Australas Phys Eng Sci Med (2012) 35:31–39 DOI 10.1007/s13246-011-0112-2
SCIENTIFIC PAPER
Respiration-rate estimation of a moving target using impulse-based ultra wideband radars Azadeh Sharafi • Mehran Baboli • Mohammad Eshghi Alireza Ahmadian
•
Received: 7 September 2010 / Accepted: 13 November 2011 / Published online: 1 December 2011 Australasian College of Physical Scientists and Engineers in Medicine 2011
Abstract Recently, Ultra-wide band signals have become attractive for their particular advantage of having high spatial resolution and good penetration ability which makes them suitable in medical applications. One of these applications is wireless detection of heart rate and respiration rate. Two hypothesis of static environment and fixed patient are considered in the method presented in previous literatures which are not valid for long term monitoring of ambulant patients. In this article, a new method to detect the respiration rate of a moving target is presented. The first algorithm is applied to the simulated and experimental data for detecting respiration rate of a fixed target. Then, the second algorithm is developed to detect respiration rate of a moving target. The proposed algorithm uses correlation for body movement cancellation, and then detects the respiration rate based on energy in frequency domain. The results of algorithm prove an accuracy of 98.4 and 97% in simulated and experimental data, respectively. Keywords Wireless detection Body movement cancellation UWB FPGA Introduction Ultra-wide band (UWB) signals have been extensively used in medical applications such as wireless tracking A. Sharafi (&) M. Baboli M. Eshghi Faculty of Electrical and Computer Engineering, Shahid Beheshti University, Tehran, Iran e-mail:
[email protected] A. Ahmadian Department of Biomedical Systems and Medical Physics, Tehran University of Medical Sciences & Research Centre for Science and Technology in Medicine, Tehran, Iran
systems [1–3] and detecting early stage of breast cancer [4, 5], after Federal Communications Commission (FCC) legalized use of them in Feb 2002 [6]. With wide bandwidth and high spatial resolution, UWB radar is eminently suitable for tracking objects in hostile environments. With their ability to penetrate through common materials, they can be used for through-wall detection. Wireless detection of vital signs such as respiration rate has always been interesting. Several methods are proposed for estimating respiration rate [7]. Recently, UWB signals have been used for monitoring patients by wireless detection of their respirations rates. Prior work in this area can be classified into two major categories: simulation and detection algorithms. The first category includes articles that are tried to simulate the UWB system including the channel model for human body. In [8], a UWB simulated framework is proposed for evaluating the effect of different parameters, such as thickness of channel layers and heart motion parameters on the accuracy of heart rate measurement. A layer channel model was considered for human chest cavity and some types of heart arrhythmias were detected. Chen et al. [9] proposed a multilayers channel model for human body was simulated and the attenuation of wave propagation while passing through different layers were taken into consideration for formulate the algorithm. However, it did not discuss detection algorithm. The second category includes research to detect heart and respiration rates by means of UWB transceivers and to propose new detection algorithms. A statistically based algorithm to detect heart and respiration rate is proposed in [10]. A mathematical framework for estimating respiration rate is presented in [11]. The main disadvantage of the algorithm described in [11] is its sensitivity to the noise. In
123
32
[12], the respiration rate is estimated by an algorithm based on continuous wavelet transform. Selection of mother wavelet in this algorithm is based on the shape of transmitted pulse; hence it is limited to applications with data taken by UWB transceivers with this same pulse shape. Another wavelet based algorithm is presented in [13]. This algorithm is very slow and it is limited to applications with data taken by UWB transceivers with this same pulse shape. Chia et al. [14] presented the structure of UWB systems and used an algorithm based on the signal energy to detect respiration rate and heart beats. In all proposed vital signs detection methods, the target has to be fixed. Therefore, these algorithms are not suitable for using in situation such as patient monitoring where the patient is not fixed all the time. An algorithm to remove body random movement using Doppler radar is described in [15]. In comparison to the microwave Doppler radars, UWB signals have better material penetration that gives them this ability to perform through even a wall. In this article, a new algorithm to detect respiration rate of a moving target using UWB signals is proposed. This algorithm is tested using simulated data, as well as real data. The results show high accuracy of the proposed algorithm. This article is organized as follows. Section Proposed algorithm for detection of respiration rate of a fixed target is about our proposed method to detect the respiration rate of a fixed target. First the UWB system is modeled using a computer simulation. Then, the algorithm is applied on real data. In Sect. Proposed algorithm for detection of respiration rate of a moving target, the algorithm is modified to be able to detect respiration rate of a moving target.
