To explore service-level appropriate network configurations efficiently, we propose a processing pipeline that generates a digital twin given only a packet capture trace from the network. Our pipeline outputs an instance of a discrete-event simulator, which can then be easily customized to explore arbitrary what-if configuration scenarios for the network. The parameters of this simulator instance are optimized to fit measured KPIs in the real-life network, using a genetic algorithm operating on a mix of actual fitness from the simulator and surrogate fitness implemented by neural networks. We show that our surrogate model can both speed up the time of finding a good fit, by 17%, as well as improve the fit, by 1.8x, compared to random parameter selection. We also evaluate our approach in an experimental testbed using what-if predictions of Wi-Fi primary channel configurations for several application workloads, showing 153% improvement in collision probability and 110% improvement in throughput predictions over a random baseline. Furthermore, we show that we only have to train our neural networks on a subset of the network workloads as it generalizes well.

The growing complexity of wireless networks has sparked an upsurge in the use of artificial intelligence (AI) within the telecommunication industry in recent years. In network slicing, a key component of 5G that enables network operators to lease their resources to third-party tenants, AI models may be employed in complex tasks, such as short-term resource reservation (STRR). When AI is used to make complex resource management decisions with financial and service quality implications, it is important that these decisions be understood by a human-in-the-loop. In this paper, we apply state-of-the-art techniques from the field of Explainable AI (XAI) to the problem of STRR. Using real-world data to develop an AI model for STRR, we demonstrate how our XAI methodology can be used to explain the real-time decisions of the model, to reveal trends about the model’s general behaviour, as well as aid in the diagnosis of potential faults during the model’s development. In addition, we quantitatively validate the faithfulness of the explanations across an extensive range of XAI metrics to ensure they remain trustworthy and actionable.

This article addresses the issue of reliability in the Industrial Internet of Things (IIoT) in case of missing sensors measurements due to network or hardware problems. We propose to support the fault detection and classification modules, which are the two critical components of a monitoring system for IIoT, with a generative model. The latter is responsible for imputing missing sensor measurements so that the monitoring system performance is robust to missing data. In particular, we adopt generative adversarial networks (GANs) to generate missing sensor measurements and we propose to fine-tune the training of the GAN based on the impact that the generated data have on the fault detection and classification modules. We conduct a thorough evaluation of the proposed approach using the extended Tennessee Eastman Process data set. Results show that the GAN-imputed data mitigate the impact on the fault detection and classification even in the case of persistently missing measurements from sensors that are critical for the correct functioning of the monitoring system.

The rapid adoption of the Internet of Things (IoT) as a means for digital transformation is sketching a new landscape of heterogeneous data and distributed, machine learning-based, applications. The intertwine of the two combined with the varying availability of data, generated in different parts of the domain, raises the need to exchange bulks of relevant data on demand across application(s) points. Data relevance escalates the role of semantics in identifying and locating suitable data; particularly at the network layer, to provide efficient mapping of data supply and demand. This paper proposes a semantic-based data networking framework, for managed IoT domains, embracing principles of information-centric networking without restrictions on the routing function. Managed semantics are used to provide flexible (label-based) data addressing scheme and a scalable semantic locator function, designed as an overlay network of distributed instances that can be realized on top of any routing or forwarding solution. Nonetheless, we outline different routing solutions and their suitability to such scenarios, to then draw a recommendation of the most suitable underlying routing fabric. We evaluate our framework over an example IoT domain of the Pervasive Nation (PN), Ireland national IoT network. Through our example, we show that the number of managed semantics in such a domain can be vastly smaller than that expected on an Internet scale. We analyze our semantic aggregation scheme over the example PN network, and show the high flexibility in mapping data while maintaining a small state in the semantic locator function.

Networked industrial systems capitalize on recent advancements in sensing, communications, computing and storage to improve productivity, operational and cost efficiency. The proliferation of effective techniques for knowledge extraction drive a paradigm shift in industrial environments and provide a fertile ground for enhanced process monitoring and control capabilities. In an effort to shed light on industrial data management operations, this paper presents two different approaches for dealing with information processing tasks of aggregated sensor measurements. Such tasks constitute part of an end-to-end process monitoring solution which is implemented in an open-source platform following a modular, scalable and interpretable procedure. A mapping of the industrial data processing components to the operational principles and architecture of a cyber-physical system reveals useful insights for an automated supervision of critical processes and workflows.

