33rd Annual International Conference of the IEEE EMBS

Boston, Massachusetts USA, August 30 - September 3, 2011

Analysis of degree of nonlinearity and stochastic nature of HRV

signal during meditation using delay vector variance method
L. Ram Gopal Reddy, Srinivas Kuntamalla

Abstract—Heart rate variability analysis is fast gaining and the increased synchronization, or respiratory sinus
acceptance as a potential non-invasive means of autonomic arrhythmia (RSA), between the breathing cycle and the heart
nervous system assessment in research as well as clinical beat during meditation [3].
domains. In this study, a new nonlinear analysis method is used In this paper, the HRV data obtained before and during
to detect the degree of nonlinearity and stochastic nature of the Chi and Kundalini meditation is analyzed for its
heart rate variability signals during two forms of meditation
(Chi and Kundalini). The data obtained from an online and nonlinearity and stochastic nature using the method of delay
widely used public database (i.e., MIT/BIH physionet vector variance. The delay vector variance (DVV) is a
database), is used in this study. The method used is the delay unified method for detecting the presence of determinism
vector variance (DVV) method, which is a unified method for and nonlinearity in a time series and is based upon the
detecting the presence of determinism and nonlinearity in a examination of local predictability of a signal [4], [5].
time series and is based upon the examination of local
predictability of a signal. From the results it is clear that there
is a significant change in the nonlinearity and stochastic nature II. METHODS
of the signal before and during the meditation (p value > 0.01). A. Data group
During Chi meditation there is a increase in stochastic nature
and decrease in nonlinear nature of the signal. There is a A publicly available RR inter beat interval database
significant decrease in the degree of nonlinearity and stochastic (www.physionet.org) consisting of data before and during
nature during Kundalini meditation. meditation, collected from eight healthy Qigong meditation
Index Terms: meditation, nonlinearity, stochastic nature, (Chi meditation) subjects (aged 29-35) and four Kundalini
heart rate variability, delay vector variance yoga meditators, 2 women and 2 men (aged 20-52) is
utilized (more information on the dataset and the meditation
method are described in [6]). The length of the time series
I. INTRODUCTION varied between 50 and 80 minutes. One characteristic case
of the RR interval time series, before and during meditation,
H EART rate variability (HRV) analysis is gaining
acceptance as a potential non-invasive means of
assessment of autonomic nervous system in research and
is presented in Figure 1, in which the evident irregular
character of the time series before meditation transcends to
clinical domains. Heart rate dynamics is nonlinear in nature smoother cyclic oscillations during meditation.
and it is proved that nonlinear analysis of HRV provides B. Surrogate Data Generation
more appropriate information for understanding and A surrogate signal is a realization of null hypothesis,
interpretation of the physiological problems associated with which in this context is that the original time series is linear.
cardiovascular system [1]. Iterative Amplitude Adjusted Fourier Transform (IAAFT)
Meditation is a complex physiological process which approach described in [7] is used to generate surrogates for
affects neural, psychological, behavioral, and autonomic this study because it yields superior results and overcomes
functions, and is considered as an altered state of the normal distribution problem of phase randomization
consciousness, differing from wakefulness, relaxation at rest, method [8] , [9]. This approach retains the amplitude
and sleep. Many meditation traditions consider breath, body spectrum and distribution of the original signal.
and mind are linked, and thus have given the breath a
central role in meditation practice. Slower respiration rate C. Delay Vector Variance method
during meditation practice induces changes in the A time series can be represented in phase space
cardiovascular activity that corresponds to an increase in the conveniently using time delay embedding. When time delay
activity of restorative parasympathetic system [2]. This is embedded into a time series it can be represented by a set
increased parasympathetic activity has also been assessed of delay vectors (DVs) of a given dimension. If m is the
through the slowing down of basal heart rate in meditators, dimension of the delay vectors then it can be expressed as X
(k) =[x (k-mτ) ……. x (k-τ)], where τ is the time lag. Now for
L. Ram Gopal Reddy is a Professor of Physics, National Institute of every DV X (k), there is a corresponding target, namely the
Technology, Warangal, INDIA (lrgreddy@gmail.com) next sample xk. A set βk (m, d) is generated by grouping
Srinivas kuntamalla is a Research Scholar in Physics Department, those DVs that are within a certain Euclidean distance (d) to
National Institute of Technology, Warangal, INDIA (ksvchary@gmail.com)

