danish hearing in noise (HINT) dataset. We show that the proposed loss function can offer improved or at par objective intelligibility (in terms of ESTOI) compared to an MSE optimized baseline while resulting in lower objective separation performance (in terms of the source to distortion ratio (SDR)). We then proceed to propose an approach where the network is first initialized with weights optimized for MSE criterion and then trained with the proposed ESTOI loss criterion. This approach mitigates some of the losses in objective separation performance while preserving the gains in objective intelligibility.

This paper proposes a novel method for separation of sound sources with ambisonic signals using multichannel non-negative matrix factorization (MNMF) for source spectrogram estimation. We present a novel frequency-independent spatial covariance matrix (SCM) model for spherical harmonic (SH) domain signals which makes the MNMF parameter estimation framework computationally feasible up to 3rd order SH signals. The evaluation is done with simulated SH domain mixtures by measuring the separation performance using objective criteria and comparing the proposed method against SH domain beamforming. The proposed method improves average separation performance over beamforming with post-filtering when using 1st and 2nd order SH signals while at higher orders performance among all tested methods is similar.

Parkinson's disease (PD) is a degenerative and long-term disorder of the central nervous system, which often causes motor symptoms, e.g., tremor, rigidity, and slowness. Currently, the diagnosis of PD is based on patient history and clinical examination. Technology-derived decision support systems utilizing, for example, sensor-rich smartphones can facilitate more accurate PD diagnosis. These technologies could provide less obtrusive and more comfortable remote symptom monitoring. The recent studies showed that motor symptoms of PD can reliably be detected from data gathered via smartphones. The current study utilized an open-access dataset named 'mPower' to assess the feasibility of discriminating PD from non-PD by analyzing a single self-administered 20-step walking test. From this dataset, 1237 subjects (616 had PD) who were age and gender matched were selected and classified into PD and non-PD categories. Linear acceleration (ACC) and gyroscope (GYRO) were recorded by built-in sensors of smartphones. Walking bouts were extracted by thresholding signal magnitude area of the ACC signals. Features were computed from both ACC and GYRO signals and fed into a random forest classifier of size 128 trees. The classifier was evaluated deploying 100-fold cross-validation and provided an accumulated accuracy rate of 0.7 after 10k validations. The results show that PD and non-PD patients can be separated based on a single short-lasting self-administered walking test gathered by smartphones' built-in inertial measurement units.

Postural Instability (PI) is a major reason for fall in geriatric population as well as for people with diseases or disorders like Parkinson's, stroke etc. Conventional stability indicators like Berg Balance Scale (BBS) require clinical settings with skilled personnel's interventions to detect PI and finally classify the person into low, mid or high fall risk categories. Moreover these tests demand a number of functional tasks to be performed by the patient for proper assessment. In this paper a machine learning based approach is developed to determine fall risk with minimal human intervention using only Single Limb Stance exercise. The analysis is done based on the spatiotemporal dynamics of skeleton joint positions obtained from Kinect sensor. A novel posture modeling method has been applied for feature extraction along with some traditional time domain and metadata features to successfully predict the fall risk category. The proposed unobstrusive, affordable system is tested over 224 subjects and is able to achieve 75% mean accuracy on the geriatric and patient population.

Atrial fibrillation (AF) is the most common type of cardiac arrhythmia. Although not life-threatening itself, AF significantly increases the risk of stroke and myocardial infarction. Current tools available for screening and monitoring of AF are inadequate and an unobtrusive alternative, suitable for long-term use, is needed. This paper evaluates an atrial fibrillation detection algorithm based on wrist photoplethysmographic (PPG) signals. 29 patients recovering from surgery in the post-anesthesia care unit were monitored. 15 patients had sinus rhythm (SR, 67.5± 10.7 years old, 7 female) and 14 patients had AF (74.8± 8.3 years old, 8 female) during the recordings. Inter-beat intervals (IBI) were estimated from PPG signals. As IBI estimation is highly sensitive to motion or other types of noise, acceleration signals and PPG waveforms were used to automatically detect and discard unreliable IBI. AF was detected from windows of 20 consecutive IBI with 98.45±6.89% sensitivity and 99.13±1.79% specificity for 76.34±19.54% of the time. For the remaining time, no decision was taken due to the lack of reliable IBI. The results show that wrist PPG is suitable for long term monitoring and AF screening. In addition, this technique provides a more comfortable alternative to ECG devices.

