This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript
Chest-Worn Health Monitor Based on a Bistatic Self-Injection-Locked Radar Fu-Kang Wang, Member, IEEE, You-Rung Chou, Yen-Chen Chiu, and Tzyy-Sheng Horng, Senior Member, IEEE Abstract—This paper presents wearable health monitors that are based on continuous-wave Doppler radar technology. To achieve low complexity, low power consumption and simultaneous wireless transmission of Doppler information, the radar architecture is bistatic with a self-injection-locked oscillator (SILO) tag and an injection-locked oscillator (ILO)-based frequency demodulator. In experiments with a prototype that was operated in the Medical Body Area Network (MBAN) and the Industrial Scientific and Medical (ISM) bands from 2.36 to 2.484 GHz, the SILO tag is attached to the chest of a subject to transform the movement of the chest due to cardiopulmonary activity and body exercise into a transmitted frequencymodulated wave. The tag consumes a very low power of 4.4 mW. The ILO-based frequency demodulator, located 30 cm from the subject, receives and processes this wave to yield the waveform that is associated with the movement of the chest. Following further digital signal processing, the cardiopulmonary activity and body exercise are displayed as time-frequency spectrograms. Promisingly, the experimental results that are presented in this paper reveal that the proposed health monitor has high potential to integrate a cardiopulmonary sensor, a pedometer and a wireless transmission device on a single radar platform.
Fig. 1. Wearable health monitor with wireless connection to mobile gadgets.
Index Terms—Wearable health monitor, Doppler continuous-wave radar, bistatic self-injection-locked (SIL) radar, time-frequency spectrogram, cardiopulmonary sensor, pedometer, wireless transmission.
F
I. INTRODUCTION
OR a long time, cardiopulmonary monitoring has been an important means of evaluating exercise intensity. The monitored heart rate variability (HRV) can be used to enhance the exercise performance and to prevent exercise-induced oxidative stress and muscle damage [1]. HRV information is commonly used in the diagnosis and treatment of autonomic imbalance and cardiovascular disease [2]. Wearable devices have gained great popularity in healthcare contexts owing to their effectiveness in the long-term monitoring of human Manuscript received Jun. 26, 2015. This work was supported in part by the Ministry of Science and Technology, Taiwan, under Grant MOST 103-2221-E-110-013-MY3 and Grant MOST 103-2221-E-110-016-MY2. Copyright (c) 2014 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending an email to [email protected]. F.-K. Wang, Y.-R. Chou, Y.-C. Chiu, and T.-S. Horng are with the Department of Electrical Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan, R.O.C. (e-mail: [email protected], [email protected])
Fig. 2. Proposed wearable Doppler radar health monitors using a SILO tag and an ILO-based frequency demodulator.
posture and physiological parameters [3], [4]. However, most currently available solutions integrate a wireless device with several sensors such as an accelerometer, an oxygen saturation meter and an electrocardiogram (ECG) monitor [5], resulting in high complexity and high power consumption. Furthermore, these sensors must be in contact with the skin so they cannot be worn on clothes. Doppler radars were developed decades ago for non-contact cardiopulmonary monitoring [6]-[8]. They have great potential as wearable health monitoring devices because they use radio waves that easily penetrate clothes. The two major classes of Doppler radars are pulse radars and continuous-wave (CW) radars. A pulse radar transmits a stream of short radio-frequency pulses, and determines the motion of a target
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript by measuring the delay taken to receive the pulses that are reflected from the target in relation to their round-trip propagation time. A monostatic pulse radar uses the same antenna for transmitting and receiving. However, it cannot receive in the period of pulse transmission. This limitation creates a blind interval that restricts the short-range detection of Doppler signals and causes difficulties in its application to wearable radar monitors. A CW radar continuously transmits a radio-frequency signal, and its receiver is always on to detect the reflected signal. Therefore, no blind interval exists. However, its major shortcoming is the need to isolate the transmitted (Tx) from the received (Rx) signals using either a circulator or separate Tx and Rx antennas. Hence, reducing the size and cost of integration of CW radars for wearable applications encounters significant challenges. Recently, the authors proposed a self-injection-locked (SIL) radar for motion detection. In its basic operation, an oscillator transmits a CW signal, which is partly reflected by a moving target and subsequently injected into the same oscillator to form a SIL state. The Doppler shifts that are associated with the motion of the target can be extracted by frequencydemodulating the output signal of the oscillator. This radar has been verified to exhibit high sensitivity with low complexity [9], and its sensitivity is highly resistant to Tx-Rx coupling, so the radar can be operated with a single antenna [10]. Additionally, the SIL radar can be made bistatic by spatially separating it into two parts - a SIL oscillator (SILO) and a frequency demodulator, each of which has its own antenna. These parts form a basic building block of a wearable monitoring system [11], [12]. Despite the superior performance of the SIL radar, movement of the body critically affects the accuracy of the monitoring of the cardiopulmonary