Radio Access Networks (RANs) are critical infrastructures that mobile operators continuously upgrade to accommodate increasing data traffic demands, stricter performance requirements, and evolutions in radio technologies. RAN updates can affect carrier-level Key Performance Indicators (KPIs) that are the foundational input to data-driven models for network management. However, to date, no study has systematically examined the dynamics of RAN deployments, and little is known about the actual prevalence of RAN updates or their impact on Machine Learning (ML) models for network automation. This paper presents a first characterization of RAN updates in a nationwide operational infrastructure composed of over 500,000 carriers. A network-side vantage point lets us (i) investigate the type and frequency of RAN modifications, (ii) assess the impact of such changes on a primary KPI for network management, i.e., the traffic volume served by individual carriers, and (iii) verify the final effects on a classical downstream ML application, i.e., traffic prediction. Our results reveal that RAN updates take place with notable frequency, e.g., occurring every few days even in medium-sized cities. Also, they affect in a significant way the demands at a considerable fraction of pre-existing carriers, where they can curb the accuracy of ML traffic forecasting models.

The rollout of 5G represents a significant advancement in the telecommunications industry, offering the potential for markedly enhanced speeds, reduced latency, and improved connectivity. Considering these anticipated advantages, it is interesting to understand the progressive adoption of the new technology by operators and their subscribers. In this paper, we analyze the evolution and current operation of the nation-wide 5G network of Orange, a leading mobile operator in France. By inspecting longitudinal data about (i) the over-five-year-long development of the country-wide 5G radio access infrastructure and (ii) the last two years of 5G traffic demands, we unveil how the operator has planned the deployment of the 5G radio access and characterize the actual usage patterns of the available 5G infrastructure. We also investigate the recent introduction of a 5G Standalone (SA) commercial service and its adoption by the mobile subscribers. We show that by mid 2025, the 5G network under study has achieved substantial coverage of populated areas and the operator has very recently started adding capacity layers to its 5G radio access. However, our investigation reveals that such massive infrastructure deployment efforts are not matched by a commensurate adoption of the technology by the end users, as the 5G capacity -especially for SA- stays largely underutilized.

Federated Learning (FL) has emerged as a pivotal privacy-preserving machine learning paradigm, enabling collaborative model train across distributed data sources. A main issue, however, is the lack of comprehensive testing environments that can accurately emulate real-world conditions at a large scale, particularly the impact of network dynamics and device capabilities on FL algorithm performance. To this end, we introduce FederNet, a novel platform designed to facilitate the development and testing of FL algorithms with realistic network and device emulation. We show how the proposed system provides a versatile platform for researchers to evaluate the performance, robustness, and scalability of FL algorithms under diverse and configurable scenarios. We describe the FederNet architecture, detail its network and device emulation capabilities, and outline potential use cases that demonstrate its utility in advancing FL research. By bridging the gap between algorithmic development and practical deployment challenges, FederNet aims to accelerate the innovation and adoption of FL technologies.

• Scanned IPv4 to find 14386 MQTT brokers and captured 3.2 billion messages • Few brokers generate most traffic (“elephants”), while many handle little (“mice”). • Brokers show low load overall, with heavy-tailed traffic patterns. • Over 88% of messages use QoS 0, prioritizing low-latency over delivery guarantees. • Topics and payloads mostly follow best practices, but outliers reveal inefficiencies. The Message Queue Telemetry Transport (MQTT) protocol is widely used in Internet of Things (IoT) applications, offering a lightweight and efficient communication model for resource-constrained devices. Despite its increasing adoption in domains such as smart homes, industrial automation, and environmental monitoring, large-scale empirical studies on MQTT traffic are rare, and existing work often focuses on controlled experiments rather than natural, in-the-wild deployments. In this paper, we address this gap by analyzing MQTT traffic “in the wild”. We developed a measurement framework to scan the IPv4 address space, identifying 14,386 active brokers. We collected about 3.2 billion messages over two weeks, enabling an in-depth study of broker throughput, topic structures, payload composition, and QoS configurations. Our findings reveal that broker throughput is generally low, suggesting limited stress in real-world usage. Topic structures vary significantly, with some brokers using deep hierarchies, which may impact distributed deployments. Structured payloads (JSON, Strings) dominate MQTT traffic, presenting opportunities for broker-side optimizations. Furthermore, QoS 0 is overwhelmingly preferred, indicating a focus on low-latency communication over reliability guarantees. These insights contribute to a better understanding of MQTT traffic patterns, which can be leveraged for protocol optimizations, scalability strategies, and security considerations for future IoT deployments.

• We use deep learning to forecast mobile traffic, with particular emphasis on traffic peaks • We integrated exogenous features with traffic input for closed-loop prediction • We generated synthetic data to address scarcity and enhance model generalization • We validated performance on real-world data, outperforming baseline methods Mobile cellular networks are experiencing rapid growth in data demand, largely driven by data-intensive applications such as video streaming. In particular, the popularity of live events can induce abrupt and localized traffic surges, often resulting in congestion and performance degradation. For these reasons, accurate traffic forecasting is expected to play an increasingly important role in future sixth-generation (6G) mobile networks, supporting both real-time operational responses and long-term capacity planning. In this work, we move beyond classical traffic forecasting approaches and propose an AI-based framework that explicitly conditions traffic predictions on contextual information available in advance, while also leveraging generative data augmentation to address data scarcity. Through a comprehensive analysis conducted on two major Italian cities and several real-world datasets spanning four years, we show that traffic dynamics exhibit strong correlations with the occurrence of football matches. Building on this observation, we design a forecasting methodology that combines historical traffic measurements with scheduled event information to forecast future traffic over the prediction horizon. To improve robustness under event-driven and high-load conditions, we further introduce a lightweight synthetic data generation strategy that mitigates the scarcity and imbalance of rare traffic patterns. Experimental results demonstrate that the proposed framework improves forecasting accuracy, particularly during peak and busy-hour regimes, compared to baseline traffic-only approaches

The rollout of 5G promises major improvements in speed, latency, and connectivity over previous-generation radio access technologies. Our study analyzes Orange's nationwide 5G network in France, combining longitudinal data on infrastructure deployment with data traffic patterns. Early results show that while 5G coverage has steadily expanded and is presently reaching the vast majority of the user population, adoption by mobile subscribers remains limited, leaving much of the new capacity underutilized.

