This paper presents a significant contribution to the field of 5G networking through the development and deployment of a Software-Defined Radio (SDR)-based 5G Stand-Alone (SA) network leveraging the Multi-Access Edge Computing (MEC) paradigm. Designed to deliver high-quality voice and video services using the Session Initiation Protocol (SIP) Kamailio Server, our implementation significantly reduces latency by integrating Virtualized Network Functions (VNFs) and Docker containerized elements onto a single workstation, enhancing the efficiency of 5G Core operations. Built entirely on open-source projects, the proposed mobile network successfully provides both Voice-overNew Radio (VoNR) and Video-over-New Radio (ViNR) services using SDR. Notably, this work introduces the first publicly available ViNR prototype for a comprehensive 5G SDR solution, thereby filling a gap in existing research and development. We further detail a complete end-to-end call flow for both VoNR and ViNR, utilizing SIP Kamailio, and provide a valuable resource for researchers and developers. The performance of the proposed network is evaluated through Quality of Service (QoS) metrics, and the computational resource allocation is analyzed. We utilize two smartphones as representatives of typical mobile users to simulate their behavior. The results are then comprehensively compared with those obtained from a high-tech radio communication test station platform, a platform commonly used for certifying terminals, chipsets, and devices against 3GPP standards. Our findings demonstrate that the proposed SDR MEC network exhibits performance closely aligned with that of the robust, established test platform, validating its effectiveness in controlled environments. This research significantly advances the state-of-the-art in 5G SDR MEC networks by providing scalable, efficient, and cost-effective solution for delivering IP multimedia services, with a particular focus on enabling and optimizing Video-over-New Radio (ViNR).
The adoption of wireless technologies in industrial environments has been a highly relevant topic of investigation, considering both the interplay between automation and industrial data communication systems as well as the goal to meet the stringent communication requirements of Industry 4.0, where network reliability and resilience are critical considerations. In such Integrated Manufacturing Systems (IMS), where people and machines interact in an industrial environment, safety aspects also play a fundamental role. In this context, this work investigates a new application framework for 5G URLLC communications in Industry 4.0, taking advantage of the low latency provided by the 5G infrastructure, through the integration of automation safety systems into an IMS architecture. In our setup, an Industrial Internet of Things, IIoT, device uses 5G connectivity to send an emergency stop command to an Industrial Autonomous Mobile Robot. Our experiment achieved a maximum latency below 1 ms with a block error rate (BLER) below 10^-5 on the air interface, including scenarios of high consumption of Physical Resource Blocks (PRBs). The results reinforce the great potential for using 5G networks for worker safety systems and automation elements in industrial environments.
Video conferencing has become a widely adopted solution for meetings involving remote participants and rapid communication. This service must comply with the required Quality of Service (QoS) parameters to meet user expectations. The primary purpose of this work is to assess the energy consumption and jitter of a real-time testbed for Video over LTE (ViLTE) and Voice over LTE (VoLTE) over an IP Multimedia Subsystem (IMS). Modern technology is deployed to recreate the best feasible scenario for this goal. Based on the results of our test experiments, VoLTE can drain the battery faster than ViLTE, which consumes up to 112% of the battery while using the LTE B7 band. Our study provides an analysis of jitter evaluation for video conversations, evaluates power usage for ViLTE and VoLTE services, and quantifies jitter values.
In recent years, the demand for reliable optical networks has intensified due to the rise of bandwidth-intensive applications. Faults in these networks can degrade transmission quality, leading to packet loss and service disruptions, making effective failure management essential for maintaining network availability and compliance with Service Level Agreements (SLAs). While most contemporary approaches rely on supervised learning (SL), which requires large amounts of often scarce fault data for training, this study explores semi-supervised learning (Semi-SL) algorithms for fault management in optical networks, which rely only on normal operating condition data, thus reducing the dependency on scarce fault data. We evaluate several cluster-based algorithms and dimensionality reduction techniques using a dataset from an optical testbed. Results based on Type I/II errors show that all techniques performed with an average accuracy exceeding 90%. The Autoencoder yielded the best fault management performance (96.46% of average accuracy), followed by Gaussian Mixture Model (92.5%), Mahalanobis Squared-Distance (92.48%), Density-based spatial clustering of applications with noise (92.42%), Fuzzy C-means (92.12%), K-means (92.09%), and Principal Component Analysis (90.22%).