signals. Currently used CW radar-based cardiopulmonary monitors that cancel the effects of random body motion include a dual-frequency radar [13], a radar with a camera [14], and two radars that are placed in the front and rear of the body [10], [15] or up and down in front of the body [16]. However, all of these monitors are based on stationary radar systems and therefore may not be wearable. Although the empirical mode decomposition method has been demonstrated to remove artifacts that arise from motion of the monitor in the extraction of heart rate [17], [18], the accuracy of the results thus obtained depends on the characteristics of the motion of the monitor. This work develops economic low-power wearable devices for tracking health and fitness. As shown in Fig. 1, a chest-worn health monitor provides information concerning respiration and heart rates, exercise intensity and the effect thereof on heart rate. Additionally, automatic wireless connection to nearby mobile gadgets does not add excessive power consumption overhead. As depicted in Fig. 2, the system consists of two main components - a SILO tag attached to the chest of the subject and an injection-locked oscillator (ILO)-based frequency demodulator that can be simply designed as part of the RF circuitry in a mobile gadget. Theoretically, the Doppler signal that is generated by the movement of the chest frequency-modulates the signal of the SILO tag, which is then
Fig. 3. Block diagram of the SILO tag.
transmitted with an RF carrier. The ILO-based frequency demodulator at the far end retrieves the Doppler signals using a frequency discrimination technique. Time-frequency analysis [19]-[21] and a moving average filter [22] in the digital signal processing (DSP) procedure enable the two concurrent chest motions, cardiopulmonary motion and motion due to body exercise, which have very different intensities, to be distinguished from each other accurately and reliably. II. BISTATIC SIL RADAR Unlike the conventional bistatic CW radar with a separate transmitter and receiver, the proposed SIL-based CW radar is bistatic in the sense that the modulator and the demodulator are separated. In the present system, shown in Fig. 2, the SILO tag and the ILO-based receiving circuit serve as the Doppler frequency modulator and demodulator, respectively. With reference to Figs. 3 and 4, when the SILO tag receives an echo signal Sinj(t), the SILO outputs a signal Sout(t) whose frequency is modulated with the Doppler information due to the injection of the echo signal. Consequently, the waves that are emitted and received by the antenna of the SILO tag contain the Doppler information. Therefore, the ILO-based frequency demodulator can receive the signal Sout(t) wirelessly via its own antenna to allow demodulation of the Doppler information at flexible locations. This is a very unique feature of the bistatic SIL radar. The two separate parts of the system described above operate according to the following principles. A. SILO Tag As shown in Fig. 3, the SILO tag uses a single antenna to transmit the CW output signal toward the subject and to receive the reflected Doppler signal without the need to isolate the signals from each other. Assume that the SILO has an inherent oscillation frequency ωosc, constant oscillation amplitude Eosc, and a tank quality factor Qtank. When a Doppler-reflected signal with amplitude Einj is injected into the SILO, the instantaneous VCO output frequency ωout(t) can be obtained using Adler’s equation [23]
out t osc LR,tag sin (t )
(1)
where
LR ,tag
osc
Einj
2Qtank Eosc
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
(2)
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript is the locking range that is associated with the SILO. In (1), α(t) equals the propagation phase delay between the transmitted signal and the injected signal, which is given by
t
2osc R x t c
(3)
(4)
where xv(t) denotes the chest displacement due to respiration and heartbeat; m is the mass of the SILO tag; ks represents the spring constant of clothes, and ab(t) is the acceleration of the body. From (1) and (3), Δx(t) can be determined by frequency-demodulating the SILO output signal. Moreover, the reflection from the antenna with coefficient a can be regarded as stationary clutter with a constant phase delay α0; accordingly, its effect on the SIL radar is to cause a small constant frequency shift equal to ωLR,tagsinα0 in the SILO, rather than a serious dc offset in the baseband, which commonly occurs in a conventional CW radar. This fact explains why the SIL radar requires only a single antenna for simultaneous transmission and reception. As discussed further below, the frequency offset that is caused by antenna reflection does not influence the baseband processing of Doppler signals by an ILO-based frequency demodulator. B. ILO-based Frequency Demodulator The ILO-based frequency demodulator, displayed in Fig. 4, is constructed from a low-noise amplifier (LNA), an ILO with a dedicated injection terminal, a pair of in-phase (I) and quadrature-phase (Q) mixers, and two low-pass filters (LPFs). The Doppler-modulated signals that are emitted from the SILO tag are captured by the Rx antenna and amplified by the LNA. Then, the output signals from the LNA are fed into the RF terminals of the IQ mixers and the injection terminal of the ILO that enters an injection-locked state after it receives the injected signals. The ILO functions as a band-pass amplifier with a time delay τIL, and with the IQ mixers, it performs frequency discrimination using an unwrapped arctangent method. Mathematically, the demodulator output is given by [11] B (t ) tan 1
Q (t ) 2 R osc IL LR ,tag IL sin osc I (t ) c
2oscLR,tag IL c
2 R cos osc x(t ) c
(5)
LPF
I (t ) RX Ant.
