Intelligent autonomous driving in off-road environments is an emerging field with great potential to impact areas such as agriculture, forestry, and rescue operations. Perception in these scenarios presents unique challenges due to the diversity of elements and weather conditions, along with the inherent ambiguity in class definitions. Consequently, off-road visual semantic segmentation datasets remain underdeveloped, roughly ten times smaller than their urban counterparts, hindering dependable performance assessment and potentially compromising the safety of autonomous systems. To address these challenges, we present a comprehensive cross-dataset evaluation of visual semantic segmentation models for autonomous off-road navigation. We propose a unified ontology that harmonizes class definitions across relevant datasets, enabling their combination for both training and testing. This approach ensures fair model comparisons and reliable assessment of generalization to unseen domains. We further benchmark models on the original datasets, analyze the impact of different ontology harmonization criteria and conversion strategies, and evaluate the trade-off between segmentation performance and computational cost. Results show that Transformer-based architectures achieve the most consistent segmentation performance across datasets. While often computationally demanding, some variants maintain real-time inference ( ≈ 12 ms) with top-tier accuracy. The unified ontology simplifies the segmentation task, yielding more reliable models and about 40% faster training convergence. Cross-dataset training further enhances generalization, improving mean IoU by up to +20% on RUGD and +13% on WildScenes compared to RELLIS-3D-only training. Overall, this study provides valuable insights for developing robust perception modules for off-road autonomous vehicles.

Crowdsensing networks for the sole purpose of performing spectrum measurements have resulted in prior initiatives that have failed primarily due to their costs for maintenance. In this paper, we take a different view and propose ORAN-Sense, a novel architecture of Internet of Things (IoT) spectrum crowd-sensing devices integrated into the Next Generation of cellular networks. We use this framework to extend the capabilities of 5G networks and localize a transmitter that does not collaborate in the process of positioning. While 5G signals can not be applied to this scenario as the transmitter does not participate in the localization process through dedicated pilot symbols and data, we show how to use Time Difference of Arrival-based positioning using low-cost spectrum sensors, minimizing hardware impairments of low-cost spectrum receivers, introducing methods to address errors caused by over-the-air signal propagation, and proposing a low-cost synchronization technique. We have deployed our localization network in two major cities in Europe. Our experimental results indicate that signal localization of non-collaborative transmitters is feasible even using low-cost radio receivers with median accuracies of tens of meters with just a few sensors spanning cities, which makes it suitable for its integration in the Next Generation of cellular networks.

This paper presents an exploration, study, and comparison of various alternatives to enhance the capabilities of an end-to-end control system for autonomous driving based on imitation learning by adding visual memory and kinematic input data to the deep learning architectures that govern the vehicle. The experimental comparison relies on fundamental error metrics (MAE, MSE) during the offline assessment, supplemented by several external complementary fine-grain metrics based on the behavior of the ego vehicle at several urban test scenarios in the CARLA reference simulator in the online evaluation. Our study focuses on a lane-following application using different urban scenario layouts and visual bird-eye-view input. The memory addition involves architectural modifications and different sensory input types. The kinematic data integration is managed with a modified input. The experiments encompass both typical driving scenarios and extreme never-seen conditions. Additionally, we conduct an ablation study examining various memory lengths and densities. We prove experimentally that incorporating visual memory capabilities and kinematic input data makes the driving system more robust and able to handle a wider range of challenging situations, including those not encountered during training, in terms of reduction of collisions and speed self-regulation, resulting in a 75% enhancement. All the work we present here, including model architectures, trained model weights, comparison tool, and the dataset, is open-source, facilitating replication and extension of our findings.

