The basic quantity of sound is the sound pressure, expressed in Pascal. There is no physical or mathematical [digital processing!] need to use a dfferent quantity. Physiologically however, the ear sensitivity shows a logarithmic character: steps in loudness correspond to multiplication of the sound pressure by a certain factor.

The similarity between silde rules and the deciBel is that both are based on logarithmic scales. While slide rules are really outdated, the deciBel is certainly not, at least, as long as people's hearing is relevant.

Looking at your link, I find the complaints leveled at dB are all based on a poor understanding of logarithms. If digital processing eliminates all logarithmic approaches to considering physical phenomena, maybe the dB will be replaced. I don't see that happening. We live in a Universe where the difference between really large and really small is beyond grasping if presented without logs. Likewise, so many phenomena are log-based (base 10, 2 or e) and exponential that the dB serves, and serves well, still. The claim that it is 'non-additive' is bunko, since it is logarithmic: the addition takes place with the logs, and the answers are conceptually easier to handle than lots and lots of zeros.

The final complaint, that because they are not additive, they impede dimensional analysis, is even more bizarre. Dimensional analysis is based on the idea that you don't have to care about the numbers (coefficients), but can calculate using the units both to check your work and to estimate (and even refine guesses). In such cases, logarithms are number ratios, and would wash out with the rest of the numbers. If you are dealing only with ratios, output/input, highpitch/lowpitch, SPL/SPL, whatever, you need to know what the units are to do Dimensional Analysis, but they're not explicitly exposed with the deciBels. If you are performing science experiments and don't know the units behind the ratios, yes, dB could be a problem...but you also probably should be in another line of work.

Finally, I really have to object to the dissing of slide rules. The age of the Slide Rule did not end when the manufacturers stopped making them (and in fact, they are still being made) because their function continues in their sister systems of nomography. Nurses still use nomographs, as do many others in other disciplines. The loss of training in slide rules can be directly tied to the lack of comfort with logarithms, the inability to grasp significant figures when entering the Sciences, and the fundamental belief of the Average Joe that MPG calculations aren't valid without 4 decimal places. I think the anti-dB folk would be just as (if not more) derisive of poor Average Joe worrying about 3 micro-Liters of gasoline as they are of slide rules.

The decibel not is a logarithmic unit neither an algorithmic quantity useful for calculation with the slide rule. In my opinion first of all it should be intended as a physical unit to measure a dimensionless physical quantity: the ratio (Attenuation and /or gain) of physical quantities with dimensions (INPUT-OUTPUT) involved in signal transmission system analysis. For example it is the same of the strain or other ratio physical quantities. In its definition the logarithm function is used. I think that there is a trivial confusion between the real conceptual meaning of the dB and the fact that it was born in the era of slide-rule and logarithmic tables. The physical unit historically can have its life and it depend on the utility not of the advantages for the experimenter and user. The scientific-technological communities decide the born, the change and the end of the physical units, not of the physical quantities. And the history of dB demonstrates this fact. Gery, I think that the end of the dB will be not now.

I think we all agree, Roberto, that the decibel is alive and kicking!

It is funny to realize, that only after the invention of electronics -where attenuation or gain in decibels was common sense- microphones, and the measurement of sound pressure levels became possible. That is another reason why the decibel was used as a measure for a physical quantity. Of course it was needed to transform this quantity to a dimensionless ratio first; this was done by introducing a reference sound pressure level p0. The value of p0 is arbitrary -as long as the whole world uses the same value- and the hearing threshold is a logical and comfortable choice: p0=20 microPascal.

We use the dB outside of sound measurement. For example it is extensively used to calculate gain of amplifier/transfer functions. Using Logarithms has its advantages since the slope of (say) a first order system Bode plot is -20dB/decade. It doesn't need replacing for DSP since there is nothing to replace it with. Also we use it for SNR measurement, quantization noise level etc. First time I have ever heard that anybody would wish to do away with dBs!

