Correlates of the shift in heart rate variability with postures and walking by time–frequency analysis

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

journal homepage: www.intl.elsevierhealth.com/journals/cmpb

Correlates of the shift in heart rate variability with postures and walking by time–frequency analysis Hsiao-Lung Chan ∗ , Ming-An Lin, Pei-Kuang Chao, Chun-Hsien Lin Department of Electrical Engineering, Chang Gung University, 259 Wenhwa 1st Road, Kweishan, Taoyuan 333, Taiwan

a r t i c l e

i n f o

a b s t r a c t

Article history:

Heart rate (HR) variability derived from electrocardiogram (ECG) can be used to assess the

Received 26 April 2006

function of the autonomic nervous system. HR exhibits various characteristics during dif-

Received in revised form

ferent physical activities attributed to the altered autonomic mediation, where it is also

27 January 2007

beneficial to reveal the autonomic shift in response to physical-activity change. In this

Accepted 14 February 2007

paper, the physical-activity-related HR behaviors were delineated using a portable ECG and body acceleration recorder based on a personal digital assistant and the smoothed

Keywords:

pseudo Wigner–Ville distribution. The results based upon eighteen subjects performing

Heart rate variability

four sequential 5-min physical activities (supine, sitting, standing and spontaneous walk-

Physical-activity

ing) showed that the high-frequency heartbeat fluctuations during supine and sitting were

Autonomic nervous system

significantly larger than during standing, and that the ratio of low- to high-frequency

Personal digital assistant

fluctuation during standing was significantly higher than during supine and sitting. This could be linked with the parasympathetic predominance during supine and sitting, and a shift to sympathetic dominance while standing. During spontaneous walking, the highfrequency fluctuation was significant lower than during supine. The low- to high-frequency ratio decreased significantly from standing to spontaneous walking, which may imply an increased vagal predominance (autonomic effect) or an increased respiratory activity (mechanical effect). © 2007 Elsevier Ireland Ltd. All rights reserved.

1.

Introduction

The variation of heart period or heart rate, generally called heart rate (HR) variability can reflect the function of the autonomic nervous system by the spectral analysis of HR variability [1,2]. In general, the low-frequency (LF) heart rate fluctuation (about 0.1 Hz) is related to the vasomotor effect, mediated by the sympathetic nervous system. The highfrequency (HF) fluctuation is synchronized with respiration, mediated by the parasympathetic nervous system. The ratio of LF to HF fluctuation behaves a sympathovagus balance index: a high LF/HF ratio shows the predominance of sympathetic activities and a low ratio for parasympathetic (vagal) dom-

∗

inance. Owing to the autonomic-linked characteristics, the analyses of HR variability were widely used to investigate the autonomic behaviors in cardiovascular dysfunctions [3,4], diabetes [5], and so on. HR exhibits various characteristics during different postures [6–9] or during daily physical activities [10,11] owing to the alteration of the autonomic mediation. The altered HR behavior in response to postural change was used as a sensitive measure of the shift in autonomic balance from parasympathetic predominance at rest to sympathetic control while standing [8]. Moreover, HR variability during dynamic physical activities was also employed to investigate autonomic shift or physiological response [12].

Corresponding author. Tel.: +886 3 2118800x5145; fax: +886 3 2118026. E-mail address: [email protected] (H.-L. Chan). 0169-2607/$ – see front matter © 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.cmpb.2007.02.003

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

2.2.

Fig. 1 – The block diagram of the ECG and body accelerations recorder.

Most researches dealing with HR variability during exercise asked subjects to walk on a treadmill [13,14] or to pedal on a cycle ergometer [15–17]. Nevertheless, HR behaviors during spontaneous dynamic physical-activity were less presented. In the present study a portable recorder based on a personal digital assistant (PDA) that can parallel record electrocardiogram (ECG) and body accelerations was developed. The actual timing of physical-activity change can be captured upon the physical-activity classification based on body accelerations. The physical-activity-related HR characteristics can be well depicted by the smoothed pseudo Wigner–Ville distribution. These aspects make the proposed method be able to give detailed HR behaviors during different phases of physicalactivity in an unrestricted space. Eighteen healthy subjects who performed four sequential physical activities (supine, sitting, standing and spontaneous walking) were included for system validation and for delineating HR characteristics during these physical activities.