Flow records, that summarize the characteristics of traffic flows, represent a practical and powerful way to monitor a network. While they already offer significant compression compared to full packet captures, their sheer volume remains daunting, especially for large Internet Service Providers (ISPs). In this paper, we investigate several lossy compression techniques to further reduce storage requirements while preserving the utility of flow records for key tasks, such as predicting the domain name of contacted servers. Our study evaluates scalar quantization, Principal Component Analysis (PCA), and vector quantization, applied to a real-world dataset from an operational campus network. Results reveal that scalar quantization provides the best tradeoff between compression and accuracy. PCA can preserve predictive accuracy but hampers subsequent entropic compression, and while vector quantization shows promise, it struggles with scalability due to the high-dimensional nature of the data. These findings result in practical strategies for optimizing flow record storage in large-scale monitoring scenarios.

As mobile data traffic continues to increase and network operators expand their infrastructures, energy consumption in Radio Access Networks (RAN) becomes a critical concern, particularly during network expansion. This paper addresses the problem of predicting RAN energy consumption in network expansion scenarios. We develop and evaluate different forecasting strategies, including per-site and network-wide models, as well as Machine Learning (ML)-based approaches, using real-world data from a LTE network. Our results show that while simple network-wide models perform well when the base station (BS) configurations remain constant, ML models are more effective in scenarios where BS configurations change during network expansion. The insights from this study can help mobile network operators improve energy efficiency by adapting their networks to traffic patterns and expansion processes, supporting both cost management and sustainability goals.

The rapid adoption of Internet of Things (IoT) devices in smart environments has led to a new era for digital forensics. As IoT devices become increasingly prevalent in homes, cities, and workplaces, they serve as silent observers of everyday human activities. Recent studies have investigated how network traces from these devices could be leveraged to support forensic investigations. However, this approach requires large-scale data collection while ensuring the confidentiality, anonymity, and integrity of the collected traces. This work advances the field of IoT forensics by introducing Chain4ensic, a blockchain-based chain-of-custody (CoC) framework designed to preserve network traces extracted from IoT devices in a secure way. The proposed framework utilizes Ethereum smart contracts and edge computing to preserve the integrity and traceability of digital evidence. Additionally, the proposed system facilitates public notifications of data access once authorized by a court of justice. The architecture is developed as an opensource solution and evaluated in a smart home scenario, demonstrating its feasibility and effectiveness in real-life applications. The results indicate low resource usage, high throughput, and small gas consumption, making it a promising approach to tackle the challenges of storing sensitive data from participants in IoT forensic investigations.

The Internet of Things (IoT) is revolutionizing connectivity by enabling everyday objects and devices to communicate over the Internet, streamlining industrial processes, improving quality of life, and driving innovation across various applications. Ensuring seamless communication between heterogeneous IoT devices remains a fundamental challenge due to differences in underlying communication protocols. This work presents a broker-based extension designed to enable direct, bidirectional interoperability between MQTT and CoAP, two of the most widely adopted IoT protocols. Unlike middleware-based solutions, this approach integrates RESTful and publish/subscribe paradigms within a single broker architecture, streamlining deployment and minimizing system complexity. The proposed solution demonstrates low-latency performance and minimal resource overhead with low median latencies of 2.15 ms and an estimated increase of only 0.53 ms over the baseline. Resource-wise, the broker extension shows a modest median CPU overhead of 2.6% and a memory usage increase of just 0.1% at higher throughputs. These results confirm that the system can efficiently manage multiple concurrent message flows and maintain scalability under varying traffic conditions. This position the broker extension as a practical, modular, and lightweight solution for enhancing IoT interoperability in constrained environments.

The rapid proliferation of Internet of Things (IoT) devices across diverse sectors has given rise to the field of IoT Forensics, which focuses on analyzing digital traces of IoT appliances for legally significant insights. This field extends traditional digital forensic methods to address the unique characteristics of IoT devices, with the goal of identifying security breaches and reconstructing human activities based on data retrieved from IoT systems. Due to the limited memory and processing capabilities of IoT devices, innovative methods for data collection and analysis are required. As an example, in the Smart Home or Smart Office scenarios intermediate network devices such as Wi-Fi access points may be leveraged for such goals, including monitoring and analysis of IoT network traffic. In this context, this paper proposes a resource optimization model for forensics tasks based on network traffic monitoring and analysis on consumer Wi-Fi access points. The model maximises the expected performance achieved for forensic tasks while balancing the storage and processing capabilities required for data collection in Wi-Fi access points. The proposed model can determine the optimal aggregation window used to group network packets for traffic analysis, the number of statistical features to extract from such packets and the bits per feature to use for data storage, in order to achieve optimal accuracy and maintain low impact on the computing device. Experimental results demonstrate the models efficacy in constrained environments, allowing us to decide on the resource allocation in network devices when a high number of tasks is involved.

The rapid evolution of artificial intelligence (AI) technologies holds transformative potential for the healthcare sector. In critical situations requiring immediate decision-making, healthcare professionals can leverage machine learning (ML) algorithms to prioritize and optimize treatment options, thereby reducing costs and improving patient outcomes. However, the sensitive nature of healthcare data presents significant challenges in terms of privacy and data ownership, hindering data availability and the development of robust algorithms. Federated Learning (FL) addresses these challenges by enabling collaborative training of ML models without the exchange of local data. This paper introduces a novel FL platform designed to support the configuration, monitoring, and management of FL processes. This platform operates on Platform-as-a-Service (PaaS) principles and utilizes the Message Queuing Telemetry Transport (MQTT) publish-subscribe protocol. Considering the production readiness and data sensitivity inherent in clinical environments, we emphasize the security of the proposed FL architecture, addressing potential threats and proposing mitigation strategies to enhance the platform's trustworthiness. The platform has been successfully tested in various operational environments using a publicly available dataset, highlighting its benefits and confirming its efficacy.

