Virtual Reality Cloud Gaming (VR-CG) is a demanding class of immersive applications that require high bandwidth, ultra-low latency, and efficient resource allocation to deliver a high-quality user experience. In this paper, we propose a scalable, QoE-aware multi-stage optimization framework for VR-CG over 6G networks. Our approach decomposes the joint resource allocation problem into three stages: (i) user association and communication resource allocation; (ii) VR-CG game engine placement with adaptive multipath routing; and (iii) attention-aware scheduling and wireless resource allocation under motion-to-photon latency constraints. For each stage, we design specialized heuristic algorithms that achieve near-optimal performance with significantly reduced computational complexity. We further introduce a user-centric QoE model based on visual attention to virtual objects, enabling adaptive selection of resolution and frame rate. Extensive evaluations using real-world datasets show that, compared to state-of-the-art approaches, the proposed framework improves QoE by up to 50%, reduces communication resource usage by 75%, and achieves up to 35% cost savings, while maintaining an average optimality gap of 5%. Moreover, the proposed heuristics solve large-scale scenarios in under 0.1 s, demonstrating their suitability for real-time deployment in next-generation mobile networks.

Millimeter wave communications are essential for modern wireless networks. It supports high data rates but suffers from severe path loss, which requires precise beam alignment to maintain reliable links. This beam management is particularly challenging in highly dynamic scenarios such as vehicle-to-infrastructure, and several methods have been presented. In this work, we propose a deep learning-based beam tracking framework based on sequential prediction using recurrent neural networks with an autoregressive inference strategy that reduces measurement overhead. The proposed architecture can support deep learning models trained for both classification and regression. In contrast to many existing studies that evaluate beam tracking under predominantly line-of-sight (LOS) conditions, our work explicitly includes highly challenging non-LOS scenarios - with up to 50% non-LOS incidence in certain datasets - to rigorously assess model robustness. Experimental results demonstrate that our approach maintains high top- K accuracy, even under adverse conditions, while reducing the beam measurement overhead by up to 66%.

The future mobile network schedulers have the complex mission of distributing radio resources among various applications with different requirements. The radio access network (RAN) slicing enables the creation of different logical networks by using dedicated resources for each group of applications. In this scenario, the radio resource scheduling (RRS) is responsible for distributing the radio resources among the slices to fulfill their requirements. Several recent studies have proposed advances in machine learning-based RRS. However, these works often evaluate their models under limited scenarios and with minimal slice diversity, raising concerns about their real-world applicability. The generalization capabilities of these models remain uncertain without rigorous testing across diverse network conditions and slice configurations, which may hinder their effectiveness upon deployment in operational networks. This paper proposes an intent-based RRS using multi-agent reinforcement learning in a RAN slicing context. The proposed method protects high-priority slices when the available radio resources are insufficient, using transfer learning to reduce the number of required training steps. The proposed method and baselines are evaluated in different network scenarios that comprehend combinations of different slice types, channel trajectories, number of active slices and users' equipment (UEs), and UE characteristics. The proposed method outperformed the baselines in protecting slices with higher priority, obtaining an improvement of 40% and, when considering all the slices, obtaining an improvement of 20% in relation to the baselines.

Interference mitigation is a common benefit claimed by disaggregated and virtualized radio access networks (vRAN). However, this benefit depends on centralizing the proper virtual network functions (VNFs) from the protocol stack of neighbor radio units (RUs). Additionally, the available computing resources and dynamic demand in RUs must be taken into consideration to obtain efficient results. Naturally, this problem also appears in O-RAN infrastructures which motivates an approach that leverages the O-RAN architecture, including its machine learning-guided design. In this work, we formulate the problem as a Markovian decision process (MDP) and solve it by employing a deep reinforcement learning (DRL) agent. We also describe how our proposal can be implemented inside the O-RAN architecture. Through simulations, we show the improved spectral efficiency provided by the DRL agent while solving the complex VNF placement considering resource constraints, RUs vicinity, and dynamic demand.