Spectrum crowdsensing systems do not provide labeled data near real-time yet. We propose a framework that addresses this challenge and relies solely on Power Spectrum Density (PSD) data collected by low-cost receivers. A major hurdle is to design a system that is computationally efficient for near real-time operation, yet using only the limited 2 MHz bandwidth of low-cost spectrum sensors. First, we present a method for unsupervised transmission detection that works with PSD data already collected by the backend of the crowdsensing platform, and that provides stable detection of transmission boundaries. Second, we introduce a data-driven deep learning solution to classify the wireless technology used by the transmitter, using transmission features in a compressed space extracted from single PSD measurements over at most 2 MHz band. We build an experimental platform, and evaluate our framework with real-world data collected from 47 different sensors deployed across Europe. We show that our framework yields an average classification accuracy close to 94.25% over the testing dataset, with a maximum latency of 3.4 seconds when integrated in the backend of a major crowdsensing network. Code and data have been released for reproducibility and further studies.

Understanding spectrum activity is challenging when attempted at scale. The wireless community has recently risen to this challenge in designing spectrum monitoring systems that utilize many low-cost spectrum sensors to gather large volumes of sampled data across space, time, and frequencies. These crowdsensing systems are limited by the uplink bandwidth available to backhaul the raw in-phase and quadrature (IQ) samples and power spectrum density (PSD) data needed to run various applications. This paper presents FlexSpec, a framework based on the Walsh-Hadamard transform to compress spectrum data collected from distributed and low-cost sensors for real-time applications. This transformation allows sensors to significantly save uplink bandwidth thanks to its inherent properties both when it is applied to IQ and PSD data. Additionally, by leveraging a feedback loop between the sensor and the edge device it connects to, FlexSpec carefully adapts the compression ratio over time to changes in the spectrum and different applications, jointly considering data size, application performance, and spectrum variations. We experimentally evaluate FlexSpec in several applications. Our results show that FlexSpec is particularly suitable for IoT transmissions and signals close to the noise floor. Compared with prior work, FlexSpec provides up to$7\times $more reduction of uplink data size for signal detection based on PSD data, and reduces up to$6\times $to$8\times $the number of undecodable messages for IQ sample decoding.

Given the availability of lightweight radio and processing technology, it becomes feasible to imagine spectrum sensing systems using weather balloons. Such balloons navigate the airspace up to 40 km, and can provide a bird's eye and clear view of terrestrial, as well as aerial spectrum use. In this paper, we present SkySense, which is an extension of the Electrosense sensing framework with mobile GPS-located sensors and local data logging. In addition, we present 6 different sensing campaigns, targeting multiple terrestrial or aerial technologies such as ADS-B, AIS or LTE. For instance, for ADS-B, we can clearly conclude that the number of airplanes that are detected is the same for each balloon altitude, but the message reception rate decreases strongly with altitude because of collisions. For each sensing campaign, the dataset is described, and some example spectrum analysis results are presented. In addition, we analyse and quantify important trends visible when sensing from the sky, such as temperature and hardware variations, increased ambient interference levels, as well as hardware limitations of the lightweight system. A key challenge is the automatic gain control and dynamic range of the system, as a radio navigating over 30km, sees a very wide range of possible signal levels. All data is publicly available through the Electrosense framework, to encourage the spectrum sensing community to further analyse the data or motivate further measurement campaigns using weather balloons.

GNSS/GPS is a positioning system widely used nowadays in our lives for real-time localization in Earth. This technology is highly vulnerable to spoofing/jamming attacks caused by malicious intruders. In the recent years, commodity and low-cost radio-frequency hardware have been used to interfere with the legitimate GPS signal. Existing spoofing detection solutions use costly receivers and computationally expensive algorithms which limit the large-scale deployment. In this work we propose a GNSS spoofing detection system that can run on spectrum sensors with Software-Defined Radio (SDR) capabilities and cost in the order of 20 euros. Our approach exploits the predictability of the Doppler characteristics of the received GPS signals to determine the presence of anomalies or malicious attackers. We propose an artificial recurrent neural network (RNN) based on Long short-term memory (LSTM) for anomaly detection. We use data received by low-cost SDR receivers that are processed locally by low-cost embedded machines such as Nvidia Jetson Nano to provide inference capabilities. We show that our solution predicts very accurately the Doppler shift of GNSS signals and can determine the presence of a spoofing transmitter.

