Search

In recent years, the field of Natural Language Generation (NLG) has been boosted by the recent advances in deep learning technologies. Nonetheless, these new data-intensive methods introduce language-dependent disparities in NLG as the main training data sets are in English. Also, most neural NLG systems use decoder-only (causal) transformer language models, which work well for English, but were not designed with other languages in mind. In this work we depart from the hypothesis that they may introduce generation bias in target languages with less rigid word ordering, subject omission, or different attachment preferences for relative clauses, so that for these target languages other language generation strategies may be more desirable. This paper first compares causal and non-causal language modeling for English and Spanish, two languages with different grammatical structures and over 1.5 billion and 0.5 billion speakers, respectively. For this purpose, we define a novel metric of average causal and non-causal context-conditioned entropy of the grammatical category distribution for both languages as an information-theoretic a priori approach. The evaluation of natural text sources (such as training data) in both languages reveals lower average non-causal conditional entropy in Spanish and lower causal conditional entropy in English. According to this experiment, Spanish is more predictable than English given a non-causal context. Then, by applying a conditional relative entropy metric to text generation experiments, we obtain as insights that the best performance is respectively achieved with causal NLG in English, and with non-causal NLG in Spanish. These insights support further research in NLG in Spanish using bidirectional transformer language models.

Mobile core networks handle critical control functions for delivering services in modern cellular networks. Traditional point-to-point architectures, where network functions are directly connected through standardized interfaces, are being substituted by service-based architectures (SBAs), where core functionalities are finer-grained microservices decoupled from the underlying infrastructure. In this way, network functions and services can be distributed, with scaling and fail-over mechanisms, and can be dynamically deployed, updated, or removed to support slicing. A myriad of network functions can be deployed or removed according to traffic flows, thereby increasing the complexity of connection management. In this context, 3GPP Release 16 defines the service communication proxy (SCP) as a unified communication interface for a set of network functions. In this paper, we propose a novel software-defined networking (SDN)-based solution with the same role for a service mesh architecture where network functions can be deployed anywhere in the infrastructure. We demonstrated its efficiency in comparison with alternative architectures.

Drones may be more advantageous than fixed cameras for quality control applications in industrial facilities, since they can be redeployed dynamically and adjusted to production planning. The practical scenario that has motivated this paper, image acquisition with drones in a car manufacturing plant, requires drone positioning accuracy in the order of 5 cm. During repetitive manufacturing processes, it is assumed that quality control imaging drones will follow highly deterministic periodic paths, stop at predefined points to take images and send them to image recognition servers. Therefore, by relying on prior knowledge about production chain schedules, it is possible to optimize the positioning technologies for the drones to stay at all times within the boundaries of their flight plans, which will be composed of stopping points and the paths in between. This involves mitigating issues such as temporary blocking of line-of-sight between the drone and any existing radio beacons; sensor data noise; and the loss of visual references. We present a self-corrective solution for this purpose. It corrects visual odometer readings based on filtered and clustered Ultra-Wide Band (UWB) data, as an alternative to direct Kalman fusion. The approach combines the advantages of these technologies when at least one of them works properly at any measurement spot. It has three method components: independent Kalman filtering, data association by means of stream clustering and mutual correction of sensor readings based on the generation of cumulative correction vectors. The approach is inspired by the observation that UWB positioning works reasonably well at static spots whereas visual odometer measurements reflect straight displacements correctly but can underestimate their length. Our experimental results demonstrate the advantages of the approach in the application scenario over Kalman fusion, in terms of stopping point detection and trajectory estimation error.

