Scribd
Upload
3 views8 pages

FT Applications - Spectrogram

The document discusses the Fourier Transform (FT) and its applications in creating spectrograms and signal fingerprints for identifying and comparing signals, such as voices or ECG patterns. It highlights challenges with signal length affecting feature representation and suggests solutions like segmenting signals into manageable chunks and ignoring phase information. The resulting spectrogram provides a visual, compact, and structured representation of signal features, making it easier to analyze and identify them.

Uploaded by

mohamed.mohamed056
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views8 pages

FT Applications - Spectrogram

The document discusses the Fourier Transform (FT) and its applications in creating spectrograms and signal fingerprints for identifying and comparing signals, such as voices or ECG patterns. It highlights challenges with signal length affecting feature representation and suggests solutions like segmenting signals into manageable chunks and ignoring phase information. The resulting spectrogram provides a visual, compact, and structured representation of signal features, making it easier to analyze and identify them.

Uploaded by

mohamed.mohamed056
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Fourier Transform

Applications
A. Spectrogram & Fingerprints
A feature that leads to an application
• Feature:
• FT decomposes the signals into some basic components (coss, sins).

• Application idea:
• So, we can use these features to:
• Compare between two signals
• Automatically identify one signal from others based on specific features, for example:
• Identify a person’s voice from other persons.
• Identify a song from other songs.
• Identify an arrhythmia ECG signal from other normal ECG signals.
• This can basically be called a “fingerprint” for the signal.
Problems with the nice dreamy idea 
• Too short signals are usually not enough to represents all the features
in the signal.

• In too long signals:


• Strong features might be repeated and may smooth out due to the phase
cancellation (destructive summation).
• imagine two parts at the same frequency but with 180 phase shift difference)
• Weaker features might be repeated and start to be comparable to the strong
ones.
• All features will eventually show up and become difficult to really identify
signal from another.

Do NOT forget: when we say “feature”, we mean “frequency component”


Solutions??
• Cut the signal into reasonable chunks that are:
• Not too short
• Not too long
• After calculating FT, ignore the phase to avoid any constructive or
destructive effect.

Example: Some ECG long


signals split into five
equal-length chunks
courtesy: [Link]
Then?
• So far, we have small chunks of signal FT
stripped out of its phase, right?
• Stack these chunks together, replace the
magnitude dimension by color value
(either gray scale or RGB values).

→ A nice colorful image


called the “Spectrogram”
of the signal.
Any issues/concerns:
• Stack the FT lines vertically or horizontally?
• Never mind. Choose something and stick to it.
• People get used to see the horizontal axis as the time so let it be ☺

• Very dependent on the picked chunk-length?


• With some experience, people start to have recommendations for each signal
source/type☺

• Where is the phase?! We lost a lot of information


• Yes! But most of it is about the dynamics of the signal (how different
frequencies are shifted through time), not the signal features (the magnitude
of each frequency)! The dynamic is surely important (even more important
than magnitude).
• Spectrogram is irreversible! We can retrieve the signal back!
So what?
• Now, we have another signal representation
that is:

• Visual (i.e. nicer to look at and comprehend


compared to 1D FT).

• Compact (i.e. you can look at the whole signal at


once compared to the very long old 1D FT of the
long signal).

• Has structure of features!


• 1D FT has the features but spectrogram structure
them!
• Looks more tempting/convenient/easier to localize
and identify them!

You might also like