MAC randomization is a widely used technique implemented on most modern smartphones to protect user’s privacy against tracking based on Probe Request frames capture. However, there exist weaknesses in such a methodology which may still expose distinctive information, allowing to track the device generating the Probe Requests. Such techniques, known as MAC de-randomization algorithms, generally exploit Information Elements (IEs) contained in the Probe Requests and use clustering methodologies to group together frames belonging to the same device. While effective on heterogeneous device types, such techniques are not able to differentiate among devices of identical type and running the same Operating System (OS). In this paper, we propose a MAC de-randomization technique able to overcome such a weakness. First, we propose a new dataset of Probe Requests captured from devices sharing the same characteristics. Secondly, we observe that the time-frequency pattern of Probe Request emission is unique among devices and can therefore be used as a discriminative feature. We embed such a feature in a two-stage clustering methodology and show through experiments its effectiveness compared to state-of-the-art techniques based solely on IEs fingerprinting. The original dataset used in this work is made publicly available for reproducible research.

Federated Learning (FL) has emerged as a promising approach for privacy-preserving machine learning, particu-larly in sensitive domains such as healthcare. In this context, the TRUSTroke project aims to leverage FL to assist clinicians in ischemic stroke prediction. This paper provides an overview of the TRUSTroke FL network infrastructure. The proposed archi-tecture adopts a client-server model with a central Parameter Server (PS). We introduce a Docker-based design for the client nodes, offering a flexible solution for implementing FL processes in clinical settings. The impact of different communication pro-tocols (HTTP or MQTT) on FL network operation is analyzed, with MQTT selected for its suitability in FL scenarios. A control plane to support the main operations required by FL processes is also proposed. The paper concludes with an analysis of security aspects of the FL architecture, addressing potential threats and to increase trustworthiness.

Recent advances in the Internet of Things are leading to a proliferation of smart devices in our daily life. Having so many connected devices around us potentially introduces new witnesses that can be a reference for forensic investigations. For these reasons, IoT Forensics has become a popular research area with the goal of extracting information from IoT devices to be used as potential evidence. This work presentsFeature-Sniffer, a framework to be installed in Wi-Fi access points with the aim of facilitating the extraction of network traffic information from IoT devices, to be later used for forensic purposes. The tool allows the on-the-fly computation of traffic features from connected IoT devices by using a simple user interface for its configuration. After presenting the tool logic and its implementation details, we present an accurate analysis of the tool computational impact on two different consumer Wi-Fi access points. Finally, we present four different IoT forensics use cases, in which network traffic features extracted with the proposed tool from consumer IoT devices are analyzed with machine learning techniques with the goal of 1) identifying the device producing the traffic; 2) recognizing the activity performed by the user; 3) detecting the user’s passage through a room door; and 4) detecting and classifying user interactions with a smart speaker. We conclude the work by presenting an analysis of possible storage optimization for evidence preservation with the use of lossy compression techniques.

The proliferation of Internet of Things (IoT) devices, coupled with the recent popularity of machine-learning and artificial intelligence has given birth to a new research field named IoT forensics. Such a new field considers network traffic from IoT devices as possible source of evidence for forensic investigations. However, the massive amount of IoT devices and traffic produced makes storage challenging, especially when this is performed on limited-resource edge devices such as e.g., WiFi access points. This paper proposes a framework to optimize the storage-accuracy trade-offs of IoT forensic analysis tasks. The goal of the framework is to find the optimal working point in terms of number of features to extract from network traffic and the number of bits used for quantizing each feature, in order to maximize the IoT forensic task accuracy under storage constraints. After presenting the framework, we validate it over two different IoT forensics tasks: IoT device identification and activity recognition from encrypted traffic of IoT cameras. Results show that with low effort it is possible to find the optimal settings to operate to maximize the analysis accuracy under given storage limitations.

Mobile network traffic forecasting is a fundamental building block for key management tasks such as resources allocation. In particular, being able to predict traffic volume peaks is of primary importance for a correct network operation. This paper proposes to exploit exogenous inputs to predict such peaks, focusing in particular on football matches. We show with an analysis conducted on 4 of the major cities in Italy for a period of 6 months that volume traffic peaks are strongly correlated with the occurrence of football matches between specific teams and we propose a methodology to exploit the football calendar to forecast traffic peaks. The proposed forecasting frameworks allows to predict more than 50 % of the peaks, improving the forecasting performance compared to a traffic signature based approach and being able to forecast the maximum traffic volume with an average overestimation below +10% of the actual value.

The dramatic growth in cellular traffic volume requires cellular network operators to develop strategies to carefully dimension and manage the available network resources. Forecasting traffic volumes is a fundamental building block for any proactive management strategy and is therefore of great interest in such a context. Differently from what found in the literature, where network traffic is generally predicted in the short-term, in this work we tackle the problem of forecasting busy hour traffic, i.e., the time series of observed daily maxima traffic volumes. We tackle specifically forecasting in the long term (one, two months ahead) and we compare different approaches for the task at hand, considering different forecasting algorithms as well as relying or not on a cluster-based approach which first groups network cells with similar busy hour traffic profiles and then fits per-cluster forecasting models to predict the traffic loads. Results on a real cellular network dataset show that busy hour traffic can be forecasted with errors below 10% for look-ahead periods up to 2 months in the future. Moreover, when clusters are available, we improve forecasting accuracy up to 8% and 5% for look-ahead of 1 and 2 months, respectively.

Network operators need to continuously upgrade their infrastructures in order to keep their customer satisfaction levels high. Crowdsourcing-based approaches are generally adopted, where customers are directly asked to answer surveys about their experience. Since the number of collaborative users is generally low, network operators rely on Machine Learning models to predict the satisfaction levels/QoE of the users rather than directly measuring it through surveys. Finally, combining the true/predicted users satisfaction labels with information on each user mobility (e.g, which network sites each user has visited and for how long), an operator may reveal critical areas in the network and drive/prioritize investments properly. In this work, we propose an empirical framework tailored to assess the quality of the detection of under-performing cells starting from subjective user experience grades. The framework allows to simulate diverse networking scenarios, where a network characterized by a small set of under-performing cells is visited by heterogeneous users moving through it according to realistic mobility models . The framework simulates both the processes of satisfaction surveys delivery and users satisfaction prediction, considering different delivery strategies and evaluating prediction algorithms characterized by different prediction performance. We use the simulation framework to test empirically the performance of under-performing sites detection in general scenarios characterized by different users density and mobility models to obtain insights which are generalizable and that provide interesting guidelines for network operators.