2.

Materials and methods

2.1.

System description and data recording

The block diagram of the portable recorder based on a PDA (Casio Cassiopeia E-200 with Window CE 3.0 operating system, Japan) is shown in Fig. 1. Two IC accelerometers (ADXL105, Analog Device, USA) affixed on the chest and thigh were used to measure the body accelerations. A microcontroller (PIC16F877, Microchip, USA) was used to acquire one-channel ECG and two acceleration signals with a sampling rate of 125 Hz. The recorded data were stored in the PDA via RS-232 communication. A graphical user interface for data recording and transmission was developed using the eMbedded Visual C++. Eighteen healthy male volunteers (age: 23.4 ± 1.6) were included for the present study. Each subject was at rest for at least 5 min before data collection, and then performed a series of physical activities: supine, sitting, standing and walking with 5-min recording for each physical-activity. There are no particular requirements for walking. The subject took a walk mildly and spontaneously. The data stored in the PDA was transmitted to personal computer for data analyses developed in the MATLAB 6.5 (The MathWorks, USA).

125

Physical-activity classification

The accelerometer can measure the orientation of the body segment in relation to the direction of gravitational acceleration, so the analyses of the accelerations measured in specific segments of the body can be used to distinguish different human postures [11,18,19]. In the present study each body acceleration was normalized to a value between −1 and 1 which maps to a gravitational acceleration from −1 to 1 g. The normalized acceleration was then filtered by a lowpass filter with a cutoff frequency of 2 Hz. The median (med) and rootmean-squared (RMS) values within every 1 s were calculated. Through the above preprocesing, four representative accelerations were generated: ac(med) and at(med) , respectively, stand for the static accelerations of the chest and thigh; ac(RMS) and at(RMS) , the dynamic accelerations. Fig. 2a shows the median accelerations obtained from one subject which performed a series of physical activities: supine, sitting, standing and spontaneous walking. The data clustering was used to classify static physical activities consisting of lying, sitting and standing, based on the vector composed of static accelerations, a = [ac(med) at(med) ]. The cluster centers (cc) for lying, sitting and standing clusters are initially set as [1 0], [0 0], and [0 1], respectively. For each acceleration vector ai , i = 1, . . ., N, the distances with respect to all cluster centers are calculated; the cluster with minimun distance is assigned, and the center of the assigned cluster (cc*) is then updated by: cc∗ = cc∗ + (ai − cc∗ ),

i = 1, . . . , N

(1)

where  is the adaption rate, set as 0.01 in the first iteration and 0.005 in the second iteration. The higher adaption rate adopted in the first interation is to provide a coarse eatimate of the cluster centers, and the adaption rate in the second interation is halved for fine adjustment. The RMS accelerations at the chest and thigh are merged into a new RMS acceleration:

 aRMS =

4a2c(RMS) + a2t(RMS) 5

(2)

The adoption of a smaller weight for at(RMS) than ac(RMS) is due to that the thigh dynamic acceleration had a larger fluctuating range than the chest dynamic acceleration, in particular during spontaneous walking. The detection of dynamic activity is performed by comparing aRMS with two thresholds. If the static acceleration is clustered as lying or sitting, its corresponding dynamic acceleration aRMS is subsequently compared with a high threshold, threshigh . As shown in Fig. 2b, the high threshold, threshigh set as 0.08 provides a strict threshold for identifying the dynamic activity. On the contrary, a low threshold, threslow = 0.02 is used to detect the dynamic activity while standing is assigned. The use of two thresholds can classify the walking slowly as a dynamic activity and avoid recognizing the sitting with high-trembling as a dynamic activity [11]. In the present study, the dynamic activity detection is improved by incorporating hysteresis into the detection based on the low threshold, which can reduce the confounding effect in distinguishing walking from standing. If the previ-

126

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

Fig. 2 – The static accelerations, dynamic acceleration, classified physical activities and heart period signals in a subject which performed supine, sitting, standing and walking.

ous activity is standing, the current activity switches to a dynamic activity as aRMS ≥ 2 threslow , whereas back to standing while aRMS < 0.5 threslow . As depicted in Fig. 2b, some acceleration data during walking can be still recognized as dynamic activities due to the hystersis-based detection even though their dynamic accelerations are less than the low threshold.

