Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Scalar features used to identify signals in the Time and Frequency domain

Introduction
Signal processing entails the use of techniques that enhance our comprehension of the
information contained in the signal. Most signals are often analyzed in the time domain,
where the time (horizontal axis) is plotted against the amplitude (vertical axis).
However, signals can also be analyzed in the frequency domain with sinusoids. In this
case, when combined together, these sinusoids provide a waveform that is equal to the
original. Then analysis is performed on the amplitudes, frequencies, and phases of the
sinusoids in relation to one another.
When analysis of a signal employs a unit of time, such as seconds or one of its multiples
(minutes or hours), as a unit of measurement, then that analysis is in the time domain.
On the other hand, if an investigation involves units such as Hertz, then it is being
conducted in the frequency domain. This paper looks at the scalar features used in the
Time and Frequency domain to analyze signals.
1. Time Domain
The time-domain representation of a signal shows the signal’s start and end time,
intensity, temporal evolution, and energy distribution along time. In this domain, a
dynamic system’s input response is a function of time, and it is possible to calculate the
system’s time response if the nature of the input and mathematical model are available.
Scalar features in the time domain

a. Mean
The mean is the most common characteristic of a wave. It calculates the mean
value of the amplitude throughout the whole duration of the signal. (Peng, Bi,
Xue, Zhang, & Wan, 2021) Let 𝑥𝑖 be the instantaneous value of the amplitude
of signal N the number of amplitude during the duration under study, we have
𝑁
1
𝑥̅ = 𝜇 = ∑ 𝑥𝑖
𝑁
𝑖=1
b. Median
The median is the middle value of the amplitude during the duration of the
signal (Tahir, Badshah, Hussain, & Khattak, 2018)
𝑁+1
𝑀𝑎𝑔𝑛𝑖𝑡𝑢𝑑𝑒( )
2
c. Range
The is a spread of the amplitude of a signal from the lowest to the highest value
(Tahir, Badshah, Hussain, & Khattak, 2018)

𝑅𝑎𝑛𝑔𝑒 = 𝑀𝑎𝑥(𝑥𝑖 ) − 𝑀𝑖𝑛(𝑥𝑖 )

d. Peak Value (Maximum Value) and Minimum Value


This is the maximum and minimum value of the amplitude of a signal
(Jiang, Cao, Yin, & Deng, 2013) (Peng, Bi, Xue, Zhang, & Wan, 2021)

1
𝑥𝑝 = 𝑚𝑎𝑥(|𝑥𝑖 |) 𝑎𝑛𝑑 𝑥𝑚𝑖𝑛 = 𝑚𝑖𝑛(|𝑥𝑖 |)

e. Peak to Peak
The peak-to-peak is the difference between the maximum positive and
maximum negative amplitude of a signal. (Peng, Bi, Xue, Zhang, & Wan, 2021)

𝑃𝑒𝑎𝑘 𝑡𝑜 𝑃𝑒𝑎𝑘 = 𝑀𝑎𝑥(|𝑥𝑖 |) − 𝑀𝑖𝑛(|𝑥𝑖 |)

f. Amplitude Square
This is the sum of the square of the instantaneous signal values (Jiang, Cao,
Yin, & Deng, 2013)
𝑛

𝑥𝑒 = ∑ 𝑥𝑖2
𝑖=1

g. Root Mean Square (RMS)


A signal’s root-mean-square (RMS) value may be determined by taking the
square root of the average value of its squared instantaneous signal values.
(LRM, et al., 2012)
𝑁
1
𝑥𝑟𝑚𝑠 = √ ∑ 𝑥𝑖2
𝑁
𝑖=1

h. Variance
It is computed by adding up the squares of the differences between a signal’s
observed value and the signal’s average value. It is a measure of statistical
dispersion that indicates the degree of variability that may be found under
certain circumstances. (LRM, et al., 2012)

𝑁
(𝑥𝑖 − 𝑥̿ )2
𝑉=∑
𝑁−1
𝑖=1
𝑥̿ = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎 𝑠𝑖𝑔𝑛𝑎𝑙
i. Standard Deviation
The Standard Deviation is gotten from the square root of the Variance.
𝑁
(𝑥𝑖 − 𝑥̿ )2
𝜎 = √∑
𝑁−1
𝑖=1

j. Skewness
Its objective is to check and calculate the symmetry of the data, hence revealing
the likelihood of variable distribution. (LRM, et al., 2012). It measures a signal’s
asymmetry or its third-order cumulative. (Altın & Er, 2016)

