Journal of Sensors

Volume 2015 (2015), Article ID 548136, 13 pages

http://dx.doi.org/10.1155/2015/548136

## Noncontact Detection and Analysis of Respiratory Function Using Microwave Doppler Radar

^{1}School of Engineering, Faculty of Science, Engineering and Built Environment, Deakin University, Geelong, VIC 3216, Australia^{2}Department of Electrical and Electronic Engineering, Melbourne University, Parkville, VIC 3010, Australia^{3}University Hospital Geelong, Geelong, VIC 3220, Australia

Received 28 October 2014; Accepted 9 January 2015

Academic Editor: Yu Chen Tsai

Copyright © 2015 Yee Siong Lee et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Real-time respiratory measurement with Doppler Radar has an important advantage in the monitoring of certain conditions such as sleep apnoea, sudden infant death syndrome (SIDS), and many other general clinical uses requiring fast nonwearable and non-contact measurement of the respiratory function. In this paper, we demonstrate the feasibility of using Doppler Radar in measuring the basic respiratory frequencies (via fast Fourier transform) for four different types of breathing scenarios: normal breathing, rapid breathing, slow inhalation-fast exhalation, and fast inhalation-slow exhalation conducted in a laboratory environment. A high correlation factor was achieved between the Doppler Radar-based measurements and the conventional measurement device, a respiration strap. We also extended this work from basic signal acquisition to extracting detailed features of breathing function (*I* : *E* ratio). This facilitated additional insights into breathing activity and is likely to trigger a number of new applications in respiratory medicine.

#### 1. Introduction

Respiration monitoring is essential in the diagnosis and treatment of conditions such as chronic obstructive pulmonary disease, heart disease, and a number of sleep related conditions [1]. Furthermore, dysfunctional respiratory patterns such as rapid or shallow breathing [2] or high frequency breathing rates have also been associated with certain psychosomatic conditions [3] all of which, at present, are typically measured via respiration rates alone. However, a more detailed analysis of breathing patterns [4–9] will provide physicians with new insights into diagnostic medicine particularly if this can be performed noninvasively. Noncontact Doppler Radar has already been considered in a variety of patient monitoring and measurement scenarios in healthcare including heartbeat and respiration monitoring in place of conventional methods such as the chest strap, photoplethysmograph [10], and ECG [11]. Research reported using Doppler Radar in measuring human physiological activity [12–18] has predominantly demonstrated the feasibility of Doppler Radar in obtaining breathing frequency or heart rate using FFT, wavelet analysis, or time-frequency analysis [14, 19, 20].

A complete respiration cycle is typically defined by inhalation (inspiration) and exhalation (expiration) states accompanied by a pause as described in [21]. Breathing rates are predominantly calculated independent of the inhalation to exhalation ratio (*I* :* E*) for each breathing cycle. For normal and spontaneous breathing, there is an abundance of time for the exhalation process from the inspired tidal volume, but in certain pathological states, for instance, asthma and COPD (chronic obstructive pulmonary disease), reduced expiratory flow would need longer time to empty the inspired lung volume [22]. Typically, for adults, a normal* I* :* E* ratio is in the range of 1 : 2 but this varies between individuals depending on the health and the physiological state of the individual [23]. Consequently more information of each component is extremely important as it can be useful in early detection of several respiratory disorders.

Another important parameter associated with breathing is respiratory tidal volume [24, 25] which can also be derived from microwave radar due to the relationship between the chest wall displacement and the tidal volume. Different types of breathing can potentially be deduced from such chest wall or abdomen displacement information during inhalation and exhalation. This information can be used to identify different types of breathing signatures such as shallow breathing, deep breathing, slow breathing, fast breathing, and other types of breathing patterns. Indeed, the displacement of the chest wall or abdomen in shallow breathing is expected to be small and the complete breathing cycle would occur in a shorter time period compared to normal breathing.

Doppler Radar operates by transmitting a radio wave signal and receiving the modulated version of the signal due to the motion triggered by the target [24, 26]. The reflected wave is in the modulated form where it undergoes a frequency shift proportional to the radial velocity that can be described using the Doppler effect. When a target has a quasi-periodic motion, the time varying position of the target can be represented as a phase modulated signal and the phase shift is directly proportional to the object’s movement. Thus, the movement of the chest wall/abdomen for respiration due to the inhalation, exhalation, and the pause states can be detected and modelled using the reflected Doppler shifted signal, the main focus of this paper. We provide a comprehensive description of the noncontact respiratory measurement via Doppler Radar which was then validated with independent measurements using a respiration belt and breathing cycle counts. We also demonstrate the different types of inhaling and exhaling states from data collected using our Doppler Radar system. The purpose of this paper is summarized as follows:(i)investigation of Doppler Radar’s feasibility in capturing different types of breathing patterns under various breathing scenarios;(ii)correction of signal imbalance and cross-validation of Doppler breathing signal with standard respiration measurement, the respiration belt (MLT1132 iezo-respiratory belt transducer);(iii)decomposition of the breathing signal (from Doppler Radar) into its respective inhalation and exhalation components, representing each component model using 4th polynomial fitting (see Table 2(a)) and classifying decomposed breathing components into its respective breathing scenarios.

