Feature Enhancement And Parameter Estimation Of Vital Signs Signal Based On Doppler Radar

Vital sign monitoring technology based on Doppler radar has attracted extensive attention of researchers because of its advantages of all-weather,all-weather,low power consumption and simple system structure.The essence of vital signs monitoring based on Doppler radar is that the radar sensor senses the weak movement of the small chest wall in the range of millimeter or centimeter by transmitting electromagnetic wave to measure the respiratory rate and heart rate.However,in the process of practical application of radar,it is difficult to ensure that the human target is in an absolute static state,and the random body movement(RBM)in the measurement process is undoubtedly a strong interference relative to the weak movement of the chest wall,This brings great challenges to high-precision vital signs monitoring.To solve this problem,this paper carries out research from two aspects: feature enhancement of vital signs and robust estimation of vital signs parameters.The main work and innovations are as follows:1.A feature enhancement algorithm of vital signs signal based on outliers robust particle filter(ORPF)is proposed.Aiming at the singular value problem introduced by the RBM of human targets,the observation noise variance is weighted by introducing noise weight to make the observation equation fit the noise variance at different times,and the expectation of a posteriori probability is estimated by Monte Carlo to update the ground particles and corresponding particle weights at each time,so as to effectively suppress the influence of singular values in vital signs,The local signal-to-noise ratio(LSNR)of respiratory rate and heartbeat rate is improved.The experimental results show that compared with the traditional method,the LSNR of respiratory frequency and heartbeat frequency in the spectrum obtained by the proposed ORPF method is increased by 2.24 d B and 5.79 d B respectively.2.A feature enhancement algorithm of vital signs signal based on deep learning is proposed.Aiming at the problem of feature enhancement of vital signs in RBM environment,the time-frequency domain vital signs are learned through convolution neural network(CNN)model,and the time-frequency map of vital signs in RBM state and quasi static state is constructed.The training set and test set are used to train and evaluate the performance of the vital signs signal feature enhancement network.The simulation data results show that compared with the traditional methods,the LSNR of respiratory frequency and heartbeat frequency in the time-frequency map obtained by the deep learning based vital signs signal feature enhancement method proposed in this paper are increased by 8.23 d B and 7.55 d B respectively.3.A robust vital sign parameter estimation technique based on Baum-Welch hidden Markov model(BW-HMM)is proposed.The hidden Markov chain is constructed to model the time-varying characteristics of the parameters of vital signs,and the parameter learning process is added to the hidden Markov model(HMM).The time-frequency power spectrum of vital signs signal is taken as the observation sequence,and the parameters of the HMM are learned through Baum-Welch algorithm.The data is divided into blocks,the correlation is established between adjacent blocks,and the hidden variables of HMM are calculated by Viterbi algorithm.Finally,the robust estimation of parameters of vital signs is obtained.The experimental results show that the accuracy of respiratory rate and heart rate estimation of BW-HMM method proposed in this paper is 94.86 % and 91.33 % respectively.Compared with the traditional method,the accuracy of respiratory rate and heart rate parameter estimation is improved by about 4 % and 5 % respectively.