Research of Beam Forming Method for Radar-based Non-invasive Pulse Wave Monitoring

We have enhanced the application of mm-wave radar for monitoring critically septic shock patients by developing an optimized beamforming algorithm. This includes fractional-order Fourier transform (FrFT)-based beamforming (FrFT based BF), combined with a self-encoder convolutional neural network for deep learning. We define the minimized mean square error (MSE) between the desired signal and the beamforming output signal as the optimization objective of beamforming. Based on the pulse wave data measured by mm-wave radar, we determined the optimal fractional order and compared it with the time-domain beamforming method. The results demonstrate that the FrFT-based beamforming method has better anti-interference ability and stability, performing better in terms of mean square error, with important features such as P-wave, QRS-wave and T-wave in pulse wave signals being observed more clearly. Additionally, the paper explores the application of autoencoders in denoising signal processing. By encoding and decoding noisy signals using an autoencoder structure, it effectively restores the original noise-free signal. Using datasets obtained from FrFT-based beamforming and time-domain beamforming as inputs to the CNN network, the former has a smaller loss function, demonstrating its potential in the field of medical signal processing.