|
What
Happens to My Recording When it's Played on the Radio?
 Frank Foti,
Omnia Audio
Robert Orban, CRL/Orban
June, 2001
Few people in the record industry really know how a radio
station processes their material before it hits the FM
airwaves. This article's purpose is to remove the many myths
and misconceptions surrounding this arcane art.
Every radio station uses a
transmission audio processor in front of its transmitter. The
processor's most important function is to control the peak
modulation of the transmitter to the legal requirements of the
regulatory body in each station's nation. However, very few
stations use a simple peak limiter for this function. Instead,
they use more complex audio chains. These can accurately
constrain peak modulation while significantly decreasing the
peak-to-average ratio of the audio. This makes the station
sound louder within the allowable peak modulation.
Garbage In-Garbage Out
Manufacturers have tuned broadcast processors to process
the clean, dynamic program material that the recording
industry has typically released throughout its history. (The
only significant exception that comes to mind is 45-rpm
singles, which often were overtly distorted.) Because these
processors have to process speech, commercials, and oldies in
addition to current material, they can't be tuned exclusively
for "hypercompressed," distorted CDs. Indeed,
experience has shown that there's no way to tune them
successfully for this degraded material.
For 20 years, broadcast processor
designers have known that achieving highest loudness
consistent with maximum punch and cleanliness requires
extremely clean source material. For more than 20 years, Orban
has published application notes to help broadcast engineers
clean up their signal paths. These notes emphasize that any
clipping in the path before the processor will cause subtle
degradation that the processor will often exaggerate severely.
The notes promote adequate headroom and low distortion
amplification to prevent clipping even when an operator drives
the meters into the red.
About three years ago, we started to
notice CDs arriving at radio stations that had been
pre-distorted in production or mastering to increase their
loudness. For the first time, we started seeing frequently
reoccurring flat topping caused by brute-force clipping in the
production process. Broadcast processors react to
pre-distorted CDs exactly the same way as they have reacted to
accidentally clipped material for more than 20 years-they
exaggerate the distortion. Because of phase rotation, the
source clipping never increases on-air loudness-it just adds
grunge.
The authors understand the reasoning
behind the CD loudness wars. Just as radio stations wish to
offer the loudest signal on the dial, it is evident that
recording artists, producers, and even some record labels want
to have a loud product that stands out against its competition
in a CD changer or a music store's listening station.
In radio broadcasting this competition
has existed for at least the last 25 years. 25 years ago,
radio stations used simple clipping to get louder, and this
25-year-old technique has now migrated to the music industry.
The following graphic shows a section of a severely clipped
waveform from a contemporary CD. The area marked between the
two pointers highlights the clipped portion. This is one of
the roots of the problem as described in this paper; the other
is excessive digital limiting that does not necessarily cause
flat-topping, but still removes transient punch and impact
from the sound.
The problem today is that we now have
sophisticated and powerful audio processing for the broadcast
transmission system and this processing does not coexist well
with a signal that has already been severely clipped.
Unfortunately, with current pop CDs, the example shown above
is more the norm than the exception.
The attack and release characteristics
of broadcast multiband compression were tuned to sound natural
with source material having short-term peak-to-average ratios
typical of vinyl or pre-1990 CDs. Excessive digital limiting
of the source material radically reduces this short-term
peak-to-average ratio and presents the broadcast processor
with a new, synthetic type of source that the broadcast
processor handles less gracefully and naturally than it
handles older material. Instead of being punchy, the on-air
sound produced from these hypercompressed sources is small and
flat, without the dynamic contours that give music its
dramatic impact. The on-air sound resembles musical wallpaper
and makes the listener want to turn down the volume control to
background levels.
There is a myth that broadcast
processing will affect hypercompressed material less than it
will more naturally produced material. This is true in only
one aspect-if there is no long-term dynamic range coming in,
then the broadcast processor's AGC will not further reduce it.
However, the broadcast processor will still operate on the
short-term envelopes of hypercompressed material and will
further reduce the peak-to-average ratio, degrading the sound
even more.
Hypercompressed material does not
sound louder on the air. It sounds more distorted, making the
radio sound broken in extreme cases. It sounds small, busy,
and flat. It does not feel good to the listener when turned
up, so he or she hears it as background music.
