Self-Identification Respiratory Disorder Based on Continuous Wave Radar Sensor System

Abstract

Contactless vital signs detection, based on the Doppler radar sensor system, has opened a great opportunity in biomedical applications. The radar sensor system can be used to provide the respiratory information of people without disturbing their comfort. This sensor system promises high accuracy in measuring breathing disorders as it escapes the touching sensors which might cause discomfort to the user and negatively affect their sleeping habits. Moreover, this sensor system does not require any special environment or depend on temperature and light conditions. In this paper, we propose a model to the end users; this model is to be built based on neural networks. Our proposed system can diagnose whether a person has a low, normal, or high breathing rate. This model can also be extended to more specific categories to help doctors to determine breathing disorders in patients. In this paper, a continuous wave radar sensor system, based on a vector network analyzer (VNA), is used to measure the breathing rate remotely. The measured signal from this radar sensor system is then processed for further purposes. Different extracted feature methods are implemented to obtain the breathing rate from the non-contact radar sensor system. A model based on the machine learning technique is investigated to classify the breathing disorder. A total of 31 people who were asked to perform low/normal/high breathing were measured by the $CW$ radar sensor. The measured data were also used to build a machine learning based model. The breathing rate measured by the $CW$ radar sensor system is compared with the reference measurement by the five-point touching Shimmer sensor. The results of the breathing rate are compatible. Two main time-frequency ($TF$ ) extraction feature methods, short-time Fourier transform (STFT) and continuous wavelet transform (CWT), were implemented in the proposed system. Under these extraction techniques, some classification approaches were employed and have shown high accuracy in categorizing the respiratory types. The research shows the possibility of building an artificial intelligence (AI) module for a non-contact radar sensor system to inform the end user of their breathing situation. This research enables a smarter and more friendly remote-detecting vital signs sensor system.

Keywords

• Machine learning
• vital signs detection
• neuron network
• classification problem