30th EPS Conference on Contr. Fusion and Plasma Phys., St. Petersburg, 7-11 July 2003 ECA Vol.

27A, P-3.7

Statistical analysis of experimentally observed transient MHD modes

G. Pokol1, G. Por1, S. Zoletnik2, É. Fekete2, A.Werner3 and W7-AS team3
1
BUTE Department of Nuclear Techniques, Association EURATOM, Budapest, Hungary
2
KFKI-RMKI, Association EURATOM, P.O.Box 49, H-1525, Budapest, Hungary
3
IPP, Association EURATOM, Boltzmannstr. 2, D-85748, Garching bei München, Germany

Transient MHD modes are often observed in the Wendelstein 7-AS stellarator with lifetimes
of some hundred microseconds. They are best seen as bursts in Mirnov coil signals. An
earlier analysis[1] revealed they are present only in the confinement region, radially
localized to some centimeters (about 1/10 of the minor radius) and they exhibit a low order
poloidal structure[2], with a mode poloidal number close to the reciprocal of the edge
rotational transform. The radial location of these modes is a sensitive function of the
rotational transform[3]. Due to the poloidal plasma rotation modes at different radial
locations appear at different frequencies in the diagnostic signals[3]. A study on temporal
variations in the amplitude of small-scale density turbulence also found[4] roughly 100
microsecond long bursts which are correlated with changes in the RMS amplitude of
poloidal magnetic field perturbations.
The above observations underline the link between MHD bursts and turbulence hence it
appears to be worth analysing the spatial and temporal structure of magnetic field
perturbations. It is assumed that these perturbations represent the MHD response of the
plasma to sudden changes caused by turbulent events. Their statistical analysis reveals the
statistics of the underlying transport events. Although the magnetic field measurements do
not have radial resolution, the bursts at different frequencies might be connected to plasma
layers at different radial locations, therefore the time
evolution of the fluctuation power at selected frequency
ranges is analyzed in ECR heated discharges with good and
bad confinement[3].
Spatial structure of MHD bursts
The spatial structure of modes is analyzed by calculating
the phase distribution of fluctuations at a fixed frequency.
Measurements were done by 16 Mirnov coils located in
one poloidal section of the stellarator. Given the
complicated geometry of the stellarator plasma the phase
distribution can be rather complicated and the
determination of the mode number is not clear. To extract
Fig.1. Phase and amplitude of the mode number and possibly other parameters as well an
fluctuations at different
empirical model was constructed of the spatial structure. It
frequencies. (Top to bottom:
power spectrum of one coil, is assumed that the modes are GAE-like[5], with radial
amplitude and phase vs. coil.) displacement of the plasma. In cylindrical geometry the
30th EPS Conference on Contr. Fusion and Plasma Phys., St. Petersburg, 7-11 July 2003 ECA Vol. 27A, P-3.7 2 of 4

complex amplitude of the velocity associated with the radial displacement is prescribed as
r − r0
−
v r ( r ,θ ) ~ e 2w
⋅ e −imθ +ωt ,
where r0, w, m and are the center radius, the radial width, the mode number and the
frequency of the mode, respectively. From this prescribed velocity profile the associated
toroidal current perturbation is calculated analytically using ideal MHD. The current density
in the poloidal cross-section of the Mirnov coils in W7-AS is prescribed by replacing the
cylindrical coordinate variables by the Reff and magnetic angle coordinates determined from
the finite beta magnetic configuration determined by the TRANS code for the actual
parameters of the discharge. From this current perturbation the phase and amplitude of
Mirnov coil signals is calculated numerically. The resulting phase distribution is compared
to the measured one and the r0 and m parameters are varied until an optimal match is
reached. At low frequencies (<50 kHz) and under favourable confinement conditions the
mode number is found to be 1/ a. At higher frequencies (~100 kHz) the mode distribution is
found to be flat and cannot be interpreted as an MHD mode. This part is most broadband
turbulence of non-MHD mode origin.

