**3. Related work**

This section presents the related work in both system design and signal processing techniques.

#### **3.1. System design**

Since the 1970s [21], microwave Doppler radar has been used in sensing physiological movement. The original work was done with heavy, bulky, and expensive components. However, it was useful for research improvements. During the 1980s, heart and respiration signals were obtained using 10.5 GHz frequency signal. Using a horn antenna placed few centimeters from the subject, the system shows capability to detect cardiopulmonary signals using a 10-mW transmitted power [22]. In 1990, systems operating at 2 and 10 GHz were tested in detecting life signs in victims under clutter. The radiated power varied between 10 and 20 mW [23]. In 1997, heartbeat and respiration signals were detected at a distance of 10 m using 24 GHz frequency system with 30 mW output power and 40 dB antenna gain [24]. In the year 2000, systems operating at 450 and 1150 MHz, with a radiated power around 300 mW, were used to detect life sings in victims under rubble [25]. In 2001, a 1.2-GHz, 70-mW quadrature superheterodyne system was used to detect breathing of a subject under 1.5-m rubble [26]. Operating at 1.6 and 2.4 GHz, direct-conversion Doppler radars have been integrated in 0.25 μm complementary metal-oxide semiconductor (CMOS) and BiCMOS technologies. The output power was estimated to be 6.5 dBm [27]. Heart and respiration activities were detected using a modified Wireless Local Area Network (LAN) Personal Computer Memory Card International Association (PCMCIA) card and a module combining the transmitted and reflected signals [28,29]. The operational power of the system was 35 mW and the distance from the subject was 40 cm. Other systems operating in the Ka-band were described in Ref. [30] using a low-power double-sideband transmission signal. For a distance of 2 m from the subject, and for 16 and 12.5 μW, respectively, the systems showed an accuracy of 80% in detecting the heartbeat rate. Recently in 2006, some measurements were performed in order to detect multiple heartbeats signals [31]. Operating at 2.4 GHz and 1 mW power, the system was able to determine the number of persons in a room. With the same characteristics, another system using single and multiple antennas systems showed the possibility of separating two respira‐ tion signals [32]. In 2007, a new study showed the possibility of detecting the presence of a person through a wall using ultra-wideband (UWB) radar [33]. Lately in 2008, some experi‐ ments are preformed to detect life signs using a 4–7-GHz band with 1 mW power and around 7 dB antenna gain. This system uses the complex signal demodulation (CSD) and the arctan‐ gent demodulation in order to cancel random body movements [34]. In 2009, a system operating at 10 GHz showed the ability to detect the heart and the respiration activity of a person behind a wall.

Recently, a system having two Vivaldi antennas, a Mini-Circuits ZHL-42 power amplifier for the transmission, and a Hittite HMC753 low-noise amplifier in the receiver is proposed [35]. The receiver is composed of a down-converter of a 20-MHz IF band, a mixer, an Agilent signal generator and a band-pass filter, and the received signal is sent to the analog-to-digital converter (ADC). A 60-MHz sampling clock provided by an external clock and synchronized with the field-programmable gate array (FPGA) reference clock for the signal digitization. Then, the sampled data are sent to the FPGA for digital down conversion. Another system composed of two antennas, an oscillator that provides both the receiver's local oscillator and the transmitted signal, and a mixer is presented in [36]. I/Q channel demodulation with calibration method is added to alleviate the null point problem and acquire an accurate phase demodulation result with high linearity. Another system presented in [37] is based on multiple transceivers, and antennas with polarization and frequency multiplexing are used to detect signals from different body orientations.

#### **3.2. Signal processing techniques**

the angle of the pump handle changes between 20° and 30°, and the rib radius changes between 10.6 and 10.8 cm. At the seventh rib, the angle of the pump handle changes between 30° and 37°, and the rib radius changes between 1.37 and 1.42 cm [18]. Comparing the chest motion in the front/back, left/right, and up/down directions shows that the largest motions correspond to the sternum and the navel. Sternum moves forward 4.3 mm with inspiration, and the navel which moves forward 4.03 mm with inspiration [19]. The relation between tidal volume and abdominal wall linear displacement is measured using a laser displacement measuring device [20]. An expansion of the abdomen is observed: 4 mm for 400 ml inspiration and 11 mm for 1100 ml inspiration. Also, during spontaneous breathing, an abdominal displacement of 12

This section presents the related work in both system design and signal processing techniques.

