Stray signals and noise
|The physical and human spects dealing with audio, video, and images
Noise stuff: Stray signals and noise · sound-related noise names · electronic non-coupled noise names · electronic coupled noise · ground loop · strategies to avoid coupled noise · Sampling, reproduction, and transmission distortions · (tape) noise reduction
device voltage and impedance, audio and otherwise ·
amps and speakers ·
basic audio hacks ·
Simple ADCs and DACs ·
digital audio ·
multichannel and surround
For more, see Category:Audio, video, images
Noise in general
Noise can mean anything you didn't want in your signal.
It can come
- from the environment - acoustically, things or people making sounds when you don't want them to
- from the environment - electronic, e.g. capacitive and inductive coupled signals from other equipment
- from recording equipment itself
- coupled in (e.g. some cheap tape decks hear their own motors)
- originating from its own components in other ways (e.g. amplified thermal noise)
- from distortion caused by overly strong signals / too strong a gain somewhere
- from the process of sampling itself
- from sensor imperfections
- voltage transients
- from imperfections of the medium you store on (if using analog media)
- and many more places.
To mathematicians and sound engineers:
- noise tends to indicate broad-spectrum signals, and usually not periodic
- such noise is often relatively uniformly present in all the sound/data,
- because most of the mentioned effects relate to general setup and don't change very quickly
- describing noise frequency content is somewhat funny
- Say, white noise cannot be accurately modeled as a function of time,
- but can be approximated well enough in frequency content.
Some noise has a uniform and randomly-distributed nature, others comes in fixed patterns.
Fixed-pattern noise tends to be a lot more noticeable than wide-spectrum noise of the same magnitude. consider
- an image sensor with misbehaving lines, columns, and/or pixels, compared to overall noise
- or mains hum, compared to white noise.
Noise floor refers to a basic level of noise present in a system or medium, usually the sum of multiple specific sources, and often mostly unavoidable due to it combing from the physical reality of electronics, wiring, environment, or whatnot. In various devices it tends to be largely thermal noise(verify).
A noise floor often doesn't change much in amplitude over time, and many sources also give a more-or-less uniform frequency content, so this tends to be on the white noise, pink noise, brown noise side of things(verify).
The noise floor settles the smallest measurement that a system can deal with certainty/consistency (which is one reason this concept of "the combination of a bunch of noise" gets a name, and relates to dynamic range).
Generating different types of noise can be useful for certain purposes, and names (often the colors) are often mentioned in such applications.
When measuring the amount of physical noise a device produced, you would often do so in an isolated room to reduce environmental noise.
Preferably an anechoic chamber so that you may primarily measure generated noise, not the reflected-and-filtered copies of the same, which would be measuring source and measurement environment.
Babble noise is a term regularly seen in auditory science dealing that involves speech, and often describes a recording in a place with many chatty people (e.g. a restaurant, or office), or sometimes an averaged spectrum of such a recording.
Noise colors and other descriptions
Noise colors are often used when describing noise of different variations. Note that not all of these necessarily describe sound.
- Wikipedia: Colors of noise
- http://en.wikipedia.org/wiki/Federal_Standard_1037C (mentions blue noise)
- The Colors of Noise, http://www.ptpart.co.uk/colors.htm
...and various others
Well settled noise colors
These three are more consistently defined than most others, sometimes in reference documents.
White noise is that in which the spectrum is flat: the pressure level for each frequency is the same.
Brownian noise, a.k.a. brown noise, is named for being produced by Brownian motion.
Its amplitude at a frequency is proportional to 1/f2. In other words, it falls about 6dB per octave, and has a dampened, soft, bassy quality.
pink noise is also called 1/f noise (see flicker noise below).
So its falloff lies between white and red/brown/Brownian noise.
Call it 3dB/octave.
Pink noise occurs both in nature and in electronic systems (where you should also look at the narrower term flicker noise, of which not all origins are well explained yet).
Pink noise is usually generated by filtering white noise.