Time variation of the signal

To analyse the time
behaviour short time
Fourier transformation
and continuous wavelet
transformation were
carried out on the
signals. This technique
selects different
frequency components
(or scaling
components). These
components are
assumed to correspond
to modes at different
radial positions. Their
amplitudes vary in
time. This variation, Fig.2. Demonstrating the efficiency of deconvolution method
using simulated signals.
i.e. the amplitude of
the different scaling components has been analysed using both autocorrelation function and
cross correlation function. The width of the autocorrelation is linked to the lifetime of the
mode, the cross correlation characterizes the coupling between them.
Our method focuses on the autocorrelation (ACF) and cross correlation (CCF) between the
frequency band integrated power variation. This means, that we evaluate the time dependent
30th EPS Conference on Contr. Fusion and Plasma Phys., St. Petersburg, 7-11 July 2003 ECA Vol. 27A, P-3.7 3 of 4

frequency content of the signal, i.e. we evaluate the time-frequency distribution first. Then
we select the meaningful frequency band(s). This leads to power variation function in the
selected frequency bands. Then we estimate the ACF and CCF for these variations. We have
to notice that the signal cannot be regarded as strongly stationary. It consists of many burst
born and dying out on several frequencies. Therefore the techniques of estimating time-
frequency distribution play an important role. We have arrived to the so called short-time
Fourier transformation (STFT) and continuous wavelet transformation (CWT) techniques
[6]. Selecting appropriate parameters for these transforms, we could easily decompose our
signals into frequency band bursts. On Figure 2., the upper two levels represent two signal
components simulated on two frequencies with burst-envelop. They are added with other
similar burst signals on different frequencies included, and their sum is shown on the third
graph. The quality of the decomposition can be seen on the bottom two graphs, comparing
them to the upper two.
Signals can be simulated as strongly
correlated or uncorrelated bursts on
those two frequencies. Their ACF,
CCF and spectra are shown on Figure
3., to convince the reader: once we get
good CCF with our method, that
means events on the given frequency
bands are correlated, having common
origin. Typically we have only a
dozen of large events during one shot.
Consequently a thorough statistical
analysis has to be performed and
Figure 3. The CCF is not altered by deconvolution errors estimated. We investigated the
method limits of the method on simulated
signals.
On experimental signals the analysis was carried out for different plasma conditions at two
frequency bands with characteristically different spatial structure: typically m=3 mode
structure at low frequency and flat phase structure around 100 kHz (see Fig.1). On Figure 4
we present the ACF, CCF and spectra for the two main frequency bands in the plasma in
well confined and poorly confined plasma states[3], where otherwise only a slight difference
in iota exists between the plasma configurations. Each plot represents the average correlation
function from 6 identical discharges. In good confinement the ACF shows a width of about
100 s, which is the temporal resolution of the signal. In bad confinement some widening is
observed. Despite of the similar ACFs substantial difference in the CCF is observed. In bad
confinement conditions a significant correlation exists between these two frequency bands,
while in good confinement the correlation is ten times less and close the error margin.
30th EPS Conference on Contr. Fusion and Plasma Phys., St. Petersburg, 7-11 July 2003 ECA Vol. 27A, P-3.7 4 of 4

Conclusions.
Due to the different spatial structure bursts in the two frequency bands are expected to
originate from different phenomena. With the CCF described above we can investigate
whether a correlation exists between their appearance. As the MHD modes appear to be
localized radially[1] the lack of correlation in good confinement indicates that there are
radial layers in the plasma where the modulation of fluctuation events is decoupled from the
bursts of broadband magnetic turbulence signal detected at the plasma edge by the Mirnov
coils. In contrast to this, in bad confinement a correlation obviously appears on Figure 4,
which is a signature of a strong radial coupling of turbulence bursts. This observation is in
good agreement with the earlier finding[1], that in bad confinement turbulence extends from
the edge to the core of the plasma. The present observation gives a hint that the reason for
confinement degradation might be the penetration of intermittent events across a larger range
of the minor radius of the plasma.

Figure 4. The two main averaged frequency bands (marked by dots on the power spectrum
plot on the top) in the case of good confinement (left hand side) and bad confinement (right).
The ACF of the bandpower signals are shown in the middle. The lower row of figures show
the crosscorrelation (CCF) between the two bands.
References
[1] S. Zoletnik, M. Anton. M. Endler, et at.; Phys. Plasmas 6 4239 (1999)
[2] M. Anton, T. Klinger, M. Häse et al.; J. Plasma Fusion Res. 1 259 (1998)
[3] S. Zoletnik, N.P. Basse, M. Saffman et al.;Plasma Phys.Control Fusion 44 1581 (2002)
[4] N.P. Basse, S. Zoletnik, M. Saffman et al.; Phys. Plasmas 9 3035 (2002)
[5] A. Weller, M. Anton, J. Geiger, et al.; Phys. Plasmas 8 931 (2001)
[6] S. Mallat: A wavelet tour of signal processing, Academic Press (2001)