Digital twins are an important technology for advancing mobile communications, specially in use cases that require simultaneously simulating the wireless channel, 3-D scenes and machine learning (ML). Aiming at contributing towards a solution to this demand, this work describes a modular co-simulation methodology called CAVIAR, for implementing the virtual counterpart of a digital twin (DT) system. Here, CAVIAR is upgraded to support a message passing library and facilitate using different 6G-related simulators. The main contributions of this work are the detailed description of different CAVIAR architectures, the implementation of this methodology to assess a 6G use case of unmanned aerial vehicle (UAV)-based search and rescue (SAR), and the generation of benchmarking data about the computational resource usage. For executing the SAR co-simulation we adopt five open-source solutions: 1) the physical and link level network simulator Sionna; 2) the simulator for autonomous vehicles AirSim; 3) scikit-learn for training a decision tree for multiple input–multiple output (MIMO) beam selection; 4) Yolov8 for the detection of rescue targets; and 5) neural autonomic transport system (NATS) for message passing. Results for the implemented SAR use case suggest that the methodology can run in a single machine, with the main demanded resources being the CPU processing and the GPU memory.

Beyond 5G (B5G) and 6G networks must offer network slicing as a service to support disruptive applications using mobile network infrastructures. Moreover, network slicing as a service should enable the orchestration and management of disaggregated radio access networks (RAN), i.e., it allows the automation and abstraction of network configurations composed of physical and virtualized components such as defined by the 3rd Generation Partnership Project (3GPP) and Open RAN (O-RAN) Alliance. Network slicing must reach the part of the network between the radio units and the distribution units, i.e., the fronthaul network. Fronthaul is essential and diverse in 5G and B5G networks and can be composed of point-to-point or multipoint connections. In this context, the literature presents several works investigating the problems of network slicing and disaggregated networks. However, no survey explores current works integrating network slicing in disaggregated networks, specifically in the fronthaul network. This article surveys network slicing on disaggregated networks and its potential use in point-to-point and multipoint fronthaul infrastructures. We cover the state-of-the-art by analyzing four fundamental research questions and discuss existing solutions, open challenges, and research opportunities in B5G and 6G networks.

Network slicing at the radio access network (RAN) domain, called RAN slicing, requires elasticity, efficient resource sharing, and customization. In this scenario, radio resource scheduling (RRS) is responsible for dealing with scarce and limited frequency spectrum resources available at the RAN domain while fulfilling the slice intents. The wide variety of scenarios supported in 5G and beyond 5G networks makes the RRS problem in RAN slicing scenario a significant challenge. This paper proposes an intent-aware reinforcement learning method to perform the RRS function in a RAN slicing scenario. The slice’s quality of service intents is described in a common intent model in a service-level agreement. The proposed method tries to prevent intent faults by making the management of radio resources available among slices. This method uses slices’ and user equipment network metrics in the observation space. The proposed method is evaluated under different network conditions and outperforms different baselines considering the slices’ intents fulfillment.

The virtualized radio access network (vRAN) placement problem consists of jointly choosing a functional split and the placement of virtualized network functions on vRAN nodes scattered in the network. The most prominent solutions present optimal approaches to solve the problem, but they are computationally expensive for large instances. Non-exact approaches emerge as alternatives to solve the vRAN placement problem, mainly using machine learning, which is largely fostered by the standardization bodies in next-generation networks. In this context, we present an approach to solve the problem using deep reinforcement learning (DRL), where the objective is to jointly minimize the number of computing resources used and maximize the vRAN centralization level. To build our DRL agent, we started from a traditional optimization formulation that guided the agent development inside a conventional DRL framework. We compare our solution with two exact optimization models from the literature, including one that has a DRL solution. Since our proposed design was based on a most advanced optimal model, it was able to outperform one of the exact optimization models and, as a consequence, its DRL agent.