Thanks for your answers!! points are well taken! so, did I misunderstand what Wikipedia states in the disadvantages section (the one in my original question) or such statement is equivocal?

Our perception threshold of 3dB, which corresponds to a doubling of the acoustic energy or divide by 2.

A variation of 3dB produces the same sensation at all sound levels. To experience increased level must double the power of the sound generator. example:

A loudspeaker product to a meter:

90 dB at 1W manufacturer data

93 dB at 2 W

96 dB at 4 w

98 db to 8 W

...................

113 dB 256 W

116 dB 512 W

119 dB 1024 W >> pain threshold

If we want to represent the frequency response of a device or our ear, the energy can not be interpreted simply.

The logarithmic scale is used in the nomograms (mechanical, chemical ...) and all physical for obvious practical reasons.

Note: these data are generally accepted average values. A change of 1 or 2 dB is negligible even if it is a lot of watts

(English google, I hope to be included ?)

Gery, Wikipedia exaggerates the disadvantages. The basic formulas stated in "physical form" (like W=I.A; diffuse, reverberant sound field) are easier to understand than those transformed to the decibel domain. Dimenional analysis belongs to the domain of the physical formulas, underlying all "decibel-formulas".

In electrical engineering (electrical power in Watts) the Wikipedia point of view seems to make more sense than in architectural acoustics.

Agree. I must also say, that digital tools are a way to represent the analog world. When working with and application, one should never forget the conection between "1000111000" and microvolts, volts, or anything. This microvolts, volts... belongs to the "sampled" signal, and this sample signal should be related to the real world through the characteristics of some transducer. In acoustics, it could be a microphone characterized by its sensitivity (conversion from Pascals to milivolts), and as the dinamic range is so wide we still need "dB" regardless of the digital signal,. Of course, the physical meaning of equations are better understood if they are kept in the physical domain without taking dBs. The transformation of physical equations into dBs expresions could lead some times to the misunderstanding of the dB equation by non-experts (example the definition of acoustic insulation, which could be thought as a difference of sound pressure levels -dB- between rooms while it is defined as a relation of indicent to trasnsmited Energy. To transform the relation of power to relation of sound pressure levels we have to take some assumptions on the kind of sound field.) . Anyway, we still need dBs to represent values of magnitudes with a wide dinamic range. That is why decibels are for.

So we cannot forget what the digital signal represents, unless we want to assume serious misconceptions. Engineers also must understand what sampling implies, and how sampling can distort the analog world (aliasing). Huge mistakes are made in the name of the digital representation of signals and spectra.

Example: how often many people FORGET that a DFT (FFT is the way to perform DFTs ) is discrete so we can miss spectral information (peaks) if it is not properly understood and handled?

Another point: Wikipedia is not a scientific discussion forum.

Gery, you have to take into account the sources of the supposed 'disadvantages' of deciBels. The several articles are not published in Electrical Engineering, but one in EE and three others, one of which is the source of the Hickling article. Let's look at them:

First, the EE article. It was published in 1954. The author, from NUSL (changed to NUSC and then to NUWC, now in Newport, RI) was involved with Sonar research, and is apparently one of those folk who are offended if a tool is used for more than one purpose. While his objection, and his recommendation to limit the dB to unit transmission loss and rename all other use of the same ratios/equations to the "Logit", rather than considering it a "Unit" is rational and even sensible, it has long been obsoleted by the fact that, in industry and in research, the dB is recognized as a unitless log of ratio. It is on the same level as objecting to a discussion of top speed capabilities and acceleration when discussing a sports car.

The other sources, i.e., the three JASA articles, are specifically aimed at SI units and are grouped, like a good shotgun pattern, around 2000. They are hardly a survey of the industry, or of science, and with the 1954 article of Hickling, constitute two unrelated objections to using a dimensionless log-ratio as a unit from very different times. This is, at best, a semantic argument. Essentially, the entire "disadvantages" set looks like someone, faced with providing some standardization in Wikipedia articles, had to come up with disadvantages and went fishing.

