The decibel

C. Groffen (cgr)
Sonion Nederland B.V.
Hoofddorp, The Netherlands
cgr@sonion.com
December 5, 2013

1 Introduction
In the everyday life the unit “decibel” (symbol: dB) is almost exclusively
used to indicate loudness of sounds. However, the decibel is a far more
flexible unit. Because at Sonion we work with audio as well as other technical
disciplines, the term dB can be confusing. This document is an attempt to
clear this up. It assumes that the reader of this article has some basic
understanding of logarithms or knows how to brush this up.

2 What is a dB?
The decibel is a tenth of a bel, just like the decimeter is a tenth of a meter.
However, the bel itself (named after Alexander Graham Bell) is not used
anymore.
The dB is not a unit like volt, meter or second but it is a ratio. When
you have 5 apples while your friend has 10, you have half as much as he has,
and this will be expressed normally as 21 or 0.5. Another much used ratio
unit is percent, so in this case 50%. In itself it has nothing to do with the
loudness of music for now.
All the earlier scales are linear, while the decibel is logarithmic This
means that the scale is “compressed”, and big numbers do not add up that
much as numbers closer to the reference value. This may seem strange, but
in fact it is more logical when expressing a ratio. For example when the
ratio is 20 to 1, so say 2000%, the opposite (so 20 times less) is 5%. With
dB’s 20 to 1 will be 26 dB while the opposite is just −26 dB.

2.1 Formal definition

The decibel is defined for power or energy units as follows:

1
 P1
 
dB = 10 · log
P2
where P2 functions as a reference power. The logarithm used in decibels is
always a base 10 log.
Often the so called “field quantities”, amplitudes, like voltage or current
needs to be compared. To make these comparable the following trick is used:
Since
V2
P =
R
gives at a constant R:
!
V12
dB = 10 · log
V22
this could also be written like:
2
V1

dB = 10 · log
V2
applying a logarithmic calculation rule:
V1
 
dB = 20 · log
V2

2.2 Frequently used numbers

With the decibel some key numbers should be remembered to get a grip on
the numbers and simplify calculations from the thop of your head.

unit fraction power units [dB] field units [dB]

√
2 1.505 3.01
2 3.01 6.021
10 10 20
20 13.01 26.021
30 14.771 29.542
100 20 40

Table 1 – Some decibel numbers rounded to 3 decimals if needed. Note

that if the fraction swaps, so from 2 to 12 for example the only thing
changing in the dB’s is the sign (it becomes a negative number).

The 6 dB and 3 dB (always rounded to the nearest integer) are used a lot
to indicate frequency ranges of amplifiers or transducers, where the term “-3
dB-point” comes from. Note that 2 times 10 is 20, giving the same output
as adding the 2 and 10 decibel units.

The decibel Page 2 of 12

December 5, 2013
2.3 Frequently used references
As the P2 or V2 in section 2.1 are used to refer to something, there are luckily
a lot of standard references.
Obviously the best known is the dBSPL, commonly referred to dB and is
about the sound pressure level (SPL). It’s a pressure (field quantity) divided
over 20 µPa. This number is empirically determined as the mininum pressure
of a 1 kHz tone to be heard. 2 pascal will give exact 100 dBSPL. 1 Pascal
can be obtained by dividing by two and this is the same as subtracting 6
dB and gives 94 (approximately) dBSPL, a frequently used number within
Sonion.
In the electrical engineering field the dBV is commonly used. It’s deci-
bels with a reference of 1 V. Taking decibels has never been so easy since
dividing by one is an operation which can be omitted.
Radio engineers often use the dBm, which uses 1 milliwatt as reference.
So this is a power unit which uses 10 log.
An own reference can be made. At Sonion for telecoils the decibel has
as reference 1 V/A/m, while microphones are referenced to 1 V/Pa. These
are all field units.

3 Why do we use the dB

The decibel is the de facto standard to describe sound pressure level. This is
because the response of the ear does fit a logarithmic scale way better than
a linear one. But the dB has more advantages.

