Professional Documents
Culture Documents
1
Neural Computation &
Engineering lab (BIOEN/EE 466/566)
2
Components of a Brain-Computer Interface
Classification
or Regression
3
Last Lecture: Frequency Domain and Wavelet
analysis of time-varying signals
Today:
Time-domain analysis and time-frequency
analysis of time-varying signals
4
Time domain analysis
✦ Signal matching
✦ Pattern of oscillations
5
Signal matching: Cross Correlation
✦ Cross correlation is used to find where two signals match
'& ∗ '& ∗
✦ 𝑤 𝑡 = ∫%& 𝑢 𝑡 𝑣 𝜏 + 𝑡 𝑑𝜏 = ∫%& 𝑢 𝜏 − 𝑡 𝑣 𝜏 𝑑𝜏
6
Signal matching: Cross Correlation
✦ u(t) is the test waveform
Example 1: Example 2:
v(t) contains u(t) with an Example 3:
unknown delay and added noise y(t) is the same as v(t) with p(t) contains –u(t)
more noise
w(t) gives a peak at the time lag z(t) can still detect the q(t) has a negative peak
where u(t-t) best matches v(t) correct time delay
7
Signal matching: Cross Correlation
✦ Cross correlation is used to find where two signals match
'& ∗ '& ∗
✦ 𝑤 𝑡 = ∫%& 𝑢 𝑡 𝑣 𝜏 + 𝑡 𝑑𝜏 = ∫%& 𝑢 𝜏 − 𝑡 𝑣 𝜏 𝑑𝜏
✦ Normalized cross-correlation
0(2)
✦ 𝑧 𝑡 =
4546
8
Signal matching: Coherence
Cross spectral density
=
𝐺89(<) Fourier transform of
𝐶89 𝑓 = the cross-
Coherence 𝐺88(<) 𝐺99(<) correlation function
✦ 0 ≤ Coh2(f) ≤ 1
9
Signal matching: Coherence
=
𝐺89(<)
𝐶89 𝑓 =
𝐺88(<) 𝐺99(<)
✦ Example:
10
Coherence as a Measures of Connectivity
11
Yazdan-Shahmorad and Silversmith et al., eLife, 2018
Current source density analysis
✦ Density of current entering or leaving the extracellular
medium
✦ The second derivative approximation of the field potential
between two adjacent electrodes
12
https://www.semanticscholar.org/
Current source density: LFP
13
Kajikawa & Schroeder, Neuron, 2011
Current source density: EEG
14
Tenke & Kayser, Clinical Neurophysiology, 2005
Five minute break
15
Time frequency analysis
16
Time-frequency analysis: Spectrogram
17
http://ssp-iiith.vlabs.ac.in/
Example: Spectrogram
18
Yazdan-Shahmorad et al, J Neur Eng, 2013
Wavelets versus Spectrograms
20
Samiee & Baillet, NeuroImage, 2017
Example 1: Phase-Amplitude Coupling
Exaggerated phase–amplitude coupling in Parkinson disease:
Gamma in motor cortex coupled to Beta in thalamus
22
He et al, 2019, J Cerebral Blood Flow and Metabolism
Quiz: Which of the following statements is
correct?
✦ A. Coherence can be used as a measure of connectivity.
23
Inducing Oscillations using optogenetics
High Gamma (60-200 Hz) Gamma (30-60 Hz) Beta (12-30 Hz)
10Hz
20Hz
30Hz
35Hz
40Hz
50Hz
70Hz
100Hz
150Hz
24
Yazdan-Shahmorad et al., IEEE EMBC, 2018
Inducing Phase-Amplitude Coupling
A B
A A 1 B B Stim Location 1 Stim Location 2
1 1 StimStim
Location 1 1
Location StimStim
Location 2 2
Location
2 High Gamma
2 2 High Gamma
(60-200
High Hz)
Gamma
(60-200 Hz)
(60-200 Hz)
Beta
M1 S1 Beta Beta
(12-30 Hz)
M1 M1S1 S1 (12-30(12-30
Hz) Hz)
Laser 1 3-200 Hz
Laser
LaserLaser
1 2 3-2003-200
Hz Hz
1 100 μV
LaserLaser
2 2 100 μV100 μV50 ms
50 ms50 ms
Index
1200
/ Hz
/ Hz
/ Hz
150 150 150
Modulation Index
1200 1200
Index
Amp freq / Hz
Amp freq / Hz
Amp freq / Hz
/ Hz
/ Hz
/ Hz
150 150 150 150 150 150
freq
freq
freq
Modulation
freq
freq
freq
100 100 100
Modulation
Amp
Amp
Amp
100 100 100 100 100 100
Amp
Amp
Amp
50 50 50
50 50 50 0
50 50 50
0 0
50 100 150 50 100 150 50 100 150
50 50 100 100 150
Phase freq150
/ Hz 50 100 100
50Phase 150 / 150
freq Hz 50 50 100
Phase 150 150
100freq / Hz
Phase freq freq
Phase / Hz / Hz Phase freq freq
Phase / Hz / Hz Phase freq freq
Phase / Hz / Hz 25
Yazdan-Shahmorad et al., IEEE EMBC, 2018
Feature extraction
✦ Frequency analysis (e.g. Power spectrum)
26
Hand kinematics from neural recording
27
Discussion: Hand kinematics from neural
recording
✦ How to implement brain control?
https://news.brown.edu 28
Decoding movement
✦ population vector algorithm
✦ linear filtering
✦ probabilistic methods
29
Kalman filter
✦ Optimal estimation algorithm
30
Kalman filter
✦ Named after Rudolf Kalman (1930-2016)
✦ Common applications:
➭ Guidance and navigation systems
➭ Computer vision
➭ Signal processing
✦ First application?
Estimate the
trajectories of the
spacecraft
31
Apollo Project
Kalman filter
Hand kinematics
(xk) +
-
Mathematical 𝑧>?
model
𝑥>?
Kalmen
filter
32
Kalman filter
✦ Estimate the state of the hand at the current time: xk
Matrix that
linearly relates the
hand state to the
neural data Coefficient matrix
Kalman filter model from training data using least squares estimation
33
Wu et al, NIPS, 2003
Kalman filter
Matrix that
linearly relates the
hand state to the
neural data Coefficient matrix
Two-step decoding:
1- prediction of the a priori state estimate
2- updating this estimate with new measurement data to produce a posteriori state estimate
34
Wu et al, NIPS, 2003
Kalman filter implementation
35