This paper proposes a method for online estimation of time-varying room impulse responses (RIR) between multiple isolated sound sources and a far-field mixture. The algorithm is formulated as adaptive convolutive filtering in short-time Fourier transform (STFT) domain. We use the recursive least squares (RLS) algorithm for estimating the filter parameters due to its fast convergence rate, which is required for modeling rapidly changing RIRs of moving sound sources. The proposed method allows separation of reverberated sources from the far-field mixture given that their close-field signals are available. The evaluation is based on measuring unmixing performance (removal of reverberated source) using objective separation criteria calculated between the ground truth recording of the preserved sources and the unmixing result obtained with the proposed algorithm. We compare online and offline formulations for the RIR estimation and also provide evaluation with blind source separation algorithm only operating on the mixture signal.

There is an emerging interest aiming at defining principles for signals on general graphs, which are analogous to the basic principles in traditional signal processing. One example is the Graph Fourier Transform which aims at decomposing a graph signal into its components based on a set of basis functions with corresponding graph frequencies. It has been observed that most of the important information of a graph signal is contained inside the low frequency band, which leads to several applications such as denoising, compression, etc. In this paper, we show that the low frequency basis functions span the salient regions in an image, which can also be considered as important regions. Motivated by this, we present a novel simple and unsupervised method to utilize a number of low-energy basis functions and show that it improves the performance of seven state-of-the-art salient object detection methods in five datasets under four different evaluation criteria, with only minor exceptions.

This paper studies the problem of full reference visual quality assessment of denoised images with a special emphasis on images with low contrast and noise-like texture. Denoising of such images together with noise removal often results in image details loss or smoothing. A new test image database, FLT, containing 75 noise-free 'reference' images and 300 filtered ('distorted') images is developed. Each reference image, corrupted by an additive white Gaussian noise, is denoised by the BM3D filter with four different values of threshold parameter (four levels of noise suppression). After carrying out a perceptual quality assessment of distorted images, the mean opinion scores (MOS) are obtained and compared with the values of known full reference quality metrics. As a result, the Spearman Rank Order Correlation Coefficient (SROCC) between PSNR values and MOS has a value close to zero, and SROCC between values of known full-reference image visual quality metrics and MOS does not exceed 0.82 (which is reached by a new visual quality metric proposed in this paper). The FLT dataset is more complex than earlier datasets used for assessment of visual quality for image denoising. Thus, it can be effectively used to design new image visual quality metrics for image denoising.

In this paper, we propose a joint framework for target localization and classification using a single generalized model for non-imaging based multi-modal sensor data. For target localization, we exploit both sensor data and estimated dynamics within a local neighborhood. We validate the capabilities of our framework by using a multi-modal dataset, which includes ground truth GPS information (e.g., time and position) and data from co-located seismic and acoustic sensors. Experimental results show that our framework achieves better classification accuracy compared to recent fusion algorithms using temporal accumulation and achieves more accurate target localizations than multilateration.

Due to the increased popularity of augmented and virtual reality experiences, the interest in representing the real world in an immersive fashion has never been higher. Distributing such representations enables users all over the world to freely navigate in never seen before media experiences. Unfortunately, such representations require a large amount of data, not feasible for transmission on today's networks. Thus, efficient compression technologies are in high demand. This paper proposes an approach to compress 3D video data utilizing 2D video coding technology. The proposed solution was developed to address the needs of 'tele-immersive' applications, such as virtual (VR), augmented (AR) or mixed (MR) reality with Six Degrees of Freedom (6DoF) capabilities. Volumetric video data is projected on 2D image planes and compressed using standard 2D video coding solutions. A key benefit of this approach is its compatibility with readily available 2D video coding infrastructure. Furthermore, objective and subjective evaluation shows significant improvement in coding efficiency over reference technology.

