ElectroSense

PSD can describe the temporal variation of wireless signal power at different frequency positions, making it suitable for constructing spectral time series prediction tasks. In this type of data, the frequency dimension can be viewed as multiple frequency sub bands or frequency nodes, while the time dimension represents continuous sampling moments. Due to the spectral correlation between adjacent or related frequency bands, as well as the obvious temporal dependence between different time points, this dataset is very suitable for verifying spectral prediction models combining graph neural networks and recurrent neural networks. In this experiment, the data collected by the spectrum sensor located in Madrid, Spain on the ElectroSense platform was selected as the research object.

The frequency range used in the experiment is 500 MHz to 800 MHz, which covers various wireless services such as broadcasting and mobile communication. The spectrum occupancy status changes frequently and can better reflect the dynamic spectrum characteristics in the actual electromagnetic environment. The frequency resolution of the data is 2 MHz, therefore a total of 151 frequency nodes are obtained in the frequency range of 500 MHz to 800 MHz; The time resolution is 15 minutes, and the monitoring time is from May 28, 2021 to June 28, 2021, with a total of 2880 time sampling points. electrosense.org