Building-Interior-Spectrum Availability Prediction Performance Using LSTM

Large deployments of sensors and short-range wireless devices are expected inside smart buildings (SB), and scenarios with disadvantages like congestion and spectrum saturation are predicted. Short-range indoor systems based on Dynamic Spectrum Access (DSA) and Cognitive Radio (CR) are proposed, and their operation is based on the search and management of spectral opportunities. This work indirectly predicts spectrum availability in buildings using a novel predictive mechanism for incumbent channels based on Long Short-Term Memory (LSTM). The results demonstrate that the proposed model outperforms linear transformation using analog modulation as a nonlinear transformation of input data. The spectral prediction has an accuracy of up to 0.8328, while the absolute error in the prediction of spectrum availability is less than 2.76%. These results allow promising applications for efficient spectrum utilization and resource allocation in wireless telecommunications, offering potential time and energy savings in spectrum management decisions for short-range cognitive systems.