Hypercompression, when combined with "major-market"
levels of broadcast processing, sucks the drama and life from
music. In more extreme cases, it sounds overtly distorted and
is likely to cause tune-outs by adults, particularly
women.
A Typical Processing Chain-What
Really Goes On When Your Recording is Broadcast:
A typical chain consists of the following elements, in the
order that they appear in the chain:
Phase rotator
The phase rotator is a chain of allpass filters (typically
four poles, all at 200Hz) whose group delay is very
non-constant as a function of frequency. Many voice waveforms
(particularly male voices) exhibit as much as 6dB asymmetry.
The phase rotator makes voice waveforms more symmetrical and
can sometimes reduce the peak-to-average ratio of voice by
3-4dB. Because this processing is linear (it adds no new
frequencies to the spectrum, so it doesn't sound raspy or
fuzzy) it's the closest thing to a "free lunch" that
one gets in the world of transmission processing.
There are a few prices to play. In the
good old days when source material wasn't grossly clipped, the
main price was a very subtle reduction in transparency and
definition in music. This was widely accepted as a valid
trade-off to achieve greatly reduced speech distortion,
because the phase rotator's effects on music are unlikely to
be heard on typical consumer radios, like car radios, boom boxes,
"Walkman"-style portables, and table radios.
However, with the rise of the clipped
CD, things have changed. The phase rotator radically changes
the shape of its input waveform without changing its frequency
balance: If you measured the frequency response of the phase
rotator, it would measure "flat" unless you also
measured phase response, in which case you would say that the
"magnitude response" was flat and the phase response
was highly non-linear with frequency. The practical effect of
this non-linear phase response is that flat tops in the
original signal can end up anywhere in the waveform after
processing. It's common to see them go right through a zero
crossing. They end up looking like little smooth sections of
the waveform where all the detail is missing-a bit like a scar
from a severe burn. This is an apt metaphor for their audible
effect, because they no longer help reduce the peak-to-average
ratio of the waveform. Instead, their only effect is to add
unnecessary grungy distortion.
There has been a myth in the recording
world that broadcast processing will modify these clipped,
over-compressed CDs less it will modify clean, dynamic CDs.
Thanks in part to phase rotation, this myth is absolutely
false. In particular, any clipping in the source material
causes nothing but added distortion without increasing on-air
loudness at all.
AGC
The next stage is usually an average-responding AGC. By
recording studio standards, this AGC is required to operate
over a very wide dynamic range-typically in the range of 25dB.
Its function is to compensate for operator errors (in live
production environments) and for varying average levels (in
automated environments). Average levels vary mainly because
the peak to average ratio of CDs themselves has varied so much
in the last 10 years or so. Therefore, normalizing hard disk
recordings (to use all available headroom) has the undesirable
side effect of causing gross variations in average levels.
Indeed, 1:1 transfers (which are also common) will also
exhibit this variation, which can be as large as 15dB.
The price to be paid is simple: the
AGC will eliminate long-term dynamics in your recording.
Virtually all radio station program directors want their
stations to stay loud always, eliminating the risk that
someone tuning the radio to their station will either miss the
station completely or will think that it's weak and can't be
received satisfactorily. Radio people often call this effect
"dropping off the dial."
AGCs can be either single-band or
multiband. If they are multiband, it's rare to use more than
two bands because AGCs operate slowly, so "spectral gain
intermodulation" (such as bass' pumping the midrange) is
not as big a potential problem as it is for later compression
stages, which operate more quickly.
AGCs are always gated in competent
processors. This means that their gain essentially freezes if
the input drops below a preset threshold, preventing noise
suck-up despite the large amount of gain reduction.
Stereo Enhancement
Not all processors implement stereo enhancement, and those
that do may implement it somewhere other than after the AGC.
(In fact, stand-alone stereo enhancers are often placed in the
program line in front of the transmission processor.)
The common purpose of stereo
enhancement is to make the signal stand out dramatically when
the car radio listener punches the tuning button. It's a
technique to make the sound bigger and more dramatic.