2
∑𝑁
𝑖=1(𝑥𝑖 − 𝑥̿ )
3
𝑆𝑘𝑒𝑤 =
(𝑁 − 1)𝜎 3
k. Kurtosis
The kurtosis measures the degree to which a distribution is flattened and
investigates whether or not it is more tapered or flattened in comparison to the
pattern that is considered normal. When the kurtosis is high, there are a larger
number of values that are significantly different from the mean. (LRM, et al.,
2012). It can also be thought of as a measure of the peakness of a probability
distribution or the cumulative measure of the fourth-order. (Altın & Er, 2016)
∑(𝑥𝑖 − 𝑥̿ )4
𝐶=
𝜎4
l. Skewness index
(Peng, Bi, Xue, Zhang, & Wan, 2021)
𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠
𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠 𝑖𝑛𝑑𝑒𝑥 =
(√𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒)3
m. Kurtosis index
(Peng, Bi, Xue, Zhang, & Wan, 2021)
𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠
𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠 𝑖𝑛𝑑𝑒𝑥 =
(√𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒)2

n. Zero Crossing (ZC)


This is the number of times the waveform becomes negative (crosses zero). In
order to reduce the amount of noise that is generated, a threshold (𝜀) is
introduced (LRM, et al., 2012).
When given two successive samples, the zero crossings, ZC, will increase if the
following conditions are met:
{𝑥𝑖 > 0 𝑎𝑛𝑑 𝑥𝑖+1 < 1} 𝑜𝑟 {𝑥𝑖 < 0 𝑎𝑛𝑑 𝑥𝑖+1 > 0 } 𝑎𝑛𝑑 |𝑥𝑖 − 𝑥𝑖+1 | ≥ 𝜀
The Zero crossing can be calculated as follows
𝑁

𝑍𝐶 = ∑[𝑠𝑔𝑛(𝑥𝑖 ∗ 𝑥𝑖+1 ) ∩ |𝑥𝑖 − 𝑥𝑖+1 | ≥ 𝜀]


𝑖=1
1, 𝑖𝑓 𝑥 ≥ 𝜀
𝑠𝑔𝑛(𝑥) = { }
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
(Altın & Er, 2016)

o. Root Amplitude( Square root amplitude)


This is calculated as shown below (Jiang, Cao, Yin, & Deng, 2013)
𝑁 2
1
𝑥𝑟 = ( ∑ 2√|𝑥𝑖 |)
𝑁
𝑖=1
p. Crest Factor ( Peak factor)
The crest factor shows the ratio between the peak values to the effective value
(Tahir, Badshah, Hussain, & Khattak, 2018)
𝑀𝑎𝑥(|𝑥|)
𝐶𝑟𝑒𝑠𝑡 𝐹𝑎𝑐𝑡𝑜𝑟(𝐶𝐹) =
𝑥𝑟𝑚𝑠
q. Shape Factor

3
It is the ratio between the RMS and the mean of the absolute value. It is
dependent on the signal shape while being independent of the signal dimensions.
(Tahir, Badshah, Hussain, & Khattak, 2018)

𝑥𝑟𝑚𝑠
𝑆ℎ𝑎𝑝𝑒 𝐹𝑎𝑐𝑡𝑜𝑟 =
1 𝑁

𝑁 𝑖=1 |𝑥𝑖 |

r. Pulse Indicator(impulse factor)


It is the ratio between the height of the peak and the mean level of the signal
(Jiang, Cao, Yin, & Deng, 2013)
𝑥𝑝
𝐼𝐹 =
𝑥̅

s. Margin factor (Crest factor)


It is the ratio of the height of a peak and the mean level of the signal
(Jiang, Cao, Yin, & Deng, 2013)
𝑥𝑝
𝐶𝐼𝐹 =
𝑥𝑟
t. Clearance factor
It is the ratio of the peak value and the squared mean value of the square roots
of the absolute amplitudes
𝑥𝑝
𝑥𝑐𝑙𝑒𝑎𝑟 =
1
( 𝑁 ∑𝑁𝑖=1 √|𝑥𝑖 |)
2

u. Waveform Length
The length of the waveform is used to measure the complexity of a signal and
is defined as the cumulative length of the waveform measured across the time
segment (Altın & Er, 2016)