In actual fact, the spurious jab at sliderules is pretty baseless as well. Yes, you could simply set hairline cursor over the larger number on the D scale, then align the C scale (on the slide) under the hairline, then either move the hairline over the index on the C scale and read straight from the L scale on a stator for voltage or current ratio in dB, or move the cursor to 2 and read the power ratio (including sound pressure level) from the L scale, in either case, mentally slipping the decimal point to the right one place. This is because the L scale, like the mantissa of a logarithm, is from 0 to .99 ( or .999 if you had really good eyes), and the usual dB equations were 20log(ratio) or 10log(ratio). Yes, slide rules made finding dB easy (or finding the voltage/power/whatever as long as you knew what the comparison value was).

There is an interesting twist, by the way: We have dBV and dBm as "standard units" in many industries which work with gain in dB. The only difference here is that the bottom of the ratio fraction is standardized: in radio to 1mw, in the audio realm, to 1V. As unit-based measurements, dBm and dBV can mess a newbie's head up a bit, and yes, you add them just like dB, but they have the benefit that you can generally figure out the actual physical value pretty easily: 30dBm, for instance, is 1Watt, because 30/10=3, and the antilog of 3 is 1000: as a ratio to 1mW, we get 1mW * 1000 = 1W. From then on, the average electronics engineer, technician, or scientist memorized the following:

30dB = x1000

3dB =x2

0dB = x1

-3dB = 1/3

etc. and maybe a few others if they came up a lot.

We still use dB today, as has been mentioned. One place it hasn't been mentioned is in lasers, especially in communication amplifiers. 3dB still means whatever you put in, light, electric/sound power, whatever, you get twice as much coming out, and for units that aren't combinations-by-multiplication of two units (Volts, Amps, etc, vs. Watts, which are V*I, or magnitude, which is generally something like V^2)

are twice the value of dB for the same ratio. Also, you'll find that EEs are unlikely to give up 3dB/octave, 10dB/decade in filters and amps.

Nothing wrong with the decibel. It's simply the log of the power ratio of the energy of a wave to a reference level. Log scales are very useful for quantities that vary over orders of magnitude. The problem is that the public is clueless about log scales. An increase in sound decibels by three represents a doubling of the energy; an increase of 10 represents a 10 fold increase. So that 104 dB fire alarm in your office is over 200 times louder than the maximum 80 dB your hearing should ever be exposed to. Try explaining that to the fire marshal.

Very fair, science presents no issue units of measurement. The decibel (dB) is like second (s), the Tesla 'T), Ampere (A) or volts (V).

It is necessary because the sensation produced by a augmention power from 1 watt to 90 dB is the same as that produced by a augmention power 512 w 116 dB in the example I gave.

The dB gives a linear scale for exponentially reality.

Christian Guilleminot, the dB is unlike every one of the units you mention for two reasons: it is not a unit, and it is logarithmic.

In every case of a unit, you can use a coefficient of 1 and have something measurable. One dB is meaningless without a reference. Many logarithmic units have been made by specifying a reference, and usually the name dBV, dBm, dBSPL, etc) indicates the reference. Even then, to represent one mW with dBm, you must use a coefficient of zero!

This is the real problem with dB: it is not a fixed-increment scale. Without knowing where you start, you cannot tell how large 1dB is.

Hi people, I like the dB discussion, and I think the big issue is not if dB makes sense or not (it definitely does make sense for comparison of variables with a high dynamic range for perceptual reasons) but if people employ it in the right way... I'm much more worried about the latter than about any units, and I'm interested in your opinion and experience on the following:

In ultrasound imaging science, people do a lot of research in many different techniques of how to improve image contrast. They compare an image before and after application of the technique, both images in dB scale. Then they say, for example: "before we had 10 dB contrast, now we have 20 dB contrast. This shows a big improvement in contrast".