Disaggregated and virtualized RANs (vRANs) offer the opportunity for flexible and efficient use of computing resources through the proper placement of the RAN Virtualized Network Functions (VNFs). However, many works neglect the necessary coordination between VNF placement and the pro-cessing of the RAN tasks inside these VNFs. This can negatively impact important tasks such as resource scheduling and interference control. In this work, we introduce a new approach for VNF placement that is aware of the wireless coverage and its associated tasks. Our solution was designed in the context of O-RAN architecture, exploring functionalities of monitoring and closed-loop decision making. Simulation results illustrate the benefits of our solution, mainly related to improvements for edge users who are exposed to the worse conditions of spectral efficiency and throughput.

Linear prediction and entropy coding are widely used in speech, audio, video, and other coding systems. Rate-distortion optimization (RDO) is another relevant coding technique, which is especially useful when the encoder performance depends on tunable parameters from distinct coding subsystems. This work presents a RDO strategy for variable-rate encoders that use linear prediction and entropy coding. The method is suitable when the signal to be compressed is short-term stationary and the prediction order varies over signal segments. The second contribution of this paper is to customize the method to the compression of time-domain OFDM-based cellular radio signals. The method enables a 4G / 5G packet-based fronthaul to flexibly trade-off rate and distortion depending on the amount of information to be transmitted and, consequently, benefit from multiplexing gain. Simulation results with LTE signals show that the proposed method achieves improved performance with a relatively low computational cost, while operating with time-domain signals under different traffic scenarios.

This paper discusses the application of long short-term memory (LSTM) neural networks to maintain the synchronization of a real-time clock in holdover operation, that is, while the timing reference input of the clock is unavailable. The approach trains the LSTM network based on timestamps acquired while the slave clock is locked to its reference input coming from a master clock. When the slave clock loses its reference and enters holdover mode, the LSTM takes over and controls the clock. We evaluate the method on a testbed consisting of IEEE 1588 Precision Time Protocol (PTP) clocks based on field-programmable gate arrays (FPGA), where we collect nanosecond-accurate timestamps for offline analysis. We evaluate two oscillator stability scenarios: when the PTP clocks rely on oven-controlled crystal oscillators (OCXOs) and when they use crystal oscillators (XOs). In both cases, we demonstrate that the algorithm can sustain the clock synchronization accuracy within reasonable limits over intervals of 1000 seconds in two different temperature scenarios.

In this work, we propose a new dataset, the UFPArk, which consists of a selection of video clips based on a surveillance camera’s footage throughout a day. The aforementioned dataset is relevant to areas such as computer vision-based object detection, human behavior detection and Internet of Things (IoT) applications, which gained momentum with recent Deep Learning (DL) breakthroughs. To validate our dataset, we conduct some experiments that illustrate the relevance of this effort and its applicability in academic and commercial environments.

Millimeter wave (mmWave) has the potential to offer high data rates for vehicle-to-everything (V2X) communication. In this paper, we provide an introduction to important use cases of V2X as they pertain to fifth generation (5G) communication networks. As 5G technology is still evolving to support vehicles, and mmWave is reflected and blocked by vehicles, it remains unclear if the target rates for the use cases can be achieved. Motivated by the different data rate requirements, we introduce a methodology for evaluating rates in 5G mmWave V2X scenarios. Our approach leverages available city CAD models, realistic traffic simulators, and industry standard ray tracing tools to allow site-specific propagation evaluation. This approach may be used to devise insight into the role of traffic density, antenna array placement, and base station density in important urban propagation settings. Results are provided that highlight the application to develop a rate map for an urban intersection. They show that rate increases in dense deployments by more than ten percent per base station, but only decreases about one percent when going from light to heavy traffic.

Provisioning of cost-effective Ethernet-based fronthaul by reusing the LAN infrastructure available in most commercial buildings is challenging predominantly in terms of the required bandwidth and synchronization. In contrast to a synchronous fronthaul, a PTP-based Ethernet network must cope with estimation noise introduced by packet delay variation (PDV) for synchronization recovery. The SYNC packet used for PTP on such networks is expected to suffer from significant PDV due to the fronthaul traffic and other background traffic. Further challenges are met when the switches adopted in the network do not support PTP and therefore synchronization can only be done by end-devices. Focusing on this scenario, this paper analyzes the problems that may affect the time offset estimation accuracy and presents schemes to mitigate these problems. The performance is evaluated through a self-developed FPGA-based testbed and the results suggest that the end-to-end PTP approach can fulfill the less strict time alignment requirements of 3GPP standards if PDV is handled properly.