Sout (t )
0
out inj
where R represents the distance from the tag to the chest; Δx(t) = xt(t)-xc(t) is the relative instantaneous displacement between the tag and the chest, and c denotes the speed of light. When the body is stationary, Δx(t) is caused by cardiopulmonary motion. However, when the body moves, Δx(t) needs adding a displacement to account for the inertia of the SILO tag. Owing to the elasticity of clothes, the proposed radar system acts as an accelerometer to detect such a displacement as a measure of body acceleration. As a result, Δx(t) is expressed as
m x t xv (t ) ab (t ) ks
Mixer I
LNA
90
ILO
Q(t ) Mixer Q
LPF
Fig. 4. Block diagram of the ILO-based frequency demodulator.
where I(t) and Q(t) are low-pass filtered baseband I and Q signals, respectively. The time delay τIL is estimated as
IL
1 2
LR, Rx osc 2
(6)
where ωLR,Rx is the locking range that is associated with the ILO, and Δωosc is the frequency difference between the inherent ILO frequency and the frequency of the injection signal. Obviously, the information of chest motion Δx(t) can be extracted from the ac component in (5). Notably, since the ILO output signal is locked to the frequency of the Doppler-modulated signal, the problem of the aforementioned frequency shift in the SILO due to stationary clutter is solved. It is worth mentioning that the demodulator shown in Fig. 4 is very similar to a quadrature receiver in a wireless communication system, except that the local oscillator is an ILO rather than a phase-locked oscillator (PLO). Therefore, it can be implemented using semiconductor technology as a standalone IC, or integrated into a wireless communication IC. Moreover, there are two ways to improve the communication distance between the tag and the demodulator. One is to increase the output power of the tag, but doing so shortens the battery life and may cause interference with other wireless devices. The other is to enhance the sensitivity of the demodulator by simply increasing the gain of the LNA, which is preferred in this application. III. DIGITAL SIGNAL PROCESSING
A. Time-Frequency Spectrogram Joint time-frequency analysis (JTFA) can reveal signal characteristics in both time and frequency domains, yielding a two-dimensional representation of the signal. This technique is appropriate for processing signals with indeterminate future behavior or with time-variant frequencies, such as speech, image and medical signals. The short-time Fourier transform (STFT) is widely applied in the JTFA of vital signs and motion signals [24]-[28] whose frequencies do not dramatically change, owing to its low computational complexity and absence of cross-term interference [20]. The sliding window function is utilized to perform the Fourier transform in a particular period, and the STFT is then defined as
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript
B w t e j 2 f d
where w is the window function. Furthermore, the spectrogram is the magnitude of the STFT squared: SP t , f STFT t , f
2 .