𝑁−1

𝑊𝐿 = ∑ |𝑥𝑖+1 − 𝑥𝑖 |
𝑖=1
v. Willison Amplitude (WAMP)
Willison amplitude is defined as the number of times the difference in amplitude
between two consecutive samples is greater than a predetermined threshold
value.
𝑁−1

𝑊𝐴𝑀𝑃 = ∑[𝑓(|𝑥𝑖+1 − 𝑥𝑖 |)]


𝑖=1

1, 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
𝑓(𝑥) = { }
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

w. Slope Sign Change (slopeSign)


The number of times that the slope of a waveform changes sign during an
analysis window is defined as the slope sign change. It defines the frequency’s

4
signal. A count threshold is often used to reduce noise-induced counts. (Tkach,
Huang, & Kuiken, 2010)

{𝑥𝑖 > 𝑥𝑖−1 𝑎𝑛𝑑 𝑥𝑖 > 𝑥𝑖+1 } 𝑜𝑟 {𝑥𝑖 < 𝑥𝑖−1 𝑎𝑛𝑑 𝑥𝑖 < 𝑥𝑖+1 } 𝑎𝑛𝑑 |𝑥𝑖 − 𝑥𝑖+1 |
≥ 𝜀 𝑜𝑟 |𝑥𝑖 − 𝑥𝑖−1 | ≥ 𝜀
x. Entropy
Entropy is used to measure the level of impureness associated with a random
variable. The entropy of a discrete variable 𝑋{𝑥1 , 𝑥2 … … … . . 𝑥𝑛 } with
probability mass function 𝑝(𝑋) is given by: (Esmael, Arnaout, Fruhwirth, &
Thonhauser, 2013)
𝑚

𝐻(𝑋) = − ∑ 𝑃(𝑥𝑖 ). log 2 𝑃(𝑥𝑖 )


𝑖=1

2. Frequency Domain
The frequency domain pertains to the analytical space in which signals are expressed
in terms of frequency instead of time. For instance, whereas a graph in the time domain
may depict changes over time, a graph in the frequency domain shows the proportion
of the signal that is present within each frequency band. Information from the time
domain can be converted to a frequency domain using a Fourier transformation. The
time function is transformed into a collection of sine waves, each of which represents a
distinct frequency. The representation of the frequency signal found in the frequency
domain is called the spectrum.
In most cases, the Fast Fourier Transform is used to carry out frequency domain
analysis, which is based on power spectral estimation. (Yadav, Yadav, & Prachi, 2013)
The frequency spectrum (spectral/frequency components) of a signal provides
information about the frequencies that are present in the signal.

The frequency domain makes it possible to express not only the qualitative behavior of
a system but also the characteristics of the system’s reaction to changes in phase shift,
gain, bandwidth harmonics, etc.

Features of Frequency of Domain


a. Mean Frequency( Spectral centroid)
The mean frequency, also known as an average frequency, is computed by

5
taking the product of the power spectrum 𝑷𝒊 at bin i and the frequency 𝒇𝒊 at bin
i and dividing that result by the entire sum of the spectrum’s intensity. (Altın &
Er, 2016)
∑𝑵𝒊=𝟏 𝒇𝒊 𝑷𝒊
𝑴𝑵𝑭 = 𝝁𝟏 = 𝑵
∑𝒊=𝟏 𝑷𝒊
Where N is the length of the frequencies
b. Median frequency
The median frequency is the frequency at which the spectrum can be split into
two sections with the same amplitude.
(Altın & Er, 2016)
𝑀𝐷𝐹 𝑀 𝑀
1
∑ 𝑃𝑖 = ∑ 𝑃𝑖 = ∑ 𝑃𝑖
2
𝑖=1 𝑖=𝑀𝐷𝐹 𝑖=1
c. Total Power of the Signal
The total power of the signal is calculated using the Fourier transform’s
output 𝑷𝒊 , which is the power spectrum (LRM, et al., 2012)
𝑵