2.4.

Time–frequency analysis

The time–frequency analysis is able to delineate the timerelated frequency components of a signal. The time–frequency distribution is referred to the general class of the Cohen distribution [22] expressed as:

 Cf (t, f, ϕ) =

2.3.



ϕ(, )z u +

Heart period preprocessing

 2





z∗ u −

 2



× e−j2(t+f −u) d du d As a QRS was detected, the parabolic interpolation was used to refine the R-wave fiducial point with an 1-kHz sampling rate [20]. An evenly spaced heart period signal was sampled at 4 Hz using the cubic-spline interpolation. To reduce the spectral leakage from very low-frequency HR components, the heart period signal was filtered by a detrending algorithm based on the discrete wavelet transform (DWT) [21] before time–frequency analysis. The wavelet coefficients corresponding to very low frequencies (0–0.0156 Hz) were first derived using a seven-level DWT incorporating the Daubechies-20 wavelet, then the heart period signal was detrended by subtracting the reconstructed signal based on very low-frequency wavelet coefficients.

(3)

where ϕ(, ) is the kernel function, and z(u) is the analytic signal of the original real signal x(u), defined as x(u) + jH{x(u)} where H is the Hilbert transform. The use of the analytic signal can avoid aliasing component at low frequencies in the quadratic time–frequency distribution [23]. By choosing various kernels, different time–frequency distributions are generated. The Wigner–Ville distribution (WVD) is derived by using the kernel as 1:



WVD(t, f ) =



z t+

 2





z∗ t −

 2



e−j2f d

(4)

The WVD provides a high-resolution representation of the signal x(t) in both time and frequency, but undesired spec-

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

tral components named as cross-term exist while analyzing a multi-component signal due to the bilinear nature of the WVD. The smoothed pseudo Wigner–Ville distribution (SPWVD) proposed by Martin and Flandrin [24] is given by:

(5)

where the window h(/2) provides frequency smoothing and the window g(u − t) gives time averaging, accordingly suppresses the cross-terms in the WVD. The length of g(u) determines the effect of cross-term suppression. The suppression of cross-term is better with a longer window, but undesirable smearing of instantaneous characteristics will be accompanied. The discrete SPWVD is expressed as:

SPWVD(n, k) = 2

L−1 

e−j2(k/L)l |hL (l)|2

l=−L+1



×

M−1 

∗

gM (m)z(n + m + l)z (n + m − l) ,

m=−M+1

k = 0, . . . , L − 1

Heart rate variability characterization

The autonomic-related HR parameters were calculated from the time–frequency spectra based on: (1) Low-frequency (LF) power, 0.04–0.15 Hz. (2) High-frequency (HF) power, 0.15–0.6 Hz. (3) The ratio of LF to HF power (LF/HF ratio).

   2           SPWVD(t, f ) = g(u − t)z u + z∗ u − h 2  2 2 × du e−j2f d

2.5.

127

(6)

In the present study the windows hL (l) and gM (m) were used as a 64-s (L = 128) Gaussian window and a 16-s (M = 32) triangular window, respectively. The use of the triangular window is to increase the weights for the quadratic products nearest to the center of the time averaging window.

The use of 0.15–0.6 Hz in the HF band is to cover possible heartbeat fluctuations related to the increased respiratory frequency during walking. The LF power reflects joint mediation of the sympathetic and parasympathetic nervous system, whereas the HF power is linked with parasympathetic modulation. The LF/HF ratio behaves an autonomic-balance index. A low LF/HF ratio indicates a dominant modulation of parasympathetic nervous system; a high ratio, sympathetic dominance.

2.6.

Statistical analysis

Repeated measures analyses of variance (RMANOVA) were performed to compare each HR dependent variable separately among four physical activities (˛ = 0.05). The HR dependent variables included mean heart period (MHP), LF power, HF power, and LF/HF ratio. For post hoc comparisons, general multivariate analysis of variance (GMANOVA) with the Scheffe test was conducted to assess the between-activity differences (˛ = 0.05).

3.

