Research On Supine Body Position Detection Method Based On Microphone Arra

Obstructive sleep apnea/hypopnea syndrome(OSAHS)is a systemic sleep apnea disease.Supine position is one of the important reasons for increasing the frequency of apnea or hypopnea events and aggravating the degree of OSAHS.Polysomnography(PSG),which is the most common clinical diagnostic technique,is very laborious,time-consuming and expensive.In addition,the limited resources render it impossible for hospitals to test all the patients in need of whole-night assessment.Besides,the existing alternative methods also have many shortcomings.Given that snoring is one of the common and earliest symptoms of OSAHS.Therefore,if the non-contact acoustic analysis of snoring can be used to detect the supine sleep position all night,it will be of positive significance for the diagnosis and study of OSAHS.According to infrared recording,we found that the head orientation is highly consistent with the sleep body position.Therefore,a detection scheme is proposed,which uses the microphone array technology to obtain whole-night head orientations,and then indirectly obtain whole-night sleep body positions.Next,two microphones on both sides of the bedhead are used to synchronously acquire the snoring signals of the patient all night.the channel ratio on both sides of the gravity frequency and the first formant are extracted to form a twodimensional set of feature samples,which can detect the sleep body positions well.At the same time,based on the spatial propagation directionality of the snoring signals,high-band energy ratio,absolute high-band spectrum variance ratio and two spatial cross-correlation deviations between the microphones on both sides of the bedhead are extracted to form four-dimensional set of feature samples,which can detect the sleep body positions better.Finally,the unsupervised clustering algorithm is used,and the snoring signals of two patients,with different degrees of OSAHS,are simulated.The performance analysis of the clustering results verifies that the scheme in this paper is feasible.The extracted four-dimensional set of spatio-temporal feature samples has better detection performance than the traditional set of time-frequency feature samples,and is more effective for the detection of mild patients.