RF Intelligence for Health: Classification of SmartBAN Signals in overcrowded ISM band

Nicola Gallucci 1 1 , Giacomo Aragnetti 1 1 , Matteo Malagrinò 1 1 , Francesco Linsalata 1 2 , Maurizio Magarini 1 2 , Lorenzo Mucchi 2 · 2026

Read RF Intelligence for Health: Classification of SmartBAN Signals in overcrowded ISM band by Nicola Gallucci 1 1 , Giacomo Aragnetti 1 1 , Matteo Malagrinò 1 1 , Francesco Linsalata 1 2 , Maurizio Magarini 1 2 , Lorenzo Mucchi 2 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-15836.txt.

Introduction

Nicola Gallucci and a large team classify health-related radio signals in a 2026 arXiv paper. Wearable health monitors rely on radio links to report data. Reliable systems need awareness of interference around them. The 2.4 GHz ISM band is overcrowded with other devices. In it, low-power medical transmissions are hard to spot. Strong co-channel interference and power asymmetry hide them.

The paper works on SmartBAN signals, a standard for body-area networks. Classifying them under heavy interference is the challenge. A classifier that spots medical traffic can help systems avoid collisions. It can also warn when conditions threaten the link. For patients who depend on wearables, that awareness is a safety feature, not a luxury.

Catalog record

CREATORNicola Gallucci 1 1 , Giacomo Aragnetti 1 1 , Matteo Malagrinò 1 1 , Francesco Linsalata 1 2 , Maurizio Magarini 1 2 , Lorenzo Mucchi 2
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-15836