The problem I see with this is that the different images were obtained using different techniques... and therefore the dB scales are not necessarily comparable. In any case... dB was defined as a logarithmic scale of power/energy/intensity.... and should only be employed with those quantities. For any technique that displays a different quantity, the dB scale seems to be the wrong choice anyway. However, people seem to extensively use dB scales with any outcome of digital processing, just for historical reasons...

Typical example: you could just square the signal amplitude, and would end up with double the dB... but this does not in any way improve contrast.

Looking at how widespread the phenomenon is at least in ultrasound research without people questioning it a lot, I think there SHOULD be a re-definition of how images should be presented, and of how to quantitatively compare the outcome of new processing techniques.

To me, the most obvious method would be statistics... what signal level do we need in order to detect it within the background? For any new processing techniques, one has to determine the noise statistics in order to answer that question. And a technique is only a good technique if it leads to better detectability... not if it leads to more "dB".

I would be very much interested if you knew any literature on this aspect of the discussion. Maybe there is even a scientifically justified general way of how to choose a scale for data presentation? This could be something similar to dB but maybe deviate from it, based on noise statistics?

Michael Jaeger, I'm not sure I'm following you, nor that it would be possible to do so.

First, your claim that doubling dB is the same as squaring "the signal amplitude": if you square the ratio between the signal amplitudes,or, for that matter, the ratio between signal power levels the dB result does double the dB. However, the ratio would have to change: If the ratio is related to a dark reading of 3, a light reading of 15 produces a ratio of 5. To get that ratio to be 25, without changing the dark level, the bright level would have to become 225. If 225:3 is the same contrast as 5:3, then there probably isn't any reason to be working to improve contrast at all, is there?

In actuality, doubling the dB squares the amplitude only of the numerator in the ratio, and this indicates a significant improvement in contrast. At the least, this does prove your contention that dB make sense, only if employed in the right way.

This comes to the question of whether contrast is a ratio or not. From a quick (and arguably too quick) survey of the display industry, I can only find one use of "contrast" which is not a simple ratio, which is RMS Contrast. In that case, statistics makes sense, because it is a measure of the Standard Deviation of pixel intensities to the average of the overall image. I would not expect the result of that calculation to be expressed in dB.

Are display scientists actively comparing noise to contrast via dB? If the measurement of pixel level is being taken in RMS, and there is a clear comparison to be made between that and noise taken in RMS, there is no reason that dB wouldn't be useful, and make sense, if there is a reference to compare it to. If dark level is based on noise, and detectability derives from RMS level over the noise background, and this is ratiometric and doesn't change with increasing noise level, there is no reason that dB can't be used. If the RMS noise background level affects detectability, then that measurement isn't going to be reasonable to describe in simple dB, although once the relationship is defined, it might be simpler to do so. In either case, dB would still make sense, as it does in signal-sensitivity for radio receivers if the dB were referenced to a specific quantity (some standard for Dark Level without Noise, perhaps).

However, regardless of how the images are obtained, if the ratio between the dark and light pixels increases, the dB result will increase, and comparison on a purely physical-measurement basis (with consideration for non-linear effects in the measuring equipment) is perfectly valid. As for the "scales not being comparable", where does the scale come from? You can't make a dB scale without specifying the reference: if you have two references, you have a ratio between them, and this results in an additive dB value that can be used to "equalize" the scales. It is an extra step, but if having the comparisons on an equal scale is important, it must be done: to skip it is to misuse the measurements. For instance, if one display's dark level, referenced to some standard minimum noise level other than zero) is .003dB, and another's is .12dB, subtracting those numbers from the contrast dB value puts them "on the same scale". For that matter, if you take the ratio of their dark pixel values, convert it to dB and subtract it from the Contrast with the higher dark value, those contrasts are now "on the same scale". If you can then show that the real perceived contrast is different, you have grounds for objecting to the use of dB... and the job of qualifying why they aren't the same and producing a way of making them comparable, which you can name after yourself.

dB in sound refers to the log of the power ratio, not the amplitude ratio. But since power goes as the square of amplitude, the log of the power ratio is twice that of the log of the amplitude ratio. Unfortunately, I suspect the distinction gets lost for many common uses of dB. I would always check how it's defined for any use other than sound.

