This paper discusses Short-Time Fourier Transform processing as a suitable technique for performing room-level location classification.
The authors of this paper report on research that used Short-Time Fourier Transform (WTFT) processing to perform room-level location classification, through the use of mobile device identification by evaluating their distinct radio frequency (RF) fingerprints that are reflected by changes in the frequency of transmitted signals. The authors collected data in eight different locations on the same floor of their engineering building, which contains indoor hallways and rooms of different sizes. Three software-defined radios (SDRs) were placed in three different locations to receive signals simultaneously but separately. The inphase and quadrature (IQ) signals and channel state information (CSI) frames were concatenated together for training a neural network. A Multi-Layer Perception (MLP) network was used to train the concatenated signals as input and their corresponding locations as labels. The authors report two challenges: that their dataset did not contain the same number of samples per location, and that several locations had insufficient training data due to signal attenuation. In order to overcome those limitations and improve the classification accuracy, they implemented an imbalanced learning method in the dataset. The classification strategy involved binary classification like individual location vs. other; using that approach, the authors obtained an average accuracy of around 95%.