Overdone, it can remix the recording. Assuming that stereo
reverb, with considerable L-R energy, was used in the original
mix, stereo enhancement, for example, can change the amount of
reverb applied to a center-channel vocalist. The moral? When
mixing for broadcast, err on the "dry" side, because
some stations' processors will bring the reverb more to the
foreground.
Because each manufacturer uses a
different technique for stereo enhancement, it's impossible to
generalize about it. The only universal constraints are the
need for strict mono compatibility (because FM radio is
frequently received in mono, even on "stereo"
radios, due to signal-quality-trigged mono blend circuitry),
and the requirement that the stereo difference signal (L-R)
not be enhanced excessively. Excessive enhancement always
increases multipath distortion (because the part of the FM
stereo signal that carries the L-R information is more
vulnerable to multipath). Excessive enhancement will also
reduce the loudness of the transmission (because of the
"interleaving" properties of the FM stereo composite
waveform, which we won't further discuss).
These constraints mean that
recording-studio-style stereo enhancement is often
incompatible with FM broadcast, particularly if it
significantly increases average L-R levels. In the days of
vinyl, a similar constraint existed because of the need to
prevent the cutter head from lifting off the lacquer, but with
CDs, this constraint no longer exists. Nevertheless, any mix
intended for airplay will yield the lowest distortion and
highest loudness at the receiver if its L-R/L+R ratio is low.
Ironically, mono is loudest and cleanest!
Equalization
Equalization may be as simple as a fixed-frequency bass boost,
or as complex as a multi-stage parametric equalizer. EQ has
two purposes in a broadcast processor. The first is to
establish a signature for a given radio station that brands
the station by creating a "house sound." The second
purpose is to compensate for the frequency contouring caused
by the subsequent multiband dynamics processing and high
frequency limiting. These may create an overall spectral
coloration that can be corrected or augmented by carefully
chosen fixed EQ before the multiband dynamics stages.
Multiband Compression and Limiting
Depending on the manufacturer, this may occur in one or
two stages. If it occurs in two stages, the multiband
compressor and limiter can have different crossovers and even
different numbers of bands. If it occurs in one stage, the
compressor and limiter functions can "talk" to each
other, optimizing their interaction. Both design approaches
can yield good sound and each has its own set of tradeoffs.
Usually using anywhere between four
and six bands, the multiband compressor/limiter reduces
dynamic range and increases audio density to achieve
competitive loudness and dial impact. It's common for each
band to be gated at low levels to prevent noise rush-up, and
manufacturers often have proprietary algorithms for doing this
while minimizing the audible side effects of the gating.
An advanced processor may have dozens
of setup controls to tune just the multiband
compressor/limiter. Drive and output gain controls for the
various compressors, attack and release time controls,
thresholds, and sometimes crossover frequencies are
adjustable, depending on the processor design. Each of these
controls has its own effect on the sound, and an operator
needs extensive experience if he or she is to tune a broadcast
multiband compressor so that it sounds good on a wide variety
of program material without constant readjustment. Unlike
mastering in the record industry, in broadcast there's no
mastering engineer available to optimize the processing for
each new source!
Pre-Emphasis and HF Limiting
FM radio is pre-emphasized at 50 microseconds or 75
microseconds, depending on the country in which the
transmission occurs. Pre-emphasis is a 6dB/octave high
frequency boost that's 3dB up at 2.1kHz (75µs) or 3.2kHz
(50µs). With 75µs pre-emphasis, 15kHz is up 17dB!
Depending on the processor's
manufacturer, pre-emphasis may be applied before or after the
multiband compressor/limiter. The important thing for mixers
and mastering engineers to understand is that putting lots of
energy above 5kHz creates significant problems for any
broadcast processor because the pre-emphasis will greatly
increase this energy. To prevent loudness loss, the processor
applies high frequency limiting to these boosted high
frequencies. HF limiting may cause the sound to become dull,
distorted, or both, in various combinations. One of the most
important differences between competing processors is how
effectively a given processor performs HF limiting to minimize
audible side effects. In state-of-the-art processors, HF
limiting is usually performed partially by HF gain reduction
and partially by distortion-cancelled clipping.