In this paper, we exploit the 3D-beamforming features of multiantenna equipment employed in fifth generation (5G) networks, operating in the millimeter wave (mmW) band, for accurate positioning and tracking of users. We consider sequential estimation of users' positions, and propose a two-stage extended Kalman filter (EKF) that is based on reference signal received power (RSRP) measurements. In particular, beamformed downlink (DL) reference signals (RSs) are transmitted by multiple base stations (BSs) and measured by user equipments (UEs) employing receive beamforming. The so-obtained beam-RSRP (BRSRP) measurements are reported to the BSs where the corresponding directions of departure (DoDs) are sequentially estimated by a novel EKF. Such angle estimates from multiple BSs are subsequently fused on a central entity into 3D position estimates of UEs by means of another (second-stage) EKF. The proposed positioning scheme is scalable since the computational burden is shared among different network entities, namely transmission/reception points (TRPs) and 5G-NR Node B (gNB), and may be accomplished with the signalling currently specified for 5G. We assess the performance of the proposed algorithm on a realistic outdoor 5G deployment with a detailed ray tracing propagation model based on the METIS Madrid map. Numerical results with a system operating at 39 GHz show that sub-meter 3D positioning accuracy is achievable in future mmW 5G networks.

Building a complete inertial navigation system using the limited quality data provided by current smartphones has been regarded challenging, if not impossible. This paper shows that by careful crafting and accounting for the weak information in the sensor samples, smartphones are capable of pure inertial navigation. We present a probabilistic approach for orientation and use-case free inertial odometry, which is based on double-integrating rotated accelerations. The strength of the model is in learning additive and multiplicative IMU biases online. We are able to track the phone position, velocity, and pose in realtime and in a computationally lightweight fashion by solving the inference with an extended Kalman filter. The information fusion is completed with zero-velocity updates (if the phone remains stationary), altitude correction from barometric pressure readings (if available), and pseudo-updates constraining the momentary speed. We demonstrate our approach using an iPad and iPhone in several indoor dead-reckoning applications and in a measurement tool setup.

This paper presents an algorithm for multiple source localization using a beamforming-inspired spatial covariance model (SCM) and complex non-negative matrix factorization (CNMF). In this work, we assume that the source signals are known in advance whereas the mixing filter is modeled by the weighted sum of direction of arrival (DOA) kernels which encode the phase and the amplitude differences between microphones for every possible source direction. The direction of arrival (i.e. azimuth and elevation) for each source is estimated using CNMF. The proposed system is evaluated for DOA estimation task using two datasets covering a large number of configurations (number of channels, number of simultaneous sources, reverberation time, microphones spacing, source types and angular positions of the sources). Finally, a comparison to other state-of-the-art methods is performed, showing the robustness of the proposed method.

Many approaches to compressive video recovery proceed iteratively, treating the difference between the previous estimate and the ideal video as residual noise to be filtered. We go beyond the common white-noise modeling by adaptively modeling the residual as stationary spatiotemporally correlated noise. This adaptive noise model is updated at each iteration and is highly anisotropic in space and time; we leverage it with respect to the transform spectra of a motion-compensated video denoiser. Experimental results demonstrate that our proposed adaptive correlated noise model outperforms state-of-the-art methods both quantitatively and qualitatively.

To meet the capacity demand of future cellular networks, which is expected to increase substantially by 2020, using mm Wave band has been suggested. Although this higher frequency band has larger amounts of spectrum and will lead to much higher capacity in 5G networks, it has some propagation constraints such as limited coverage and sensitivity to Line of Sight (LoS) blockage. A possible solution to these issues could be small cell densification combined with utilizing access points carried by Unmanned Aerial Vehicles (UAVs). This paper presents a performance evaluation of the UAV-assisted mmWave network in urban environments using ns-3 simulations.

Data clustering is a fundamental machine learning problem. Community structure is common in social and biological networks. In this article we propose a novel data clustering algorithm that uses this phenomenon in mutual k - nearest neighbor (MKNN) graph constructed from the input dataset. We use the authentic scores-a metric that measures the strength of an edge in a social network graph-to rank all the edges in the MKNN graph. By removing the edges gradually in the order of their authentic scores, we collapse the MKNN graph into components to find the clusters. The proposed method has two major advantages comparing to other popular data clustering algorithms. First, it is robust to the noise in the data. Second, it finds clusters of arbitrary shape. We evaluated our algorithm on synthetic noisy datasets, synthetic 2D datasets and real-world image datasets. Results on the noisy datasets show that the proposed algorithm clearly outperforms the competing algorithms in terms of Normalized Mutual Information (NMI) scores. The proposed algorithm is the only one that does not fail on any data in the the synthetic 2D dataset, which are specifically designed to show the limitations of the clustering algorithms. When testing on the real-world image datasets, the best NMI scores achieved by the proposed algorithm is more than any other competing algorithm. The proposed algorithm has computational complexity of O(k3n+kn\log (kn)) and space complexity of O(kn), which is better than or equivalent to the most popular clustering algorithms.