3.1 Using large next to small numbers

When handling big numbers next to small numbers it is cumbersome to work
with a linear scale. Plotting numbers this way require a very large graph, or
some lines are squished.
In figure 1 and 2 the difference in noise performance hence THD in a
loopback measurement (output directly coupled to the input) between a
LabVIEW and Rohde & Schwarz measurement system is shown. By “com-
pressing” the high number with the logarithmic decibel all traces become
visible. A rough estimate of the performance difference can be made by
subtracting the decibel numbers, in this case −40 − −100 = 60 dB. This is
a factor of 1000 (3 times 20 dB).
Other examples are figures 3 and 4, where a receiver is fed with a single
frequency. In the linear volt scale the graph is really boring. There is not
much to see apart from a peak at 1 kHz and the rest of the line is hugging
the frequency axis.
With the decibels (noted with dBSPL because it handles sound pressure
level) all harmonics and their relative loudness can be observed. Even the
receiver response can be seen with peaks at 1.5, 4 and 7 kHz.

The decibel Page 3 of 12

December 5, 2013
 3.5

20 mV RS
3 20 mV LV
60 mV RS
60 mV LV
1 V RS
2.5
1 V LV

2
THD [%]

1.5

0.5

0
100 1000 10000
f [Hz]

Figure 1 – The amount of THD in a loopback measurement with two dif-

ferent measurement systems at three different voltages. The THD plotted
in percent leave the bottom three traces unreadible.

-20

-30

-40

-50

-60
THD [dB]

-70
20 mV RS
20 mV LV
-80 60 mV RS
60 mV LV
1 V RS
-90
1 V LV

-100

-110

-120
100 1000 10000
f [Hz]

Figure 2 – The same graph and data as in figure 1 but now is the THD
plotted in dB. All traces are visible now, and the relation between them
is now visible (they show the same rolloff for example).

3.2 Ease of computational operations

With logarithms come certain identities to describe them. One of them
already showed up in this article, which is:

log (xa ) = a log (x)

This made it possible to write the field quantities like 20 log instead of 10.
Another identity is:

log(x · y) = log(x) + log(y)

The decibel Page 4 of 12

December 5, 2013
 0.5

0.45

0.4

0.35

0.3
Output [V]

0.25

0.2

0.15

0.1

0.05

0
100 1000 10000
f [Hz]

Figure 3 – 4 mA input @ 1 kHz in a 31A007 @ 2CC ITE. Pure reference

microphone output is plotted in volts. A well trained eye can see small
bumps at 2 and 3 kHz.

120

100

80

60
dBSPL

40

20

-20
100 1000 10000
f [Hz]

Figure 4 – The same graph and data as in figure 3 but now plotted in
dBSPL. The harmonics are well visible as well as the noise floor.

This states that if you needed to multiply certain numbers (i.e. total gain of
a cascaded chain of amplifiers) with decibels you simply have to add them!
Not only with amplifier cascading but also in other situations this property
comes in very handy.

4 How are calculations performed

The pure calculation is very simple. In Excel for example it can be done
with =20*LOG10(V1/V2) where V1 and V2 are numbers of your own choice.
Going the other way around goes as follows:

The decibel Page 5 of 12

December 5, 2013
 V1
 
dB = 20 log
V2
dividing by 20:
dB V1
 
= log
20 V2
taking away the log:
V1
10( 20 ) =
dB

V2
The numerator or denominator alone of the right hand side of the equation
can be retrieved in the normal way. In Excel this exercise would look like
this, having V1/V2 as the answer: =10^(dB/20) where dB is your beloved
decibel number of course.
Further in this section various examples will be given which will be used
at Sonion. Be careful with rescaling actions since it only works when the
device is linear. However, receivers can be seen as linear up to 5% distortion
and microphones up to 3%. Electronic devices act completely linear up to
their maximum relative to our transducers.

4.1 Receiver sensitivity rescaling

When a receiver is acoustically measured, most of the time it’s measured
with a constant voltage from a low output impedance source (voltage drive).
The drive level is derived from a certain power, e.g. 0.35 mVA @ 500 Hz.
But now another receiver is measured with a different drive for example

0.5 mVA @ 500 Hz. If you want to compare them, simply add 10 log 0.35 0.5
≈
1.55 dB to the 0.35 mVA @ 500 Hz measured graphand they are now at the
same level (0.50 mVA @ 500 Hz). Note that 10 log 0.35 0.5
≈ −1.55 dB so be
on guard by choosing the correct sign adding/subtracting. Remind yourself
that a receiver is playing louder with more power. This will solidify your
calculation.
This same method applies when you do not have the power, but a drive
level. This works only with one type of receiver every time because of the
impedance. But you have your receiver measured on 150 mVrms (side note:
with sine waves the difference between rms and peak is 3 dB), while your
customer did a measurement on 190 mVrms. Is there a need to  remeasure

the SPL? No, the SPL curves can be compared by adding 20 log 190 150
≈ 2.05
dB to the 150 mVrms curve. Note that 20 log is used here in contrary to the
previous example.

