Colors of noise

This article may be confusing or unclear to readers. Please help clarify the article. There might be a discussion about this on the talk page. (December 2021) (Learn how and when to remove this message)

In

Various noise models are employed in analysis, many of which fall under the above categories.

The color names for these different types of sounds are derived from a loose analogy between the spectrum of frequencies of sound wave present in the sound (as shown in the blue diagrams) and the equivalent spectrum of light wave frequencies. That is, if the sound wave pattern of "blue noise" were translated into light waves, the resulting light would be blue, and so on.^[citation needed]

10 seconds of white noise

---

Problems playing this file? See media help.

The frequency spectrum of

Since there are an infinite number of logarithmic bands at both the low frequency (DC) and high frequency ends of the spectrum, any finite energy spectrum must have less energy than pink noise at both ends. Pink noise is the only power-law spectral density that has this property: all steeper power-law spectra are finite if integrated to the high-frequency end, and all flatter power-law spectra are finite if integrated to the DC, low-frequency limit.^[citation needed]

10 seconds of pink noise

---

Problems playing this file? See media help.

Brownian noise, also called Brown noise, is noise with a power density which decreases 6.02 dB per octave (20 dB per decade) with increasing frequency (frequency density proportional to 1/f²) over a frequency range excluding zero (DC). It is also called "red noise", with pink being between red and white.

Brownian noise can be generated with temporal integration of white noise. "Brown" noise is not named for a power spectrum that suggests the color brown; rather, the name derives from Brownian motion, also known as "random walk" or "drunkard's walk".

10 seconds of Brown noise

---

Problems playing this file? See media help.

Blue noise is also called azure noise. Blue noise's power density increases ${\displaystyle 10\log _{10}2=}$ 3.01 dB per octave with increasing frequency (density proportional to f ) over a finite frequency range.^[5] In computer graphics, the term "blue noise" is sometimes used more loosely as any noise with minimal low frequency components and no concentrated spikes in energy. This can be good noise for dithering.^[6] Retinal cells are arranged in a blue-noise-like pattern which yields good visual resolution.^[7]

Cherenkov radiation is a naturally occurring example of almost perfect blue noise, with the power density growing linearly with frequency over spectrum regions where the permeability of index of refraction of the medium are approximately constant. The exact density spectrum is given by the Frank–Tamm formula. In this case, the finiteness of the frequency range comes from the finiteness of the range over which a material can have a refractive index greater than unity. Cherenkov radiation also appears as a bright blue color, for these reasons.

10 seconds of blue noise

---

Problems playing this file? See media help.

Violet noise is also called purple noise. Violet noise's power density increases 6.02 dB per octave with increasing frequency^[8][9] "The spectral analysis shows that GPS acceleration errors seem to be violet noise processes. They are dominated by high-frequency noise." (density proportional to f ²) over a finite frequency range. It is also known as differentiated white noise, due to its being the result of the differentiation of a white noise signal.

Due to the diminished sensitivity of the human ear to high-frequency hiss and the ease with which white noise can be electronically differentiated (high-pass filtered at first order), many early adaptations of dither to digital audio used violet noise as the dither signal.^{[citation needed]}

Acoustic thermal noise of water has a violet spectrum, causing it to dominate hydrophone measurements at high frequencies.^[10] "Predictions of the thermal noise spectrum, derived from classical statistical mechanics, suggest increasing noise with frequency with a positive slope of 6.02 dB octave⁻¹." "Note that thermal noise increases at the rate of 20 dB decade⁻¹"^[11]

10 seconds of violet noise

---

Problems playing this file? See media help.

10 seconds of grey noise

---

Problems playing this file? See media help.

Velvet noise is a sparse sequence of random positive and negative impulses. Velvet noise is typically characterised by its density in taps/second. At high densities it sounds similar to white noise; however, it is perceptually "smoother".^[12] The sparse nature of velvet noise allows for efficient time-domain convolution, making velvet noise particularly useful for applications where computational resources are limited, like real-time reverberation algorithms.^[13][14] Velvet noise is also frequently used in decorrelation filters.^[15]

2 seconds of velvet noise.

---

Problems playing this file? See media help.

There are also many colors used without precise definitions (or as synonyms for formally defined colors), sometimes with multiple definitions.

• A synonym for Brownian noise, as above.^[16][17] That is, it is similar to pink noise, but with different spectral content and different relationships (i.e. 1/f for pink noise, while 1/f² for red noise, or a decrease of 6.02 dB per octave).
• In areas where terminology is used loosely, "red noise" may refer to any system where power density decreases with increasing frequency.^[18]

10 seconds of Wisniewski's version of green noise

---

Problems playing this file? See media help.

