Dynamics 1 Dynamics 2

1 Introduction
SGN-14006 / A.K. SGN-14006 / A.K.

Dynamic range control !  Sound pressure level of natural sounds varies over time
Source: Zölzer, ”Digital audio signal processing,” Wiley & Sons, 2008. –  For example in music and in movies, the loudness of sounds
works as an effect and is part of the content
–  When recording, the distance to microphone may vary, which
causes unwanted level variation
!  When audio is played back, the useful dynamic range may
Contents:
be limited due to background noise for example
!  Introduction
!  Static curve
!  Dynamic behaviour
!  Implementation
!  Stereo processing
!  Bandwise dynamic control

Dynamics 3 Dynamics 4
Introduction
Introduction
SGN-14006 / A.K. SGN-14006 / A.K.

Applications of dynamic range control

!  Figure: block diagram of dynamic range control
!  When recording, it is desirable to use the full amplitude 1.  Measure the level of input signal
range optimally 2.  Multiply the delayed input signal by factor g(n)
–  Also to protect AD converters from overloading (clipping) y(n) = g (n) ⋅ x(n − D)
!  Noise gates are used to suppress low-amplitude noise !  Delaying signal x(n) compared to the control signal g(n) allows
= only audio signals exceeding a certain level will be passed through ”predictive” level control (gain goes down before any ”big bangs”)
!  Value of gain factor g(n) is calculated in two steps
!  In record production, dynamic range variation can be –  Static curve defines desired output level corresponding to the input level
limited at will –  Temporal varation of g(n) is smoothed using a lowpass filter
!  In various recording formats, the limited range of
amplitudes should be optimally used
!  When playing back music or speech e.g. in a car, dynamic
range variation has to be matched to background noise
–  Listening becomes easier / possible
 Dynamics 5 Dynamics 6
Static curve
2 Static curve
SGN-14006 / A.K. SGN-14006 / A.K.

2.1 Operation regions

!  Figure: relationship between input level and weighting level is !  Limiter limits the output level when the input level exceeds
determined by a static level curve G[dB] = f ( X[dB] )
the limiter threshold LT
–  Output level and weighting level are functions of input level
–  All input levels above the threshold lead to a constant output level
–  Thresholds: LT=limiter threshold, CT=compressor thr., ET=expander thr.,
NT=noise gate thr. !  Compressor maps a change of input level to a smaller
change in the output level
!  Expander increases changes in the input level to larger
changes in the output level
!  Noise gate is used to suppress low-level signals and to
reduce noise
!  Threshold values in different parts of the static curve
determine the lower limit for limiter and compressor and
upper limit for expander and noise gate

Dynamics 7 Dynamics 8
Static curve SGN-14006 / A.K. Static curve SGN-14006 / A.K.

2.2 Compression ratio Compression ratio

!  Compression ratio is visible in the logarithmic !  From figure: line equation Y = CT + (1 R ) ⋅ ( X − CT )
representation of the static curve: ratio of input level and compression ratio R = ( X − CT ) (Y − CT )
–  CT is input threshold (dB), where compression starts
change ΔPI to output level change ΔPO
!  Switching from logarithmic to linear representation we obtain
R = ΔPI ΔPO log10 ( x / cT )
!  Typical compression ratios: R = ∞ (limiter), R=
log10 ( y / cT )
R > 1 (compressor), 0 < R < 1 (expander),
where x and y are linear levels, ja cT is linear compression threshold
R = 0 (noise gate at threshold)
!  From here we can solve linear output level y as a function of the input
level x
y cT = 10(1 R )⋅log10 ( x cT ) = ( x cT )1 R
1−1 R
y = cT ⋅ x1 R
!  The control factor g(n) can be calculated as

g (n) = y x = ( x cT )1 R −1
 Dynamics 9 Dynamics 10
Dynamic behaviour
3 Dynamic behaviour
SGN-14006 / A.K. SGN-14006 / A.K.

3.1 Level measurement

!  Besides static curve based level control, the dynamic !  There are two widely-used methods for level measurement: PEAK and
RMS
(time-varying) behaviour of the control factor (attack and
–  Functionally interchangeable (same task)
release time) plays a significant role in sound quality –  However PEAK is typically used in the limiter, whereas RMS is used in
!  The rapidity of input signal level measurement using compressor, expander and noise gate
PEAK or RMS algorithms also affects the rapidity of the !  Let us denote the input signal by x(n)
overall dynamic range control !  PEAK algorithm (produces a level measurement xpeak(n)):
•  Coefficient AT determines the attack time, coefficient RT the release time
!  In the following, we first consider the level measurement
if |x(n)| > xpeak(n-1)
and then the control of attack and release times by xpeak(n) = (1- AT ) xpeak(n-1) + AT · |x(n)|
filtering control factor values over time else
xpeak(n) = (1 – RT) · xpeak(n – 1)
end
AT = 1 - exp( -2.2 Ts / (ta/1000)), where ta=attack time (ms), Ts=1/Fs is samp. interval
RT = 1 - exp( -2.2 Ts / (tr/1000)), where tr = release time (ms)
Typically ta = 0.02ms…10ms and tr = 1ms…5000ms

Dynamics 11 Dynamics 12

Level measurement Level measurement

SGN-14006 / A.K. SGN-14006 / A.K.