(8)
The window length influences the resolution of both time and frequency. Generally, a longer window is required to distinguish more effectively signals with similar frequencies; however, an irregular signal is then more likely to become too strong and possibly destroy the harmonic pattern [24]. This fact explains why a shorter window (such as 0.25 s [25] or 0.4 s [28]) is favored for analyzing motion whereas a longer window is favored for examining vital signs (such as 5 s [24] or 8 s [26]). Moreover, the overlap between window lengths, which can reduce the edge aliasing in the spectrogram, should be increased (such as to 39 % [24] or 50 % [28]) to favor the detection of violent motions. Notably, the dynamic range of a conventional motion sensor is limited, and to the best of the authors’ knowledge, no literature exists on the use of a spectrogram to display the motion associated with exercise and cardiopulmonary activities at the same time. In this paper, a Kaiser window of sufficient window length and overlap is adopted in the STFT-based spectrogram of the body motion and cardiopulmonary signals that were detected using the proposed bistatic SIL radar system. The various signals in the spectrogram can be individually emphasized by applying the moving average filter that is described in the following subsection. B. Moving Average Filter The moving average filter operates by averaging a number of points from the input signal to produce each point in the output signal. Mathematically, [22]
Bk [i]
1 N
N 1
Bk 1[i j]
(9)
j 0
where Bk-1[ ] and Bk[ ] are the input and output signals in the kth iteration, and N is the number of points that are used in the averaging. Equation (9) is a low-pass moving average filter. Differently, this work uses a band-pass moving average filter as given below:
Bk [i] Bk 1[i ]
1 N
N 1
Bk 1[i j] .
(10)
j 0
The frequency response of (10) can be derived as
sin( N f t ) hk ( f ) 1 N sin( f t )
k
(11)
where Δ t is the sampling time interval. This frequency response yields a maximum value at
3 2 1 N . 2 3 2 f0t
0
(7)
(12)
Attenuation (dB)
STFT t, f
10
20
30 k = 3 40
0
1
k = 5
2
3
k = 11 4
5
f / f0 Fig. 5. Attenuation responses of the moving average filters with 3, 5 and 11 iterations.
(a)
(b)
Fig. 6. Experimental setup of the proposed wearable Doppler radar health monitors. (a) Experimental site. (b) Chest-worn SILO tag.
Notably, in approximation (12), f0 is the frequency of a specific signal of interest. The suppression of signals of other frequencies can be enhanced by increasing the number of iterations k. Substituting (12) into (11) and assuming fΔt
TBME-00132-2015.R1 – Final Manuscript
Chest-Worn Health Monitor Based on a Bistatic Self-Injection-Locked Radar Fu-Kang Wang, Member, IEEE, You-Rung Chou, Yen-Chen Chiu, and Tzyy-Sheng Horng, Senior Member, IEEE Abstract—This paper presents wearable health monitors that are based on continuous-wave Doppler radar technology. To achieve low complexity, low power consumption and simultaneous wireless transmission of Doppler information, the radar architecture is bistatic with a self-injection-locked oscillator (SILO) tag and an injection-locked oscillator (ILO)-based frequency demodulator. In experiments with a prototype that was operated in the Medical Body Area Network (MBAN) and the Industrial Scientific and Medical (ISM) bands from 2.36 to 2.484 GHz, the SILO tag is attached to the chest of a subject to transform the movement of the chest due to cardiopulmonary activity and body exercise into a transmitted frequencymodulated wave. The tag consumes a very low power of 4.4 mW. The ILO-based frequency demodulator, located 30 cm from the subject, receives and processes this wave to yield the waveform that is associated with the movement of the chest. Following further digital signal processing, the cardiopulmonary activity and body exercise are displayed as time-frequency spectrograms. Promisingly, the experimental results that are presented in this paper reveal that the proposed health monitor has high potential to integrate a cardiopulmonary sensor, a pedometer and a wireless transmission device on a single radar platform.
Fig. 1. Wearable health monitor with wireless connection to mobile gadgets.
Index Terms—Wearable health monitor, Doppler continuous-wave radar, bistatic self-injection-locked (SIL) radar, time-frequency spectrogram, cardiopulmonary sensor, pedometer, wireless transmission.