This paper presents a method for improving the accuracy of extended GNSS satellite orbit predictions with convolutional neural networks (CNN). Satellite orbit predictions are used in self-assisted GNSS to reduce the Time to First Fix of a satellite positioning device. We describe the models we use to predict the satellite orbit and present the improvement method that uses CNN. The CNN estimates future prediction errors of our model and these estimates are used to correct our orbit predictions. We also describe how the neural network can be implemented into our prediction algorithm. In tests with GPS and BeiDou data, the method significantly improves orbit prediction accuracy. For example, the 68% error quantile of 7 day orbit prediction errors of GPS satellites was reduced by 45% on average.

We have focused our paper on the aspects important in adapting an Information System (IS) to the user's cultural background. We are interested both in the factors related to IS development and in the use of IS. Increasingly, ISs are being developed and used in a global context. We have perceived differences in expectations of functionalities, architecture, structural properties, information search practices, web-based system properties, and user interfaces. One conclusion would be that a high quality IS reflects user behavior in its use context. In that case, the system has to model its user one way or another. Until now, the topic has been handled without meaningful effort to model user behavior. Current publications cover a wide variety of rules on how to take into account cultural differences in the IS context. In this paper, our aim is to study the current state-of-the-art of user modeling - modeling the human being as an IS user. We start with general aspects related to the role of the user in IS development and alternatives to adaptable systems. The findings are applicable in the educational context as well. More and more, the use of computers and ISs is becoming an essential part of studies: the use of MOOCs (Massively Open Online Courses) as a part or replacement for traditional face-to-face classes; flipped learning methodology emphasizing the significance of self-learning; and blended learning, including quite often computerized study content. Our focus is on the global context, in which students represent different cultures and the IS is globally available.

Close to 100% employment of students and easy access to abundance of information on Internet has essentially changed student's learning practices and their earlier knowledge background, especially on rapidly progressing field of Software Engineering. On workplace they have to use technologies, which are used in practice of their employing enterprise, but often do not understand the scientific and/or technological principles on which these technologies are based. They seek explanations on Internet, but information on Internet is often low quality, one-sided and presented with business targets on mind - to get more users to technologies developed and sold by a business enterprise. Thus university has to explain basic principles of technologies what students already know and have used and correct some popular beliefs, which are supported by software vendors and based their business interests. Non-formal sources of knowledge - workplace training and Internet - do not reduce teacher's task, but force teachers constantly study all new which appears in this field, thus increase teachers workload. Students increasing use of non-formal sources of knowledge imply need for flipping the process - instead of teaching students are set to learn from provided detailed tutorials. Use of Internet and work has made self-study, seeking information from Internet sources very customary for current students, thus such flipping worked very well in a game programming course provided by the first author.

Originating from civilian/commercial wireless networks, the progressive concept of same frequency simultaneous transmission and reception (SF-STAR), a.k.a. in-band full-duplex operation, has high potential also at the future battlefield. The prospects of a military full-duplex radio (MFDR) are not limited to enhancing the spectral efficiency of tactical communications, which would already be a significant advancement considering the universal congestion of electromagnetic spectrum. Perhaps even more importantly, armed forces could gain a major technical advantage by employing multifunction MFDRs that are capable of jointly conducting signals intelligence, electronic warfare, and tactical communications owing to their SF-STAR capability. This study focuses on one specific promising application, where a radio transceiver performs spectrum monitoring and signal surveillance for potential hostile transmissions when simultaneously performing an electronic attack against opposing forces' receivers at the same frequency band. In particular, we demonstrate by experiments in a laboratory environment that the MFDR technology can be successfully used for detecting an attempt to control remotely an improvised explosive device while also preventing its activation by transmitting a jamming signal.