Wireless connectivity is essential for industrial production processes and workflow management. Moreover, the connectivity requirements of industrial devices, which are usually long-term investments, are diverse and require different radio interfaces. In this regard, the 3GPP has studied how to support heterogeneous radio access technologies (RATs) such as Wi-Fi and unlicensed cellular technologies in 5G core networks. In some cases, these technologies coexist in the same spectrum. Dynamic spectrum sharing (DSS), which has already been proven to increase spectrum efficiency in licensed bands, can also be applied to this scenario. In this paper, we propose two solutions for mobile network operators (MNOs) or service providers to dynamically divide (multiplex) the radio resources of a shared channel between a Wi-Fi basic service set (BSS) and one or several carriers of scheduled wireless networks, such as cellular technologies, with a configurable level of sharing granularity. These solutions do not require modifications to the current commercial off-the-shelf (COTS) end devices. We adapt the existing IEEE 802.11 procedures to notify the Wi-Fi stations that they must share channels with different access networks. We demonstrate that our dynamic sharing proposals are also advantageous over direct coexistence and evaluate each of them quantitatively and qualitatively to determine when one or the other is preferable. The evaluation is particularized for IEEE 802.11ac and long-term evolution (LTE) license assisted access (LAA), but the solutions can be easily extended to 5G new radio-unlicensed (5G NR-U) or to any other wireless technology in which the network side schedules end device transmissions.

Short texts are omnipresent in real-time news, social network commentaries, etc. Traditional text representation methods have been successfully applied to self-contained documents of medium size. However, information in short texts is often insufficient, due, for example, to the use of mnemonics, which makes them hard to classify. Therefore, the particularities of specific domains must be exploited. In this article we describe a novel system that combines Natural Language Processing techniques with Machine Learning algorithms to classify banking transaction descriptions for personal finance management, a problem that was not previously considered in the literature. We trained and tested that system on a labelled dataset with real customer transactions that will be available to other researchers on request. Motivated by existing solutions in spam detection, we also propose a short text similarity detector to reduce training set size based on the Jaccard distance. Experimental results with a two-stage classifier combining this detector with a SVM indicate a high accuracy in comparison with alternative approaches, taking into account complexity and computing time. Finally, we present a use case with a personal finance application, CoinScrap, which is available at Google Play and App Store.

Queued Radio Frequency Identification (RFID) networks arise naturally in many applications, where tags are grouped into batches, and each batch must be processed before the next reading job starts. In these cases, the system must be able to handle all incoming jobs, keeping the queue backlogs bounded. This property is called stability. Besides, in RFID networks, it is common that some readers cannot operate at the same time, due to mutual interferences. This fact reduces the maximum traffic that readers can process since they have to share the channel. Synchronous networks share the channel using a TDMA approach. The goal of this work is to analytically determine whether a synchronous queued RFID network attains stable operation under a given incoming traffic. Stability depends on the service rate, which is characterized in this paper using an exact numerical method based on a recursive analytical approach, overcoming the limitations of previous works, which were based on simplifications. We also address different flow optimization problems, such as computing the maximum joint traffic that a network can process stably, selecting the minimal number of readers to process a given total load, or determining the optimal timeslot duration, which are novel in the RFID literature.

Traditionally, the cellular spectrum is allocated to operators (OPs) through auctions, as ideal mechanisms to discover market prices and allocate scarce resources. Even though spectrum is indeed scarce in sub-6 GHz bands, it becomes abundant in millimeter-wave (mmWave) bands. Interestingly, in that context, it is base station (BS) density which is limiting, and thus a critical factor, due to the outage phenomena in urban environments. Facing BS scarcity is one of the main reasons to foster virtualization techniques aimed at improving utilization and lowering costs. We consider a scenario with an infrastructure provider (InP) owner of a number of BSs and a set of OPs with their users (UEs). We propose a three-phase framework to price network infrastructure slices (NISs) and allocate them to OPs and to efficiently associate UEs with those NISs. The framework stages are: 1) an initial association, 2) a distributed auction mechanism across the BSs to allocate resources to Ops, and 3) a re-association process where the OPs can optimize the NISs they are awarded. The auction incentivizes OPs to bid truthfully and the outcome yields both socially optimal NISs and Vickrey-Clarke-Groves (VCG) prices. For the re-association phase, we propose deterministic and stochastic exchange-matching algorithms and demonstrate their convergence to stable matching and stable-optimal matching, respectively.