Large scale multiple-input multiple-output (MIMO) or Massive MIMO is one of the pivotal technologies for future wireless networks. However, the performance of massive MIMO systems heavily relies on accurate channel estimation. While the acquisition of channel state information (CSI) in such systems requires an increasingly large amount of training overhead as the number of users grows. To tackle this issue, in this paper, we propose a deep learning assisted blind channel estimation technique for orthogonal frequency division multiplexing (OFDM) based massive MIMO systems. We prove that by exploiting the asymptotic orthogonality of the massive MIMO channels, the channel distortion can be averaged out without the prior knowledge of channel impulse responses, and after some mathematical manipulation, different users' transmitted data symbols can be extracted. Thus, by deploying a denoising convolutional neural network algorithm (DnCNN), we mitigate a remaining channel and noise effect to accurately detect the transmitted data symbols at the channel sounding stage. Using the detected data symbols as virtual pilots, we estimate the CSI of all the users at each BS antennas. Our simulation results testify the efficacy of our proposed technique and demonstrate that it can provide a mean square error (MSE) performance which coincides with that of the data-aided channel estimation technique.

Ultradense Internet of Things (IoT) mesh networks and machine-to-machine communications herald an enormous opportunity for new computing paradigms and are serving as a catalyst for profound change in the evolution of the Internet. The collective computation capability of these IoT devices is typically neglected in favor of cloud or edge processing. We propose a framework to tap into this resource pool by coupling data communication and processing. Raw data captured by sensing devices can be aggregated and transformed into appropriate actions as it travels along the network toward actuating nodes. This paper presents an element of this vision, whereby we map the operations of an artificial neural network onto the communication of a multihop IoT network for simultaneous data transfer and processing. By exploiting the principle of locality inherent to many IoT applications, the proposed approach can reduce the latency in delivering processed information. Furthermore, it improves the distribution of energy consumption across the IoT network compared to a centralized processing scenario, thus mitigating the “energy hole” effect and extending the overall lifetime of the system.

Ultra-dense Internet of Things (IoT) networks and machine type communications herald an enormous opportunity for new computing paradigms and are serving as a catalyst for profound change in the evolution of the Internet. We explore leveraging the communication within IoT to serve data processing by appropriately shaping the aggregate behavior of a network to parallel more traditional computation methods. This paper presents an element of this vision, whereby we map the operations of an artificial neural network onto the communication of an IoT network for simultaneous data processing and transfer. That is, we provide a framework to treat a network holistically as an artificial neural network, rather than placing neural networks within the network. The operation of components of a neural network, neurons and connections between neurons, are performed by the various elements of the IoT network, i.e., the devices and their connections. The proposed approach reduces the latency in delivering processed information and supports the locality of information inherent to IoT by removing the need for transfer to remote data processing sites.

Studies of clustering in Wireless Sensor Networks (WSNs) usually tackle the problems of designing new algorithms and compare them based on a set of properties (e.g. energy efficiency, scalability), lacking the understanding of the underlying mechanisms and communication patterns that lead to these properties. Our approach tackles this lack of understanding by applying techniques developed by complex systems scientists. Functional topology graphs, which describe the interactions between system parts, are used to represent different implementations of clustering in WSNs. We employ a complexity metric - functional complexity (CF) - to quantify the potential of the functional topology to transport information. Our analysis highlights the trade-off between scalability and energy efficiency, showing that higher values of CFindicate higher scalability and lower energy efficiency.

In this work we propose a virtualized network architecture for an infrastructure provider that shares the physical resources of a Massive MIMO cell among several virtual network operators (VNOs) using spatial multiplexing. In this architecture the infrastructure provider allocates spatial streams to the VNOs, which enables each VNO to select its own scheduling policy and user priority to differentiate its service from the other VNOs. To assign the spatial streams to the VNOs that value them the most, we propose an auction-based spatial stream allocation approach. We show that the proposed auction- based approach performs very close to the optimal (fixed) approach in the case of homogeneous static VNOs demand. In case of heterogeneous demands, the auction mechanism is able to dynamically allocate the resources according to the needs of different VNOs.