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

RF Intelligence for Health: Classification of SmartBAN Signals in overcrowded ISM band Nicola Gallucci 1 1 , Giacomo Aragnetti 1 1 , Matteo Malagrinò 1 1 , Francesco Linsalata 1 2 , Maurizio Magarini 1 2 , Lorenzo Mucchi 2 Abstract Accurate classification of Radio-Frequency (RF) signals is essential for reliable wearable health-monitoring systems, providing awareness of the interference conditions in which medical protocols operate. In the overcrowded 2.4 GHz ISM band, however, identifying low-power transmissions from medical sensors is challenging due to strong co-channel interference and substantial power asymmetry with coexisting technologies. This work introduces the first open source framework for automatic recognition of SmartBAN signals in Body Area Networks (BANs). The framework combines a synthetic dataset of simulated signals with real RF acquisitions obtained through Software-Defined Radios (SDRs), enabling both controlled and realistic evaluation. Deep convolutional neural networks based on ResNet encoders and U-Net decoders with attention mechanisms are trained and assessed across diverse propagation conditions. The proposed approach achieves over 90% accuracy on synthetic datasets and demonstrates consistent performance on real over-the-air spectrograms. By enabling reliable SmartBAN signal recognition in dense spectral environments, this framework supports interference-aware coexistence strategies and improves the dependability of wearable healthcare systems. Keywords — RF Signals Classification, SDR, Images Segmentation, ISM band, Deep Learning I Introduction Body Area Networks (BANs) are becoming essential in modern healthcare, enabling continuous and non-invasive monitoring of vital signs through wearable medical devices [ 1 ] . These technologies are increasingly integrated into daily life, with smartwatches and fitness trackers as familiar examples [ 2 , 3 ] . Fig. 1 : An example of a crowded RF signals environment with different co-existing technologies at ISM band. To ensure interoperability and efficient spectrum usage, the ETSI Smart Body Area Network (SmartBAN) technical committee was established in 2013 to define standards for ultra-low-power physical (PHY) and medium access control (MAC) layers, as well as lightweight data formats [ 4 , 3 ] . The latest specifications include an ultra-low-power PHY (TS 103 326 [ 5 ] ), a low-complexity MAC (TS 103 325 [ 6 ] ), and interoperable data representation and management formats. SmartBAN typically adopts a star network topology, where a central coordinator manages control and network operations while acting as a gateway to external systems. Intelligence is further introduced through semantic data models and coexistence management strategies, such as cognitive channel sensing and dynamic frequency selection [ 4 ] . Despite these advances, SmartBAN transmissions operate in the crowded 2.4–2.48 GHz ISM band, where they often coexist with higher-power technologies such as WLAN, ZigBee, and Bluetooth, as depicted in Fig. 1 . The extremely low power of SmartBAN medical signals makes them particularly vulnerable to interference, threatening the reliable delivery of critical health data [ 7 ] . AI-based signal detection methods have garnered significant attention for wireless classification in dense ISM environments. Previous studies have applied machine learning techniques such as support vector machines (SVMs) and convolutional neural networks (CNNs) to recognize Wi-Fi, ZigBee, and Bluetooth transmissions, even in overlapping conditions. For instance, Rashidpour and Bahramgiri employed SVM and K-Nearest Neighbors algorithms to classify Wi-Fi and Bluetooth protocols, achieving classification accuracies of 97.83% and 98.12%, respectively [ 8 ] . Similarly, He et al. utilized deep learning techniques to identify ISM band signals, demonstrating the efficacy of deep learning in this domain [ 9 ] . However, these approaches primarily target relatively strong signals with distinct spectral features. In contrast, SmartBAN medical waveforms present a more challenging scenario due to their extremely low power density and subtle spectral footprint. Recent advancements have sought to address these challenges. For example, Zhang et al. [ 10 ] developed deep neural networks to detect coexisting signal types based on I/Q samples, achieving competitive accuracies even in the presence of multiple signal types. Additionally, Xu et al. proposed a composite time-frequency analysis method combined with a Siamese neural network for interference classification in frequency hopping systems, enhancing classification accuracy in complex environments [ 11 ] . Furthermore, Software-defined radios (SDRs) enable realistic, over-the-air evaluation of classification networks by generating and capturing signals under real channel conditions, hardware impairments, and coexisting interference. A notable SDR-validated study by Zhang et al. [ 10 ] transmits coexisting Wi-Fi, LTE, and 5G New Radio compliant waveforms with USRPs and classifies raw I/Q and signal features using a convolutional neural network architecture, reporting high accuracy in over-the-air tests. However, these evaluations focus on relatively strong wireless standards and favorable signal conditions; in contrast, our work targets extremely low-power SmartBAN medical waveforms in realistic ISM coexistence scenarios. These studies underscore the potential of AI-based methods in classifying weak and overlapping signals in ISM bands and the notable use of SDRs for real testing. However, the classification of SmartBAN signals remains an open challenge due to their low power density and subtle spectral characteristics. Further research is needed to develop robust AI-based frameworks capable of accurately detecting and classifying such weak signals in dense ISM environments. In this context, distinguishing SmartBAN transmissions from stronger coexisting signals is a key enabler for interference-aware spectrum access. Accurate classification provides the foundation for adaptive channel selection and coexistence control, ensuring reliable data exchange in wearable medical applications operating under severe power asymmetry and dense spectral overlap. Paper Contributions To address this issue, we propose Smart RF Signal Recognition , an open source [ 12 ] deep learning framework for the detection and classification of SmartBAN medical waveforms. The main contributions of this work are as follows: • Provide an analysis of the spectral behaviour of SmartBAN transmissions in the 2.4–2.48 GHz ISM band, emphasizing the difficulties introduced by strong co-channel interference from WLAN, ZigBee, and Bluetooth. • Design multiple neural architectures that combine ResNet encoders with U-Net decoders enhanced by attention mechanisms, optimized for dynamic and heterogeneous channel conditions. • Develop a hybrid evaluation methodology that integrates synthetic spectrogram simulations with real-world RF captures, enabling comprehensive assessment in mixed-signal environments. • Demonstrate the classification performance and robustness of the proposed framework, even under adverse channel conditions. Paper Organization The rest of this paper is structured as follows: Section II describes the generation of the dataset used for training and evaluating the proposed deep learning framework. Building on the system model introduced in the previous section, Section III presents the design of a deep learning framework for the semantic segmentation of spectrogram images. Section IV reports the numerical and experimental results obtained using both synthetic and realistic spectrograms. Finally, Section V concludes the paper. II Signal and Interference Model In this section, we describe the generation of the dataset used for training and evaluating the proposed deep learning framework. The dataset is designed to capture the coexistence of multiple wireless technologies within the congested 2.4–2.48 GHz ISM band. II-A Time-Frequency Signal and Channel Model Let s_{i}(t) denote the baseband waveform of the i -th wireless technology, with i\in\{\text{Wi-Fi, Bluetooth, ZigBee, SmartBAN}\} . Each signal propagates through a multipath fading channel h_{i}(t,\tau) and is affected by additive white Gaussian noise n(t) . The received signal is expressed as r(t)=\sum_{i=1}^{N}\int s_{i}(t-\tau)h_{i}(t,\tau)d\tau+n(t), (1) where N is the number of simultaneously active signals, and h_{i}(t,\tau) represents the time-varying channel impulse response. The fading

Read the complete text