Thomas, thanks for shedding some light on this. In order to explain doubling of the log ratio, you have to square _both_ the signal amplitude and the reference amplitude, which is analogous to comparing, in electricity, for instance, the power from a constant impedance caused by that voltage. In that case, yes, the entire ratio is squared, and that's where Voltage dB get the factor of 2 from.

This touches the famous dilemma between acoustics and psychoacoustics. As nobody else mentioned here, I suggest you take a look on the classical Fletcher-Munson equal loudness curves. These ones were measured for the hearing of simple sounds (sine waves). For complex sounds (i.e. practically all sounds in nature), there were some efforts done in trying to come up with an equal loudness curve for "any sound", however they mostly bump into individual differences (different ears hear differently ... not mentioning the brain). There is also a trade on the music industry known as "loudness war" as the marketing guys realized that louder songs tend to sell more, but how to put more perceived intensity (loudness) into the same 16bits limit imposed by the digital audio CD standard? The answer has been mostly trial and error, and the human artistry of mixing professionals.

dB is very practical in many situations; like percentage, it describes a magnitude relationship. As Jose Fornari hints at, the 'dilemma' between psychoacoustics and acoustics has a long history - human sensation is complex. A great reference is the work by Eberhard Zwicker and Hugo Fastl, e.g. http://www.amazon.com/Psychoacoustics-Models-Springer-Information-Sciences/dp/3540231595. 3rd edition is also available as PDF online.

The decibel (dB) is a logarithmic unit used to express the ratio between two values of a physical quantity, often power or intensity. One of these quantities is often a reference value, and in this case the decibel can be used to express the absolute level of the physical quantity. The number of decibels is ten times the logarithm to base 10 of the ratio of the two power quantities.[1] A decibel is one tenth of a bel, a seldom-used unit named in honor of Alexander Graham Bell.

The definition of the decibel is based on the measurement practices in telephony of the early 20th century in the Bell System in the United States. Today, the unit is used for a wide variety of measurements in science and engineering, most prominently in acoustics, electronics, and control theory. In electronics, the gains of amplifiers, attenuation of signals, and signal-to-noise ratios are often expressed in decibels. The decibel confers a number of advantages, such as the ability to conveniently represent very large or small numbers, and the ability to carry out multiplication of ratios by simple addition and subtraction. However, the unit has been criticized by professionals as being cumbersome and confusing.

Decibel is only a relative unit mainly used by communities working on electromagnetic waves, optical waves and sound waves with typically (Radar, Sonar, Lidar, Sodar and in optical fibers) . This unit is a very useful relative unit (a ratio) as for example to estimate a level upon a noise and is not an absolute unit.

Therefore a decibel can be an amplitude ratio but of course for acoustic waves the acoustic pressure the relevant unit is in Pascal but dB can be also an acoustic power ratio.

For Radar waves, often we need the Radar power and in this case one speaks of dBm or dBmW which are the power ratio between measured power and one mW : 0 dBm= 5. 10-4 Watts, 3 dBm= 2. 10-3 Watts.

It is also useful to remember and this also justifies the use of dB, that human perception and relative subjective sensations can be measured in dB following the Weber-Fechner law.

A decibel (1/10 of a Bel, both invented by and named for Alexander Graham himself) is just a way to represent numbers that range from very small to (relatively) very large.  It is not more "old" than such concepts as fractions and logarithms, without which, we would still be hunting and gathering.