The cyclic prefix adopted by discrete multitone (DMT) systems may impose a data rate reduction or be insufficient to prevent inter-symbol and inter-carrier interference from distorting the received symbols and degrading the system performance. Thus, techniques that are able to prevent this distortion while maintaining low power consumption and reasonable complexity are attractive. This works develops a structure to mitigate insufficient cyclic prefix distortion by combining a transmitter-based precoder and a receiver equalizer operating in the time domain DMT symbols. The proposal explores complexity reduction strategies to improve the trade off between the data rate loss imposed by cyclic prefixes and the additional computational cost for accomplishing precoding and equalization. Simulation results show that the proposed method has a computational cost significantly below the cost presented by previous methods.

The use of vectoring for crosstalk cancellation in the new ITU-T G.fast standard for next generation DSL systems becomes essential for efficient utilization of the extended bandwidth (up to 200 MHz). In VDSL2 (up to 30 MHz), a zero-forcing-based linear precoder is used in downstream which approaches single-line performance. However, at high frequencies, the linear precoder may amplify the signal power substantially since the crosstalk channel is much stronger than at lower frequencies. Performance could be significantly degraded by power normalization to keep the PSD below the mask. In this work, we extended a per-line power normalization scheme by linear programming (LP) optimization. By simulations using measured cable data it is shown how the LP-based scheme further improves the linear precoder and it is also capable of balancing the data rate between lines. Further, the simulations also show the non-linear Tomlinson-Harashima precoder performs better than the linear precoders.

This work describes the current version of newGASpeech, a framework centered on analysis-by-synthesis and genetic algorithms for automatically estimating the input parameters of Klatt's speech synthesizer. The goal is to speed up the process of speech imitation (or utterance copy), where one has to find the model parameters that lead to a synthesized speech sounding close enough to the target speech. The main focus is speech pathology research. The proposed system is compared with WinSnoori and it outperforms this baseline by a large margin with respect to, for example, mean square error and PESQ scores.

G.fast is a new standard from the International Telecommunication Union, which targets 1 Gb/s over short copper loops using frequencies up to 212 MHz. This new technology requires accurate parametric cable models for simulation, design, and performance evaluation tests. Some existing copper cable models were designed for the very high speed digital subscriber line spectra, i.e., frequencies up to 30 MHz, and adopt assumptions that are violated when the frequency range is extended to G.fast frequencies. This paper introduces a simple and causal cable model that is able to accurately characterize copper loops composed by single or multiple segments, in both frequency and time domains. Results using G.fast topologies show that, apart from being accurate, the new model is attractive due to its low computational cost and closed-form expressions for fitting its parameters to measurement data.

There are several commercial text-to-speech (TTS) systems that generate speech signals that sound very natural. A distinct problem is utterance copy, which consists in taking speech as input (instead of text, as in TTS) and find the input parameters that would drive a speech synthesizer to generate speech that mimics the target speech with respect to contents and speaker identity. Utterance copy is a difficult task due to the need of adjusting several parameters and their nonlinear relation to the output. Genetic algorithms (GA) have been used in this task embedded in an analysis-by-synthesis loop, which requires solving several optimization processes, one for each short segment of speech. The contribution of this work is to present two new strategies, namely the voicing gene and the lookahead mechanism, which consistently improve the performance of the previous GA-based architecture. The results show that the proposed improvements reduced the mean squared error by more than 40%.

