Development And Implementation Of Vital Signs Monitoring System Based On Fmcw Radars

Vital signs monitoring is an important part of daily health monitoring.With low power,high distance resolution,low interception probability and simple structure,FM continuous wave(FMCW)radar has broad application prospects in the fields of vital sign detection,road traffic and motion recognition.However,most of the existing studies have neglected the influence of target state on vital sign monitoring,and there is a lack of options for signal separation algorithms for different states.The aim of this paper is to investigate and implement FMCW radar-based vital signs monitoring.Firstly,the identification of targets in three different states:standing,sitting and lying;secondly,vital signs monitoring is carried out based on the state identification results,and the influence of different states on vital signs monitoring is investigated and a suitable signal separation method is selected to extract respiratory and heartbeat signals;finally,apnoea detection is achieved by setting the respiratory signal energy threshold in the sleep scenario.This paper firstly proposes a target state feature extraction method based on signal pre-processing and establishes a state recognition dataset.The method extracts spatial location information from the target signal by multi-dimensional Fourier transform and uses time-accumulation averaging to reduce the interference caused by environmental and human micro-movements.A distance-angle map is then used as a representation of the current state signal features.Through multiple methods for experimental comparison,the final recognition of three different state targets,standing,sitting and lying,was achieved using the Resnet50 network model,with an F1 score of 0.97.Secondly,the paper compares the signal separation effects of two algorithms,empirical modal decomposition(EMD)and variational modal decomposition(VND),based on different target states,and proposes the complementary ensemble empirical modal decomposition(DN-CEEMD)algorithm based on the envelope distribution to determine the noise,in order to solve the problem of modal confusion and signal residual noise.The algorithm first adjusts the noise amplitude according to the envelope variation to improve the polar distribution and reduce the residual noise,and then constructs multiple sets of signals for cumulative averaging based on the selected noise values.The algorithm is then used to decompose the modal components containing respiration and heartbeat frequencies to achieve different states of vital signs monitoring.The experimental data show that the target lying state has a greater impact on vital signs monitoring,and the DN-CEEMD algorithm outperforms both EMD and VMD in terms of target signal separation in the standing or lying state,with an accuracy rate of 91.1%and 86.4%respectively.Finally,the paper correlates vital sign indicators with the target sleep state and performs sleep apnea detection based on the respiratory signal.Firstly,a long time correction threshold setting method was proposed to dynamically adjust the threshold using the mean value of energy over multiple time windows,and to determine whether apnoea occurred based on the duration of energy below the threshold.By counting the number of apnoeas occurring in each hour,an assessment of the target sleep state can be achieved.Experimental results show that the method is able to accurately determine the onset and termination points of apnoea with a range similarity of 87.5%.