• The mid-frequency component of white noise, used in
  dithering^[19]
• Bounded Brownian noise
• Vocal spectrum noise used for testing audio circuits[20]
• Joseph S. Wisniewski writes that "green noise" is marketed by producers of ambient sound effects recordings as "the background noise of the world". It simulates the spectra of natural settings, without human-made noises. It is similar to pink noise, but has more energy in the area of 500 Hz.^[20]

• Silence
• Infrasound^[21]
• Noise with a 1/f^β spectrum, where β > 2. This formula is used to model the frequency of natural disasters.^{[22][clarification needed]}
• Noise that has a frequency spectrum of predominantly zero power level over all frequencies except for a few narrow bands or spikes. Note: An example of black noise in a facsimile transmission system is the spectrum that might be obtained when scanning a black area in which there are a few random white spots. Thus, in the time domain, a few random pulses occur while scanning.^[23]
• Noise with a spectrum corresponding to the
  violet noise. At the same time, Hawking radiation of black holes may have a peak in hearing range, so the radiation of a typical stellar black hole
  with a mass equal to 6 solar masses will have a maximum at a frequency of 604.5 Hz – this noise is similar to green noise. A formula is: ${\displaystyle f_{\text{max}}\approx 3627\times {{\text{M}}_{\odot } \over {\text{M}}}}$ Hz. Several examples of audio files with this spectrum can be found here.^{[citation needed]}

In

• In facsimile or display systems, such as television, a nonuniformity in the white area of the image, i.e., document or picture, caused by the presence of noise in the received signal.
• A signal or
  record medium
  .

In

• In facsimile or display systems, such as television, a nonuniformity in the black area of the image, i.e., document or picture, caused by the presence of noise in the received signal.
• A signal or
  record medium
  .

Colored noise can be computer-generated by first generating a white noise signal, Fourier-transforming it, then multiplying the amplitudes of the different frequency components with a frequency-dependent function.^[26] Matlab programs are available to generate power-law colored noise in one or any number of dimensions.

Identifying the dominant noise type in a time series has many applications including clock stability analysis and market forecasting. There are two algorithms based on autocorrelation functions that can identify the dominant noise type in a data set provided the noise type has a power law spectral density.

The first method for doing noise identification is based on a paper by W.J Riley and C.A Greenhall.^[27] First the lag(1) autocorrelation function is computed and checked to see if it is less than one third (which is the threshold for a stationary process):

${\displaystyle R_{1}={\frac {{\frac {1}{N}}\sum _{t=1}^{N-1}(z_{t}-{\bar {z}})*(z_{t+1}-{\bar {z}})}{{\frac {1}{N}}\sum _{t=1}^{N}(z_{t}-{\bar {z}})}}}$

where ${\displaystyle N}$ is the number of data points in the time series, ${\displaystyle z_{t}}$ are the phase or frequency values, and ${\displaystyle {\bar {z}}}$ is the average value of the time series. If used for clock stability analysis, the ${\displaystyle z_{t}}$ values are the non-overlapped (or binned) averages of the original frequency or phase array for some averaging time and factor. Now discrete-time fractionally integrated noises have power spectral densities of the form ${\displaystyle (2sin(\pi f))^{-2\delta }}$ which are stationary for ${\displaystyle \delta <.25}$. The value of ${\displaystyle \delta }$ is calculated using ${\displaystyle R_{1}}$:

${\displaystyle \delta ={\frac {R_{1}}{1+R_{1}}}}$

where ${\displaystyle R_{1}}$ is the lag(1) autocorrelation function defined above. If ${\displaystyle \delta >.25}$ then the first differences of the adjacent time series data are taken ${\displaystyle d}$ times until ${\displaystyle \delta <.25}$. The power law for the stationary noise process is calculated from the calculated ${\displaystyle \delta }$ and the number of times the data has been differenced to achieve ${\displaystyle \delta <.25}$ as follows:

${\displaystyle p=-2(\delta +d)}$

where ${\displaystyle p}$ is the power of the frequency noise which can be rounded to identify the dominant noise type (for frequency data ${\displaystyle p}$ is the power of the frequency noise but for phase data the power of the frequency noise is ${\displaystyle p+2}$).

This method improves on the accuracy of the previous method and was introduced by Z. Chunlei, Z. Qi, Y. Shuhuana. Instead of using the lag(1) autocorrelation function the lag(m) correlation function is computed instead:^[28]

${\displaystyle R_{m}={\frac {{\frac {1}{N}}\sum _{t=1}^{N-m}(z_{t}-{\bar {z}})*(z_{t+m}-{\bar {z}})}{{\frac {1}{N}}\sum _{t=1}^{N}(z_{t}-{\bar {z}})}}}$

where ${\displaystyle m}$ is the "lag" or shift between the time series and the delayed version of itself. A major difference is that ${\displaystyle z_{t}}$ are now the averaged values of the original time series computed with a moving window average and averaging factor also equal to ${\displaystyle m}$. The value of ${\displaystyle \delta }$ is computed the same way as in the pervious method and ${\displaystyle \delta <.25}$ is again the criteria for a stationary process. The other major difference between this and the previous method is that the differencing used to make the time series stationary (${\displaystyle \delta <.25}$) is done between values that are spaced a distance ${\displaystyle m}$ apart:

${\displaystyle z_{1}=z_{1+m}-z_{1},z_{2}=z_{2+m}-z_{2}...,z_{N-m}=z_{N}-z_{N-m}}$

The value of the power is calculated the same as the previous method as well.

• Mains hum (also known as the AC power hum)
• Whittle likelihood