Device-Free Human Activity Recognition Based on Dual-Channel Transformer Using WiFi Signals

Algorithm 1

The pseudocode of preprocessing module.

Input: The raw CSI data .

   The element in represents the CSI data of subcarrier that received by spatial stream.

Output: Temporal features , frequency features .

 1: Extracting the moving average of CSI amplitude. , where is the length of sliding window.

 2: Utilizing a six-order Butterworth low-pass filter along the second dimension of to remove high-frequency noise.

 3: Calculating the activity indicator for based on Equation (5) and finding the peak index of activity indicator .

 4: Extracting the CSI segment from , in which and represents the length of CSI segment.

 5: Taking as the output of temporal CSI features .

 6: Doing PCA for .

 7: Calculating the Doppler spectrum of based on STFT, in which and represents the number of FFT points in STFT.

 8: Discarding the frequency bins above 30 Hz in . Then obtaining the output of frequency features , where is the number of frequency bins below 30 Hz.

 9: return and