In this paper, we propose a new regularization scheme for the well-known Support Vector Machine (SVM) classifier that operates on the training sample level. The proposed approach is motivated by the fact that Maximum Margin-based classification defines decision functions as a linear combination of the selected training data and, thus, the variations on training sample selection directly affect generalization performance. We show that the exploitation of the proposed regularization scheme is well motivated and intuitive. Experimental results show that the proposed regularization scheme outperforms standard SVM in human action recognition tasks as well as classical recognition problems.

The paper proposes an image segmentation method for lossless compression of plenoptic images. Each light-field image captured by the plenoptic camera is processed to obtain a stack of subaperture images. Each subaperture image is encoded by using a gradient-base detector which classifies the image edges and designs refined contexts for an improved prediction and segmentation. The paper's main contribution is a new segmentation method which generates a preliminary segmentation, either by scaling the intensity differences or by using a quantum cut based algorithm, and merges it with an edge ranking-based segmentation. The results show around 2% improved performance compared to the state-of-the-art for a dataset of 118 plenoptic images.

In this paper, a computational color constancy method is proposed via estimating the illuminant chromaticity in a scene by pooling from many local estimates. To this end, first, for each image in a dataset, we form an image pyramid consisting of several scales of the original image. Next, local patches of certain size are extracted from each scale in this image pyramid. Then, a convolutional neural network is trained to estimate the illuminant chromaticity per-patch. Finally, two more consecutive trainings are conducted, where the estimation is made per-image via taking the mean (1^st training) and median (2^nd training) of local estimates. The proposed method is shown to outperform the state-of-the-art in a widely used color constancy dataset.

Traditional challenges for deploying end-to-end streaming systems are made harder when considering 360-degree media content. One of these challenges relates to the lack of commonly accepted standardized methodologies for subjective 360-degree video quality assessment, especially oriented towards streaming services. The contribution of this paper falls in the area of subjective assessment of 360-degree video. - From traditional standardized test methodologies originally designed for 2D/3D video, we tailored a methodology more oriented towards Virtual Reality (VR) streaming services. The methodology inherits a lot from existing ITU standards for video subjective quality evaluation. The additions incorporate the special properties of 360-video, namely omnidirectionality, as opposed to traditional video. - With this goal in mind, a new metric called Similarity Ring Metric (SRM) is introduced. It measures the degree of similarity in watching patterns of a single subject or between different subjects for several subjective assessment tests. This metric enables an inclusion or rejection criteria for test results in subjective assessment sessions. We also present visual fatigue results related to a subjective quality experiment of 360-degree video.

In this paper, we present a new software tool, called HTGS Model-based Engine (HMBE), for the design and implementation of multicore signal processing applications. HMBE provides complementary capabilities to HTGS (Hybrid Task Graph Scheduler), which is a recently-introduced software tool for implementing scalable workflows for high performance computing applications. HMBE integrates advanced design optimization techniques provided in HTGS with model-based approaches that are founded on dataflow principles. Such integration contributes to (a) making the application of HTGS more systematic and less time consuming, (b) incorporating additional dataflow-based optimization capabilities with HTGS optimizations, and (c) automating significant parts of the HTGS-based design process. In this paper, we present HMBE with an emphasis on novel dynamic scheduling techniques that are developed as part of the tool. We demonstrate the utility of HMBE through a case study involving an image stitching application for large scale microscopy images.

Detection of targets using low power embedded devices has important applications in border security and surveillance. In this paper, we build on recent algorithmic advances in sensor fusion, and present the design and implementation of a novel, multi-mode embedded signal processing system for detection of people and vehicles using acoustic and seismic sensors. Here, by "multi-mode", we mean that the system has available a complementary set of configurations that are optimized for different trade-offs. The multimode capability delivered by the proposed system is useful to supporting long lifetime (long term, energy-efficient "standby" operation), while also supporting optimized accuracy during critical time periods (e.g., when a potential threat is detected). In our target detection system, we apply a strategically-configured suite of single-and dual-modality signal processing techniques together with dataflow-based design optimization for energyefficient, real-time implementation. Through experiments using a Raspberry Pi platform, we demonstrate the capability of our target detection system to provide efficient operational tradeoffs among detection accuracy, energy efficiency, and processing speed.