In recent years machine learning has been used for automatic medical image diagnosis. Solutions leveraging machine learning can help physicians in the diagnosis process and also reduce the time spent by medical experts analyzing images and video frames in order to conclude their assessments. In the application scenario considered in this work, the image is captured in a resource-constrained remote basic health facility and sent to the inference model on the cloud that returns the result to the physician. We investigate the impact of straightforward low-bit-rate image coding solutions on the classification performance of neural network models targeted at cloud-based image diagnosis solutions. Our experiments show that it is possible to lower the bit rate needs without significant harm to the prediction accuracy of the models using both downsizing and compression. Thus, providing evidence for the viability of deploying automated diagnostic systems as a Service over constrained communication infrastructure to assist remote areas.
Accurately localizing drones in complex environments remains a significant challenge, with important implications for defense, law enforcement, and autonomous systems. This study addresses the problem of estimating drone localization in environments characterized by strong reflections and noise. We employ Time Difference of Arrival (TDOA) techniques for localization estimation and compare them with a specialized machine learning regression model. While previous works have considered neural networks for audio source localization, they often suffer from limited generalization across different environments. To address this, we propose a novel method that enhances the TDOA vector by incorporating both primary and secondary peaks of the cross-correlation, guided by the Zero Cyclic Sum condition. Additionally, we introduce optimization strategies that selectively reduce the number of TDOA inputs based on a least-squares cost function. We present a comparative analysis of TDOA-based optimization techniques with a machine learning method that utilizes the environment's reverberation fingerprint as input features for training. Experimental results demonstrate that the proposed TDOA-based method achieves a localization accuracy of 0.55±0.35 meters, showcasing its effectiveness and practical applicability in challenging acoustic environments.
An innovative formulation, intended by the authors, for the calculation and analysis of ground-reflection of ray paths under conditions of horizontally variable refractivity, is introduced in this letter. The deduced formulation is incorporated within a modified Ray Tracing (RT) technique that considers atmospheric refractivity effect in radiopropagation modeling. Canonical tests were carried out, where the proposed analytical solution was applied to predict coverage in a long-distance scenario that assumed different configurations of horizontally stratified refractivity changes along the investigated environment. The results achieved were confronted with the numerical solution of the Split-Step Fourier Parabolic Equation (SSPE) at 10 GHz. It was possible to observe that the results obtained from a modified RT with horizontally variable refractive and the SSPE method have a similar behavior when estimating the received power. These innovations contribute to the development of algorithms for modeling and analyzing wireless networks in more realistic environments.
This letter assesses the interference between terrestrial 5G network (TN-5G) and non-terrestrial 5G network (NTN-5G) operating in millimeter wave Ka-band. Specifically, the impact that NTN-5G causes on TN-5G system in the 28 GHz band is analyzed. Simulations using Monte-Carlo method are performed to obtain the interference power in the victim system, for different guard bands. Also, experiments using laboratory equipment are performed to analyze the BLER and Throughput of the user equipment (UE) under different interference power levels. The results showed that interference from NTN-5G may impact the TN-5G, depending on the adopted guard band. Ideally, guard bands above 100 and 180 MHz are recommended for TN-5G operating with 50 and 100 MHz bandwidths, respectively. On the other hand, experiments showed that UE performance can be severely degraded by interference power above -108 dBm. Finally, it is concluded that TN-5G and NTN-5G can coexist since the appropriate guard bands are considered, depending mainly on the bandwidth of the victim system.
This paper explores the application of deep learning techniques to enhance physical layer security in 5G networks by detecting spoofing attacks through the analysis of MIMO beam patterns. The study focuses on developing and evaluating three machine learning models—Deep Autoencoder (DAE), Convolutional DAE, and Convolutional Neural Networks (CNNs)—to identify unique beam signatures caused by manufacturing imperfections in antenna arrays. Using datasets generated via simulations, the models are trained to distinguish legitimate devices from intruders based on their transmission beam patterns. The performance of the models is evaluated using metrics such as accuracy, precision, recall, and specificity. The results demonstrate the potential for using neural networks to implement robust security measures in 5G networks by leveraging intrinsic physical characteristics of antennas, providing a new layer of protection against unauthorized access.
