wavelet transformation emrah duzel institute of cognitive neuroscience ucl
TRANSCRIPT
Wavelet transformation
Emrah Duzel
Institute of Cognitive Neuroscience
UCL
Why analyse neural oscillations?
• Temporal code of information processing (versus rate code)
• Functional coupling
• Interareal synchrony
• Local field potentials and their correlation with fMRI
• Functional specificity of oscillations
Large scale neural dynamics of higher cognitive processes: At least three types of stimulus-responses
• Evoked response:Evoked response: the addition of response amplitude to the ongoing brain activity in a time-locked manner.Schah et al., 2004, Cereb Cortex
• Phase resetting response:Phase resetting response: the resetting of ongoing oscillatory brain activity without concomitant changes in response amplitude.Penny, Kiebel, Kilner, Rugg, 2002, Trends in Cog Sci. / Makeig et al., 2002, Science
• Induced response:Induced response: the addition of response amplitude that is not time-locked to stimulus onset.Tallon-Baudry and Bertrand, 1998, Trends in Cog Sci.
Makeig et al., 2004
8 trials Phase-resetting of a 10 Hz oscillation
Phase resetting
ERP power
10
Measure of phase alignment
Penny, Kiebel, Kilner, Rugg, 2002, Trends in Cog Sci. / Makeig et al., 2002, Science / Klimesh et al., 2001, Cog Brain Res. / Burgess and Gruzelier, 2000, Psychophys.
Single subject analyses of M400 old/new effectsClear evidence of evoked responses in some subjects
Overview
• Basics of digital signal processing– Sampling theory
• Fourier Transforms– Discrete Fourier Transforms
• Wavelet Analysis
• Applications and online demonstrations
Digital signal processing
• Decompose a signal into simple additive components
• Process these components in a useful manner
• Synthesize them into a final result
Sampling theory• Nyquist theorem• Sample rate• Nyquist frequency• Aliasing
• With each signal there are 4 critical parameters:– Highest frequency in the signal (determined by low-pass filter) – Twice this frequency– Sampling rate– SR / 2 (nyquist frequency/rate)
Sampling theoryNyquist theoremNyquist theorem: a signal can be properly sampeld only if it does not contain
frequencies above ½ sampling frequency
• AliasingAliasing: if a signal contains frequencies above the Nyquist frequency.
– Loss of information
– Introduces wrong information (waves take on different ‚identities‘
– Loss of phase information (phase shift)
Single-epoch wavelet transforms
x Spectral analysis
Wavelet
averaging
+Phase
ERP
Wavelettransformation
Different morlet wavelets
Better time resolution
Good compromise
Better freq. resolution
Time-frequency resolution of a standard Morlet-wavelet
Time-frequency resolution of a standard Morlet-wavelet
Time-frequency resolution of a standard Morlet-wavelet
convolution
Matlab demo
• Create an artificial signal composed of several frequencies of varying time/amplitude modulation
– continuous delta [2Hz]– continuous alpha [10 Hz]– continuous beta [20Hz]– theta-burst [5Hz, +200 ms] – gamma_burst [40 Hz, -200] – gamma_burst [67 Hz, -100] – gamma_burst [67 Hz, +200]
• Create a wavelet• Convolve wavelets and signal
– highlight the issue of amplitude normalization– highlight limits of time/frequency resolution
• Plot a time/frequency spectrogramm• Illustrate phase resetting
-500 +500
67hz
40hz
theta
beta
alpha
delta
Matlab demo
• Create an artificial signal composed of a linear combination of several sinusoids with different frequencies and time/amplitude modulations
whereω is the angular frequency or angular speed (measured in radians per second), T is the period (measured in seconds), f is the frequency (measured in hertz)
e.g. if T = 50 ms = 0.05 sec then f = 1/0.05 = 20 Hz
angular frequency
delta=sin(2*pi*1/500*(t))t=-500:500
A*sin(2 pi ω t)
Matlab demo
• Create a wavelet
wavelet_beta=sin(2*pi*t/50).*exp(-(t/50/strecth).^2)
Complex numbers
Euler’s formula
trigonometric form
exponential form
r
In a Cartesian coordinate systemeach point z is determined by two axes
In polar notation
each point z is determined by an angle φ and a distance r
central pointis ‘pole’
r is called the absolute value or modulus of z
Frequency resolution
Time resolution