We present an automatic text expansion system to generate English sentences, which performs automatic Natural Language Generation (NLG) by combining linguistic rules with statistical approaches. Here, “automatic” means that the system can generate coherent and correct sentences from a minimum set of words. From its inception, the design is modular and adaptable to other languages. This adaptability is one of its greatest advantages. For English, we have created the highly precise aLexiE lexicon with wide coverage, which represents a contribution on its own. We have evaluated the resulting NLG library in an Augmentative and Alternative Communication (AAC) proof of concept, both directly (by regenerating corpus sentences) and manually (from annotations) using a popular corpus in the NLG field. We performed a second analysis by comparing the quality of text expansion in English to Spanish, using an ad-hoc Spanish-English parallel corpus. The system might also be applied to other domains such as report and news generation.

Vehicle automation is driving the integration of advanced sensors and new applications that demand high-quality information, such as collaborative sensing for enhanced situational awareness. In this work, we considered a vehicular sensing scenario supported by 5G communications, in which vehicle sensor data need to be sent to edge computing resources with stringent latency constraints. To ensure low latency with the resources available, we propose an optimization framework that deploys User Plane Functions (UPFs) dynamically at the edge to minimize the number of network hops between the vehicles and them. The proposed framework relies on a practical Software-Defined-Networking (SDN)-based mechanism that allows seamless re-assignment of vehicles to UPFs while maintaining session and service continuity. We propose and evaluate different UPF allocation algorithms that reduce communications latency compared to static, random, and centralized deployment baselines. Our results demonstrated that the dynamic allocation of UPFs can support latency-critical applications that would be unfeasible otherwise.

The unstoppable adoption of the Internet of Things (IoT) is driven by the deployment of new services that require continuous capture of information from huge populations of sensors, or actuating over a myriad of “smart” objects. Accordingly, next generation networks are being designed to support such massive numbers of devices and connections. For example, the 3rd Generation Partnership Project (3GPP) is designing the different 5G releases specifically with IoT in mind. Nevertheless, from a security perspective this scenario is a potential nightmare: the attack surface becomes wider and many IoT nodes do not have enough resources to support advanced security protocols. In fact, security is rarely a priority in their design. Thus, including network-level mechanisms for preventing attacks from malware-infected IoT devices is mandatory to avert further damage. In this paper, we propose a novel Software-Defined Networking (SDN)-based architecture to identify suspicious nodes in 4G or 5G networks and redirect their traffic to a secondary network slice where traffic is analyzed in depth before allowing it reaching its destination. The architecture can be easily integrated in any existing deployment due to its interoperability. By following this approach, we can detect potential threats at an early stage and limit the damage by Distributed Denial of Service (DDoS) attacks originated in IoT devices.

Smart cities are expected to improve the quality of life of citizens by relying on new paradigms, such as the Internet of Things (IoT) and its capacity to manage and interconnect thousands of sensors and actuators scattered across the city. At the same time, mobile devices widely assist professional and personal everyday activities. A very good example of the potential of these devices for smart cities is their powerful support for intuitive service interfaces (such as those based on augmented reality (AR)) for non-expert users. In our work, we consider a scenario that combines IoT and AR within a smart city maintenance service to improve the accessibility of sensor and actuator devices in the field, where responsiveness is crucial. In it, depending on the location and needs of each service, data and commands will be transported by an urban communications network or consulted on the spot. Direct AR interaction with urban objects has already been described; it usually relies on 2D visual codes to deliver object identifiers (IDs) to the rendering device to identify object resources. These IDs allow information about the objects to be retrieved from a remote server. In this work, we present a novel solution that replaces static AR markers with dynamic markers based on LED communication, which can be decoded through cameras embedded in smartphones. These dynamic markers can directly deliver sensor information to the rendering device, on top of the object ID, without further network interaction.