This work explores the versatility of complex-valued signals present in data representation in diverse applications. Integrating complex values into signals and processing structures proves powerful in system modeling, demonstrating effectiveness in many scenarios. The inherent mathematical efficiency of complex numbers enhances both elegance and practical utility, simplifying calculations and expanding capabilities. However, the use of complex notation introduces considerations related to the intrinsic geometry of the data. This tutorial unravels the theoretical foundations and practical applications in adaptive filtering of complex-valued signals, emphasizing the importance of understanding the displayed profiles of the studied signals and data. Using complex augmented notation and widely linear processing, superior performance is attainable in specific cases, providing valuable information to readers interested in signal processing.
Kernel methods and Support Vector Machine (SVM) are widely used in machine learning. However, when multidimensional data are used, the classical vector-based kernel functions must vectorize the inputs, which breaks down the original tensor structure, leading to performance loss. To avoid this problem, tensor kernel functions can be used. In the present work, three novel tensor kernel functions are presented. The proposed methods are based on the core tensors of the Higher-Order Singular Value Decomposition (HOSVD) and Tensor-Train Decomposition (TTD). Two of the presented methods are fast kernel functions that ignore the factor matrices of these tensor decompositions, alleviating the time complexity burden. The presented techniques were evaluated in the classification of hand movements. A low-cost "smart glove" with accelerometers and gyroscopes was developed, generating tensor input samples with modes related to sensors, channels and features. The experiments showed a good performance of the proposed techniques when compared to state-of-the-art tensor kernel functions.
This work proposes an angle of arrival (AoA) estimator for analog beam refinement in cellular systems operating in millimeter wave (mmWave) bands. Differently from previous works, the proposed method exploits the polarization domain and the concept of virtual arrays to estimate the AoA through a cross-correlation function. The proposed AoA estimator requires only a polarization-multiplexed measurement to estimate the AoA at the receiver. To cope with polarization leakage issues, this work proposes a channel equalization in the polarization domain conditioned on the estimation of the cross-channel gain. Simulation results indicate that the beam refinement procedure with the proposed AoA estimator has similar performance to the classical beam sweeping, but using less time-multiplexed measurements.
Providing high-resolution video streaming services with a high Quality of Experience (QoE) for end users is a challenge today. This is due to limitations in both the hardware of user devices and the channel’s bandwidth. In parallel, there is a noticeable decrease in people’s attention span, which can make it difficult for them to perceive details on the screen. This paper introduces Transcoding Resolution Induced by Custom Keyframes (TRICK), an adaptive transcoding strategy that reduces the video resolution in frames where this reduction might not be noticed, saving bandwidth and possibly not compromising the QoE, which has a time complexity of O(N). An opinion survey compares videos with TRICK to videos in full 480p and 1080p. Videos with TRICK are liked by more than 84% of people, and offer a similar experience to that of 1080p videos, with savings of up to 18% in bandwidth in these transmissions.
The surge in ransomware attacks in recent years has elevated this malware to one of the foremost cybersecurity threats. This article presents a dynamic ransomware classification approach, leveraging the malware analysis environment provided by Cuckoo Sandbox and machine learning techniques. We introduce a methodology encompassing steps for malicious code sample collection, environment configuration for sample execution, data collection, and dataset construction for ransomware detection and experimentation. Six machine learning classifiers were employed to identify ransomware families and individual cases, furnishing valuable tools for threat detection. The results underscore the effectiveness of tree-based methods, such as Random Forests and Decision Trees, in delineating between different ransomware families.
Atrial fibrillation (AF) is a common cardiac arrhythmia associated with various cardiovascular diseases and has a significant impact on mortality around the world. This work focuses on the detection of AF using data collected from cardiac monitoring through the Electrocardiogram (ECG), proposing new attributes for the prediction of AF. In particular, the present work proposes novel convergence and optimization indicators derived from Block-term Decomposition (BTD), applied to five RRI ECG segments, combined with RRI intervals (RRI) to improve the detection of AF. These features were used with treebased machine learning algorithms to classify signals as Atrial Fibrillation (AF) or Normal Sinus Rhythm (NSR). The study also discusses data acquisition from three different ECG databases: Atrial Fibrillation Database (AFDB), Long-term Atrial Fibrillation Database (LTAFDB), and Normal Sinus Rhythm Database (NSRDB).