F
I. INTRODUCTION
OR a long time, cardiopulmonary monitoring has been an important means of evaluating exercise intensity. The monitored heart rate variability (HRV) can be used to enhance the exercise performance and to prevent exercise-induced oxidative stress and muscle damage [1]. HRV information is commonly used in the diagnosis and treatment of autonomic imbalance and cardiovascular disease [2]. Wearable devices have gained great popularity in healthcare contexts owing to their effectiveness in the long-term monitoring of human Manuscript received Jun. 26, 2015. This work was supported in part by the Ministry of Science and Technology, Taiwan, under Grant MOST 103-2221-E-110-013-MY3 and Grant MOST 103-2221-E-110-016-MY2. Copyright (c) 2014 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending an email to [email protected]. F.-K. Wang, Y.-R. Chou, Y.-C. Chiu, and T.-S. Horng are with the Department of Electrical Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan, R.O.C. (e-mail: [email protected], [email protected])
Fig. 2. Proposed wearable Doppler radar health monitors using a SILO tag and an ILO-based frequency demodulator.
posture and physiological parameters [3], [4]. However, most currently available solutions integrate a wireless device with several sensors such as an accelerometer, an oxygen saturation meter and an electrocardiogram (ECG) monitor [5], resulting in high complexity and high power consumption. Furthermore, these sensors must be in contact with the skin so they cannot be worn on clothes. Doppler radars were developed decades ago for non-contact cardiopulmonary monitoring [6]-[8]. They have great potential as wearable health monitoring devices because they use radio waves that easily penetrate clothes. The two major classes of Doppler radars are pulse radars and continuous-wave (CW) radars. A pulse radar transmits a stream of short radio-frequency pulses, and determines the motion of a target
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript by measuring the delay taken to receive the pulses that are reflected from the target in relation to their round-trip propagation time. A monostatic pulse radar uses the same antenna for transmitting and receiving. However, it cannot receive in the period of pulse transmission. This limitation creates a blind interval that restricts the short-range detection of Doppler signals and causes difficulties in its application to wearable radar monitors. A CW radar continuously transmits a radio-frequency signal, and its receiver is always on to detect the reflected signal. Therefore, no blind interval exists. However, its major shortcoming is the need to isolate the transmitted (Tx) from the received (Rx) signals using either a circulator or separate Tx and Rx antennas. Hence, reducing the size and cost of integration of CW radars for wearable applications encounters significant challenges. Recently, the authors proposed a self-injection-locked (SIL) radar for motion detection. In its basic operation, an oscillator transmits a CW signal, which is partly reflected by a moving target and subsequently injected into the same oscillator to form a SIL state. The Doppler shifts that are associated with the motion of the target can be extracted by frequencydemodulating the output signal of the oscillator. This radar has been verified to exhibit high sensitivity with low complexity [9], and its sensitivity is highly resistant to Tx-Rx coupling, so the radar can be operated with a single antenna [10]. Additionally, the SIL radar can be made bistatic by spatially separating it into two parts - a SIL oscillator (SILO) and a frequency demodulator, each of which has its own antenna. These parts form a basic building block of a wearable monitoring system [11], [12]. Despite the superior performance of the SIL radar, movement of the body critically affects the accuracy of the monitoring of the cardiopulmonary signals. Currently used CW radar-based cardiopulmonary monitors that cancel the effects of random body motion include a dual-frequency radar [13], a radar with a camera [14], and two radars that are placed in the front and rear of the body [10], [15] or up and down in front of the body [16]. However, all of these monitors are based on stationary radar systems and therefore may not be wearable. Although the empirical mode decomposition method has been demonstrated to remove artifacts that arise from motion of the monitor in the extraction of heart rate [17], [18], the accuracy of the results thus obtained depends on the characteristics of the motion of the monitor. This work develops economic low-power wearable devices for tracking health and fitness. As shown in Fig. 1, a chest-worn health monitor provides information concerning respiration and heart rates, exercise intensity and the effect thereof on heart rate. Additionally, automatic wireless connection to nearby mobile gadgets does not add excessive power consumption overhead. As depicted in Fig. 2, the system consists of two main components - a SILO tag attached to the chest of the subject and an injection-locked oscillator (ILO)-based frequency demodulator that can be simply designed as part of the RF circuitry in a mobile gadget. Theoretically, the Doppler signal that is generated by the movement of the chest frequency-modulates the signal of the SILO tag, which is then
Fig. 3. Block diagram of the SILO tag.