Future wireless communications are heading towards an all-Internet Protocol (all-IP) design, and will rely on the Session Initiation Protocol (SIP) to manage services, such as voice over IP (VoIP). The centralized architecture of traditional SIP has numerous disadvantages for mobile ad hoc services that may be possibly overcome by advanced peer-to-peer (P2P) technologies initially developed for the Internet. In the context of mobile sensing, P2PSIP protocols facilitate decentralized and fast communications with sensor-enabled terminals. Nevertheless, in order to make P2PSIP protocols feasible in mobile sensing networks, it is necessary to minimize overhead transmissions for signaling purposes, which reduces the battery lifetime. In this paper, we present a solution to improve the management of wireless overlay networks by defining an adaptive algorithm for the calculation of refresh time. The main advantage of the proposed algorithm is that it takes into account new parameters, such as the delay between nodes, and provides satisfactory performance and reliability levels at a much lower management overhead than previous approaches. The proposed solution can be applied to many structured P2P overlays or P2PSIP protocols. We evaluate it with Kademlia-based distributed hash tables (DHT) and dSIP.

The extension of the network lifetime of Wireless Sensor Networks (WSN) is an important issue that has not been appropriately solved yet. This paper addresses this concern and proposes some techniques to plan an arbitrary WSN. To this end, we suggest a hierarchical network architecture, similar to realistic scenarios, where nodes with renewable energy sources (denoted as primary nodes) carry out most message delivery tasks, and nodes equipped with conventional chemical batteries (denoted as secondary nodes) are those with less communication demands. The key design issue of this network architecture is the development of a new optimization framework to calculate the optimal assignment of renewable energy supplies (primary node assignment) to maximize network lifetime, obtaining the minimum number of energy supplies and their node assignment. We also conduct a second optimization step to additionally minimize the number of packet hops between the source and the sink. In this work, we present an algorithm that approaches the results of the optimization framework, but with much faster execution speed, which is a good alternative for large-scale WSN networks. Finally, the network model, the optimization process and the designed algorithm are further evaluated and validated by means of computer simulation under realistic conditions. The results obtained are discussed comparatively.

In this paper, we propose a solution for gunshot location in national parks. In Spain there are agencies such as SEPRONA that fight against poaching with considerable success. The DiANa project, which is endorsed by Cabañeros National Park and the SEPRONA service, proposes a system to automatically detect and locate gunshots. This work presents its technical aspects related to network design and planning. The system consists of a network of acoustic sensors that locate gunshots by hyperbolic multi-lateration estimation. The differences in sound time arrivals allow the computation of a low error estimator of gunshot location. The accuracy of this method depends on tight sensor clock synchronization, which an ad-hoc time synchronization protocol provides. On the other hand, since the areas under surveillance are wide, and electric power is scarce, it is necessary to maximize detection coverage and minimize system cost at the same time. Therefore, sensor network planning has two targets, i.e., coverage and cost. We model planning as an unconstrained problem with two objective functions. We determine a set of candidate solutions of interest by combining a Sensors 2009, 9 9494 derivative-free descent method we have recently proposed with a Pareto front approach. The results are clearly superior to random seeding in a realistic simulation scenario.

In Narrowband Internet of Things (NB-IoT), the control of uplink transmissions is a complex task involving device scheduling, resource allocation in the carrier, and the configuration of link-adaptation parameters. Existing heuristic proposals partially address the problem, but reinforcement learning (RL) seems to be the most effective approach a priori, given its success in similar control problems. However, the low sample efficiency of conventional (model-free) RL algorithms is an important limitation for their deployment in real systems. During their initial learning stages, RL agents need to explore the policy space selecting actions that are, in general, highly ineffective. In an NB-IoT access network this implies a disproportionate increase in transmission delays. In this paper, we make two contributions to enable the adoption of RL in NB-IoT: first, we present a multi-agent architecture based on the principle of task division. Second, we propose a new model-based RL algorithm for link adaptation characterized by its high sample efficiency. The combination of these two strategies results in an algorithm that, during the learning phase, is able to maintain the transmission delay in the order of hundreds of milliseconds, whereas model-free RL algorithms cause delays of up to several seconds. This allows our approach to be deployed, without prior training, in an operating NB-IoT network and learn to control it efficiently without degrading its performance.