Message queuing telemetry transport (MQTT), one of the most popular application layer protocols for the Internet of Things, works according to a publish/subscribe paradigm where clients connect to a centralized broker. Sometimes (e.g., in high scalability and low-latency applications), it is required to depart from such a centralized approach and move to a distributed one, where multiple MQTT brokers cooperate together. Many MQTT brokers (both open source or commercially available) allow to create such a distributed environment: however, it is challenging to select the right solution due to the many available choices. This article proposes, therefore benchmarking framework for distributed MQTT brokers (BORDER), a framework for creating and evaluating distributed architectures of MQTT brokers with realistic and customizable network topologies. Based on isolated Docker containers and emulated network components, the framework provides quantitative metrics about the overall system performance, such as End-to-End latency as well as network and physical resources consumed. We use BORDER to compare five of the most popular MQTT brokers that allow the creation of distributed architectures and we release it as an open-source project to allow for reproducible researches.

Continuously monitoring the network activity to proactively recognise possible problems and prevent users QoE degradation is a major concern for network operators, for both mobile radio and home networks. Considering video streaming applications, which generate the majority of overall Internet traffic, monitoring the chunk requests from the video client to the video server is of particular interest, as they not only indicate that a download burst is imminent, but their type (e.g., request of an audio or video chunk) and frequency also allow to estimate which and how much data will be downloaded to the client. In this work, we propose a machine-learning based video streaming traffic monitoring architecture able to i) predict when next uplink request will be issued by the video client and ii) classify the type of next uplink request. We evaluate the system performance on a dataset of more than 900 HTTP adaptive streaming sessions and 15,000 request-response exchanges, where both the predictor of the next request arrival and the request type classifier are fed with lightweight features extracted from encrypted traffic in an online fashion, both in the uplink and downlink directions of the traffic. Results show that i) the system is able to classify the type of a HAS uplink requests with an accuracy greater than 95 % and ii) pipe-lining request type classification and prediction of next request arrival time improves the final prediction performance.

Vehicle-to-everything (V2X) is expected to become one of the main drivers of 5G business in the near future. Dedicated network slices are envisioned to satisfy the stringent requirements of advanced V2X services, such as autonomous driving, aimed at drastically reducing road casualties. However, as V2X services become more mission-critical, new solutions need to be devised to guarantee their successful service delivery even in exceptional situations, e.g. road accidents, congestion, etc. In this context, we propose π-ROAD, a deep learning framework to automatically learn regular mobile traffic patterns along roads, detect non-recurring events and classify them by severity level. π-ROAD enables operators to proactively instantiate dedicated Emergency Network Slices (ENS) as needed while re-dimensioning the existing slices according to their service criticality level. Our framework is validated by means of real mobile network traces collected within 400 km of a highway in Europe and augmented with publicly available information on related road events. Our results show that π-ROAD successfully detects and classifies non-recurring road events and reduces up to 30% the impact of ENS on already running services.

Monitoring the Quality of Experience (QoE) of the customer base is a key task for Mobile Network Operators (MNOs), and it is generally performed by collecting users feedbacks through directed surveys. When such feedbacks are few in number, a MNO may predict the users QoE starting from objective network measurements, gathered directly from the users equipments through crowdsourcing. In this work, we compare such a traditional approach with a different one, where the data used for predicting the users QoE is gathered directly at the network access, using Key Performance Indicators (KPI) available on each base station. Although such KPIs are aggregated by design (i.e., they refer to the distribution of a population of users rather than to a single individual), we show through experiments with a country-wide dataset that their predictive power is comparable and in some cases superior than the one of crowdsourcing. Such a result is particularly attractive for MNOs, since network KPIs are generally much easily obtainable than crowdsourcing data.

MQTT, one of the most popular protocols for the IoT, works according to a publish/subscribe pattern in which multiple clients connect to a single broker, generally hosted in the cloud. However, such a centralised approach does not scale well considering the massive numbers of IoT devices forecasted in the next future, thus calling for distributed solutions in which multiple brokers cooperate together. Indeed, distributed brokers can be moved from traditional cloud-based infrastructure to the edge of the network (as it is envisioned by the upcoming MEC technology of 5G cellular networks), with clear improvements in terms of latency, for example. This paper proposes MQTT-ST, a protocol able to create such a distributed architecture of brokers, organised through a spanning tree. The protocol uses in-band signalling (i.e., reuses MQTT primitives for the control messages) and allows for full message replication among brokers, as well as robustness against failures. We tested MQTT-ST in different experimental scenarios and we released it as open-source project to allow for reproducible research.

The Cloud-RAN (C-RAN) paradigm is envisioned to increase the efficiency of future mobile networks by moving the computational resources needed at the Remote Radio Heads (RRH) to the cloud infrastructure. In this work, we provide a framework that optimizes the number of allocated virtual resources by considering both the computational requirements of the RRH and the Quality of Service of users, which could experience loss of service due to reassociations between the RRH and the virtual machines. The provided optimization framework is supported by data coming from a real mobile network of a middle-sized European city, which provides an estimate for the computational loads coming from the RRH. We evaluate the performance of the framework in different scenarios, analyzing the impact of different forecasting algorithms as well as different look-ahead intervals for the predictions (short-term / long-term). The results obtained by our framework can be used to assist network operators in the optimization of C-RAN resources and shed some light on the interplay between forecasting errors and overall performance.

Accurate and efficient resource utilization predictions are of vital importance for the future generation of mobile wireless networks. By anticipating network resource demand, the operator can perform proactive resource allocation and predictive network control to improve network resource efficiency. In this paper, we exploit deep and transfer learning algorithms for multi-step resource utilization prediction in radio networks. In particular, we propose long short-term memory network-based architectures with transfer learning for the multi-step prediction task, in order to address scalability, computation time and data storage limitations of current implementations for large-scale networks. We carry out extensive experiments on a dataset collected from an LTE field network. When predicting physical resource block percentage utilization, our approach achieves state of the art results with root mean square error below 12 for a four-hour-ahead prediction, in half of the computation time required by deep learning methods without transfer learning.

We analyze a city-wide dataset of 4G mobile network traffic obtained directly from user-side logs, allowing fine-grained analyses of different application services over time and space. We group applications in classes and analyze their traffic patterns: the analysis reveals great heterogeneity in the usage of different applications and in their space/time correlations, with important implications for future networking services such as network slicing and resource allocations.