Clipping
In most processors, the clipping stage is the primary
means of peak limiting. It's crucial to broadcast processor
performance. Because of the FM pre-emphasis, simple clipping
doesn't work well at all. It produces difference-frequency IM
distortion, which the de-emphasis in the radio then
exaggerates. (The de-emphasis is flat below 2-3kHz, but rolls
off at 6dB/octave thereafter, effectively exaggerating energy
below 2-3kHz.) The result is particularly offensive on cymbals
and sibilance ("essses" become "efffs").
In the late seventies, one of the
authors of this article (R.O.) invented distortion-cancelled
clipping. This manipulates the distortion spectrum added by
the clipper's action. In FM, it typically removes the
clipper-induced distortion below 2kHz (the flat part of the
receiver's frequency response). This typically adds about 1dB
to the peak level emerging from the clipper, but, in exchange,
allows the clipper to be driven much harder than would
otherwise be possible.
Provided that it doesn't introduce
audibly offensive distortion, distortion-cancelled clipping is
a very effective means of peak limiting because it affects
only the peaks that actually exceed the clipping threshold and
not surrounding material. Accordingly, clipping does not cause
pumping, which gain reduction can do, particularly when gain
reduction operates on pre-emphasized material. Clipping also
causes minimal HF loss by comparison to HF limiting that uses
gain reduction. For these reasons, most FM broadcast
processors use the maximum practical amount of clipping that's
consistent with acceptably low audible distortion.
Real-world clipping systems can get
very complicated because of the requirement to strictly
band-limit the clipped signal to less than 19kHz despite the
harmonics that clipping adds to the signal. (Bandlimiting
prevents aliasing between the stereo main and subchannel,
protects subcarriers located above 55kHz in the FM stereo
composite baseband, and protects the stereo pilot tone at
19kHz). Linearly filtering the clipped signal to remove energy
above 15kHz causes large overshoots (up to 6dB in worst case)
because of a combination of spectral truncation and time
dispersion in the filter. Even a phase-linear lowpass filter
(practical only in DSP realizations) causes up to 2dB
overshoot. Therefore, state-of-the-art processors use complex
overshoot compensation schemes to reduce peaks without
significantly adding out-of-band spectrum.
Some chains also apply composite
clipping or limiting to the output of the stereo encoder. The
stereo encoder is the circuit that encodes the left and right
channels into the single multiplex signal that drives the
transmitter, and it's actually the peak level of this signal
that government broadcasting authorities regulate. Composite
clipping or limiting has long been a controversial technique,
but the latest generation of composite clippers or limiters
has greatly reduced the interference problems characteristic
of earlier technology.
Conclusions
Broadcast processing is complex and sophisticated, and was
tuned for the recordings produced using practices typical of
the recording industry during almost all of its history. In
this historical context, hypercompression is a short-term
anomaly and does not coexist well with the
"competitive" processing that most pop-music radio
stations use. We therefore recommend that record companies
provide broadcasters with radio mixes. These can have all of
the equalization, slow compression, and other effects that
producers and mastering engineers use artistically to achieve
a desired "sound." What these radio mixes should not
have is fast digital limiting and clipping. Leave the
short-term envelopes unsquashed. Let the broadcast processor
do its work. The result will be just as loud on-air as
hypercompressed material, but will have far more punch,
clarity, and life.
A second recommendation to the record
industry is to employ studio or mastering processing that
provides the desired sonic effect, but without the undesired
extreme distortion component that clipping creates. The
alternative to brute-force clipping is digital look-ahead
limiting, which is already widely available to the recording
industry from a number of different manufacturers (including
the authors' companies). This processing creates lower
modulation distortion than clipping and also avoids blatant
flat-topping of waveforms. Compared to clipping, it is
therefore substantially more compatible with broadcast
processing. Nevertheless, even digital limiting can have a
deleterious effect on sound quality by reducing the
peak-to-average ratio of the signal to the point that the
broadcast processor responds to it in an unnatural way, so it
should be used conservatively. Ultimately, the only way to
tell how one's production processing will interact with a
broadcast processor is to actually apply the processed signal
to a real-world broadcast processor and to listen to its
output, preferably through a typical consumer radio.
Top
|