Results

As shown in Fig. 2d, the HR characteristic varies as the physical-activity is changed. The corresponding

Fig. 3 – The time–frequency distribution (upper) and the instantaneous low- and high-frequency power (lower) of the heart period signal shown in Fig. 2d. The spectra and spectral power during transient segments (between 10 s before and 40 s after postural changes) are not shown. Sup, supine; Sit, sitting; Sta, standing; Wal, walking.

128

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

Fig. 4 – The averaged spectra of heart rate variability during supine, sitting, standing and walking derived from the time–frequency distribution depicted in Fig. 3.

time–frequency distribution of HR variability is illustrated in Fig. 3. There are apparent high-frequency heartbeat fluctuations during supine and sitting, and shift to low-frequency dominance while standing. In order to obtain a less postural change interfered time–frequency representation, the spectra during transient segments defined as the time duration between 10 s before and 40 s after postural change were excluded from the time–frequency distribution. Based on the classified physical-activity, the time– frequency spectra within the time between 40 s after previous postural change and 10 s before next postural change were averaged as the representative HR spectrum for this physical-activity. Fig. 4 illustrates the HR spectra during four physical activities in the subject same as Fig. 3. The high-frequency HR fluctuations attenuated during standing, and shifted to a higher frequency during walking

which could be attributed to the increase of respiration rate. Significant differences in HR parameters among different physical activities were showed in MHP, LF/HF ratio, and HF power (p < 0.05), but not in LF power (p > 0.05). The results of the post hoc tests are shown in Table 1 and Fig. 5, and are summarized as follows: (1) MHP: The subjects had significantly larger MHP during supine than all other activities (p < 0.05). MHP in sitting was also significantly larger than standing (p < 0.05). No significant difference in MHP was found between standing and walking. Therefore, MHP is primarily affected by posture changes. (2) LF/HF ratio: LF/HF ratio in standing was significantly larger than in all other activities (p < 0.05). LF/HF ratio in walking

Table 1 – Heart rate behaviors during different physical activities Supine MHP LF, ln HF, ln LF/HF, ln

904 7.01 6.53 0.48

± ± ± ±

99 0.55 0.57 0.55

Sitting 803 7.12 6.23 0.89

± ± ± ±

97* 0.54 0.81 0.51

Standing 709 7.10 5.49 1.61

± ± ± ±

90* , & 0.56 0.73* , & 0.54* , &

Walking 724 6.84 5.81 1.03

± ± ± ±

92* 0.49 0.76* 0.55* ,†

Values are means ± S.D. MHP, mean heart period; LF, low-frequency power; HF, high-frequency power; LF/HF, ratio of low- to high-frequency power; ln, nature logarithm. ∗ & †

p < 0.05, vs. supine. p < 0.05, vs. sitting. p < 0.05, vs. standing.

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

129

Fig. 5 – The statistical behaviors of heart rate variability in nine healthy subjects during supine, sitting, standing and walking.

was also significantly larger than in supine (p < 0.05). The significant difference between standing and walking indicates not only static postures but also dynamic motions can influence LF/HF ratio. (3) HF power: HF power in standing was significantly smaller than in supine and sitting (p < 0.05). HF power in walking was also significantly smaller than in supine (p < 0.05). Together with the non-significant finding in LF power, HF variability seems to be the primary contributor to the changes of LF/HF ratio.

4.

Discussion

The short-term analysis of HR variability during different postures or physical activities provides a tool to investigate the autonomic mediation in response to postural or physicalactivity changes in healthy subjects [6–8,14,15,17] and patients with cardiovascular dysfunction [9,13,16]. Since the HR characteristic may be time-varying, especially during exercise, the time-related HR variability would be better delineated by the quadratic time–frequency analysis than by the short-time Fourier transform [25,26]. In particular, the capability of capturing the actual timing of physical-activity changes and the employment of quadratic time–frequency analysis make the proposed system be able to give detailed and less postural change interfered HR behaviors in different phase of physicalactivity. In the present study, the larger HF power and lower LF/HF ratio during supine and sitting than standing suggest higher parasympathetic mediation and vagal dominance dur-