The use of a hybrid copper and fiber architecture is attractive in both fixed access and mobile backhauling scenarios. This trend led the industry and academia to start developing the fourth generation broadband system, which aims at achieving bit-rates of 1 Gb/s over short copper loops. In this context, accurate models of short twisted-pair cables operating at relatively high frequencies are key elements. This work describes new parametric cable models that incorporate four important characteristics: support of frequencies up to 200 MHz, few parameters, causal impulse responses and require relatively easy fitting procedures. The results show that the models achieve good accuracy for single segments, which are the expected topology for G.fast deployments.

The evolving broadband access systems using copper networks are currently deployed in a frequency band that goes up to 30 MHz, as specified in VDSL2. As hybrid fiber-copper architectures become more important in the industry and academia, using shorter loop lengths (i.e. up to 250 meters) from the last distribution point to users enables adopting even higher frequencies to achieve very high data rates of 500 Mbps and beyond, as is the case with the G.fast standard under development by ITU-T. In this work, a time-domain simulator has been developed to evaluate G.fast system performance. System capacity is evaluated with different cyclic extension lengths and different reference loop topologies specified by ITU-T. The simulation results show that G.fast systems are robust to bridgetaps and capable of providing very high data rates for all simulated loop topologies to support next generation ultra high speed broadband services.

Dual-ended line testing (DELT) is a common capability in most current modems and can be used for digital subscriber line (DSL) qualification and monitoring purposes. In spite of that, this feature remains largely unexplored in the literature. This paper proposes a new method based on wavelets and DELT measurements for estimating the total length of the line under test and identification of bridged-taps, two important line parameters that affect the maximum bit rate reached by a DSL line. The proposed method was tested with measurements employing real twisted-pair cables and obtained reasonably accurate results for the analyzed cases.

The proper management of the signal to noise ratio margin is important for improving the stability of digital subscriber line (DSL) networks and increasing the customer satisfaction regarding services such as triple-play. This work presents a novel algorithm for multiuser margin optimization that benefits from the relationships between power and margin adaptation. The proposed multiuser margin maximization (MMM) is a meta algorithm that repeatedly invokes a spectrum balancing base algorithm. The base algorithm corresponds to a dynamic spectrum management (DSM) algorithm of level one or two, and is responsible for part of the overall spectra optimization procedure. Simulation results illustrate the performance of the proposed algorithm when using two base algorithms: iterative water-filling (IWF) and iterative spectrum balancing (ISB). The results show that the proposed MMM is able to design better noise margins and spectra, providing performance improvements when compared to a previous margin optimization algorithm that is restricted to using DSM level 1 coordination.

This paper presents a performance analysis of an ultra wideband (UWB) communication system based on Multiband Orthogonal Frequency Division Multiplexing (MB-OFDM) in the presence of narrowband interference. Although OFDM-based systems are known to be capable of withstanding some level of narrowband interference, mitigation techniques may have to be used when interference levels are high. Furthermore, as typical UWB systems employ low resolution analog-to-digital converters, the effect of narrowband interference may be exacerbated in the process of analog-to-digital conversion. In order to improve the performance of MB-OFDM UWB systems in the presence of narrowband interference, we consider the use of analog notch filters. This paper proposes an analytical model and formulation that allows for the evaluation of the probability of symbol error of an MB-OFDM UWB system when an analog notch filter is used for narrowband interference mitigation. It is shown that using an analog notch filter can significantly improve the performance of MB-OFDM UWB systems in high-interference scenarios.

Spectrum balancing (SB) techniques optimize transmission and can significantly improve digital subscriber lines (DSL) services. In the literature, the DSL system optimization is typically formulated as a rate maximization problem. However, there is an increasing interest in minimizing the considerable amount of power consumed by telecommunication networks. Few works in the SB literature have explored algorithms for power minimization. It is known that some existing solutions for rate maximization can be converted into power minimization algorithms. This relation has not been fully explored and, consequently, the area lacks results regarding what can be achieved with power minimization SB algorithms. This work aims to diminish this gap. First, the equivalence between rate maximization and power minimization problems is formalized. Second, extended versions of some rate maximization SB algorithms are proposed for power minimization purposes and evaluated through simulations. In addition, the power-usage capabilities and convergence characteristics of each extended SB algorithms is discussed.