4.1.1 Putting impedance into the equation

Previous section the impedance was either way assumed the same or one
frequency was taken. But of course this could all be taken into account

The decibel Page 6 of 12

December 5, 2013
when comparing receivers.
Imagine we have two receivers, with different drive levels, different sen-
sitivities and different impedances (not only in height, but as well with a
different inductance for example), how can we lump the data together? We
can get the curve at the height as if driven with 1 mW as follows:

1000 · V 2
!
dBSPL1mW = dBSPL − 10 log
|Z|
 
What basically happens here is adding with dB = 10 · log P1
P2
. P2 in this
case is 1 mW, so 0.001 W. We can rewrite this as:

dB = 10 · log (1000 · P1 )
P1 is equal to V 2 /|Z| (follows out of Ohms law), giving the rest of the
formula.
110 1000

105 900

100 800

95 700

E50DA020 dBSPL
90 600
2303 dBSPL

|Z| [ohm]
dBSPL

85 E50DA020 |Z| 500

2303 |Z|

80 400

75 300

70 200

65 100

60 0
100 1000 10000
f [Hz]

Figure 5 – Two different receivers measured both with 0.35 mVA @ 500
Hz, however, their impedance shape is different. The 2303 has a higher
coil ratio, meaning that it is more inductive, tilting the SPL curve towards
the low frequencies.

4.2 Receiver vibration measurements

The vibration of a receiver is generated by the armature and membrane,
so the moving mass of the system, being accelerated and decelerated. This
vibration could be measured with a laser vibrometer, giving output in m/s,
or with an artificial mastoid. The latter one uses a heavy block with an
accelerometer rigidly attached in it. The receiver is clamped onto the block
which transfers all his vibrating energy. Since the heavy block is way more
heavy than the receiver, the mass of the heavy block times the acceleration
equals the force generated by the receiver:

F~receiver = mblock · ~ablock

The decibel Page 7 of 12

December 5, 2013
 120

115

110

105

100
dBSPL

95

E50DA020 dBSPL
90

2303 dBSPL
85

80

75

70
100 1000 10000
f [Hz]

Figure 6 – The same data as in figure 5 but modified to 1 mW according

to the formula given in the text. At 500 Hz this curve is 10 log(0.35) ≈
4.56 dB higher.

The sound pressure level in our application is proportional to the membrane

displacement times the membrane surface area. This displacement is gener-
ated with the armature and is linear with the current fed to the coil.
This means that with our approximately flat sound response curve up to
the first resonance frequency the vibration should have a slope of around 40
dB/decade. This is because acceleration is the double derivative of displace-
ment and in the frequency domain this means that the value is multiplied
with (2πf )2 .
With velocity the output rises linear with frequency. This means that
double the frequency is double the output. Ten times the frequency is ten
times the output. 20 log(10) is exact 20. So the slope with one differentiation
is 20 dB/decade or 6 dB/octave. Doing this again, we get a slope of 40
dB/decade. 20 log(10 × 10) = 40.
So we have all kinds of receivers with different armature sizes/masses
and different membrane sizes. But we want to compare different sizes. With
the vibration expressed in dB relative to 1 N we only have to subtract the
receiver output. See figure 8. We can add 100 dB to get the number flat
equalized for 100 dBSPL to get more realistic values (not done in the figure)

4.3 Microphone frequency response

Microphones are measured by placing a speaker in a small cavity, together
with a reference microphone and the microphone under test. Feeding a
constant voltage over the speaker gives the results as in figure 9.
The sensitivity of Sonion microphones are expressed in dB units re-
ferred to 1V/Pa. The dBSPL is referred to 20 µPa. 1 pascal therefore
is 20 log 0.00002 = 93.979 . . . ≈ 94 dBSPL.
1

If we want to set the microphone sensitivity right as in figure 10 we

The decibel Page 8 of 12

December 5, 2013
 0 130

-10 125

-20 120

-30 115
Vibration [dB re. 1 N]

-40 110

dBSPL
-50 105

-60 100

-70 95

-80 90

-90 85

-100 80
100 1000 10000
f [Hz]

Figure 7 – 26UA01C @ 400 mVrms. The blue line is the vibration in

the membrane direction, the red line is the acoustical output at a 0.5 cc
coupler with a short tubing. Note the rising blue line up to the resonance
frequency.