#### 2. Methods

##### 2.1. Respiration Monitoring via Microwave Doppler Radar

The Doppler effect occurs when there is a shift in the frequency of the wave either reflected or radiated, received by an object in motion [27]. Consider a transmitted sine wave signal with an angular frequency , where is the transmitted signal, is the time, and is the arbitrary phase shift. Assume that the target is stationary at a distance of from the radar and the transmission time from radar to target is where is the wave propagation velocity. The target range at time is given by equation below , where is the range of the target from the radar and (velocity) is the rate of change of and is the time at . The received signal at the stationary target is the same as the transmitted signal at the time which can be given as

The received signal from the target at time would have been sent seconds prior to time . This can be represented as . Referring to (1), signal can be depicted in the same formulation given as Substituting into (3), the received signal is further represented as For a target moving (radially) with respect to the radar, the distance will vary and by using and , the received signal can be further derived as where the frequency of the reflected signal is shifted by and the phase angle by . Therefore, the Doppler shift can also be denoted by , where is the Doppler shift in Hertz and is the transmitted frequency. Using , can be written as where the negative sign accounts for the fact that if is negative (when the target is approaching), the Doppler frequency will be positive or vice versa [27]. From (5), the phase angle of the received signal is given as . Therefore, the transmitted wave from the radar to the target will be reflected to the receiver with some phase shifting and can be represented as phase modulation given as

The measurement model for human respiration using Doppler Radar can be derived as follows. Generally, the Doppler shift in frequency is given by where is the velocity of the target, is the wavelength of the transmitted signal, and is velocity of the propagating wave. Assuming the target to be stationary or undergoing a periodic movement of with no net velocity, the Doppler frequency shift can be represented in the form of nonlinear phase modulation as the phase signal given by where is the displacement of the chest wall or abdomen. Using a continuous wave (CW) radar, the transmitted signal is represented by where is the transmitted signal and is the arbitrary phase shift or the phase noise of the signal source if the transmitted wave is reflected by the target/subject at a nominal distance with a time varying displacement of which is caused by the movement of the torso (abdomen). Thus, the distance [28] between the transmitter and the target is given as . The measurement of the time delay between the transmitter and the target is denoted as the distance travelled over the signal’s propagation velocity given as . Thus, due to the movement of the abdomen during the process of respiration, the distance between the antenna and the abdomen at the time of reflection is denoted by and the round trip time can be further derived as .

Using the similar formulation shown in (3) along with and , the received signal can be represented as and further approximated as Demodulation of the phase is used to determine the motion signature which can be detected at the receiver. In the direct conversion system, the received signal will be mixed with local oscillator to obtain the baseband output given as In a quadrature receiver system, the received signal will be split into two forms which are an in-phase () and a quadrature phase () signal where the phase difference will be . Therefore, general two orthogonal baseband outputs of the quadrature receiver system can be denoted by Here, is the constant phase shift dependent on the nominal distance to the target and is the residual phase noise. The benefit of using a quadrature receiver is to overcome the null problem [11] where at least one output (either ) is not null when the other is null.

##### 2.2. Signal Processing, Decomposition, and Identification

A complete breathing cycle is comprised of inhalation (), exhalation (), and pause components where the ratio of* I* :* E* can certainly be asymmetric [23]. Therefore, computation of breathing rates purely based on simple single frequency signatures computed via fast Fourier transforms (FFT) is not sufficient to provide detailed breathing pattern features, particularly for the identification and analysis of respiratory conditions. Firstly, the basic received signal is sent to the (in-phase and quadrature phase) demodulator for direct conversion into its baseband differential signal and then sampled at 1000 Hz using NI-DAQ (National Instrument Data Acquisition System). The differential signals were then converted to a single ended baseband signal, removing any DC components of the raw signals, and then processed in two different approaches. In the first approach, the preprocessed raw data was modelled using a piecewise linear least squares approach [29]. In the second approach, the raw data was processed using a SG (Savitzky-Golay polynomial least square) [30] smoothing filter and further analysed using Fourier filtering [31]. The first approach offers a simple method applicable for real-time processing while the second approach offers more accurate identification of the respiration cycle components and their properties, the main focus in this paper.