You could point out their relations a little mathematically like:
- white noise is 1/f0 (=1/1, i.e. constant)
- pink is 1/f1 noise, and
- brown noise is 1/f2,
Moderately settled noise colors
Blue noise (sometimes azure noise, and (more ambiguously) violet noise(verify))
Blue noise has a power density that increases 3dB per octave (density proportional to f)
Also used to refer to other noise/distributions with minimal low frequencies.
Violet noise has a power density that increases 6dB per octave (density proportional to f2)
White noise may be physically all the same frequencies, our eyes amount as a filter that lets mids through most, and little of the very high and very low.
Grey noise is often meant to be (white) noise subjected to (the inverse of) an equal loudness filter, e.g. A-weight, boosting the highs and lows so that our ears report a similar amount of all frequencies.
Most noticeably, this takes out a lot of mids, and boosts very low frequencies.
Not-so-well-defined, sometimes ambiguous noise colors
These are less official in that they are not exactly defined, and/or are defined differently by different sources.
Roughly from best to worst defined, they include:
Regularly refers to Brownian noise.
But also sometimes used refer to pink noise, though.
Making the term sort of meaningless - figure out what the author meant.
Black noise is seen referring to multiple things, usually one of:
- a few spikes in silence, as e.g. in faxes (see also impulse noise, noisy white/black)
- noise outside human-audible frequencies
- noise with faster falloff than red noise: 1/fx noise for x≥3
Noisy white, noisy black
Mostly seen in the context of analog transfer of images.
Refers to an image area that should be white, respectively black, but has enough noise to visibly not be so.
- often described as something like "a quasi-stationary noise with a finite power spectrum with a finite number of small bands of zero energy dispersed throughout a continuous spectrum."
- the zero-energy bands coincide with notes in a scale/system of choice/interest, It leaves out-of-tune frequencies, which makes this less-than-pleasing noise.
- reason for name seems to be association with sour, citrus; orange
- "Orange noise is most easily generated by a roomful of primary school students equipped with plastic soprano recorders." :)
- not very official at all, (according to wikipedia) usually one of:
USASI refers to the United States of America Standards Institute,
USASI noise noise is a white noise source (i.e. noise with equal energy at all frequencies) that is filtered by (1) a 100 Hz, 6 dB per octave high-pass network and (2) a 320 Hz, 6 dB per octave low-pass network.
- NRSC-1-A Standard (and previously in EIA-549)
A lot of electronic noise comes from other things, coupled via resistance, inductance, or capacitance -- see Electronic coupled noise for that.
There are other sources of noise,many of them imperfections of electronic components, or their real-world behaviour. Some of these are detailed below (though note some have implications on coupling as well)
For some (further) names that name specific causes, see Electronic_coupled_noise#More_specific_names
Thermal noise, a.k.a. Johnson–Nyquist noise, refers to agitation of electrons in charge carriers that happens due to whatever their temperature is.
Warmer things do this more, colder things do this less.
Basically, if it's not at 0 Kelvin, you'll get some of this.
The amount of this at room temperature is small enough to be insignificant in most situations.
It becomes relevant only sometimes, e.g. where sensors work with signals so weak that the signal is on the same rough order as room-temperature thermal noise.
For example, digital optical cameras sensors have to think about this.
The amount of thermal noise is small enough that this only matters in the dark - in a bright situation it's such a large factor below the signal from actual light that you may barely be able to store it.
But in the dark, the charge from thermal noise (and other reasons), accumulated into each pixel well over some time, is roughly comparable to the light coming in.
This is why cameras will clear any charge from their image sensors just before they start capture.
At the same time, they can't do much about the same during capture, which is part of why low-light shots are often noisy, why e.g. astrophotography sometimes bothers to cool their image sensors, and why long exposure helps a little (averaging means random noise lessens, while signal doesn't. It's diminishing returns for multiple technical reasons, though).
(Sometimes called dark noise, as in image sensors this leads to seeing false signal in compete darkness. This name is potentially confusing as there are other noises that are related to darkness and/or called dark)
The frequency content of thermal noise is roughly that of white noise. (verify)
Cooling the electronics can make this effect smaller. How cold it needs to be depends on various things, often largely the integration interval: how long you sample.