Proposed algorithm for detection of respiration rate of a fixed target First an UWB system is modeled and the detection algorithm, which we proposed in [16], is applied to the simulated data. Next, the experiments are carried out using an UWB transceiver; and the detection algorithm is applied to real data. System modeling Transmitter The Hermit polynomial of order n, Eq. 1, [17] is selected and simulated as one of typical UWB pulse in the transmitter part of simulation.
123
Australas Phys Eng Sci Med (2012) 35:31–39
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ½n=2 X 1i ðt=sÞn2i En 2 2 pffiffiffiffiffiffiet =4s n! yðtÞ ¼ 2 ðn 2iÞ!i! sn! 2p i¼0
ð1Þ
Channel model A well known model for characterization of UWB channels is a discrete time, multipath, impulse response model. The phenomenon of multipath propagation is represented mathematically by the discrete impulse response of the channel [17, 18], shown in 2. XL1 hð t Þ ¼ a dðt lTm Þ ð2Þ l¼0 l where al is the amplitude attenuation factor on path l and it is a function of time and distance between the transmitter and receiver. The parameter Tm is the minimum resolution time of the pulse, L is the number of resolvable multipath components, and d(t) is the Dirac delta function. Motion model of target The motion has been modeled by changing the attenuation and delay of corresponding multipath component of the target in channel. It is assumed that the target is located in front of a transceiver, as shown in Figs. 1 and 2. In this situation, the delay of corresponding multipath component is calculated by 3 tobj ¼
dto þ dro c
ð3Þ
where c is the speed of the light, dt-o is the distances between the target and the transmitter, and dr-o is the distances between the target and the receiver (Fig. 2). If the target moves in the linear path (Fig. 3), the new delay, tubs (new), is calculated by 4–6. sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2ffi d tr 2 ð4Þ h ¼ dto 2 s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi dtr 2 ð5Þ dtoðnewÞ ¼ droðnewÞ ¼ þ ð x þ hÞ 2 2 tobjðnewÞ ¼
dtoðnewÞ þ droðnewÞ c
ð6Þ
where x is the target displacement. It this step, it is assumed that the target doesn’t have any movement but his/her chest cavity. It is also assumed that the change in lung volume and the displacement of the chest cavity is a sinusoidal function of time, and the distance traveled by the corresponding multipath component also varies periodically with a period dependent on the respiration rate fr. Hence, the displacement of the target and chest cavity is modeled by 7.
Australas Phys Eng Sci Med (2012) 35:31–39 Fig. 1 Transmitted simulated signal (Hermit polynomial order 3) in a Time, b frequency domain
33
3
1
0.8
1
Power(dB)
Amplitude
2
0 -1
0.4
0.2
-2 -3 -400
0.6
-200
0
200
400
0
0
2
4
6
8
Time(ps)
Frequency(GHz)
(a)
(b)
10
12
In this case, we have series of harmonics of the respiration rate. Receive data and analysis algorithm The transmitted pulse train is correlated with impulse response of the channel and the receiver. The antennas are simulated by derivative operator in this simulation. Each received waveform is recorded in one row of received matrix, R. Then, the detection algorithm, which we have presented in [16] is applied on data. The block diagram of this algorithm is shown in Fig. 4.
Fig. 2 Measurement setup
Simulation results A complete simulation is needed to model the UWB system. In this simulation, the bandwidth of UWB pulses is considered to be 3.2 GHz, time between two consecutive pulses is 100 ms, target attenuation is 10 dB, and the total simulation time is 10 s. The parameters used in simulation are shown in Table 1. For this experiment, the motion frequency is set to 0.416 Hz. The result of applying the detection algorithm to received data is shown in Fig. 5. The estimated frequency is 0.4118 which differs about 1% from the expected frequency.
Fig. 3 Target motion model
xðsÞ ¼ Asinð2pfr sÞ
ð7Þ
Where A is the maximum displacement of chest cavity and fr is respiration rate. In the case of non sinusoidal respiration, as described in [11] if the respiration is a zero-mean periodic function with frequency fr, we can expand it in terms of the Fourier Sine Series which is shown in 8. X1 x ð sÞ ¼ A sinð2pifr sÞ ð8Þ i¼0 i
Experimental results After simulation, the algorithm is applied on real data. The experiments are carried out using an UWB transceiver with 3.2 GHz of Bandwidth. The target is located 1.5 m away from transceiver, as shown in Fig. 2. For each transmitted UWB pulse, a waveform including all multipath from existing objects is recorded. A sample of received waveform is shown in Fig. 6.