Flexible licensing model is a necessary enabler of the technical and procedural complexities of Spectrum Access System (SAS)-based sharing framework. The purpose of this study is to explore the effectiveness of 3.5GHz Licensing Framework which is based on census tracts as license area units. Their main characteristic is population - as such, the boundary of census tract does not follow the edge of wireless network coverage. We demonstrate why census tracts are not suitable for small cell networks licensing, by: (1) gathering and analysing the official census data, (2) exploring the boundaries of census tracts which are in the shape of non-convex polygons, and (3) giving a measure of effectiveness of the licensing scheme through metrics of area loss and the number of people per census tract with access to spectrum. Results show that census tracts severely impact the effectiveness of the licensing framework since almost entire strategically important cities in the U.S. will not avail from spectrum use in 3.5 GHz band. Our paper does not seek to challenge the core notion of geographic licensing concept, but seeks a corrective that addresses the way the license is issued for a certain area of operation. The effects that inappropriate size of the license has on spectrum assignments lead to spectrum being simply wasted in geography, time and frequency. The corrective is necessary since the main goal of promoting innovative sharing in 3.5 GHz band is to put spectrum to more efficient use.

This paper addresses the channel selection problem for Long Term Evolution Unlicensed (LTE-U). Channel selection is a frequency-domain mechanism that facilitates the coexistence of multiple networks sharing the unlicensed band. In particular, the paper considers a fully distributed approach where each small cell autonomously selects the channel to set-up an LTE-U carrier. The problem is modeled using a non-cooperative repeated game and the Iterative Trial and Error Learning - Best Action (ITEL-BA) learning algorithm is used to drive convergence towards a Nash Equilibrium. The proposed approach is evaluated by means of simulations in different situations analyzing both the throughput performance and the convergence behavior.

This paper proposes a two-stage process for assigning fungible orthogonal channel sets to multiple cognitive radios (CRs) for opportunistic spectrum access. Assigning orthogonal channel sets to the CRs eliminates the possibility of collision among them, and allows the CRs to focus on avoiding collisions with the primary user (PU). In particular, each CR uses a learning-based dynamic channel selection (DCS) algorithm to maximize the exploitation of the assigned channels. We propose a neural network that can accurately estimate the performance of the adopted learning-based DCS algorithm on a set of channels, using the duty cycle and the complexity of the PU's behavior on the channels. Our simulations on synthetic and real measurement data sets show that the proposed channel sets allocation algorithm, together with the neural network, significantly outperforms a method that selects channels with the lowest duty cycle.

Telecommunication systems have evolved, from being simple monolithic structures to complex ones. The present paper is an attempt to recast telecom systems in the language of complex systems. As a specific application of complex systems science to telecom systems, we tackle the problem of self-organizing frequency allocation. For this purpose, we model the problem of frequency assignment within the framework of cellular automata and we present an algorithm that reaches an interference-free allocation of the channels in a finite number of steps. By applying a measure of spatial structure and pattern to the two-dimensional lattice representing the frequency allocation, our study shows that autonomous networks assigning frequencies to their cells in a self-organized way, can be defined and therefore studied as complex systems. On the contrary, traditional networks based on centralized pre-planned frequency allocation cannot be considered as complex, according to the meaning given to the word by complex systems science. The present paper is to be considered as a step towards a comprehensive and rigorous study of Complex Communication Systems (CCS), adopting in communication networks design and analysis, the results and philosophy underlying the emerging multi-disciplinary field of Complex Systems Science.

This paper suggests that the regulatory tool known as the Block Edge Mask (BEM), which is used to define the technical conditions governing a spectrum license, can be used as an instrument to jointly drive spectrum and energy efficiency in a network. The paper explores this concept and goes on to show how networks can generate waveforms which comply with the energy limits of the BEM and at the same time make best use of the spectral resources available to them. The focus of the paper is on orthogonal frequency division multiplexing (OFDM) waveforms. The spectrum- and energy-efficient BEM-compliant waveforms are generated using a process that optimally combines symbol shaping and subcarrier-power loading to create a waveform that minimizes the difference between the transmitted wave and BEM. Practical considerations relating to peak-to-average-power-ratio (PAPR) reduction and time and frequency synchronization issues are explored to ensure the BEM-compliant waveforms can be transmitted with minimal changes to the transceiver. It is envisaged that these techniques will be used by transmitters that change their frequency of operation and in doing so need to shape their transmitted waveforms to comply with the different BEMs governing the use of the different spectrum blocks they wish to access.

Carrier aggregation is a key feature of next generation wireless networks to deliver high-bandwidth links. This paper studies carrier aggregation for autonomous networks operating in shared spectrum. In our model, networks decide how many and which channels to aggregate in multiple frequency bands, hence extending the distributed channel allocation framework. Moreover, our model takes into the account physical layer issues, such as the out-of-channel interference in adjacent frequency channels and the cost associated with inter-band carrier aggregation. We propose learning algorithms that converge to Nash equilibria in a reasonable number of iterations under the assumption of incomplete and imperfect information.

