Digital Distortion in CD’s and DVD’s:

The Consequences of Traditional Digital Peak Meters

A white paper by Nika Aldrich for Trillium Lane Labs

Introduction

The advent of the Red Book CD standard in the early 1980’s was heralded as a
breakthrough in consumer audio quality. Many discerning listeners however, claim that
modern CD’s have actually decreased in audio quality. This white paper explores the
impact of industry demands for the loudest possible mixes and the resulting effect on
digital audio quality as one possible reason for the perceived decrease in quality in
modern CD’s.

History

The digital sampling theory was first proposed in 1928 in a paper by H. Nyquist entitled
“Certain Topics in Telegraph Transmission Theory” [Nyquist 1928] The theory was
mathematically proven by mathematician Claude Shannon in his paper, “Communication
in the Presence of Noise” [Shannon 1949]. According to Shannon’s theorem:

“Theorem 1: If a function f(t) contains no frequencies higher than W cps [cycles per
second, or “Hz”], it is completely determined by giving its ordinates at a series of points
spaced 1/2W seconds apart.”

July 2003 Page 1

The significance of this statement is often overlooked. Shannon tells us that the entire
waveform, in proper amplitude, frequency, and phase can be recreated (as he says,
“completely determined”) with only sampling points given at greater than half the highest
frequency to be sampled. A key observation is that the entire waveform is represented by
the sampling points, but a reconstruction process still needs to occur in order to recreate
the waveform represented. One cannot simply ‘connect the dots’ between sample points
and yield the original waveform. Shannon says that the waveform (function) can be
“completely determined” but does not describe that process – he simply alludes that it can
be done.

July 2003 Page 2

The process of recreating the original waveform involves a filter called a reconstruction
filter. This filter removes all content above the Nyquist frequency (half the sample rate).
The range below the Nyquist frequency defines the ‘legal’ range of allowed frequencies
as frequencies in this range can be accurately reproduced. All frequencies above the
Nyquist frequency do not adhere to Nyquist or Shannon’s theorems regarding allowable
frequencies, cannot be reproduced and are therefore considered illegal frequencies.
Because of mathematical realities observed by Fourier in the 1800’s, and subsequently by
Shannon in 1948, when a waveform has all frequencies removed above the Nyquist
frequency, the resulting waveform will be the original waveform that was sampled.

This process is significantly more involved than simply ‘connecting the dots’ between
sample points. Today it involves extremely sophisticated means of reconstructing the
waveform, using filters that are highly complex mathematical systems utilizing
‘oversampling,’ ‘upsampling,’ ‘linear phase, equiripple FIR’ designs and much more.
The result is that today’s digital to analog converters get closer to the original than ever
before, making music played on systems today as accurate as possible. Even today’s
inexpensive components such as off-the-shelf CD players have drastically improved
filters, and thus better reconstruction abilities than in years past.

July 2003 Page 3

Application

Most contemporary audio recording is done with Digital Audio Workstations (DAWs),
although digital mixing systems in the form of outboard digital mixers are also very
popular. To the user, these digital systems appear similar to traditional audio tools, and
are designed order to emulate the operation of a conventional analog recording system.

One familiar analog tool that has been bought over to the digital realm is a ‘peak meter’
that tells the amplitude of the waveform’s peaks. In the analog realm, peak signal was an
indicator that would tell the audio engineer when the peak signal level was getting too
high. A peak signal in analog recording would cause the tape to saturate, creating
distortion. In an analog system however, this type of distortion was often deliberately
engineered into tracks in order to achieve a certain sound. In the digital realm this type of
meter is important and more vital, because if the amplitude of a waveform exceeds the
top of the measurable scale (full scale, or ‘full code’) the signal will ‘clip’ causing
unwanted and unpleasant distortion rather than the traditional distorted sound of analog.
This digital clipping occurs because the waveform is ‘lopped off’ and the data is changed.
When the waveform is reconstructed it cannot be accurately done so in order to represent
the original waveform. Instead, it has a significant amount of inharmonic distortion
caused by aliasing. For this reason, digital recording has a maximum level at which
signals can be recorded. Anything exceeding this level (full scale) has undesirable
consequences.

The method used for computing the peak value inside the system however, is not
particularly accurate. DAW and digital mixer manufacturers typically take the amplitude
of the samples and use these as the basis for the peak meter. The problem with this
approach is easily identified: the samples themselves do not represent the peak value of
the waveform. The waveform is only complete after the reconstruction process. Until
this process has been completed, the waveform is inaccurately represented by the
samples. This is the reason that in most DAWs the waveform is represented on the
screen as a ‘dot to dot’ connection between sample points. They do not undergo the
reconstruction process inside the system, so all that can be represented is the sample
points, and for the sake of visual ease, they connect the dots between them with straight
lines. They save the reconstruction process for the digital to analog converters and show
the user inaccurate information instead.