123
34
Australas Phys Eng Sci Med (2012) 35:31–39
Each received waveform is recorded in one row of the received matrix, R, then, the detection algorithm which is shown in Fig. 4 is applied on data. The measured respiration rate in this experiment is 25 breaths/min (or 0.416 Hz). The estimated rate of the proposed system is 0.411 which differs about 1.2% from expected rate, as shown in Fig. 7c.
Proposed algorithm for detection of respiration rate of a moving target System modeling In this section, all parts of the model are the same as fixed target except the target motion model. For this section, it is assumed that the target has a linear displacement which can be modeled by 9. 1 xðsÞ ¼ as2 þ ts 2
ð9Þ
where a is the acceleration and v is the speed of target. In addition to linear displacement, the thorax has another displacement because of respiration. As mentioned in Sect. Motion Model of Target the respiration can be modeled as sinusoidal or non-sinusoidal motion: Fig. 4 Block diagram of detection algorithm in a fixed target Table 1 Parameters used in simulation Bandwidth
3.2 GHz
Ts (Time between two consecutive transmitted pulse)
100 ms
Total time of experiments Target attenuation
10 s 10 dB
1 xðsÞ ¼ as2 þ ts 2 þ Asinð2pfr sÞ
ðSinusoidal respirationÞ
1 xðsÞ ¼ as2 þ ts 2 X1 A sinð2pifr sÞ þ i¼0 i
ð10Þ
ðNon-sinusoidal respirationÞ ð11Þ
Fig. 5 Simulation result in fixed target scenario. The expected rate is 0.416 Hz. a Matrix R, b matrix R after removing background clutter. c The final result of detection algorithm
123
Australas Phys Eng Sci Med (2012) 35:31–39
35
500
first, simulation is run using new target model. Next, experiments are carried on moving target using UWB transceiver. The proposed detection algorithm of a moving target
0
-500
0
2
4
6
Time(ns)
Fig. 6 Received waveform
8
10
Tracking the object in its environment is the first step for detecting motion rate that is done by comparing the received waveforms to the base waveform. This base waveform is received when the target does not exist in the environment. For removing background clutter, the base waveform is subtracted from the received waveform that is stored in Matrix R. Matrix R, before and after removing background clutter is shown in Fig. 8 in 3D format. In this figure, the x axis is the row and y axis is the column of
Fig. 7 Estimating respiration rate of a fixed target. Measured rate is 0.416 Hz. a Matrix R, b Matrix R after removing background clutter. c The final result of algorithm
Fig. 8 Matrix R a before b after removing background clutter in time domain
123
36
Australas Phys Eng Sci Med (2012) 35:31–39 Table 2 Parameters used in simulation
Fig. 9 Matrix R after body movement cancellation
matrix R. L represents the number of rows and it is equal to the number of received waveforms. K represents the number of columns and it is equal to the number of samples in each received waveforms. The second step is removing the effect of target movement from data. Based on transmitted pulse bandwidth, the receiver resolution and the size of target, several multipaths are received from the target. Some of these multipaths are correspond to the chest cavity and the rests are reflected from another part of body. As shown in Fig. 8, the delays of corresponding multipath components change uniformly when target moves, but some of them change when the
Fig. 10 Block diagram of proposed algorithm
123
Bandwidth
3.2 GHz
Ts (Time between two consecutive transmitted pulse)
100 ms
Total time of experiments
10 s
Target attenuation
10 dB
chest cavity moves. For removing body movement, the cross correlation between each row of matrix R, (each received waveform) and the first row, which is considered as base waveform in time domain, is calculated. Then, each row is circularly shifted to the point where the amount of its correspondent cross correlation is maximized. The periodic movement of the chest cavity is not removed because of the multipath correspondent to fixed part of body. The result of this step is shown in Fig. 9. By removing the effect of target movement, the respiration rate can be detected using our detection algorithm for fixed target (Fig. 4).In this algorithm, the main part is detecting the point which contains the motion that is done as follows. First of all, the Discrete Fourier Transform of matrix R called Rd is calculated. The spectral of the point that contains chest cavity data has a high peak in frequency corresponded to respiration rate. In order to find this point in the Rd matrix, first, the proportion of peak’s energy to total energy of signal at each column j, called w(j) is calculated by 10.