Since the 1970s [21], microwave Doppler radar has been used in sensing physiological movement. The original work was done with heavy, bulky, and expensive components. However, it was useful for research improvements. During the 1980s, heart and respiration signals were obtained using 10.5 GHz frequency signal. Using a horn antenna placed few centimeters from the subject, the system shows capability to detect cardiopulmonary signals using a 10-mW transmitted power [22]. In 1990, systems operating at 2 and 10 GHz were tested in detecting life signs in victims under clutter. The radiated power varied between 10 and 20 mW [23]. In 1997, heartbeat and respiration signals were detected at a distance of 10 m using 24 GHz frequency system with 30 mW output power and 40 dB antenna gain [24]. In the year 2000, systems operating at 450 and 1150 MHz, with a radiated power around 300 mW, were used to detect life sings in victims under rubble [25]. In 2001, a 1.2-GHz, 70-mW quadrature superheterodyne system was used to detect breathing of a subject under 1.5-m rubble [26]. Operating at 1.6 and 2.4 GHz, direct-conversion Doppler radars have been integrated in 0.25 μm complementary metal-oxide semiconductor (CMOS) and BiCMOS technologies. The output power was estimated to be 6.5 dBm [27]. Heart and respiration activities were detected using a modified Wireless Local Area Network (LAN) Personal Computer Memory Card International Association (PCMCIA) card and a module combining the transmitted and reflected signals [28,29]. The operational power of the system was 35 mW and the distance from the subject was 40 cm. Other systems operating in the Ka-band were described in Ref. [30] using a low-power double-sideband transmission signal. For a distance of 2 m from the subject, and for 16 and 12.5 μW, respectively, the systems showed an accuracy of 80% in detecting the heartbeat rate. Recently in 2006, some measurements were performed in order to detect multiple heartbeats signals [31]. Operating at 2.4 GHz and 1 mW power, the system was able to determine the number of persons in a room. With the same characteristics, another system using single and multiple antennas systems showed the possibility of separating two respira‐ tion signals [32]. In 2007, a new study showed the possibility of detecting the presence of a person through a wall using ultra-wideband (UWB) radar [33]. Lately in 2008, some experi‐

mm is observed.

**3. Related work**

34 Advanced Biosignal Processing and Diagnostic Methods

**3.1. System design**

In Doppler cardiopulmonary monitoring, the heartbeat and the respiration signals are laid together. Hence, a processing technique is needed in order to determine the characteristics of each signal. The signal processing part includes the separation of the cardiopulmonary signals and the extraction of the heartbeat rate. The amplitude of the respiration signal is much greater than the amplitude of the heartbeat signal. Therefore, the respiration rate can be determined without filtering. On the other hand, determining the heartbeat rate needs a processing technique. At rest, the heartbeat rate varies between 50 and 90 beats per minute [38]; this corresponds to a frequency between 0.83 and 1.5 Hz, respectively. On the other hand, the resting respiration rate varies between 9 and 24 breaths per minute [38]; this corresponds to a frequency between 0.15 and 0.4 Hz. Due to the difference of the frequencies that correspond to the heartbeat and the respiration rates, the average heartbeat rate could be determined upon extracting the frequency components of the cardiopulmonary signals. This allows determining the average heartbeat rate over a specific window of time. On the other hand, determining the heartbeats variation over time needs a peak-finding technique.

Several techniques were used in processing the cardiopulmonary signals. This processing includes separating the heartbeat signal from the respiration signal and then finding the heartbeat rate. Some measurements were performed while holding the breath. This eliminates the isolation process of the heartbeat signal, but a filtering approach is still needed in order to remove noise and distorting signals.