ing supine and sitting, and a withdrawal of vagal activity when the posture was changed to standing. This is concordant with the general physiological response. Our result also showed that high-frequency heartbeat fluctuation during spontaneous walking was maintained in a comparable scale between sitting and standing, and the LF/HF ratio significantly decreased from standing to walking. Mourot et al. showed decreased LF and HF power, and increased LF/HF ratio from standing to exercise on a cycle ergometer in control subjects before endure training, but with no significant differences [17]. Brown et al. also found a similar trend in patients after coronary artery bypass graft surgery under a ergometer test [16]. On the contrary, Fei et al. presented significant and large decreases of LF power, HF power and LF/HF ratio during walking on a treadmill in normal subjects and survivors after sudden cardiac death [13]. The different results may be due to different exercise scales and different subjects. Furthermore, some researchers pointed out that the increased respiratory activity for increased metabolic need during exercise would be responsible for the maintenance of HF modulation [12,15]. Therefore, the decreased LF/HF ratio from standing to spontaneous walking in the present study may be attributed to increased vagal predominance (autonomic effect), or be linked to the maintenance of HF modulation due to the increased respiratory activity (mechanical effect). Based upon the basis of the presented study, future study using similar framework is needed to investigate the HR behaviors related to other daily physical activities and to obtain more insight the pathogenesis of diseased states. In addition, the use of a PDA-based system has an advantage of high extensibility in applications. For instance, the wire-

130

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 6 ( 2 0 0 7 ) 124–130

less capability of the PDA such as Bluetooth make it can be extended to a real-time wireless physiological monitoring system with less disturbing personnel daily physical-activity. Moreover, its powerful programming capability make the heart rate variability analysis can be further implemented on the PDA.

[11]

[12]

Acknowledgement [13]

The authors would like to express sincere appreciation to the grant support from National Science Council, Taiwan under Grant NSC90-2213-E-182-006.

references

[1] S. Akselrod, D. Gordon, F.A. Ubel, D.C. Shannon, A.C. Barger, R.J. Cohen, Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control, Science 213 (1981) 220–222. [2] M. Pagani, F. Lombardi, S. Guzzetti, O. Rimoldi, R. Furlan, P. Pizzinelli, G. Sandrone, G. Malfatto, S.D. Orto, E. Piccaluga, M. Turiel, G. Baselli, S. Cerutti, A. Malliani, Power spectral analysis of heart rate and arterial pressure variabilities as a marker of sympatho-vagal interaction in man and conscious dog, Circ. Res. 59 (1986) 178–193. [3] J.L. Lin, H.L. Chan, I.N. Lin, C.C. Du, C.W. Lai, K.T. Lin, C.P. Wu, Y.Z. Tseng, W.P. Lien, Chronic, ␤-Blocker therapy improves autonomic nervous regulation in advanced congestive heart failure—a longitudinal heart rate variability study, Am. Heart J. 137 (1999) 658–665. [4] E.B. Schroeder, D. Liao, L.E. Chambless, R.J. Prineas, G.W. Evans, G. Heiss, Hypertension, blood pressure, and heart rate variability the Atherosclerosis Risk In Communities (ARIC) study, Hypertension 42 (2003) 1106–1111. [5] H. Oka, S. Mochio, K. Sato, H. Ssato, K. Katayama, S. Watanabe, T. Nohara, T. Hasunuma, K. Houi, Y. Isogai, Spectral analyses of R–R interval and systolic blood pressure in diabetic autonomic neuropathy, J. Autonom. Nerv. Syst. 52 (1995) 203–211. [6] B. Pomeranz, R.J.B. Macaulay, M.A. Caudill, I. Kutz, D. Adam, D. Gordon, K.M. Kilborn, A.C. Barger, D.C. Shannon, R.J. Cohen, H. Benson, Assessment of autonomic function in humans by heart rate spectral analysis, Am. J. Physiol. 248 (1985) H151–H153. [7] S.M. Pikkujamsa, T.H. Makikallio, K.E. Airaksinen, H.V. Huikuri, Determinants and interindividual variation of R–R interval dynamics in healthy middle-aged subjects, Am. J. Physiol. Heart Circ. Physiol. 280 (2001) H1400–H1406. [8] M.R. Carnethon, D. Liao, G.W. Evans, W.E. Cascio, L.E. Chambless, G. Heiss, Correlates of the shift in heart rate variability with an active postural change in a healthy population sample: the Atherosclerosis Risk In Communities study, Am. Heart J. 143 (2002) 808–813. [9] V. Vuksanovic, V. Gal, J. Kalanj, S. Simeunovic, Effect of posture on heart rate variability spectral measures in children and young adults with heart disease, Int. J. Cardiol. 101 (2005) 273–278. [10] R. Furlan, S. Guzzetti, W. Crivellaro, S. Dassi, M. Tenelli, G. Baselli, S. Cerutti, F. Lombardi, M. Pagani, A. Malliani,

