Signal & Data Analysis in Neuroscience

2024
Filters & Windows

Izhar Bar-Gad
Room: 408 Phone: 7141 Email: [email protected]

Outline – Frequency Domain

 Introduction
 Fourier Transform
 Sampling Theory

 Systems

 Filters

 Windows

 Spectral Analysis

Filter in signal analysis

 Filters are used to eliminate unwanted

frequencies from an input signal or to select a
desired frequency range among many others.

 Filters form a sub-group of the systems.

 There are analog and digital filters. We will

focus on the digital filters.

1
3
 Filters

|X(ω)|
Original frequency domain
ω

|X(ω)|

 Low Pass Filter (LPF)

ω
|X(ω)|
 High Pass Filter (HPF)
ω
|X(ω)|
 Band Pass Filter (BPF)
ω
|X(ω)|
 Band Stop Filter (BSF)
ω

Input – Filter – Output

 Given the input X(ω) and desired output

Y(ω), how can we describe H(ω) ?

 Moreover, how can we transfer it to the

time domain and get a representation of
the three components?

Ideal Filtering

Since Y(ω)=X(ω)∙H(ω)

H(ω) X(ω)
|H(ω)| |X(ω)|
∙
ω ω

Y(ω)
|Y(ω)|

2 Remember : No frequency domain systems exist !!

6
 Ideals problem : F(rect)=sinc

Ideal frequency response  sinc in the time domain !

The sinc function

sin(𝑥)
 The classical “unnormalized” definition 𝑠𝑖𝑛𝑐 𝑥 =
𝑥
sin(𝜋𝑥)
 Signal analysis “normalized” definition 𝑠𝑖𝑛𝑐 𝑥 =
𝜋𝑥

sinc problem

 The sinc function is infinite, forever oscillating

around zero.

 Inserting Irrelevant samples into the filtered signal

 Since both F(rect)=sinc & F(sinc)=rect

this problem is relevant for rectangular windows in
both time and frequency domains

3
9
 Filter Design - Amplitude

(Example of a python design tool is the “pyfda” package)

10

Truncated sinc filter

11

Filter families

 There are many filter families possessing

different characteristics
 In this example a 10 pole Chebyshev &
Butterworth low-pass filters

4
12
 Number of poles

 Butterworth filter of order 1  5

13

Special case: Notch Filter

 Special case of band-stop filter.

 Remove a narrow band from the signal
 Usually used for network noise (50,60 hz)
 BEWARE:
 Drastically distorts phase, esp. when analog
 Even if not, cancels most of the energy in the narrow
band instead of just the noise part of it

14

Filter example I
Original extracellular signal

Extracellular recording, containing the summation of

different processes. Some relevant (neural origin)
and some irrelevant (non neural origin).
0.015

0.01

0.005
Voltage (mV)

-0.005

-0.01

5 -0.015
0 100 200 300 400 500
Time (ms)
600 700 800 900 1000

15
 Filter example II
Following notch filter (~1800Hz)

External (non biological noise) may have a very

narrow band. Both local field potential (LFP) and
action potentials (spikes) are apparent.

0.01

0.005

Voltage (mV)
-0.005

-0.01

-0.015
0 100 200 300 400 500 600 700 800 900 1000
Time (ms)

16

Filter example III

Following high pass (~300Hz)

Removal of the high frequency components

removes the local field potential, leaving the spiking
activity.
-3
x 10
5

0
Voltage (mV)

-5

-10
0 100 200 300 400 500 600 700 800 900 1000
Time (ms)

17

Filter Design - Phase

LTI system: H ( )  A( )ei ( )

Therefore, in order to achieve constant delay

(that all frequencies will be delayed by the
same phase) we would like to have a linear
phase response.
[or at least at the relevant frequency range]
6
18
 Semi-Log plot

H(ω) 50

Magnitude (dB)
0

-50

-100

-150
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Frequency ( rad/sample)

5000

Phase (degrees)
0

-5000

-10000

-15000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Frequency ( rad/sample)

matlab: freqz

19

Phase distortion
Extracellular recording

(Yael & Bar-Gad., 2017)

20

Phase distortion
Linear vs. Non-linear phase

7
21
 Filters
Avoiding sampling problems

when X(ω)  X[k] looks like this,

You have an aliasing problem…
Anti-Aliasing filter
What is there to do?
X(ω)
And if we can not increase ωs ?

X(ω)
cut

X[k]

22

Online vs. Offline Filtering

 Online filtering –
 Live streaming (usually analog) data
 Delay limit
 Limited computational resources

 Offline filtering –
 Sampled, stored, digital data
 No time limit
 Less limited computational resources

23

Analog Filters

RC Circuit = the simplest LPF

8
24
 Filters in Neuroscience

 Cell Membrane as LPF

 Slice Bath as LPF

25

2D Fourier transform

 Continuous 2D transform

 Discrete 2D transform

(http://fourier.eng.hmc.edu/e101/lectures/Image_Processing/Image_Processing.html)

26

2D Fourier transform

9
27
 Spatial filtering

 Spatial featufiltersres may be applied on

2D (or even 3D) images.
(https://medium.com/@hicraigchen/digital-image-processing-using-fourier-transform-in-python-bcb49424fd82)

Original Low pass High pass

(from OpenCV github)

28

Spatial filtering

Ideal Butterworth
filter filter

l Low pass

High pass

(from OpenCV github)

29

Outline – Frequency Domain

 Introduction
 Fourier Transform

 Sampling Theory

 Systems

 Filters

 Windows

 Spectral Analysis

10
30
 sinc: time & frequency duality

Time Domain Frequency Domain

31

Finite sinc

32

Ideal Filter Example

2 2
signal  sin( t )  sin(10 * t)
1000 1000
2

1 . 5

0 . 5

- 0 . 5

- 1

- 1 . 5

- 2
0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0 2 5 0 0 3 0 0 0 3 5 0 0 4 0 0 0 4 5 0 0 5 0 0 0

Convolution with a rectangular window

result = conv(ones(1,120), signal)
1 5 0

1 0 0

5 0

- 5 0

11 - 1

- 1
0

5
0

0
0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0 2 5 0 0 3 0 0 0 3 5 0 0 4 0 0 0 4 5 0 0 5 0 0 0

33
 Windows

34

Windows frequency response

frequency time
Triangular window

W[n]=

-M ≤ n ≤ M
Hanning window

Hamming window

35

Leakage windows I

12
36
 Leakage windows II

37

Windows and filters

38

DFT Frequencies

 Discrete 2k
j n 2
 Range : 0-2, e N
 k 
N
 Periodic

But…
What about frequencies in between?

13
39
 DFT Leakage

Say we had a sinusoid of 2.3 Khz and we still

sampled it at 8000 samples/s :

The output frequency bins = 0Kkz, 1Khz, …, 7Khz

2.3 Khz bin is missing!!

What happens is that this frequency component

shows up (leaks) in all the other frequency bins

Remedy
Windowing Techniques

40

14