Scribd
Upload
44 views14 pages

Chapter 10 Dot Product and Convolution Correlation

This document discusses the dot product and convolution. It defines the dot product as the sum of elements in one vector weighted by elements of another vector of equal length. Convolution is defined as an extension of the dot product where the dot product is computed repeatedly over time by shifting one vector relative to the other. Convolution can be used to analyze time-frequency properties of signals like EEG data by convolving the signal with wavelets at different frequency bins to reveal time-specific activity at each frequency.

Uploaded by

Arunmozhli
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
Download as pdf or txt
44 views14 pages

Chapter 10 Dot Product and Convolution Correlation

This document discusses the dot product and convolution. It defines the dot product as the sum of elements in one vector weighted by elements of another vector of equal length. Convolution is defined as an extension of the dot product where the dot product is computed repeatedly over time by shifting one vector relative to the other. Convolution can be used to analyze time-frequency properties of signals like EEG data by convolving the signal with wavelets at different frequency bins to reveal time-specific activity at each frequency.

Uploaded by

Arunmozhli
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
Download as pdf or txt
Download as pdf or txt
You are on page 1/ 14

THE DOT PRODUCT AND

CONVOLUTION
MICHAEL GOLDSTEIN
P S Y 6 9 6 B – N E U R A L T I M E S E R I E S A N A LY S I S
SPRING 2014
THE EQUATION

dotproductab

THE INTERPRETATIONS

•  Signal-processing: “sum of elements in one vector


weighted by elements of another vector”

•  Statistics: “covariance or similarity between two


vectors”

•  Geometry: “mapping between vectors (product of


the magnitudes of the two vectors scaled by the
cosine of the angle between them”

•  In any case… two vectors of equal length


GEOMETRY VISUALIZATION
MATLAB (YAYYYY!)
X
TIME-FREQUENCY (SQUIGGLY LINE) ANALYSIS
DEFINITION

•  “Extension of the dot product, in which the dot


product is computed repeatedly over time”

•  Algorithm: “compute the dot product between two


vectors, shift one vector in time relative to the other
vector, compute the dot product again, and so
on.”

•  Terminology (a la MXC):
•  Signal = EEG data
•  Kernel = wavelet or sine wave
THE INTERPRETATIONS

•  Signal-processing: “time series of one signal


weighted by another signal that slides along the first
signal”

•  Statistics: “cross-variance (similarity between two


vectors over time”

•  Geometry: “time series of mappings between two


vectors”

•  Other: “frequency filter”


HOW TO CONVOLVE: BASIC

1.  Create a kernel (e.g. 1Hz


sine wave)
2.  Flip kernel backwards
(why?)
3.  Compute dot product at
beginning of signal
4.  Move across the signal,
computing dot products
along the way
HOW TO CONVOLVE: ADVANCED

1.  Create a kernel (e.g. 1Hz sine


wave)
2.  Flip kernel backwards
3.  Zero-pad signal at beginning
and end
4.  Compute dot product at
beginning of signal (yielding a
point at center of kernel)
5.  Move across the signal,
computing dot products
along the way
6.  Remove zero-padding
CONVOLUTION VS. CROSS-COVARIANCE

•  Convolution: kernel is reversed

•  Cross-correlation (cross-covariance scaled by the


variances): kernel kept in original orientation
APPLICATION TO EEG DATA ANALYSIS

•  Use wavelets consisting of a sine wave for each frequency bin


across the frequency spectrum

•  Convolution for each frequency bin provides info of bin-


specific and time-specific activity

•  In MXC words, “…it reveals when and to what extent the EEG
data contain features that look like the wavelet.”

http://www.bem.fi/book/13/13.htm
MATLAB (YAYYYY!)

You might also like