To improve their ability to find spectrum opportunities, intelligent secondary radios (SR) can learn from their past observations and predict possible spectrum opportunities. However, because of the diverse behavior of primary users (PU) in different spectrum bands, spectrum holes exhibit diverse characteristics, which in turn affect the performance of a learning algorithm. This paper studies the effect of the PU's activity on channel predictability. In particular, we introduce a Markov process-based learning algorithm, and we investigate the dependency of its spectrum decisions on the duty cycle (DC) and on the complexity of each channel activity, for both synthetic and real data. Our findings show that the probability of finding a free channel among a group of considered channels strongly depends on the DC and the complexity of the channel activity. Moreover, it is possible to reduce the number of observed channels without compromising the probability of finding a free channel, by only considering the more informative channels.

The concept of cancellation carriers (CCs) has been proposed in the literature for sidelobe suppression for orthogonal frequency division multiplexing (OFDM) systems. Subcarriers at the edges of the OFDM spectrum are used for sidelobe reduction while the remaining subcarriers are used for data transmission. Existing CCs techniques require performing complex optimization that should be applied for each OFDM symbol which is not suitable for real-time applications. In this paper, we propose a heuristic approach for CCs. The proposed algorithm involves few computations compared with all other techniques proposed in the literature. Simulation results show that sidelobe reduction performance can be close to the optimal. Moreover, the proposed algorithm is implemented in a software defined radio and implementation results prove that it can be introduced for real-time applications.

We explore the question of when learning improves the performance of opportunistic dynamic channel selection by characterizing the primary user (PU) activity using the concept of Lempel-Ziv complexity. We evaluate the effectiveness of a reinforcement learning algorithm by testing it with real spectrum occupancy data collected in the GSM, ISM, and DECT bands. Our results show that learning performance is highly correlated with the level of PU activity and the amount of structure in the use of spectrum. For low levels of PU activity and/or high complexity in its utilization of channels, reinforcement learning performs no better than simple random channel selection. We suggest that Lempel-Ziv complexity might be one of the features considered by a cognitive radio when deciding which channels to opportunistically explore.

This paper considers spatial spectrum sharing-based carrier aggregation (SSS-CA) from a game theoretic perspective. In SSS-CA, a network operator can not only transmit on its own licensed spectrum but it can also access and aggregate the licensed spectrum of other operators on payment of a certain price. The difference between operators and the aggregators in each licensed spectrum makes this network heterogeneous. We first model the pairing problem between potential operators as a pairing game and then derive the condition for which both operators are incentivized to form an SSS-CA pair. We then introduce the power control game to derive the optimal transmit power of each spectrum aggregator. Finally, we consider the pricing optimization problem by forming a pricing adjustment game. We observe that these three problems are linked by the price function of the operators and hence can be jointly optimized by using a hierarchical game theoretic framework. We derive the Stackelberg equilibrium for the pricing and power joint optimization problem and present the numerical results to compare the performance improvement brought by our proposed joint optimization method.

This paper presents an approach for the simultaneous learning and recognition of places applied to autonomous robotics. While noteworthy results have been achieved with respect to off-line training process for appearance-based navigation, novel issues arise when recognition and learning are simultaneous and unsupervised processes. The approach adopted here uses a Gaussian mixture model estimated by a novel incremental MML-EM to model the probability distribution of features extracted by image-preprocessing. A place detector decides which features belong to which place integrating odometric information and a hidden Markov model. Tests demonstrate that the proposed system performs as well as the ones relying on batch off-line environmental learning.

This paper deals with the design and implementation of a visual control of a robotic system composed of a dexterous hand and video camera. The aim of the proposed system is to reproduce the movements of a human hand in order to learn complex manipulation tasks or to interact with the user. A novel algorithm for robust and fast fingertips localization and tracking is presented. A suitable kinematic hand model is adopted to achieve a fast and acceptable solution to an inverse kinematics problem. The system is part of a cognitive architecture for posture learning that integrates the perceptions by a high-level representation of the scene and of the observed actions. The anthropomorphic robotic hand imitates the gestures acquired by the vision system in order to learn meaningful movements, to build its knowledge by different conceptual spaces, and to perform complex interactions with the human operator.

The paper deals with the design and implementation of a visual control of a robotic system composed of a dexterous hand and stereo cameras. The aim of the proposed system is to reproduce the movements of a human hand in order to learn complex manipulation tasks. A novelty algorithm for a robust and fast fingertips localization and tracking is presented. Moreover, a simulator is integrated in the system to give useful feedbacks to the users during operations, and provide robust testing framework for real experiments.