!  Output of the PEAK algorithm for a rectangular pulse !  RMS algorithm (produces level measurement xrms(n)):
•  Square of the input, averaging with first-order lowpass filter
•  Temporal length of the averaging is determined by coefficient TAV
x2rms(n) = (1 – TAV) · x2rms (n – 1) + TAV · [x(n)]2
•  TAV = 1-exp( -2.2 Ts / (tM / 1000))
–  tM is the averaging time in milliseconds.
–  In the figure tM=100 ms (green line)
–  For peak (red line): on the left, ta=10ms, on the right ta=0.02ms

Time(s)
 Dynamics 13 Dynamics 14
Dynamic behaviour SGN-14006 / A.K. 4 Implementation SGN-14006 / A.K.

3.2 Gain factor smoothing 4.1 Example: limiter

!  Attack and release times can also be implemented by !  Figure: limiter block diagram
1.  Level xpeak(n) of input signal x(n) is measured using the PEAK algorithm
smoothing the temporal variation of the control factor g(n)
2.  Value log2[xpeak(n)] is compared to limiter threshold LT
1.  Control factor f(n) is compared to its previous value 3.  If the value is above the threshold (difference is positive)
(control factor, calculated using the static curve, was •  Difference is multiplied by the negative slope –LS of the limiter
•  Take antilogarithm 2G
denoted by g(n) above) •  The resulting control factor f(n) is smoothed using first-order lowpass filter
–  Determine whether the control factor is in attack or release state SMOOTH
4.  If the value is not above the threshold, factor f(n) is set to value 1.
–  Accordingly, assign variable k either value AT or RT
5.  Delayed input signal x(n – D1) is multiplied with the smoothed control
2.  Value of the control factor is obtained by filtering the factor g(n) to give the output y(n)
control factor values with first-order lowpass filter:
g(n) = (1 – k) · g(n – 1) + k · f(n)

Dynamics 15 Dynamics 16
Implementation SGN-14006 / A.K. Implementation SGN-14006 / A.K.

4.2 Combination system Combination system: example

!  The basic structure of compressor, expander, and noise gate is similar to the
limiter !  Figure:
–  Level measurement can be made using RMS-based value 0.5 · log2[xrms(n)] –  Input x(n)
!  Practical implementation is a cascade, where each part implement one of the
basic operations (limiter / compressor / expander)
–  Appropriate parameters can be chosen for each stage
–  Because the maximum output level is reduced by the limiter, the static curve in its
entirety can be shifted up with constant gain
!  Figure: static curve of the combination system (additional static gain 10 dB)

–  Output y(n)

–  Control factor
g(n)
 Dynamics 17 Dynamics 18
Implementation SGN-14006 / A.K. Implementation SGN-14006 / A.K.

Combination system: example 4.3 Computational aspects

!  It is seen that signal with high amplitude are compressed !  The sampling rate of the level measurements xpeak(n) and
and those with low amplitudes are expanded xrms(n) can be reduced by factor four
–  An additional gain of 12 dB (control factor maximum value is 4) –  Because the signals are by their nature already bandlimited, we
–  g(n) = 4 : operating in the linear region of the static curve can simply pick every fourth value (withouth aliasing)
–  1 < g(n) < 4 : compressing
" Achieving a louder output signal overall
!  The computation can be spread / shared as follows:
–  0 < g(n) < 1 : expanding –  Levels PEAK/RMS are updated for each input sample
" Dynamic range of low-amplitude signals is increased –  The following modules are executed one at a time every four
–  Operation of the noise gate is not visible in the example cycles: (1) LD(x): logarithm, (2) CURVE: static curve,
(3) 2x : antilogarighm, (4) SMO : control factor smoothing
!  Note: the control factor indeed decreases the level of the
expanded part of the signal (if the additional gain of –  Values of the control factor are interpolated (upsampled) by
repeating each value 4 times and by multiplying (MULT) the input
+12 dB is not taken into account)
signal to the output
–  That way compression ratio R = ΔPI ΔPO becomes larger than 1
and level changes in input are mapped to larger changes in output
–  Visible in the static curve of G [dB] in §4.2

Dynamics 19 Dynamics 20

5 Stereo signals 6 Bandwise dynamic range control

SGN-14006 / A.K. SGN-14006 / A.K.

!  For stereo signals, a commong control factor g(n) is needed !  When processing complex audio signals, dynamic range
–  That way the stareo balance is not displaced control is sometimes performed bandwise
–  Input signal level is measured from channel average !  For example 4-band filterbank followed by compression at
!  Figure: dynamic range control of a stereo signal each band separately
!  De-esser: dynamic range control at sibilants (/s/
phoneme) frequency range (about 5 kHz)
–  Bandpass filter at 2-6 kHz is used to detect prominent phonemes.
–  If a threshold value is crossed, a so-called peak filter is used to cut
the level of /s/ phonemes in the same frequency range (peak filter:
has unity response everywhere else except around its center
frequency, where either boost or cut can be applied)