Australas Phys Eng Sci Med (2012) 35:31–39
37
Fig. 11 Simulation result in moving target scenario. The expected rate is 0.4 Hz. a matrix R b matrix R after background and body movement cancellation. c The final result of algorithm
Table 3 Parameters used in experiments PRF(Pulse repetition frequency)
9.6 MHz
Center frequency
4.7 GHz
Bandwidth
3.2 GHz
EIRP
-12.8 dBm
Power consumption
5.7 W
Dimension (housing w/o antenna)
16.5 9 10.2 9 5.1 cm3
Raw data rate
9.6 Mbps
Ts
100 ms
Total time of experiments
10 s Fig. 12 A virtual thorax motion phantom
max ðrd ði; jÞÞ2 wð jÞ ¼ PL 2 i¼1 ðrd ði; jÞÞ
ð12Þ
where rd is an element of matrix Rd, i is row index and L is the total number of rows in this matrix. Then, maximum of w(j) for j = 1,2,….,L is detected and the index of this column is called tp. The frequency of peak in frequency spectrum of column tp corresponds to the respiration rate. The block diagram of complete process is shown in Fig. 10. Simulation result A complete simulation is needed to model the UWB system. In this simulation, the bandwidth of UWB pulses is considered to be 3.2 GHz, time between two consecutive pulses is 100 ms, target attenuation is 10 dB, and total time of experiment is considered to 10 s. The parameters used in simulation are shown in Table 2. The result of applying each step of the proposed algorithm is shown in Fig. 11. In
this simulation, the expected frequency of the respiration rate considered to be 0.4 Hz. The estimated value of this frequency is 0.3937 Hs, which has 1.6% error to the expected frequency. Experimental results The experiments are carried out using the Time Domain PulsON P220 Evaluation Kit [19]. Its specification and the parameters which were set in transceiver are shown in Table 3. For fully controlled on target frequency a virtual thorax motion phantom with capability of setting frequency and movement range was built and used in the experiment. In this phantom, a motion is generated using a stepper motor with precision of 3.2 per step and transmitted to the movable part by timing belt system. The frequency and range of motion was controlled using an ATMEGA32 microcontroller. For setting initial position of movement, the system uses infra-red sensor. The phantom is shown in Fig. 12. In addition to its cyclical
123
38
Australas Phys Eng Sci Med (2012) 35:31–39
Fig. 13 Estimating respiration rate of a moving target. Measured rate is 24 breaths per min (0.4 Hz). a matrix R in 3D format, b matrix R after removing background clutter and body movement cancellation. c The final result of algorithm
Table 4 Experiment result:measured and estimated rates Directly measured frequency (Hz)
Estimated frequency (Hz)
Error (%)
0.4
0.4086
2.15
0.4
0.4050
1.27
0.4
0.4118
2.95
0.3333
0.3253
2.34
0.416
0.4133
0.8
0.45
0.44045
2.12
0.35 0.3
0.3544 0.2929
1.28 2.34
0.2166
0.2207
1.87
motion, the phantom is moved in linear path while recording data to simulate the human movement. The result of each step of algorithm is shown in Fig. 13. The expected and estimated frequency is 0.4 Hz and 0.4118 Hs, respectively. More results are shown in Table 4. To prove the effectiveness of our body cancelation algorithm, we illustrates the result of applying the proposed respiration detection algorithm [16] in Fig. 14 when the body cancellation algorithm is not applied. Other suggested algorithms in literature [10], [11], [13], [14] also are not able to detect the respiration rate without body movement cancellation.
Fig. 14 Comparing the results of applying body movement cancellation algorithm. The expected rate is 1.2 Hz. a matrix R in 3D formats. b The result of algorithm without body movement cancellation. c The result of algorithm with body movement cancellation
123
Australas Phys Eng Sci Med (2012) 35:31–39
Conclusion The goal of this article was to present an algorithm to detect respiration rate. The algorithm was first developed and applied on simulated and experimental data to detect respiration rate of a fixed target. Next, the algorithm was expanded to detect respiration rate of a moving target. In the proposed algorithm, correlation is used for body movement cancellation, and using an algorithm based on energy in frequency domain respiration rate is detected. The results of algorithm in this step proved an accuracy of 98.4% and 97% in simulated and experimental data, respectively. The main advantage of the proposed algorithm for detecting respiration rate, compared to others is its body cancellation ability. Furthermore, the proposed algorithm is fast speed and has simple structure. Hence, this algorithm can be used in real-time and online process and hardware implementation. In the future, we plan to use more comprehensive channel model by considering other effects and phenomena such as refraction, diffraction, clutter, reflection, aperture-medium coupling loss and absorption; and also use more complex model for body movement. Additionally, future work may concentrate on carrying out some experiments on human target with more complex movement path and implement the algorithm on FPGA to achieve real time processing.