transmitted with an RF carrier. The ILO-based frequency demodulator at the far end retrieves the Doppler signals using a frequency discrimination technique. Time-frequency analysis [19]-[21] and a moving average filter [22] in the digital signal processing (DSP) procedure enable the two concurrent chest motions, cardiopulmonary motion and motion due to body exercise, which have very different intensities, to be distinguished from each other accurately and reliably. II. BISTATIC SIL RADAR Unlike the conventional bistatic CW radar with a separate transmitter and receiver, the proposed SIL-based CW radar is bistatic in the sense that the modulator and the demodulator are separated. In the present system, shown in Fig. 2, the SILO tag and the ILO-based receiving circuit serve as the Doppler frequency modulator and demodulator, respectively. With reference to Figs. 3 and 4, when the SILO tag receives an echo signal Sinj(t), the SILO outputs a signal Sout(t) whose frequency is modulated with the Doppler information due to the injection of the echo signal. Consequently, the waves that are emitted and received by the antenna of the SILO tag contain the Doppler information. Therefore, the ILO-based frequency demodulator can receive the signal Sout(t) wirelessly via its own antenna to allow demodulation of the Doppler information at flexible locations. This is a very unique feature of the bistatic SIL radar. The two separate parts of the system described above operate according to the following principles. A. SILO Tag As shown in Fig. 3, the SILO tag uses a single antenna to transmit the CW output signal toward the subject and to receive the reflected Doppler signal without the need to isolate the signals from each other. Assume that the SILO has an inherent oscillation frequency ωosc, constant oscillation amplitude Eosc, and a tank quality factor Qtank. When a Doppler-reflected signal with amplitude Einj is injected into the SILO, the instantaneous VCO output frequency ωout(t) can be obtained using Adler’s equation [23]
out t osc LR,tag sin (t )
(1)
where
LR ,tag
osc
Einj
2Qtank Eosc
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
(2)
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript is the locking range that is associated with the SILO. In (1), α(t) equals the propagation phase delay between the transmitted signal and the injected signal, which is given by
t
2osc R x t c
(3)
(4)
where xv(t) denotes the chest displacement due to respiration and heartbeat; m is the mass of the SILO tag; ks represents the spring constant of clothes, and ab(t) is the acceleration of the body. From (1) and (3), Δx(t) can be determined by frequency-demodulating the SILO output signal. Moreover, the reflection from the antenna with coefficient a can be regarded as stationary clutter with a constant phase delay α0; accordingly, its effect on the SIL radar is to cause a small constant frequency shift equal to ωLR,tagsinα0 in the SILO, rather than a serious dc offset in the baseband, which commonly occurs in a conventional CW radar. This fact explains why the SIL radar requires only a single antenna for simultaneous transmission and reception. As discussed further below, the frequency offset that is caused by antenna reflection does not influence the baseband processing of Doppler signals by an ILO-based frequency demodulator. B. ILO-based Frequency Demodulator The ILO-based frequency demodulator, displayed in Fig. 4, is constructed from a low-noise amplifier (LNA), an ILO with a dedicated injection terminal, a pair of in-phase (I) and quadrature-phase (Q) mixers, and two low-pass filters (LPFs). The Doppler-modulated signals that are emitted from the SILO tag are captured by the Rx antenna and amplified by the LNA. Then, the output signals from the LNA are fed into the RF terminals of the IQ mixers and the injection terminal of the ILO that enters an injection-locked state after it receives the injected signals. The ILO functions as a band-pass amplifier with a time delay τIL, and with the IQ mixers, it performs frequency discrimination using an unwrapped arctangent method. Mathematically, the demodulator output is given by [11] B (t ) tan 1
Q (t ) 2 R osc IL LR ,tag IL sin osc I (t ) c
2oscLR,tag IL c
2 R cos osc x(t ) c
(5)
LPF
I (t ) RX Ant.
Sout (t )
0
out inj
where R represents the distance from the tag to the chest; Δx(t) = xt(t)-xc(t) is the relative instantaneous displacement between the tag and the chest, and c denotes the speed of light. When the body is stationary, Δx(t) is caused by cardiopulmonary motion. However, when the body moves, Δx(t) needs adding a displacement to account for the inertia of the SILO tag. Owing to the elasticity of clothes, the proposed radar system acts as an accelerometer to detect such a displacement as a measure of body acceleration. As a result, Δx(t) is expressed as
m x t xv (t ) ab (t ) ks
Mixer I
LNA
90
ILO
Q(t ) Mixer Q
LPF
Fig. 4. Block diagram of the ILO-based frequency demodulator.