##### 2.3. Correction of Amplitude and Phase Imbalance

Two orthogonal outputs ( and ) are obtained from a quadrature receiver system but in practice (due to the imperfection of components in the hardware design), it suffers from amplitude and phase imbalance which affects the accuracy of the recovered data at the output [32]. Consequently, phase and amplitude corrections are necessary to increase accuracy. There are a number of approaches to correct the amplitude and phase imbalance [33, 34]. In [34], a final form of two orthonormal vectors using a method similar to the Gram Schmidt orthogonalization (GSO) [32] has been proposed as shown in (17). The derivation of this is as follows. The ideally received signal is defined by where and are the in-phase and quadrature phase of the information signal respectively. In our approach, with the presence of amplitude imbalance and phase offset, the received signal at the mixer can be represented as where and are the amplitude and phase imbalance. Demodulation of received signal is as follows: Expanding the derivation After the low pass filtering and ignoring the term , representation of orthogonal and in matrix form Using (17), correction on amplitude and phase imbalance can be performed. Simulation results of using this approach will be discussed in Section 3.

##### 2.4. The Piecewise Linear Fitting Method

This method fits nonlinear, typically noisy waveforms by choosing an optimal segmentation of the waveform and then fitting each segment with a linear function [29]. Here the segmentation process is critical and, in this case, appropriate lengths of nonoverlapping segments were used. Also, we used fixed nonoverlapping segments of 200 ms to accommodate the Doppler Radar signal.

##### 2.5. The Savitzky-Golay Method and Fourier Filtering

The Savitzky-Golay filter is a least square polynomial filter [30]. By applying the filter to the noisy data obtained from the chemical spectrum analysers, Savitzky and Golay demonstrated how it reduces noise while preserving the shape and height of waveform peaks. Here, the SG filter was used to smooth the input raw data after the DC components were removed. The output from the SG filter improved the shape of the signal significantly where noise and redundancy were filtered extensively as shown in Figure 3 (data set 1) ((a) and (c)).

The signals were smoothed by SG filter and then reconstructed using Fourier filtering. This was to extract absolute maxima and minima points of the breathing curve that denotes each of the inhalation and exhalation components. Fourier filtering from [31] has already been used as one of the processing algorithms to further eliminate noise and to reconstruct the signals. It is a filtering function that manipulates specific frequency components of a signal by taking the Fourier transform of the corresponding signals which later either attenuate or amplify frequencies of interest. In this paper, the Fourier filter was used to eliminate noise employing a band pass filter depending on the desired breathing frequency range while not distorting the signal significantly. The shape of the Fourier filtered signal was quite similar to the resulting signal from piecewise linear fitting but was smoother and local minima and maxima were prominent.

##### 2.6. Breathing Signal Decomposition

For the breathing cycles obtained from Doppler Radar we assumed that the transition from local minima to local maxima on the curve represents the inhalation component and vice versa for exhalation component, respectively. A peak detection algorithm was then used to determine the maximum and minimum points of each transition defining the inhalation and exhalation components, respectively. These components were extracted separately and represented by a fourth-order polynomial. We then computed the average representation for normal and fast breathing components (inhalation and exhalation) to be used as a model for component identification as discussed in Section 5.3.2.

##### 2.7. Identification-Dynamic Time Warping

Dynamic time warping (DTW) is used to optimally align two time series where one time series is transformed to best fit the other [35]. This technique has been extensively used in speech recognition to identify the similarity of spoken phases from two waveforms as the duration of each spoken sound can vary with similar overall waveform shapes. DTW has also been used in other areas such as data mining and gait recognition [36]. Typically, similarity between two time series for the purpose of classification often requires distance measurement between the two. Computation of Euclidean distance between the two time series may not yield accurate results if one of the two time series is slightly shifted along the time axis. To overcome this limitation, DTW was introduced as described in [35]. Here, we use DTW for registering and comparing breathing components to determine temporal features (extracted breathing component model).

#### 3. Experiment Mechanism

Measurement of humans respiration was approved by the Faculty of Science and Technology Ethics Subcommittee HEAG (Faculty Human Ethics Advisory Groups), Deakin University, and all participants provided their written informed consent to participate in this study.

A Doppler Radar system (Figure 1(a)) has a continuous wave (CW) that operates at 2.7 GHz with 2.14 dBm, two panel antennae where one is (Tx) and the other (Rx), demodulator (Analog Device AD8347), and a data acquisition module (NI-DAQ) were used. The received signals were directly converted into decomposition using AD8347 where the demodulated signal was then sent to a DAQ for further processing using MATLAB.