The development and assessment of spectrum management methods for the copper access network are usually conducted under the assumption of accurate channel information. Acquiring such information implies, in practice, estimation of the crosstalk coupling functions between the twisted-pair lines in the access network. However, this estimation is not supported or required by current digital subscriber line (DSL) standards. In this work, we investigate the impact of non-ideal crosstalk estimation on the dynamic spectrum management (DSM) performance. Two different crosstalk estimators are considered: a conventional model-based estimator and a novel estimation procedure. The DSM performance is evaluated based on the obtained crosstalk estimates for two different network scenarios consisting of real twisted-pair cables. For a reference comparison, the crosstalk channels are measured with a network analyzer. The simulation results indicate that the novel estimation procedure achieves DSM performance results close to the ones obtained with the network analyzer and that the model-based estimator can lead to an inaccurate rate estimation.

Digital subscriber lines (DSL) are today one of the most important means for delivering high-speed data transmission. An emerging technique for dealing with one of the technology's most harmful problems, crosstalk, is dynamic spectrum management (DSM). DSM literature already counts with some half a dozen important solutions. These solutions can be classified according to four different aspects: optimality, computational cost, distribution and required crosstalk channel information. In this work we present an algorithm, named semi- blind spectrum balancing, which achieves a compelling trade-off between these four aspects. The scheme is based on the idea of optimization against a virtual line, a fictitious line to represent the damage caused to other users in the network. This line is adjusted with the aid of limited message-passing between modems and a central agent and very simple crosstalk channel information. Crosstalk channel knowledge required should be much simpler to obtain than full channel estimation.

Data mining can play a fundamental role in modern power systems. However, the companies in this area still face several difficulties to benefit from data mining. A major problem is to extract useful information from the currently available non-labeled digitized time series. This work focuses on automatic classification of faults in transmission lines. These faults are responsible for the majority of the disturbances and cascading blackouts. To circumvent the current lack of labeled data, the alternative transients program (ATP) simulator was used to create a public comprehensive labeled dataset. Results with different preprocessing (e.g., wavelets) and learning algorithms (e.g., decision trees and neural networks) are presented, which indicate that neural networks outperform the other methods.

We investigate feature selection applied to automatic speech recognition (ASR) systems. We focus on systems based on support vector machines (SVM), which can naturally use features optimized for each classifier. We present a new method for feature selection based on the AdaBoost algorithm. This method was an order of magnitude faster than a similar one, while leading to equivalent accuracy. Experiments with phone classification using TIMIT and a total of 760 features (PLP, MFCC, Seneff's, formants, etc.) indicated that the proposed method automatically discovered important information in the data. When using only 25 selected features per SVM, the accuracy was higher than when using a homogeneous set of 118 features based on PLP (perceptual linear prediction) coefficients.

Many applications require classification of examples into one of several classes. A common way of designing such classifiers is to determine the class based on the outputs of several binary classifiers. We consider some of the most popular methods for combining the decisions of the binary classifiers, and improve ex-isting bounds on the error rates of the combined classifier over the training set. We also describe a new method for combining binary classifiers. The method is based on stacking a neural network and, when used with support vector machines as the binary learners, substantially decreased the error rate in two vowel classification tasks. 1.

We propose a new architecture for deploying speech recognition over the Internet. The client performs the recognition, but is assisted by the server who computes the speech parameters. To demonstrate the architecture, we developed a Java-based Web-navigation system where the precomputed HMM models of the hyperlinked words are stored on the Web page and downloaded by the client. We tested the system on a digit-recognition example. The results show that with quantization and compression of the speech parameters, good recognition can be achieved in acceptable download and calculation time even on clients with modest connection speeds and computational powers.

A predictive vector quantization (PVQ) structure is proposed, where the encoder uses a predictor based on an intrablock support region, followed by a modified vector quantizer stage. Simulation results show that a modification on a previously published PVQ system led to an improvement of 1 dB in PSNR for Lenna.