Framework Achieves 90% Accuracy Classifying SmartBAN Signals in Crowded ISM Band

Researchers are tackling the critical challenge of reliably identifying vital signals from wearable health monitors in increasingly congested wireless environments. Nicola Gallucci, Giacomo Aragnetti, and Matteo Malagrinò, from DEIB at Politecnico di Milano, alongside colleagues including Francesco Linsalata, Maurizio Magarini, and Lorenzo Mucchi from the Università di Firenze, present the first open-source framework for automatically recognising Smart Body Area Network (BAN) signals, a significant step towards dependable wearable healthcare. Their innovative approach combines simulated and real-world radio frequency (RF) data, utilising deep learning to achieve over 90% accuracy in classifying these low-power medical transmissions, even amidst strong interference in the 2.4GHz ISM band. This breakthrough promises to improve interference-aware coexistence strategies and ultimately enhance the reliability of wireless health monitoring systems for everyone.

4GHz ISM band, where identifying the low-power transmissions from medical sensors is complicated by strong interference and power imbalances with other wireless technologies. This work introduces the first framework capable of accurately classifying these signals, paving the way for improved reliability in wearable healthcare. The team achieved this breakthrough by combining a meticulously crafted synthetic dataset of simulated signals with real-world radio frequency (RF) acquisitions captured using Software-Defined Radios (SDRs), allowing for both controlled experimentation and realistic performance evaluation.

Researchers designed deep convolutional neural networks, specifically utilising ResNet encoders and U-Net decoders enhanced with attention mechanisms, to process and interpret the complex RF signals. These networks were rigorously trained and assessed under a variety of propagation conditions, demonstrating a consistent ability to discern SmartBAN signals from background noise and interference. Experiments show the proposed approach achieves over 90% accuracy on synthetic datasets, a remarkable feat given the challenging signal characteristics. Crucially, the framework also maintains consistent performance when analysing real over-the-air spectrograms, validating its effectiveness in practical scenarios.
The study unveils a hybrid evaluation methodology, integrating simulated data with real-world RF captures, which provides a comprehensive assessment of the framework’s capabilities in mixed-signal environments. By accurately identifying SmartBAN signals in dense spectral environments, this research supports the development of interference-aware coexistence strategies, enabling medical devices to operate reliably alongside other wireless technologies. This innovation directly addresses the vulnerability of low-power medical signals to interference, ensuring the dependable delivery of critical health data. The work opens exciting possibilities for more robust and trustworthy wearable healthcare systems, improving patient monitoring and care.

This breakthrough establishes a foundation for adaptive channel selection and coexistence control, guaranteeing reliable data exchange in wearable medical applications operating under severe power asymmetry and dense spectral overlap. The framework’s open-source nature encourages further development and collaboration within the research community, accelerating the advancement of wireless health technologies. Furthermore, the ability to accurately classify SmartBAN signals is not only vital for improving the performance of existing devices but also for enabling the development of new and innovative healthcare applications that rely on seamless wireless connectivity0.4GHz ISM band. The team engineered a hybrid dataset combining synthetically generated signals with real-world radio frequency (RF) acquisitions captured using Software-Defined Radios (SDRs), allowing for both controlled experimentation and realistic evaluation of performance. This approach enables rigorous testing under diverse propagation conditions, mirroring complex over-the-air scenarios.

Researchers meticulously generated synthetic spectrograms to represent SmartBAN signals alongside interfering technologies like WLAN, ZigBee, and Bluetooth, precisely modelling the 2.4, 2.48GHz ISM band. These simulations incorporated realistic channel conditions and power levels, accurately reflecting the substantial power asymmetry between medical sensors and coexisting devices. Simultaneously, the study harnessed SDRs to capture actual over-the-air RF signals, providing a crucial layer of realism absent in purely simulated environments. The SDR system facilitated the generation and capture of signals under genuine channel impairments and interference, ensuring the framework’s adaptability to real-world complexities.

Experiments employed deep convolutional neural networks, specifically ResNet encoders paired with U-Net decoders, to analyse the generated spectrogram images. The architecture was deliberately chosen to effectively process the subtle spectral features of low-power SmartBAN waveforms, which are often obscured by stronger signals. Training and assessment were conducted across a range of propagation conditions, rigorously evaluating the network’s ability to maintain accuracy in dynamic and heterogeneous environments. The proposed approach consistently achieved over 90% accuracy on synthetic datasets and demonstrated robust performance on real over-the-air spectrograms, validating the effectiveness of the combined synthetic and real-world evaluation methodology.

This methodological innovation, integrating synthetic data with real RF captures, provides a comprehensive assessment platform for interference-aware coexistence strategies. By accurately classifying SmartBAN signals, the framework supports adaptive channel selection and control, ultimately improving the dependability of wearable healthcare systems operating in dense spectral environments. The study pioneers a robust solution for detecting extremely low-power medical waveforms, a significant advancement over previous work focused on stronger wireless standards and favourable signal conditions.