Standard LoRaWANs leverage pure ALOHA at the medium access control layer, which is proved to be a performance bottleneck as the network size scales up. Stimulated by this fact, this paper studies the applicability and the performance of listen before talk (LBT) medium access schemes in the context of LoRaWANs. We consider two different implementations of LBT: physical layer LBT based on energy detection only and MAC layer LBT based on layer 2 frame decoding, and we propose a Markovian framework to evaluate the performance of LoRaWANs under such setting in terms of data extraction rate and average delay experienced by transmitted uplink messages. The proposed framework is also leveraged to assess the performance of “mixed” LoRaWAN scenarios, where some devices access the channel according to the standard-compliant ALOHA protocol, while other devices transmit according to LBT.

The Message Queueing Telemetry Transport (MQTT) publish/subscribe protocol is the de facto standard at the application layer for IoT, M2M and wireless sensor networks applications. This demonstration showcases MQTT+, an advanced version of MQTT which provides an enhanced protocol syntax and enriches the broker with data filtering, processing and aggregation functionalities. Such features are ideal in all those applications in which edge devices are interested in performing processing operations over the data published by multiple clients, where using the original MQTT protocol would result in unacceptably high network bandwidth usage and energy consumption for the edge devices. MQTT+ is implemented starting from an open source MQTT broker and evaluated in different application scenarios which are demonstrated live using the Node-RED IoT prototyping framework.M

We set ourselves from the perspective of a LoRaWAN network operator and we introduce a mathematical programming framework to jointly optimize network layout and network configuration at design time. The proposed framework returns the most cost-effective network layout in terms of gateways position, gateways backhaul configuration and LoRaWAN physical parameters configuration under tight constraints of end node coverage, end-to-end message transmission latency and message extraction rate. Numerical results obtained on realistic network instances demonstrate that the proposed approach leads to network configuration with superior performance with respect to coverage-only classical design policies.

In this work we address the analysis of a LoRaWAN network where some devices access the channel according to the standard-compliant ALOHA protocol, while other devices transmit according to a Listen Before Talk paradigm based on the CSMA/CA mechanism. To analyze this scenario, we propose a mathematical model to obtain the Data Extraction Rate both for CSMA/CA and ALOHA devices, as well as the average delay experienced by messages transmitted by CSMA/CA devices. Simulation results show the accuracy of our model, as well as the benefits of introducing CSMA/CA devices into the network, even when not all the devices implement this mechanism and must coexist with ALOHA devices.

Fifth generation wireless networks (5G) will face key challenges caused by diverse patterns of traffic demands and massive deployment of heterogeneous access points. In order to handle this complexity, machine learning techniques are expected to play a major role. However, due to the large space of parameters related to network optimization, collecting data to train models for all possible network configurations can be prohibitive. In this paper, we analyze the possibility of performing a knowledge transfer, in which a machine learning model trained on a particular network configuration is used to predict a quantity of interest in a new, unknown setting. We focus on the tilt-dependent received signal strength maps as quantities of interest and we analyze two cases where the knowledge acquired for a particular antenna tilt setting is transferred to (i) a different tilt configuration of the same antenna or (ii) a different antenna with the same tilt configuration. Promising results supporting knowledge transfer are obtained through extensive experiments conducted using different machine learning models on a real dataset.

We focus on the problem of managing a shared physical wireless sensor network (WSN) where a single network infrastructure provider leases the physical resources of the networks to application providers to run/deploy specific applications/services. In this scenario, we solve jointly the problems of application admission control (AAC), that is, whether to admit the application/service to the physical network, and wireless sensor network slicing (SNS), that is, to allocate the required physical resources to the admitted applications in a transparent and effective way. We propose a mathematical programming framework to model the joint AAC-SNS problem which is then leveraged to design effective solution algorithms. The proposed framework is thoroughly evaluated on realistic WSNs infrastructures.

Sensor network virtualization enables the possibility of sharing common physical resources to multiple stakeholder applications. This paper focuses on addressing the dynamic adaptation of already assigned virtual sensor network resources to respond to time varying application demands. We propose an optimization framework that dynamically allocate applications into sensor nodes while accounting for the characteristics and limitations of the wireless sensor environment. It takes also into account the additional energy consumption related to activating new nodes and/or moving already active applications. Different objective functions related to the available energy in the nodes are analyzed. The proposed framework is evaluated by simulation considering realistic parameters from actual sensor nodes and deployed applications to assess the efficiency of the proposals.

Several computer vision tasks exploit a succinct representation of the visual content in the form of sets of local features. Given an input image, feature extraction algorithms identify keypoints and assign to each of them a descriptor, based on the characteristics of the surrounding visual content. Several tasks might require local features to be extracted from a video sequence, on a frame-by-frame basis. Although temporal downsampling has been proven to be an effective solution for mobile augmented reality and visual search, high temporal resolution is a key requirement for time-critical applications such as object tracking, event recognition, pedestrian detection, surveillance. In recent years, more and more computationally efficient visual feature detectors and descriptors have been proposed. Nonetheless, such approaches are tailored to still images. In this paper we propose a fast keypoint detection algorithm for video sequences, that exploits the temporal coherence of the sequence of keypoints. According to the proposed method, each frame is preprocessed so as to identify the parts of the input frame for which keypoint detection and description need to be performed. Our experiments show that it is possible to achieve a reduction in computational time of up to 40%, without significantly affecting the task accuracy.

Visual Sensor Networks (VSNs) have been recently used for implementing automatic visual analysis tasks where local image features, instead of images, are compressed and transmitted to a central controller. Such features may also be compressed in a multi-view fashion, exploiting the redundancy between overlapping views. In this paper we analyze the problem of multi-view coding and routing of features in VSNs. We empirically analyze the relationship between the bitrate reduction obtained with a practical multi-view local features encoder and several geometry-based, image-based and feature-based predictors. The purpose of this analysis is to identify the most accurate, yet compact predictor of the achievable compression efficiency when jointly encoding correlated streams of local features. Then, we propose a robust optimization framework that exploits the aforementioned predictors. The proposed mathematical problem maximizes the amount of data extracted from the VSN by properly routing the streams of features, subject to capacity, interference and energy constraints, explicitly considering the uncertainty in the compression efficiency estimation. Extensive experiments on simulated VSNs show that multi-view coding maximizes the amount of data extracted from camera nodes, while the robust optimization approach provides significant improvement in uncertain scenarios compared to the optimal solution of a deterministic approach.

