Projecte assignat
Títol: Deep Learning-Based LPWAN Technology Identification under Channel Effects and Hardware Impairments
Director/a: GARCÍA LOZANO, MARIO
Departament: TSC
Títol: Deep Learning-Based LPWAN Technology Identification under Channel Effects and Hardware Impairments
Data inici oferta: 03-09-2026 Data finalització oferta: 03-04-2027
Estudis d'assignació del projecte:
MU MASTEAM 2015
| Tipus: Individual | |
| Lloc de realització: EETAC | |
| Paraules clau: | |
| LPWAN, Deep learning, Technology identification, Channel impairments, Hardware impairments | |
| Descripció del contingut i pla d'activitats: | |
| The increasing coexistence of multiple Low-Power Wide-Area Network (LPWAN) technologies in sub-GHz bands creates a growing need for automatic mechanisms capable of identifying the radio access technology associated with a received signal. This capability is particularly relevant in spectrum monitoring, dynamic spectrum access, cognitive radio systems, and multi-protocol gateways. In this context, deep learning techniques operating directly on signal samples have shown significant potential for wireless technology identification without requiring technology-specific demodulation procedures.
In real-world scenarios, wireless signals are simultaneously affected by radio channel effects and by hardware impairments in the transmitter and receiver, such as carrier frequency offsets, I/Q imbalance, nonlinearities, gain variations, and phase distortions. The combination of these effects can significantly alter signal characteristics and consequently affect the robustness and generalization capability of automatic classifiers. The main objective of this Master's Thesis is to investigate and develop automatic LPWAN technology identification techniques capable of operating under realistic propagation conditions and in the presence of hardware impairments. To this end, the individual and combined impact of different signal degradations on classification performance will be analyzed. Different deep learning architectures will be explored, with convolutional approaches based on ResNet and Transformer-based architectures considered as primary candidate solutions. The study will address different signal-to-noise ratios, channel conditions, and hardware impairment configurations in order to determine which factors most significantly limit technology identification performance. In addition to classification accuracy, the work will consider aspects related to computational complexity and the feasibility of deploying the resulting models on resource-constrained processing platforms. The ultimate goal is to assess the trade-off between classification performance, robustness under realistic signal conditions, and computational requirements, contributing towards LPWAN technology identification solutions suitable for practical radio systems. |
|
| Orientació a l'estudiant: | |
| Horari d'atenció a estudiants per a l'assignació de projecte: |
|