978-1-4244-4122-8/11/$26.00 ©2011 IEEE 2720

DV X(k).This Euclidean distance will be varied in a manner series is non-linear. Thus the deviation from the bisector line
standardized with respect to the distribution of pair wise is an indication of nonlinearity, and can be quantified as root
distances between DVs. Now for a given embedding mean square error (RMS error) between the σ*2 s of the
dimension m, a measure of unpredictability σ*2 (target original time series and the σ*2 s averaged over the DVV
variance) is computed over all sets of β k. The variation of the plots of the surrogate data.
standardized distance enables the complete range of pair
wise distances to be examined. The procedure for Delay
Vector Variance method can be summarized as below [4],
[5].
1.5
RR interval(sec)

0.5

0
0 500 1000 1500 2000
beat number

1.2
RR interval (sec)

1
0.8
0.6
0 500 1000 1500 2000
beat number
Fig. 1. RR intervals before (top plot) and during meditation
(bottom plot)

The mean μd and the standard deviation σ d are computed Fig. 2. DVV plots ( top plot - before meditation, bottom plot – during
over all pair wise Euclidean distances between DVs given by meditation) for a subject (c7) in chi meditation group (solid curve) and
average DVV plots computed over 25 surrogates (dashed curve)
. The sets βk (m, d) are generated
such that k i.e. , sets which In this paper, optimum time delay embedding dimension
consist of all DVs that lie closer to X(k) than a certain is considered as to ensure a common dimension for all the
distance d, taken from the interval [ μd -nd σ d ; μd +nd σ d] subjects for comparison. Therefore, all the parameters are
where nd is a parameter controlling the span over which to computed for given dimension (m=5) and given distance
perform DVV analysis. For every set βk (m, d) the variance parameter (nd=5) and τ =1. Twenty five surrogates are
of the corresponding targets σk 2(m, d) is computed. The generated using IAAFT method. As reported by Temujin
average over all sets βk (m, d) is divided by the variance of Gautama et al. [5], the length of the time series must be
the time series signal , σk gives the inverse measure of greater than 1000, otherwise there is a profound effect on
predictability, namely target variance σ*2(m, d). The minimum target variance. The length of time series
variance is computed only if there are at least 30 DVs in a considered in this paper is 2000 samples. The feature
set βk (m, d). extraction and implementation of method is done in Matlab
The plot of target variance σ*2(m, d) as a function of 7.3.
spans d for a given dimension m is called DVV plot and as
the distance is standardized it is easy to interpret the plot. At III. RESULTS
the extreme right, the DVV plots smoothly converges to The parameters extracted from DVV plots and DVV
unity, since for maximum spans, all DVs belong to the same Scatter diagrams are tested for null hypothesis using a
universal set, and the variance of the targets is equal to the significance test (T-test). T-test is the most commonly used
variance of the time series. method to evaluate the differences in means between the two
DVV scatter diagrams are produced using the original groups. The significance level for rejection of null
time series and its surrogates. If the surrogate signals yield hypothesis is set to 0.01 in this study. The p value < 0.01 is
DVV plots similar to that of the original time series, the considered to be statistically significant.
‗DVV scatter diagram‘ coincides with the bisector line and The value of minimum target variance for a Chi
original signal is linear. If the surrogate time series yield meditator (c7) is 0.0116 before meditation and 0.2595
DVV plots not similar to that of the original time series, the during meditation (Fig. 2.). The minimum target variance,
curve will deviate from the bisector line and original time is a measure of noise present in the time series.

2721
The amount of noise is indicative of the stochastic 0.0658(±0.0313) during meditation and 0.3139(±0.0961)
component prevalent. The presence of strong deterministic before meditation (p value > 0.01). The average value of
component will lead to small target variance for small spans. RMS error for Kundalini meditation is 0.0242(±0.0143)
This shows that Chi meditation increases stochastic nature of during meditation and 0.0607(±0.0298) before meditation (p
HRV. From Fig. 3., the value of RMS error of a Chi value > 0.01).
meditator (c7) is 0.0609 before meditation and 0.0384
during meditation. Thus there is an increased linearity during
Chi meditation.