In recent years we are experiencing an important growth of interest for sensing the electromagnetic spectrum and making its access more agile. Emerging initiatives use low-cost receivers in large deployments for sensing the radio spectrum or collecting air-traffic signals at large scale. One of the major drawbacks of low-cost spectrum receivers is their limited sampling rate, which does not allow to decode wideband signals. In order to circumvent the hardware limitations of single receivers, we envision a scenario where non-coherent receivers sample the signal collaboratively to cover a larger bandwidth than the one of the single receiver and then, enable the signal reconstruction and decoding in the backend. We present a methodology to enable the signal reconstruction in the backend by multiplexing in frequency a certain number of noncoherent receivers in order to cover a signal bandwidth that would not otherwise be possible using a single receiver. We propose a method that does not use the knowledge of the modulation scheme, and has been designed to be transparent to the subsequent decoding process. As such, it is equivalent to the reception of the signal by a high-end receiver. We demonstrate and evaluate our approach with two non-coherent receivers which collaboratively sample an aviation signal of almost twice the bandwidth of each receiver. The experimental results show that, using two non-coherent receivers, our method is able to reconstruct and decode correctly more than 80% of data.

Precise Time-of-Arrival (TOA) estimations of aircraft and drone signals are important for a wide set of applications including aircraft/drone tracking, air traffic data verification, or self-localization. Our focus in this work is on TOA estimation methods that can run on low-cost software-defined radio (SDR) receivers, as widely deployed in Mode S / ADS-B crowdsourced sensor networks such as the OpenSky Network. We evaluate experimentally classical TOA estimation methods which are based on a cross-correlation with a reconstructed message template and find that these methods are not optimal for such signals. We propose two alternative methods that provide superior results for real-world Mode S / ADS-B signals captured with low-cost SDR receivers. The best method achieves a standard deviation error of 1.5 ns.

Crowdsourced signal monitoring systems are gaining attention for capturing the wireless spectrum at large geographical scale. Yet, most of the current systems are still limited to simple power spectrum measurements reported by each sensor. Our objective is to enhance such systems with signal decoding capabilities performed in the backend while retaining the original vision of a low-cost and crowdsourced setup. We propose a distributed system architecture for collaborative radio signal monitoring and decoding that builds on $12 low-cost radio frequency (RF) frontends and embedded boards and that takes into consideration the limited network bandwidth from the sensors to the backend. We present a distributed time multiplexing mechanism to sample the spectrum in a coordinated fashion that exploits the similarity of the radio signal received by more than one RF frontend in the same radio coverage. We address the strict time synchronization required among sensors to reconstruct the signal from the samples they receive when in the same radio coverage. We study and implement techniques to identify and overcome errors in the timing information in the presence of noise sources and decode the data in the backend. We provide an evaluation based on simulations and on real signals transmitted by Long-Term Evolution (LTE) base stations. Our results show that we can reliably reconstruct and decode radio signals received by low-cost crowdsourced sensors.

Today's radio frequency (RF) spectrum measurements are mainly performed by governmental agencies which drive around using bulky and expensive specialized hardware. This approach does not scale well, providing us with only a poor situational awareness of the actual RF spectrum usage around us. We have developed a wideband spectrum monitoring sensor for remote operation that builds upon portable and low-cost commercial off-the-shelf (COTS) hardware components with a total cost per sensor device below $100. This results in a stunning cost reduction factor of 50 to 500 comparing to professional equipment. Our sensor platform adopts the software-defined radio paradigm and performs all signal processing steps on the CPU and GPU of a low-cost single-board computer. We address the challenges of large frequency errors and long scanning times due to the hardware constraints by proposing new correction and optimization methods, providing a satisfactory level of accuracy in indoor and outdoor environments. Our remote sensing platform is envisioned to be used at larger scale for various applications such as dynamic spectrum access in cognitive radios, detecting regions with elevated electro-smog, or for policy enforcement in the electromagnetic space.