-130

-140 1700
1900
2000
-150
2300
2600U
-160 2600WB
3100
dB re. 1 N/20 µPa

3500
-170
E25S
4100
-180 40 dB/dec

-190

-200

-210

-220
100 1000 10000
f [Hz]

Figure 8 – Vibration of a lot of single receiver families within Sonion,

subtracted over the output. Receivers of all sizes do not differ that much
when referred to the output. A 40 dB/dec line is plotted for reference.

basically have to set the dBV output while the reference output is 94 dB.
So in formula form:

dBcorrected = dBVmicrophone − dBSPLreference + 94

4.4 Microphone THD measurements

The distortion in a microphone is determined by the used electronics (hy-
brid). Before the microphone membrane collapses, around 160 dBSPL should

The decibel Page 9 of 12

December 5, 2013
 100 -20

95 -25

90 -30

85 -35

dBV
dBSPL

80 -40

Reference
microphone
75 -45
65GC30

70 -50

65 -55

60 -60
100 1000 10000
f [Hz]

Figure 9 – Typical microphone measurement with as device under test

a 65GC30 supplied with 1 V. The current curve is pretty hard to read.

-20

-22

-24

-26
Sensitivity [dB re. 1V/Pa]

-28

-30

-32

-34

-36

-38

-40
100 1000 10000
f [Hz]

Figure 10 – Typical microphone measurement with as device under test a

65GC30 supplied with 1 V. The reference microphone curve is subtracted
giving a nice curve.

be reached, while the hybrid normally doesn’t go higher than 120 dBSPL
equivalent.
The word equivalent is used because a hybrid is not sensitive to sound
anyway. But basically the hybrid amplifies 1:1 the voltage coming in from
the cartridge (the membrane and backplate assembly) up to the point where
it’s supply voltage (most of the times 0.9 V) or ground is limiting any further
undistorted output.
Since the cartridge determines the sensitivity of the microphone, the
lower sensitivity microphone, the more sound input it can handle before
saturation because it generated less volts with the same sound input.
Therefore at figure 11 a big difference can be seen in THD. The “knee
point” of microphone 1 is more to the right in the graph, meaning there is

The decibel Page 10 of 12

December 5, 2013
more SPL needed to get the microphone in distortion. But since microphone
1 is less sensitive that number 2, this is fairly logical because the hybrid gives
less output at the same dBSPL input level of the microphone.
In figure 12 therefore microphone 1 is compensated by easily subtracting
the difference in sensitivity of the microphones, and the difference disappears.

10

6
THD [%]

5 Microphone 1

4 Microphone 2

0
100 105 110 115 120 125
dBSPL

Figure 11 – THD of two microphones of the same type. Microphone 1

has a sensitivity of −38.61 and microphone 2 of −34.03 dB re. 1V/Pa.
Measuring frequency is 1 kHz.

10

6
THD [%]

5
Corrected microphone 1

4 Microphone 2

0
100 105 110 115 120 125
dBSPL

Figure 12 – Same graph as in figure 11 but the curve of microphone 1

is shifted 4.58 dB to the left to compensate for the sensitivity difference.

5 Concluding remarks
The arithmetic shown in secion 4 works for far more examples than the ones
shown there. Basically the decibel allows to do multiplications in the linear
domain by simple addition and substraction, thereby easing up calculations.

The decibel Page 11 of 12

December 5, 2013
However, adding in the linear domain is not possible to do in the logarithmic
domain, so unfortunately decibel numbers should be transformed to the
linear domain first to allow adding.
It “compresses” values so that wide-ranged value sets are easily inter-
preted and plotted in a single graph.

The decibel Page 12 of 12

December 5, 2013