This work proposes a framework to discover the topology of a non-collaborative packet-based wireless network using radio-frequency (RF) sensors. The methodology developed is blind, allowing topology sensing of a network whose key features (i.e., number of nodes, physical layer signals, and medium access control (MAC) and routing protocols) are unknown. Because of the wireless medium, over-the-air signals captured by the sensors are mixed; therefore, blind source separation (BSS) and measurement association are used to separate traffic patterns. Then, to infer the topology, we detect directed data flows among nodes by identifying causal relationships between the separated transmitted patterns. We propose causal inference methods such as Granger causality (GC), transfer entropy (TE), and conditional transfer entropy (CTE) that use the times series of traffic profiles, and a solution based on a neural network (NN) that exploits distilled time-based features. The framework is validated on an ad-hoc wireless network accounting for MAC protocol, packet collisions, nodes mobility, the spatial density of sensors, and channel impairments, such as path-loss, shadowing, and noise. Numerical results reveal that the proposed approach reaches a high probability of link detection and a moderate false alarm rate in mild shadowing regimes and low to moderate network nodes mobility.

We investigate the packet collision probability among uncoordinated devices, in the uplink of Direct-to-Satellite Internet of Things (DtS-IoT). Both the satellite spot shape and its motion along its orbital path are considered in the analysis. We analyze the probability of no uplink collision under two DtS-IoT settings: 1) unconfirmed ALOHA over a single channel, e.g., long-range wide-area network (LoRaWAN) class A with long-range chirp spread spectrum (LoRa-CSS) modulation and 2) unconfirmed ALOHA with frequency-hopping compliant with LoRaWAN class A with long-range frequency-hopping spread spectrum (LR-FHSS). A closed-form solution is derived for the former, while an upper bound is found for the latter. The analytical results are validated by comparison with the outcomes of extensive simulations, showing that the obtained closed-form expressions accurately predict the probability of no collision. Moreover, the upper bound for the frequency-hopping case is proven to be tighter when there are less than 35 hopping channels. Finally, we provide a concise performance comparison between LoRa-CSS and LR-FHSS in DtS-IoT, showing that LR-FHSS significantly increases the number of devices that can simultaneously transmit during a satellite pass, thereby enhancing uplink capacity.

In this letter, we propose a method for passive human activity classification exploiting ground vibrations observed by a biaxial geophone. The solution is grounded on the idea that some activities can be better analyzed by the horizontal channel (bicycle and car) and others by the vertical one (walk and run). Thus, two solutions are proposed: a) joint processing of the vertical and horizontal data by a single classifier, b) cascade processing by two classifiers, that analyze the two channels separately. Numerical results based on real data show that while a parametric method such as a support vector machine (SVM) performs well in both cases, a non-parametric method such as the k-nearest neighbors (k-NN) reaches a higher accuracy in cascade processing. Besides, the results are compared with those obtained using a monoaxial geophone only.

Grant-free random access protocols are among the enabling techniques for mMTC, where a large number of devices activate sporadically and transmit short packets, typically containing a preamble (or a pilot sequence), without any resource allocation from the BS. One of the critical tasks to be accomplished by the BS is thus the preamble-based detection of the transmitted packets. This letter proposes a DL-based solution for detecting preambles in an asynchronous grant-free random access uplink scenario, assuming multiple antennas at the BS. The DL-based approach outperforms the classical correlator-based approach.

CF-mMIMO networks leverage seamless cooperation among numerous access points to serve a large number of users over the same time/frequency resources. This paper addresses the challenges of pilot and data power control, as well as pilot assignment, in the uplink of a CF-mMIMO network, where the number of users significantly exceeds that of the available orthogonal pilots. We first derive the closed-form expression of the achievable uplink rate of a user. Subsequently, harnessing the universal function approximation capability of ANN, we introduce a novel multi-task DL-based approach for joint power control and pilot assignment, aiming to maximize the minimum user rate. Our proposed method entails the design and unsupervised training of a DNN, employing a custom loss function specifically tailored to perform joint power control and pilot assignment, while simultaneously limiting the total network power usage. Extensive simulations demonstrate that our method outperforms the existing power control and pilot assignment strategies in terms of achievable network throughput, minimum user rate, and per-user energy consumption. The model versatility and adaptability are assessed by simulating two different scenarios, namely a UMa and an industrial one.

The massive and autonomous structural health monitoring (SHM) of bridges is a problem that is of growing interest due to its importance and topicality. However, a considerable amount of data must be elaborated and managed in such an application. This paper proposes a set of machine learning (ML) tools to detect anomalies in a bridge from vibrational measurements using the minimum amount of data. The proposed framework starts from the fundamental frequencies extracted through operational modal analysis (OMA) and clustering, followed by a density-based time-domain tracking algorithm. The fundamental frequencies extracted are then fed to one-class classification (OCC) algorithms that perform anomaly detection. Then, to reduce the amount of data, we analyze the effect of the number of sensors, the number of bits per sample, the observation time, and the measurement noise on damage detection performance. As a case study, the Z-24 bridge is considered because of the extensive database of accelerometric measurements in both standard and damaged conditions. A comparison of OCC algorithms, such as principal component analysis (PCA), kernel principal component analysis (KPCA), Gaussian mixture model (GMM) and one-class classifier neural network (OCCNN) $$^2$$2 is performed, and their robustness to data shrinking is evaluated. In many cases, OCCNN $$^2$$2 increases the performance with respect to classical anomaly detection techniques in terms of accuracy.