First measurements for heartbeat and respiration were performed separately. Holding the breath allows detecting the heartbeat signal [21]. Another study shows the possibility of measuring the heartbeat and respiration activities successively where a low-pass filter with 4 Hz cutoff frequency was used to remove unwanted frequencies [23]. In 2000, heartbeat and respiration signals were measured simultaneously. The output signal is fed through a bandpass filter (BPF) with passing band between 0.1 and 4 Hz. The heartbeat and respiration rates are obtained by applying fast Fourier transform (FFT) to the original signal. The dominant peak in the frequency domain was taken as the breathing frequency, and the second dominant peak was taken as the heartbeat frequency [25]. In 2002, separated measurements for heartbeat and respiration were performed. The respiration signal was filtered with a BPF (0.03–0.3 Hz), and the heart signal was filtered with a BPF (1–3 Hz) [27]. Another work tended to detect the heartbeat signal using a 12-dB/octave high-pass filtering at 0.03 Hz in order to remove DC offset, and a 12-dB/octave low-pass filtering at 3 Hz was used to avoid aliasing error. The heart signal was further isolated with an additional 12 dB/octave HPF at 1 Hz [39]. Also in 2002, measurements for breathing persons were performed. The respiration signal was isolated using a fourth-order low-pass Butterworth filter with cutoff frequency at 0.7 Hz. The heartbeat signal was isolated using a fourth-order band-pass Butterworth filter with cutoff frequencies at 1 and 3 Hz. The rate determination is based on the use of auto-correlation. A spatial zeroforcing filter is applied so that the DC is removed from the measured received signal [40]. In 2003, a wireless LAN PC card was used. A low-pass resistor-capacitor (RC) filter having a cutoff frequency 100 Hz is used to filter the baseband output of the receiver. This helps denoising the signal as well as avoiding aliasing error. The filtered signal is then converted to digital in order to be processed in a notebook PC. The prefiltered, digitized signal was filtered further in the digital domain to separate the heart and breathing signals. The heart signal was isolated using a 0.75–5 Hz band-pass filter for 10 s interval. Based on the periodicity of the autocorrelation function, the heartbeat rate was estimated [28]. In 2006, a system using a signal processing part similar to some previous work is stated. The heartbeat signal was first separated from the respiration signal by a Butterworth BPF with passband from 0.7 to 3 Hz. The filtered signal was then windowed and auto-correlated. Then, FFT was applied to the autocorrelated signal to obtain the heartbeat rate [41].

Recently, other processing techniques are used for cardiorespiratory separation. In [35], FPGAs are used to process either time- or frequency-domain signals in human sensing radar appli‐ cations. It is applied for continuous wavelet (CW) and UWB radars. In CW Doppler radar, a novel superheterodyne receiver is used to suppress low-frequency noise and includes a digital down-converter module implemented in an FPGA. In [36], compact quadrature Doppler radar sensor is used: Continuous wavelet filter and ensemble empirical mode decomposition (EEMD) based algorithms are applied for cardiorespiratory signal to separate the cardiac and respiratory signals. The accurate beat-to-beat interval can be acquired in time domain for heart rate variability (HRV) analysis. A curvelet transform is applied in [42] in order to remove the direct coupling wave and background clutters. Life signals are denoised using a singular value decomposition. Both the FFT and the Hilbert-Huang transform are applied in order to separate and extract the frequencies of the human vital sign as well as the characteristics of micro-Doppler shift for an UWB radar. Least mean square adaptive harmonic cancellation algorithm is proposed in [43] to separate the breathing and heartbeat signal from biological Doppler radar. The respiration signal is used as a model reference input while the radar signal due to body motion is considered as the original input of the model. A model is designed and

validated experimentally with commercial motion detector [44]. A low-pass filter with 0.7 Hz cutoff frequency is used to extract the respiration signal, while a band-pass filter between 0.9 and 2.5 Hz is used to extract the heartbeat signal. In [37], complex technique is discussed; a complex signal demodulation technique is proposed to eliminate the null detection point problem in non-contact vital sign detection. This technique is robust against DC offset in a direct conversion system. Hence, a random body movement cancellation technique is devel‐ oped to cancel out strong noise caused by random body movement in non-contact vital sign monitoring. The complex signal is software reconstructed in real time by *S*(*t*) = *I*(*t*) + j *Q*(*t*). System setup of random body movement cancellation technique is designed of two transceiv‐ ers, one in front of and the other behind the human body, which are transmitting and receiving signals with different polarization and wavelength. The two complex signals are multiplied. This multiplication corresponds to convolution and frequency shift in frequency domain, thus canceling the Doppler frequency drift and only keeping the periodic Doppler phase effects. In [45], fast acquisition of HR is proposed, the length of the time window is less than 5 s and the accuracy is significantly degraded due to insufficient spectrum resolution. In [37], CSD is used for vital sign detection. A time-window-variation technique is developed for fast acquisition of HR from short-period time windows and measuring HR variation using CSD. The proposed method has also proved to be able to measure HR variation using CSD.