Enhanced mobile broadband is one of the three major operating scenarios of 5G, directly impacting the average mobile user. Therefore, capacity planning is crucial for providing excellent user experience. The 3GPP, in its TS 38.306 specification, established a theoretical model to estimate maximum throughput values. However, this equation presents challenges for real usage analysis because capacity is tracked in practice using key performance indicators (KPIs) that do not entirely correspond to the model’s input parameters. This paper presents a case study of real 5G gNodeB throughput, integrating practical KPIs within the 3GPP model framework. Various tests were conducted in different usage scenarios, comparing real measured throughput with predictions using a proposed hybrid version of the 3GPP model. A good correlation between predictions and measurements was observed regarding hourly throughput variation. Additionally, it was found that considering the actual use of the available bandwidth, rather than assuming all resource blocks are fully occupied at all times, applying a scale factor can improve the predictions, leading to a good convergence with the measurements.
Blind Source Separation (BSS) is a well-known problem in signal processing and still receives attention from the scientific community, given its applicability in different areas. This work presents a theoretical background overview of the Kalman Filter formulation and its applicability to the BSS problem as a parameter estimator in two different approaches: joint (JEKF) and dual parameter estimation (DEKF). These approaches are evaluated in different scenarios for first-order autoregressive source signals, with analysis of the initialization details, presenting simulation results and performance comparison with classic algorithms, SOBI and SONS, evaluated by SIR, MER and MSE. The results showed that both the JEKF and DEKF algorithms can perform separation in a two-source-two-mixture scenario. In general, over the scenarios studied, DEKF presented a better performance when compared to the JEKF on the evaluated metrics. However, neither algorithm correctly estimated the parameters for mixtures involving more than two sources, showing convergence issues and sensitivity to initialization for an increased number of sources.
Integrated access and backhaul (IAB) technology is a flexible solution for network densification. IAB nodes can also be deployed in moving nodes such as buses and trains, i.e., mobile IAB (mIAB). As mIAB nodes can move around the coverage area, the connection between mIAB nodes and their parent macro base stations (BSs), IAB donor, is sometimes required to change in order to keep an acceptable backhaul link, the so called topology adaptation (TA). The change from one IAB donor to another may strongly impact the system load distribution, possibly causing unsatisfactory backhaul service due to the lack of radio resources. Based on this, TA should consider both backhaul link quality and traffic load. In this work, we propose a load balancing algorithm based on TA for IAB networks, and compare it with an approach in which TA is triggered based on reference signal received power (RSRP) only. The results show that our proposed algorithm improves the passengers worst connections throughput in uplink (UL) and, more modestly, also in downlink (DL), without impairing the pedestrian quality of service (QoS) significantly.
Electronic systems in general can be impaired by impulsive noise generated by a variety of sources. Spectrum sensors are of particular interest herein, since their probabilities of detection and false alarm can be severely degraded under this impairment. Several models for impulsive noise have been studied in the literature, all of them having the common characteristic of being well represented by heavy-tailed probability density functions, like Laplace and some Stable distributions. This article addresses the performances of state-of-the-art detectors for cooperative spectrum sensing when the received signal is impaired by Laplacian noise. This is made by means of estimating the probability of detection for a fixed false alarm rate, when important system parameters are varied. It is demonstrated that the robustness against impulsive noise varies significantly depending on the adopted detection strategy.
Speech conversion is a technique that modifies the identity of the voice in a speech signal without changing the spoken content. Accurate pitch conversion is a requirement the best speech conversion systems must address, as this characteristic is essential to the correct identification of the target speaker. This work proposes a pitch-controlled end-to-end voice conversion model that combines state-of-the-art ideas from both speaking and singing voice conversion with a novel cost function to ensure artifact-free pitch tracking. The model is trained in Brazilian Portuguese, overcoming the lack of high-quality data by improving a large but flawed dataset with filtering operation. Our model mostly outperforms other popular open source models in both listening tests and objective measurements. In particular, on a 5-point MOS, we obtained the highest speaker similarity score (4.05), and a naturalness score of 3.48, second only to a system whose similarity score was 2.62.