In this work, we analyze the use of a WiFi network deployed in a large-scale technical university. To this extent, we leverage three weeks of WiFi traffic data logs and characterize the spatio-temporal correlation of the traffic at different granularities (each individual access point, groups of access points, entire network). The spatial correlation of traffic across nearby access points is also assessed. Then, we search for distinctive fingerprints left on the WiFi traffic by different situations/conditions; namely, we answer the following questions: Do students attending a lecture use the wireless network in a different way than students not attending a lecture?, and Is there any difference in the usage of the wireless network during architecture or engineering classes? A supervised learning approach based on Quadratic Discriminant Analysis (QDA) is used to classify empty vs. occupied rooms and engineering vs. architecture lectures using only WiFi traffic logs with promising results.

We propose a method for classifying Wi-Fi enabled mobile handheld devices (smartphones) and non-handheld devices (laptops) in a completely passive way, that is resorting neither to traffic probes on network edge devices nor to deep packet inspection techniques to read application layer information. Instead, classification is performed starting from probe requests Wi-Fi frames, which can be sniffed with inexpensive commercial hardware. We extract distinctive features from probe request frames (how many probe requests are transmitted by each device, how frequently, etc.) and take a machine learning approach, training four different classifiers to recognize the two types of devices. We compare the performance of the different classifiers and identify a solution based on a Random Decision Forest that correctly classify devices 95% of the times. The classification method is then used as a pre-processing stage to analyze network traffic traces from the wireless network of a university building, with interesting considerations on the way different types of devices uses the network (amount of data exchanged, duration of connections, etc.). The proposed methodology finds application in many scenarios related to Wi-Fi network management/optimization and Wi-Fi based services.

We present a complete, open-source framework for rapid experimentation of visual sensor network (VSN) solutions. From the software point of view, we base our architecture on open-source and widely known C++ libraries to provide the basic image processing and networking primitives. The resulting system can be leveraged to create different types of VSNs, characterized by the presence of multiple cameras, relays and cooperator nodes, and can be run on any Linux-based hardware platform, such as the BeagleBone Black. To demonstrate the flexibility of the proposed framework, we describe two different application scenarios typical of VSNs, namely object recognition and parking monitoring. The framework is then used to evaluate the benefits of two complementary paradigms for networked visual analysis recently discussed in the literature. In the traditional compress-then-analyze (CTA) paradigm, compressed images are transmitted from camera nodes to a central controller, where they are analyzed. In the novel analyze-then-compress (ATC) paradigm, camera nodes extract and compress local features from the acquired images. Such features are transmitted to the central controller and used to perform visual analysis. We show that the ATC paradigm outperforms CTA from the consumed energy point of view, at the same target analysis accuracy in both the application scenarios.

Visual sensor networks (VSNs) have attracted the interest of researchers worldwide in the last few years, and are expected to play a major role in the evolution of the Internet-of-Things (IoT). When used to perform visual analysis tasks, VSNs may be operated according to two different paradigms. In the traditional compress-then-analyze paradigm, images are acquired, compressed and transmitted for further analysis. Conversely, in the analyze-then-compress paradigm, image features are extracted by visual sensor nodes, encoded and then delivered to a remote destination where analysis is performed. The question this paper aims to answer is What is the best visual analysis paradigm in VSNs?To do this, first we empirically characterize the rate-energy-accuracy performance of the two aforementioned paradigms. Then, we leverage such models to formulate a resource allocation problem for VSNs. The problem optimally allocates the specific paradigm used by each camera node in the network and the related transmission source rate, with the objective of optimizing the accuracy of the visual analysis task and the VSN coverage. Experimental results over several VSNs instances demonstrate that there is no “winning” paradigm, but the best performance are obtained by allowing the coexistence of the two and by properly optimizing their utilization.

This work focuses on Visual Sensor Networks (VSNs) which perform visual analysis tasks such as object recognition. There, the goal is to find the image in a reference database which is the closest match to the image captured by camera sensor nodes. Recognition is performed by relying on visual features extracted from the acquired image, which are matched against a database of labeled features in order to find the closest image match. The matching functionalities are often implemented at a central controller outside the VSN. In contrast, we study the performance trade-offs involved in distributing the matching functionalities inside the VSN by letting sensor nodes performing parts of the matching process. We propose an optimization framework to optimally distribute the matching task to in-network sensor nodes with the goal of minimizing the overall completion time of the recognition task. The proposed optimization framework is then used to assess the performance of distributed matching, comparing it to a traditional, centralized approach in realistic VSN scenarios.

Distributed visual analysis applications, such as mobile visual search or Visual Sensor Networks (VSNs) require the transmission of visual content on a bandwidth-limited network, from a peripheral node to a processing unit. Traditionally, a “Compress-Then-Analyze” approach has been pursued, in which sensing nodes acquire and encode the pixel-level representation of the visual content, that is subsequently transmitted to a sink node in order to be processed. This approach might not represent the most effective solution, since several analysis applications leverage a compact representation of the content, thus resulting in an inefficient usage of network resources. Furthermore, coding artifacts might significantly impact the accuracy of the visual task at hand. To tackle such limitations, an orthogonal approach named “Analyze-Then-Compress” has been proposed [1]. According to such a paradigm, sensing nodes are responsible for the extraction of visual features, that are encoded and transmitted to a sink node for further processing. In spite of improved task efficiency, such paradigm implies the central processing node not being able to reconstruct a pixel-level representation of the visual content. In this paper we propose an effective compromise between the two paradigms, namely “Hybrid-Analyze-Then-Compress” (HATC) that aims at jointly encoding visual content and local image features. Furthermore, we show how a target tradeoff between image quality and task accuracy might be achieved by accurately allocating the bitrate to either visual content or local features.

This work studies how visual analysis tasks based on feature extraction can be speeded up in the context of Visual Sensor Networks. The main catch is for the camera node to leverage the presence of neighboring sensor nodes and offload the task, thus parallelizing its execution. We propose two mathematical programming formulations for the optimal visual task offloading problem: the first one targets the minimization of the overall task completion time while enforcing energy consumption constraints onto the nodes; the second maximizes the overall sensor network lifetime subject to a temporal constraint on the task completion time. The aforementioned formulations are used to characterize the achievable speed-up and consequent energy consumption in representative visual sensor network topologies.

