The Effect of Window Size and Window Shape in STFT for Pre-Processing FMCW Radar Data in Human Activity Recognition Based on Bi-LSTM

Abstract

Many studies use radars for Human Activity Recognition (HAR), and numerous techniques for preprocessing FMCW radar data have been explored to improve HAR performances. Our approach employs 1-D radar to classify four human activities, i.e., walking, standing, crouching, and sitting. We use Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT) with Kaiser window to generate range-time and Doppler-time data from inphase and quadrature radar signal. The choice of windowing parameters, i.e., window size and window shape represented by the beta parameter in Kaiser window, is considered to have significant impacts on the performances of deep learning LSTM models, including the F1-score. However, our study in this paper, including statistical analysis using t-tests, shows otherwise. Our results consistently support the null hypothesis, which mean that variations in window size and window shape do not significantly affect the F1-score. In essence, our findings underscore the robustness of our preprocessing methodology, emphasizing the stability and reliability of the selected configurations. This research provides valuable insights into the preprocessing techniques for radar data in the context of human activity recognition, enhancing the consistency and credibility of deep learning models in this domain.

Keywords

• radar signal processing
• deep learning