...and also on practical details like that at some point you just make another noise source more significant and there's no point to cooling further. In most cases -25C to -50C is already a lot. More than that sometimes happens, e.g. in astronomy, electron microscopy, and other (often-stationary) scientific applications.
Cooling is impractical for e.g. handheld cameras, but does help, which is why some DIY astrophotography does it.
Infrared cameras don't require cooling to work at all, but it's necessary for higher precision and/or speed.
Shot noise, also known as Poisson noise, is a result of the discrete nature of what we are sensing (electrons, photons, or such), and of statistics.
Consider a digital photo sensor, and ignore thermal noise for a moment.
Consider a single pixel, for simplicity.
The brighter it is, the more you get an integration of many photons, and one more or fewer is not going to be noticeable at all.
The darker it is, and the shorter the exposure, the fewer photons you actually count in a pixel well, the more that the amount of photons, and the interval of their arrival, matters to the nature of the numbers that come out.
Seen another way: Depending on your timing, the brightness will vary a little over time.
And for an image sensor, that's going to happen unpredictably and slightly different for each pixel, so show up as overall noise.
Not linearly - the relation is apparently apparently sqrt(intensity)(verify).
Dark shot noise
Dark shot noise (or dark-current shot noise) - refers to noise that originates from current leaking from image sensor's pixels charge wells.
For example, structural flaws may mean specific rows, columns, or pixels always drain faster.
If a subset of pixels do so consistently, this can be corrected to a good degree: measuring the effective gain, and multiplying every image by that.
Note that this mainly just puts their response on the same magnitude (so this won't look like dark pixels / lines). This is good enough for photography, while in scientific use you may wish to consider the lower response may mean lower resolution
Flicker noise happens when a component or circuit shows a noticeably variation in resistance (intended workings, unavoidable imperfections, etc), leading to differences in sensed current or voltage.
It is often seen as imperfections in electronics, and alongside other electronic imperfections.
It is more noticeable for DC and low frequencies.
Because the above is a vague definition, and because it has a 1/f (pink) spectrum, it is sometimes called 1/f noise -- though "1/f noise" is a wider concept because this is not the only reason for 1/f style noise.
Impulse noise, burst noise
Impulse noise, burst noise, popcorn noise, bi-stable noise, random telegraph signal are all closely related, and used as near-synonyms.
Definitions (and synonymity) vary with context.
Often, it refers to either
- near-instant ticks in a signal, e.g. caused by certain types of signal interference
- step-like transitions in signal levels (e.g. in semiconductors)
Also sometimes spike noise, and some causes are closely related to impulse noise.
Named for (digital) images - sparse white pixels (salt noise) and/or sparse black pixels (pepper noise), which often come from bad gain, ADC errors, or near-zero response (dead pixels).
It can groups a few more specific things (e.g. dark shot noise), and some of them can be corrected for.
Has a fat-tail distribution.(verify)
Readout noise - one of various imperfections in readout electronics.
Typically presented as a sum of all such noise on a chip, though often one or two are the major contributors, and then often mainly the amplification step.
Specific readout noise sources include:
- amplifier shot noise
- amplifier 1/f noise
- column amplifier noise
- pixel noise
- Dark current - Non-uniformity rather than noise: the reproducible amount of reaction that happens in response to no input signal.
- Can be dealt with decently by getting an average of many dark images (dark frame, a.k.a. zero frame, sometimes bias frames) and subtracting that from all later images.
A reference to avalance breakdown in semiconductors. The very short version is that when this starts to happen, the current flow is ragged.
Sometimes used intentionally as a source of randomness, audio noise (it's pretty wideaband white noise), or even (relaxation) oscillation.
Noise in electronics
And many more, including perhaps:
Comfort noise refers to intentionally adding noise when its absence is perceived to be worse.
This applies mostly to communication. For example, classical phone lines had inherent and audible noise on the line. These days, improvements in microphones, processing, and the fact it's digital means that mobile phones and can be entirely silent, leading to a lot of "wait hello are you still there?" or "I think you're muted, can you say something?"
So various systems will generate noise at the receiving end, to play when the sender is quiet to serve as indication.
And it turns out that noise (white, brown) noise is probably the gentlest, least interruptive way to do that.