Binary local features represent an effective alternative to real-valued descriptors, leading to comparable results for many visual analysis tasks while being characterized by significantly lower computational complexity and memory requirements. When dealing with large collections, a more compact representation based on global features is often preferred, which can be obtained from local features by means of, e.g., the bag-of-visual word model. Several applications, including, for example, visual sensor networks and mobile augmented reality, require visual features to be transmitted over a bandwidth-limited network, thus calling for coding techniques that aim at reducing the required bit budget while attaining a target level of efficiency. In this paper, we investigate a coding scheme tailored to both local and global binary features, which aims at exploiting both spatial and temporal redundancy by means of intra- and inter-frame coding. In this respect, the proposed coding scheme can conveniently be adopted to support the analyze-then-compress (ATC) paradigm. That is, visual features are extracted from the acquired content, encoded at remote nodes, and finally transmitted to a central controller that performs the visual analysis. This is in contrast with the traditional approach, in which visual content is acquired at a node, compressed and then sent to a central unit for further processing, according to the compress-then-analyze (CTA) paradigm. In this paper, we experimentally compare the ATC and the CTA by means of rate-efficiency curves in the context of two different visual analysis tasks: 1) homography estimation and 2) content-based retrieval. Our results show that the novel ATC paradigm based on the proposed coding primitives can be competitive with the CTA, especially in bandwidth limited scenarios.

Local features represent a powerful tool which is exploited in several applications such as visual search, object recognition and tracking, etc. In this context, binary descriptors provide an efficient alternative to real-valued descriptors, due to low computational complexity, limited memory footprint and fast matching algorithms. The descriptor consists of a binary vector, in which each bit is the result of a pairwise comparison between smoothed pixel intensities. In several cases, visual features need to be transmitted over a bandwidth-limited network. To this end, it is useful to compress the descriptor to reduce the required rate, while attaining a target accuracy for the task at hand. The past literature thoroughly addressed the problem of coding visual features extracted from still images and, only very recently, the problem of coding real-valued features (e.g., SIFT, SURF) extracted from video sequences. In this paper we propose a coding architecture specifically designed for binary local features extracted from video content. We exploit both spatial and temporal redundancy by means of intra-frame and inter-frame coding modes, showing that significant coding gains can be attained for a target level of accuracy of the visual analysis task.

Local visual features are commonly adopted to accomplish analysis tasks such as object recognition/tracking and image retrieval. Recently, several visual features extraction algorithms tailored to low-power architectures have been proposed, in order to enable image analysis on energy-constrained devices such as smart-phones or Visual Sensor Networks (VSN). In this work, we dissect and analyze BRISK, a state-of-the-art low-power visual feature extractor, in order to evaluate the impact of its individual building blocks on the overall energy consumption. For each building block, we propose a solution to limit the energy consumption without affecting the overall analysis performance. The resulting BRISKOLA (BRISK Optimized for Low-power ARM architectures) feature extractor exhibits energy savings up to 30% with respect to the original implementation.

This demo showcases some of the results obtained by the GreenEyes project, whose main objective is to enable visual analysis on resource-constrained multimedia sensor networks. The demo features a multi-hop visual sensor network operated by BeagleBones Linux computers with IEEE 802.15.4 communication capabilities, and capable of recognizing and tracking objects according to two different visual paradigms. In the traditional compress-then-analyze (CTA) paradigm, JPEG compressed images are transmitted through the network from a camera node to a central controller, where the analysis takes place. In the alternative analyze-then-compress (ATC) paradigm, the camera node extracts and compresses local binary visual features from the acquired images (either locally or in a distributed fashion) and transmits them to the central controller, where they are used to perform object recognition/tracking. We show that, in a bandwidth constrained scenario, the latter paradigm allows to reach better results in terms of application frame rates, still ensuring excellent analysis performance.

A robust hash, or content-based fingerprint, is a succinct representation of the perceptually most relevant parts of a multimedia object. A key requirement of fingerprinting is that elements with perceptually similar content should map to the same fingerprint, even if their bit-level representations are different. In this work we propose BAMBOO (Binary descriptor based on AsymMetric pairwise BOOsting), a binary local descriptor that exploits a combination of content-based fingerprinting techniques and computationally efficient filters (box filters, Haar-like features, etc.) applied to image patches. In particular, we define a possibly large set of filters and iteratively select the most discriminative ones resorting to an asymmetric pair-wise boosting technique. The output values of the filtering process are quantized to one bit, leading to a very compact binary descriptor. Results show that such descriptor leads to compelling results, significantly outperforming binary descriptors having comparable complexity (e.g., BRISK), and approaching the discriminative power of state-of-the-art descriptors which are significantly more complex (e.g., SIFT and BinBoost).

Wireless visual sensor networks (VSNs) are expected to play a major role in future IEEE 802.15.4 personal area networks (PANs) under recently established collision-free medium access control (MAC) protocols, such as the IEEE 802.15.4e-2012 MAC. In such environments, the VSN energy consumption is affected by a number of camera sensors deployed (spatial coverage), as well as a number of captured video frames of which each node processes and transmits data (temporal coverage). In this paper we explore this aspect for uniformly formed VSNs, that is, networks comprising identical wireless visual sensor nodes connected to a collection node via a balanced cluster-tree topology, with each node producing independent identically distributed bitstream sizes after processing the video frames captured within each network activation interval. We derive analytic results for the energy-optimal spatio-temporal coverage parameters of such VSNs under a priori known bounds for the number of frames to process per sensor and the number of nodes to deploy within each tier of the VSN. Our results are parametric to the probability density function characterizing the bitstream size produced by each node and the energy consumption rates of the system of interest. Experimental results are derived from a deployment of TelosB motes and reveal that our analytic results are always within 7% of the energy consumption measurements for a wide range of settings. In addition, results obtained via motion JPEG encoding and feature extraction on a multimedia subsystem (BeagleBone Linux Computer) show that the optimal spatio-temporal settings derived by our framework allow for substantial reduction of energy consumption in comparison with ad hoc settings.