𝑷𝒕𝒐𝒕𝒂𝒍 = ∑ 𝑷𝒊
𝒊=𝟏

d. The Frequency at 20%, 50%,80%, 95%

Frequency of 20%: It is the frequency for which 20 percent of the total power
of 𝑷𝒊 (𝑷𝟐𝟎). is below it (LRM, et al., 2012)
𝑁

(𝑷𝟐𝟎) = ∑ 𝑷𝒊 (𝒊 ∗ 𝚫𝒇) ≤ 𝟎. 𝟐𝟎 ∗ 𝑷𝒕𝒐𝒕𝒂𝒍


𝑖=1
Frequency of 50%: 50% is the median frequency, which divides the spectrum
into two equal parts (LRM, et al., 2012)
𝑁

(𝑷𝟓𝟎) = ∑ 𝑷𝒋 (𝒊 ∗ 𝚫𝒇) ≤ 𝟎. 𝟓𝟎 ∗ 𝑷𝒕𝒐𝒕𝒂𝒍


𝑖=1
Frequency of 80%: It is the frequency for which 80 percent of the total power
of 𝑷𝒋 (𝑷𝟖𝟎). is below it (LRM, et al., 2012)
𝑁

(𝑷𝟖𝟎) = ∑ 𝑷𝒋 (𝒊 ∗ 𝚫𝒇) ≤ 𝟎. 𝟖𝟎 ∗ 𝑷𝒕𝒐𝒕𝒂𝒍


𝑖=1

Frequency of 95%: It is the frequency for which 95 percent of the total power
of 𝑷𝒋 (𝑷𝟗𝟓). is below it (LRM, et al., 2012)
𝑁

(𝑷𝟗𝟓) = ∑ 𝑷𝒋 (𝒊 ∗ 𝚫𝒇) ≤ 𝟎. 𝟗𝟓 ∗ 𝑷𝒕𝒐𝒕𝒂𝒍


𝑖=1

e. Root Mean Square frequency


(Wang, Zheng, Zhao, & Wang, 2015)

6
𝟐
∑𝑵
𝒊=𝟏 𝒇𝒊 𝑷𝒊
𝟏
𝑭𝒓𝒎𝒔𝒇 = ( 𝑵 )𝟐
∑𝒊=𝟏 𝑷𝒊
f. Root Variance frequency (Spectral spread)
It is the standard deviation around the spectral centroid (Mathworks, 2022)
(Wang, Zheng, Zhao, & Wang, 2015)
∑𝑵 𝟐
𝒊=𝟏(𝒇𝒊 − 𝑭𝒎𝒇 ) 𝑷𝒊 𝟏
𝑭𝒓𝒗𝒇 = 𝝁𝟐 = ( )𝟐
∑𝑵𝒊=𝟏 𝑷𝒊

𝑷
With 𝑭𝒎𝒇 = ∑𝑵
𝒊=𝟏 𝑵
𝒊

g. Power spectrum Deformation


The Power Spectrum Deformation ratio is sensitive to variations in spectral
symmetry and is an indicator of spectral deformation. It also offers a measure
of power spectrum distortion. (Altın & Er, 2016)
𝑀
√ 2
𝑀0
Ω= 𝑤𝑖𝑡ℎ 𝑀𝑛 𝑏𝑒𝑖𝑛𝑔 𝑡ℎ𝑒 𝑛𝑡ℎ 𝑠𝑝𝑒𝑐𝑡𝑟𝑎𝑙 𝑚𝑜𝑚𝑒𝑛𝑡.
𝑀1
𝑀0
𝑖𝑚𝑎𝑥
𝑀𝑛 = ∑𝑖=0 𝑃𝑖 𝑓𝑖 where 𝑃𝑖 is the power spectral density at the frequency 𝑓𝑖𝑛
𝑛

h. Spectral Skewness ( skewness of band power)


Spectral skewness is a measure of the degree of symmetry around the centroid.
It is a third-moment order (Mathworks, 2022)

∑𝑁 3
𝑖=1(𝒇𝒊 −𝝁𝟏 ) 𝑃𝑖
.𝜇3 = 𝝁𝟐 3 ∑𝑁
𝑖=1 𝑷𝒊

i. Spectral Kurtosis ( Kurtosis of band power)