References 1. Chang C, Sahai A (2004) Object tracking in a 2D UWB sensor network, In: Conference record of the thirty-eighth asilomar conference on signals, systems and computers, vol 1. Pacific Grove, pp 1252–1256 2. Ergut S, Rao RR, Dural O, Sahinoglu Z (2008) Localization via TDOA in a UWB sensor network using neural networks. In: ICC ‘08 IEEE international conference on communications, Beijing, pp 2398–2403 3. Wang Z, Li X, Fei Y (2006) Moving target position with throughwall radar. In: CIE’06 international conference on radar, Shanghai, pp 1–4 4. Xu L, Bond EJ, Van Veen BD, Hagness SC (2005) An overview of ultra-wideband microwave imaging via space-time beamforming for early-stage breast-cancer detection. Antennas Propag Mag IEEE 47:19–34
39 5. Winters DW, Shea JD, Madsen EL, Frank GR, Van Veen BD, Hagness SC (2008) Estimating the breast surface using UWB microwave monostatic backscatter measurements. IEEE Trans Biomed Eng 55:247–256 6. Buehrer RM, Safaai-Jazi A, Davis W, Sweeney D (2004) Ultrawideband propagation measurements and modeling final report, DARPA NETEX program 7. Folke M, Cernerud L, Ekstro¨m M, Ho¨k B (2003) Critical review of non-invasive respiratory monitoring in medical care. Med Biol Eng Comput 41:377–383 8. Baboli M, Sharafi A, Ahmadian A, KarimiFard S (2009) A framework for simulation of UWB system for heart rate detection. In: ICBPE ‘09 international conference on biomedical and pharmaceutical engineering, pp 1–5 9. Yifan CPR (2008) Human respiration rate estimation using ultrawideband distributed cognitive radar system. Int J Autom Comput 5:325–333 10. Baboli M, Sharafi A, Ahmadian A, Nambakhsh MS (2009) An accurate and robust algorithm for detection of heart and respiration rates using an impulse based UWB signal. In: ICBPE ‘09 international conference on biomedical and pharmaceutical engineering, Singapore, pp 1–4 11. Venkatesh S, Anderson CR, Rivera NV, Buehrer RM (2005) Implementation and analysis of respiration-rate estimation using impulse-based UWB. In: Military communications conference, 2005. IEEE MILCOM 2005, vol 5. Atlantic City, pp 3314–3320 12. Ossberger G, Buchegger T, Schimback E, Stelzer A, Weigel R (2004) Non-invasive respiratory movement detection and monitoring of hidden humans using ultra wideband pulse radar. In: Ultra wideband systems, 2004 international workshop on ultra wideband systems joint with conference on ultra wideband systems and technologies (Joint UWBST & IWUWBS), Kyoto, pp 395–399 13. Baboli M, Ghorashi SA, Saniei N, Ahmadian A (2009) A new wavelet based algorithm for estimating respiratory motion rate using UWB radar. In: international conference on biomedical and pharmaceutical engineering, 2009 ICBPE ‘09, Singapore, pp 1–3 14. Chia MYW, Leong SW, Sim CK, Chan KM (2005) Through-wall UWB radar operating within FCC’s mask for sensing heart beat and breathing rate. In: European radar conference EURAD 2005, Paris, pp 267–270 15. Changzhi L, Jenshan L (2008) Random body movement cancellation in doppler radar vital sign detection. IEEE Trans Microw Theory Tech 56:3143–3152 16. Sharafi A, Baboli M, Eshghi M (2010) A new algorithm for detection motion rate based on energy in frequency domain using UWB signals, In: 4th international conference on bioinformatics and biomedical engineering (iCBBE), Chengdu, pp 1–4 17. Ghavami M, Kohno R, Michael LB (2007) Ultra wideband signals and systems in communication engineering. Wiley, Chichester 18. Saleh AAM, Valenzuela R (1987) A statistical model for indoor multipath propagation. IEEE J Sel Areas Commun 5:128–137 19. http://www.timedomain.com/datasheets/P220aEVK.php
123