#### **3.3. Discussion**

Hz cutoff frequency was used to remove unwanted frequencies [23]. In 2000, heartbeat and respiration signals were measured simultaneously. The output signal is fed through a bandpass filter (BPF) with passing band between 0.1 and 4 Hz. The heartbeat and respiration rates are obtained by applying fast Fourier transform (FFT) to the original signal. The dominant peak in the frequency domain was taken as the breathing frequency, and the second dominant peak was taken as the heartbeat frequency [25]. In 2002, separated measurements for heartbeat and respiration were performed. The respiration signal was filtered with a BPF (0.03–0.3 Hz), and the heart signal was filtered with a BPF (1–3 Hz) [27]. Another work tended to detect the heartbeat signal using a 12-dB/octave high-pass filtering at 0.03 Hz in order to remove DC offset, and a 12-dB/octave low-pass filtering at 3 Hz was used to avoid aliasing error. The heart signal was further isolated with an additional 12 dB/octave HPF at 1 Hz [39]. Also in 2002, measurements for breathing persons were performed. The respiration signal was isolated using a fourth-order low-pass Butterworth filter with cutoff frequency at 0.7 Hz. The heartbeat signal was isolated using a fourth-order band-pass Butterworth filter with cutoff frequencies at 1 and 3 Hz. The rate determination is based on the use of auto-correlation. A spatial zeroforcing filter is applied so that the DC is removed from the measured received signal [40]. In 2003, a wireless LAN PC card was used. A low-pass resistor-capacitor (RC) filter having a cutoff frequency 100 Hz is used to filter the baseband output of the receiver. This helps denoising the signal as well as avoiding aliasing error. The filtered signal is then converted to digital in order to be processed in a notebook PC. The prefiltered, digitized signal was filtered further in the digital domain to separate the heart and breathing signals. The heart signal was isolated using a 0.75–5 Hz band-pass filter for 10 s interval. Based on the periodicity of the autocorrelation function, the heartbeat rate was estimated [28]. In 2006, a system using a signal processing part similar to some previous work is stated. The heartbeat signal was first separated from the respiration signal by a Butterworth BPF with passband from 0.7 to 3 Hz. The filtered signal was then windowed and auto-correlated. Then, FFT was applied to the auto-

Recently, other processing techniques are used for cardiorespiratory separation. In [35], FPGAs are used to process either time- or frequency-domain signals in human sensing radar appli‐ cations. It is applied for continuous wavelet (CW) and UWB radars. In CW Doppler radar, a novel superheterodyne receiver is used to suppress low-frequency noise and includes a digital down-converter module implemented in an FPGA. In [36], compact quadrature Doppler radar sensor is used: Continuous wavelet filter and ensemble empirical mode decomposition (EEMD) based algorithms are applied for cardiorespiratory signal to separate the cardiac and respiratory signals. The accurate beat-to-beat interval can be acquired in time domain for heart rate variability (HRV) analysis. A curvelet transform is applied in [42] in order to remove the direct coupling wave and background clutters. Life signals are denoised using a singular value decomposition. Both the FFT and the Hilbert-Huang transform are applied in order to separate and extract the frequencies of the human vital sign as well as the characteristics of micro-Doppler shift for an UWB radar. Least mean square adaptive harmonic cancellation algorithm is proposed in [43] to separate the breathing and heartbeat signal from biological Doppler radar. The respiration signal is used as a model reference input while the radar signal due to body motion is considered as the original input of the model. A model is designed and

correlated signal to obtain the heartbeat rate [41].

36 Advanced Biosignal Processing and Diagnostic Methods

Systems used in these works lack determining the most appropriate parameters for these applications. These parameters are the operational frequency, the radiated power, and the optimal signal processing technique. The proposed system shows the ability of tuning both the operational frequency and transmitted power. Hence, it is able to determine better emitted frequency with less power that detects heartbeat accurately. On the other hand, most of the processing techniques tend to extract an average heartbeat rate of the subject. This does not give information about the variation of the heartbeat rate and requires a long-duration window which makes the real-time processing not possible. The proposed signal processing technique shows the ability to detect the variation of the heart activity in time.