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22] [23]

[24]

[25]

[26]

Continuous 24-h assessment of the neural regulation of systemic arterial pressure and RR variabilities in ambulant subjects, Circulation 81 (1990) 537–547. H.L. Chan, S.C. Fang, Y.L. Ko, M.A. Lin, H.H. Huang, C.H. Lin, Heart rate variability characterization in daily physical activity using wavelet analysis and multi-layer fuzzy activity clustering, IEEE Trans. Biomed. Eng. 53 (2006) 133–139. R. Perini, A. Veicsteinas, Heart rate variability and autonomic activity at rest and during exercise in various physiological conditions, Eur. J. Appl. Physiol. 90 (2003) 317–325. L. Fei, M.H. Anderson, D.J. Statters, M. Malik, A.J. Camm, Effects of passive tilt and submaximal exercise on spectral heart rate variability in ventricular fibrillation patients without significant structural heart disease, Am. Heart J. 129 (1995) 285–290. ¨ M.P. Tulppo, R.L. Hughson, T.H. Makikallio, K.E.J. Airaksinen, ¨ T. Seppanent, H.V. Huikuri, Effects of exercise and passive head-up tilt on fractal and complexity properties of heart rate dynamics, Am. J. Physiol. Heart Circ. Physiol. 280 (2001) H1081–H1087. R. Perini, S. Milesi, N.M. Fisher, D.R. Pendergast, A. Veicsteinas, Heart rate variability during dynamic exercise in elderly males and females, Eur. J. Appl. Physiol. 82 (2000) 8–15. C.A. Brown, L.A. Wolfe, S. Hains, G. Ropchan, J. Parlow, Heart rate variability following coronary artery bypass graft surgery as a function of recovery time, posture, and exercise, Can. J. Physiol. Pharm. 82 (2004) 457–464. L. Mourot, M. Bouhaddi, S. Perrey, J.D. Rouillon, J. Regnard, Quantitative Poincare´ plot analysis of heart rate variability: effect of endurance training, Eur. J. Appl. Physiol. 91 (2004) 79–87. K. Aminian, Ph. Robert, E.E. Buchser, B. Rutschmann, D. Hayoz, M. Depairon, Physical activities monitoring based on accelerometry: validation and comparison with video observation, Med. Biol. Eng. Comput. 37 (1999) 304–308. P.H. Veltink, H.B.J. Bussmann, W. de Vries, W.L.J. Martens, R.C. Van Lummel, Detection of static and dynamic activities using uniaxial accelerometers, IEEE Trans. Rehab. Eng. 43 (1996) 375–385. M. Merri, D.C. Farden, J.G. Mottley, E.L. Titlebaum, Sampling frequency of the electrocardiogram for the spectral analysis of heart rate variability, IEEE Trans. Biomed. Eng. 37 (1990) 99–106. U. Wiklund, M. Akay, U. Niklasson, Short-term analysis of heart-rate variability by adapted wavelet transform, IEEE Eng. Med. Biol. (1997) 113–119, Sept/Oct. L. Cohen, Time–frequency distributions—a review, IEEE Proc. 77 (1989) 941–979. B. Boashash, Note on the use of the Wigner distribution for time–frequency signal analysis, IEEE Trans. Acoust. Speech Sig. Proc. 36 (1988) 1518–1521. W. Martin, P. Flandrin, Wigner–Ville spectral analysis of non-stationary processes, IEEE Trans. Acoust. Speech Sig. Proc. 33 (1985) 1461–1470. P. Novak, V. Novak, Time/frequency mapping of the heart rate, blood pressure and respiratory signals, Med. Biol. Eng. Comput. 31 (1993) 103–110. S. Pola, A. Macerata, M. Emdin, C. Marchesi, Estimation of the power spectral density in non-stationary cardiovascular time series: assessing the role of the time–frequency representations (TFR), IEEE Trans. Biomed. Eng. 43 (1996) 46–59.