Visual features are successfully exploited in several applications (e.g., visual search, object recognition and tracking, etc.) due to their ability to efficiently represent image content. Several visual analysis tasks require features to be transmitted over a bandwidth-limited network, thus calling for coding techniques to reduce the required bit budget, while attaining a target level of efficiency. In this paper, we propose, for the first time, a coding architecture designed for local features (e.g., SIFT, SURF) extracted from video sequences. To achieve high coding efficiency, we exploit both spatial and temporal redundancy by means of intraframe and interframe coding modes. In addition, we propose a coding mode decision based on rate-distortion optimization. The proposed coding scheme can be conveniently adopted to implement the analyze-then-compress (ATC) paradigm in the context of visual sensor networks. That is, sets of visual features are extracted from video frames, encoded at remote nodes, and finally transmitted to a central controller that performs visual analysis. This is in contrast to the traditional compress-then-analyze (CTA) paradigm, in which video sequences acquired at a node are compressed and then sent to a central unit for further processing. In this paper, we compare these coding paradigms using metrics that are routinely adopted to evaluate the suitability of visual features in the context of content-based retrieval, object recognition, and tracking. Experimental results demonstrate that, thanks to the significant coding gains achieved by the proposed coding scheme, ATC outperforms CTA with respect to all evaluation metrics.

Visual features provide a convenient representation of the image content, which is exploited in several applications, e.g., visual search, object tracking, etc. In several cases, visual features need to be transmitted over a bandwidth-limited network, thus calling for coding techniques to reduce the required rate, while attaining a target efficiency for the task at hand. Although the literature has recently addressed the problem of coding local features extracted from still images, in this paper we propose, for the first time, a coding architecture designed for local features extracted from video content. We exploit both spatial and temporal redundancy by means of intra-frame and inter-frame coding modes. In addition, we propose a coding mode decision based on rate-distortion optimization. Experimental results demonstrate that, in the case of SIFT descriptors, exploiting temporal redundancy leads to substantial gains in terms of coding efficiency.

Binary descriptors have recently emerged as low-complexity alternatives to state-of-the-art descriptors such as SIFT. The descriptor is represented by means of a binary string, in which each bit is the result of the pair-wise comparison of smoothed pixel values properly selected in a patch around each keypoint. Previous works have focused on the construction of the descriptor neglecting the opportunity of performing lossless compression. In this paper, we propose two contributions. First, design an entropy coding scheme that seeks the internal ordering of the descriptor that minimizes the number of bits necessary to represent it. Second, we compare different selection strategies that can be adopted to identify which pair-wise comparisons to use when building the descriptor. Unlike previous works, we evaluate the discriminative power of descriptors as a function of rate, in order to investigate the trade-offs in a bandwidth constrained scenario.

We compare two paradigms for image analysis in visual sensor networks (VSN). In the compress-then-analyze (CTA) paradigm, images acquired from camera nodes are compressed and sent to a central controller for further analysis. Conversely, in the analyze-then-compress (ATC) approach, camera nodes perform visual feature extraction and transmit a compressed version of these features to a central controller. We focus on state-of-the-art binary features which are particularly suitable for resource-constrained VSNs, and we show that the “winning” paradigm depends primarily on the network conditions. Indeed, while the ATC approach might be the only possible way to perform analysis at low available bitrates, the CTA approach reaches the best results when the available bandwidth enables the transmission of high-quality images.

This interactive demo presents and compares two different paradigms for image analysis in visual sensor networks (VSN), using a testbed based on battery-operated Beagle-Bone platforms with sight and wireless communication capabilities.

We consider the problem of allocating the resources in a wireless sensor network, which is designed to perform visual analysis (e.g. object recognition). We depart from the traditional compress-then-analyze paradigm, in which nodes sense, compress and transmit visual data to a sink node. Instead, we study the case in which nodes extract and lossy code local features from pixel-domain representations of the sensed visual scene. The formulation of the allocation problem entails maximizing the lifetime of the visual sensor network subject to a target accuracy of the analysis task, together with energy, bandwidth and routing constraints. To this end, we contribute with the definition of a rate-accuracy model, which plays the role of the traditional rate-distortion model commonly adopted in visual communication. The proposed model captures the impact of: i) the number of selected local features; ii) the number of bits used for quantizing local features; iii) the criterion used to select the subset of local features to be transmitted. We verify the correctness of the models on two widely adopted visual dataset and we demonstrate the network lifetime gain that can be achieved by an optimal allocation of the resources.

Efficient coding of local image descriptors is of paramount importance when they need to be transmitted to a remote destination on bandwidth constrained networks. This is a case that arises, e.g., in mobile visual search and visual wireless sensor networks. In this work we consider SURF, a popular descriptor suitable for low-complexity devices, and we provide a comparative study of lossy coding schemes operating at low bitrate (e.g., less than 128 bits / descriptor). Our investigation covers schemes that address both intra- and inter-descriptor redundancy, including methods that have not been tested before in this context, e.g., sparse coding, lifting-based coding on trees, and hybrid intra and inter-descriptor coding. The experimental evaluation is carried out on two publicly available datasets, in terms of both rate-distortion and rate-accuracy, for the specific task of object recognition. Our results show that a rate saving of 15-30% can be achieved by exploiting intra-descriptor redundancy. On the other side, addressing inter-descriptor redundancy does not lead to substantial gains when applied alone, whereas it leads to marginal gains (up to 3%) when used in hybrid schemes jointly with intra-descriptor coding.

Computational auditory scene analysis exploits signals acquired by means of microphone arrays. In some circumstances, more than one array is deployed in the same environment. In order to effectively fuse the information gathered by each array, the relative location and pose of the arrays needs to be obtained solving a problem of geometric inter-array calibration. We consider the case where the arrays do not share a synchronous clock, which impairs the use of time-difference of arrival measures across arrays. Conversely, each array produces an acoustic image, which describes the energy of acoustic signals received from different directions. We jointly consider acoustic images acquired by the different arrays and adapt computer vision techniques to solve the calibration problem, thus estimating the location and pose of microphone arrays sensing the same auditory scene. We evaluate the robustness of the calibration process in a simulated environment and we investigate the effect of the various system parameters, namely the number of probing signal locations, the resolution of the acoustic images, the non-ideal intra-array calibration.