where I(t) and Q(t) are low-pass filtered baseband I and Q signals, respectively. The time delay τIL is estimated as
IL
1 2
LR, Rx osc 2
(6)
where ωLR,Rx is the locking range that is associated with the ILO, and Δωosc is the frequency difference between the inherent ILO frequency and the frequency of the injection signal. Obviously, the information of chest motion Δx(t) can be extracted from the ac component in (5). Notably, since the ILO output signal is locked to the frequency of the Doppler-modulated signal, the problem of the aforementioned frequency shift in the SILO due to stationary clutter is solved. It is worth mentioning that the demodulator shown in Fig. 4 is very similar to a quadrature receiver in a wireless communication system, except that the local oscillator is an ILO rather than a phase-locked oscillator (PLO). Therefore, it can be implemented using semiconductor technology as a standalone IC, or integrated into a wireless communication IC. Moreover, there are two ways to improve the communication distance between the tag and the demodulator. One is to increase the output power of the tag, but doing so shortens the battery life and may cause interference with other wireless devices. The other is to enhance the sensitivity of the demodulator by simply increasing the gain of the LNA, which is preferred in this application. III. DIGITAL SIGNAL PROCESSING
A. Time-Frequency Spectrogram Joint time-frequency analysis (JTFA) can reveal signal characteristics in both time and frequency domains, yielding a two-dimensional representation of the signal. This technique is appropriate for processing signals with indeterminate future behavior or with time-variant frequencies, such as speech, image and medical signals. The short-time Fourier transform (STFT) is widely applied in the JTFA of vital signs and motion signals [24]-[28] whose frequencies do not dramatically change, owing to its low computational complexity and absence of cross-term interference [20]. The sliding window function is utilized to perform the Fourier transform in a particular period, and the STFT is then defined as
0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2453255, IEEE Transactions on Biomedical Engineering
TBME-00132-2015.R1 – Final Manuscript
B w t e j 2 f d
where w is the window function. Furthermore, the spectrogram is the magnitude of the STFT squared: SP t , f STFT t , f
2 .
(8)
The window length influences the resolution of both time and frequency. Generally, a longer window is required to distinguish more effectively signals with similar frequencies; however, an irregular signal is then more likely to become too strong and possibly destroy the harmonic pattern [24]. This fact explains why a shorter window (such as 0.25 s [25] or 0.4 s [28]) is favored for analyzing motion whereas a longer window is favored for examining vital signs (such as 5 s [24] or 8 s [26]). Moreover, the overlap between window lengths, which can reduce the edge aliasing in the spectrogram, should be increased (such as to 39 % [24] or 50 % [28]) to favor the detection of violent motions. Notably, the dynamic range of a conventional motion sensor is limited, and to the best of the authors’ knowledge, no literature exists on the use of a spectrogram to display the motion associated with exercise and cardiopulmonary activities at the same time. In this paper, a Kaiser window of sufficient window length and overlap is adopted in the STFT-based spectrogram of the body motion and cardiopulmonary signals that were detected using the proposed bistatic SIL radar system. The various signals in the spectrogram can be individually emphasized by applying the moving average filter that is described in the following subsection. B. Moving Average Filter The moving average filter operates by averaging a number of points from the input signal to produce each point in the output signal. Mathematically, [22]
Bk [i]
1 N
N 1
Bk 1[i j]
(9)
j 0
where Bk-1[ ] and Bk[ ] are the input and output signals in the kth iteration, and N is the number of points that are used in the averaging. Equation (9) is a low-pass moving average filter. Differently, this work uses a band-pass moving average filter as given below:
Bk [i] Bk 1[i ]
1 N
N 1
Bk 1[i j] .
(10)
j 0
The frequency response of (10) can be derived as
sin( N f t ) hk ( f ) 1 N sin( f t )
k
(11)
where Δ t is the sampling time interval. This frequency response yields a maximum value at
3 2 1 N . 2 3 2 f0t
0
(7)
(12)
Attenuation (dB)
STFT t, f
10
20
30 k = 3 40
0
1
k = 5
2
3
k = 11 4
5
f / f0 Fig. 5. Attenuation responses of the moving average filters with 3, 5 and 11 iterations.
(a)
(b)
Fig. 6. Experimental setup of the proposed wearable Doppler radar health monitors. (a) Experimental site. (b) Chest-worn SILO tag.
Notably, in approximation (12), f0 is the frequency of a specific signal of interest. The suppression of signals of other frequencies can be enhanced by increasing the number of iterations k. Substituting (12) into (11) and assuming fΔt