Fig. 4. DVV plots ( top plot- during meditation, bottom plot – before
meditation) for a subject (y2) in kundalini meditation group (solid curve)
and average DVV plots computed over 25 surrogates (dashed curve)

Fig. 3. DVV Scatter plots ( top plot- before meditation, bottom plot – From the plots and results, it is clear that there is a
during meditation) for a subject (c7) in chi meditation group significant change in the nonlinearity and stochastic nature
The value of minimum target variance for a Kundalini of the signal before and during the meditation (p value >
meditator (y2) is 0.2107 before meditation and 0.0349 0.01). There is a decrease in the degree of nonlinearity
during meditation (Fig. 4). The stochastic component of during the two forms of meditation while the determinism is
HRV is therefore, significantly reduced during Kundalini increased in Kundalini yoga and decreased in Chi
meditation. DVV Scatter plots (top plot - before meditation, meditation. These changes are may be because of state of
bottom plot – during meditation) for a subject (y2) in induced deep mental relaxation, which brings about the
Kundalini meditation group are shown in Figure 5. The changes in the cardiovascular system function. It is proved
value of RMS error is 0.0614 before meditation and 0.0144 by many investigators that meditation brings about greater
during meditation. Kundalini yoga seems to greatly increase coordination and harmony between the mind and body [2].
the linearity of HRV signal. Figures 6 and 7 show Box – With increased inward attention during meditation there is
Whiskers plot for the parameters determined from DVV more withdrawal of the body and mind from external senses
plots and DVV scatter diagrams for Chi and Kundalini which yields control to the vagal innervation. The significant
meditation groups. decrease observed in the nonlinearity of HRV signal during
the two forms of meditation should probably be a reflection
IV. CONCLUSION of this i.e., the underlying mechanism responsible for HRV
is relatively simpler during meditation. During the Kundalini
The average value of minimum target variance for Chi
meditation the nonlinearity and stochastic nature of HRV are
meditation is 0.0918(±0.0518) before meditation and
markedly decreased, while the nonlinearity is decreased and
0.1708(±0.0421) during meditation (p value > 0.01). The
the stochastic nature of HRV is increased during chi
average value of RMS error for Chi meditation is
meditation. The increase in stochasticity is note worthy. This
0.0227(±0.0207) before meditation and 0.011(±0.0095)
might have probably arised owing to the difference in the
during meditation (p value > 0.01). The average value of
processes adopted in Chi and Kundalini meditation. While
minimum target variance for Kundalini yoga is
2722
chi meditation employs visualization technique, mantras
are chanted in Kundalini yoga.

Fig. 7. Box – Whiskers plot for the parameters determined from DVV
plots and DVV scatter plots for kundalini meditation group.

REFERENCES
[1] C. Karmakar, et al., "Defining asymmetry in heart rate variability
signals using a Poincaré plot," Physiological measurement, vol. 30,
2009.
[2] Shr-Da Wu and Pei_Chen Lo, ―Inward-attention meditation increases
parasympathetic activity: a study based on heart rate variability‖,
Biomedical Research, 29(5), pp 245-250, 2008.
[3] B.S. Raghavendra and D. Narayan Dutt, ― Nonlinear Dynamical
characterization of heart rate variability time series of meditation‖,
International journal of biological and life sciences, vol 8 no. 3 ,
pp167-177, 2012.
[4] Temujin Gautama, Danilo P. Mandic , Marc M. Van Hulle, ―The
delay vector variance method for detecting determinism and
Fig. 5. DVV Scatter plots ( top plot- before meditation, bottom plot – nonlinearity in time series‖, Physica D, 190, pp 167–176, 2004.
during meditation) for a subject (y2) in kundalini meditation group
[5] Temujin Gautama, Danilo P. Mandic , Marc M. Van Hulle, ―A Novel
method for determining the nature of time series‖, IEEE trans.
Biomed. Engg., vol. 51, no. 5, pp 728-736, 2004.
ACKNOWLEDGMENT [6] C-K Peng et al, ― Exaggerated heart rate oscillations during two
meditation techniques‖, International journal of cardiology, vol 70,
The authors are thankful to Danilo P. Mandec et al. pp 101-107, 1999.
for providing the DVV toolbox for Matlab under GNU
[7] T. Schreiber and A. Schmitz, ―Improved surrogate data for
general public license. nonlinearity tests‖, Phys.Rev. Lett., 77, pp 635-638,1996.
[8] T. Schreiber and A. Schmitz, ―Surrogate time series‖, Physica D, vol.
142, no.3-4, pp346-382, 2000.
[9] Theiler, S. Eubank, A. Longtin, B. Galdrikian and J.D. Farmer,
―Testing for nonlinearity in time series: The method of surrogate
data‖, Physica D, 58, pp 77-94, 1992.

Fig. 6. Box – Whiskers plot for the parameters determined from DVV
plots and DVV scatter plots for Chi meditation group.

2723