The spectral kurtosis measures the flatness of the spectrum around the centroid.
It indicates the peakiness of the spectrum (Mathworks, 2022)
∑𝑁 4
𝑖=1(𝒇𝒊 −𝝁𝟏 ) 𝑃𝑖
.𝜇4 = 𝝁𝟐 4 ∑𝑁
𝑖=1 𝑷𝒊

j. Spectral crest (Relative Spectral peak per band)


It is the ratio of the maximum spectrum to the arithmetic mean of the spectrum
(Mathworks, 2022)
max (𝑃𝑖 )
𝐶𝑟𝑒𝑠𝑡 =
1 𝑁

𝑁 𝑖=1 𝑷𝒊

Conclusion
Time and Frequency domains play a vital role in the analysis of signals as signals can
be converted between the frequency and the time domain. When it comes to areas of
application, The time domain is used to investigate the transient response of a system
which also makes it possible to get a deeper comprehension of the flow of both

7
mechanical and electrical energy. This includes phenomena like wave propagation, the
structural changes that occur inside a system, and the generation of electric potential.
On the other hand, the frequency domain is often used as a visualization tool, as is the
case of a spectrum analyzer. Additionally, the use of transforms in certain signal
processing methods leads to the production of a time-frequency domain. The choice as
to using either the Time domain or the Frequency domain is dependent on the type of
signal and its use. However, the use of both methods instead of just one yields more
insight when processing signals.

8
References
Altın, C., & Er, O. (2016). Comparison of Different Time and Frequency Domain Feature
Extraction Methods on Elbow Gesture’s EMG. European Journal of Interdisciplinary
Studies, Volume 2, Issue 3 .
Cerna, M., & Harvey, A. F. (2000). The Fundamentals of FFT-Based Signal Analysis and
Measurement . National Instruments, Application Note 041.
Esmael, B., Arnaout, A., Fruhwirth, R. K., & Thonhauser, G. (2013). A Statistical Feature-
Based Approach for Operations Recognition in Drilling Time Series . International
Journal of Computer Information Systems and Industrial Management Applications,
ISSN 2150-7988 Volume 5 pp. 454-461.
Jiang, L.-L., Cao, Y.-X., Yin, H.-K., & Deng, K.-S. (2013). An Improved Kernel K-Mean
Cluster Method and Its Application in Fault Diagnosis of Roller Bearing. Scientific
Research, Engineering, 2013, 5, 44-49.
Kossakowska, J., Bombinski, S., & Ejsmont, K. (2021). Analysis of the Suitability of Signal
Features for Individual Sensor Types in the Diagnosis of Gradual Tool Wear in Turning.
Energies, 14, 6489 https://doi.org/10.3390/en14206489.
LRM, P., AA, P., MFS, A., GL, C., ST, M., & AO, A. (2012, June). Analysis of the relationship
between EEG signal and aging through Linear Discriminant Analysis (LDA. Brazilian
Journal of Biomedical Engineering, pp. Volume 28, Número 2, p. 155-168, 2012.
http://dx.doi.org/10.4322/rbeb.2012.023.
Mathworks. (2022, 05 16). Spectral Descriptors. Retrieved from MathsWorks.com:
https://www.mathworks.com/help/audio/ug/spectral-descriptors.html
Peng, B., Bi, Y., Xue, B., Zhang, M., & Wan, S. (2021). Multi-View Feature Construction
Using Genetic Programming for Rolling Bearing Fault Diagnosis. IEEE
COMPUTATIONAL INTELLIGENCE MAGAZINE, 79-94.
Tahir, M. M., Badshah, S., Hussain, A., & Khattak, M. A. (2018). Extracting accurate time
domain features from vibration signals for reliable classification of bearing faults.
International Journal of Advanced and Applied Sciences, 156-163.
Tkach, D., Huang, H., & Kuiken, T. A. (2010). Study of stability of time-domain features for
electromyographic pattern recognition. Journal of NeuroEngineering and
Rehabilitation, Article number: 21 (2010.
Wang, X., Zheng, Y., Zhao, Z., & Wang, J. (2015). Bearing Fault Diagnosis Based on Statistical
Locally Linear Embedding. B. Sensors (Basel, Switzerland)., 15. 16225-47.
10.3390/s150716225. .
Yadav, G. S., Yadav, S., & Prachi. (2013). Time and Frequency Exploration of ECG Signal.
International Journal of Computer Applications, Volume 67– No.4, April 2013.

You might also like