July 2003 Page 4

The consequence of the way in which DAW’s treat waveforms is that the meter inside the
DAW or other digital mixers inevitably shows inaccurate information. It is virtually a
mathematical certainty that the waveform will actually exceed the amplitude of the
samples in any sampling system. The samples themselves only represent a waveform, It
is important to understand that the amplitude of the waveform will invariably exceed the
sample values.

Manifestation

One may ask why this poses a problem. For various reasons, mostly having to do with
marketing demands and industry trends, recordings made and mastered in today’s
recording environment are mixed and mastered as ‘hot’ as is possible, pushing the levels
up to the highest tolerable amount, supposedly just short of clipping. Sophisticated digital
tools allow music to be highly compressed, then recompressed, compressed even more so
with multi-band compressors, limited, normalized, and maximized to get the audio to
play as loud as possible out of a consumer’s system. Hence, it is very common for
popular music CDs to be full of digital samples that are at, or nearly at full scale.

The problem is realized in that while going through these digital gyrations and utilizing
digital tools to amplify the signal as much as possible, both during mixing and during
mastering, the ‘peak value’ of the sample points is closely watched to ensure that it does
not get to full scale. Since, the peak meters in said DAW and digital mixing systems are
inaccurate, and do not actually indicate the peak values of the resulting waveform, the
result is that while the samples themselves do not exceed full scale, and are carefully
monitored to insure this, the resulting waveforms represented by the samples may exceed
full scale throughout any standard CD!

July 2003 Page 5

While the digital mixing system is not clipping the music or distorting the music, the
digital to analog converters that have the task of recreating the audio through digital
reconstruction filters are clipping repeatedly throughout most CDs on the market. The
result is that most CDs and DVDs end up distorting with regularity when they are asked
to reconstruct and play back audio that appears to be completely ‘legal’ because not a
single sample actually clipped.

In a recent paper [Nielsen 2003], seven consumer CD players were subjected to tests
designed to analyze their ability to reproduce and reconstruct signal levels above full
scale (0dBFS). All of the players experienced difficultly dealing with signal levels this
high, further showing that, while all of the samples can be legal, the level can still be
hotter than is legal the result being that a CD player can be unable to reproduce the audio
accurately.

It is nearly certain that this constant barrage of distortion that we, the consumers, are
hearing on compressed and mastered CDs contributes to the ‘digital harshness’ still
reported by the more sensitive audiophiles in the music industry. According to industry
insiders, not a single off-the-shelf digital to analog converter chip made today can
accurately pass a signal wherein the samples are under full scale but the waveform that
they represent exceeds full scale. Only a few high end converters in the professional
market can do this. This means that the preponderance of consumer (and professional
audio) playback equipment is not designed to deal with these ‘hotter than full scale’
signal levels.

Monitoring in most mastering studios is typically performed using high end digital
converters. Consequentially, audio and mastering engineers are often putting out music
that cannot be accurately recreated by consumer playback equipment. In some cases the
reconstruction sounds ‘perfect’ to the mastering engineer – because the engineer’s
equipment actually can reproduce the waveforms properly.

July 2003 Page 6

Potential solutions

There are several potential solutions to this problem. On the reproduction side of the
equation, one solution would be to equip the consumers’ home and car playback
equipment with modified digital to analog conversion so that it can actually recreate
illegal waveforms that exceed full scale at their peaks. The magnitude of this chore is not
only unrealistic, but unnecessary as well. In truth, the waveforms we are asking the
machines to recreate are illegal and contain information that exceeds the bounds specified
by the systems being used. Rather than change the equipment, a more appropriate
solution can be found in merely changing the content.

On the content creation side, there are also alternatives for audio and mastering
engineers. Changing the content requires not putting out mastered music that exceeds full
scale. This means putting out music that, even though the samples themselves hardly
ever exceed full scale, does not have the waveforms exceeding full scale.

One alternative for engineers is simply to turn down the level by a fixed amount at the
mastering stage, to ensure the waveform will not clip when reconstructed. This is an
imperfect solution for two reasons. First, it sacrifices potentially unused dynamic range
and second, it is unlikely to be acceptable to clients given current industry demands for
the loudest possible mixes.

The second is to monitor the reconstructed waveform for clipping at the final mix and
mastering stages and make appropriate adjustments without sacrificing overall level or
dynamic range. This requires a digital mixing and mastering system that has peak meters
that simulate the reconstruction filters used in digital to analog converters throughout the
professional audio and consumer industries. This task is difficult without appropriate
metering tools that allow mixing and mastering engineers to know what the actual signal
level of their music is while they work on it, rather than just the level of the samples used
to represent it.

Impact on mixing and mastering

The inevitable result is that, in order to comply with the actual legal range of the digital
audio system, mixes will have to be reduced in amplitude so that when a specific
waveform exceeds the sample values there is enough headroom for it to be reconstructed
below full scale. Studies have shown that waveforms can exceed full scale (considering
the reconstruction filters on most digital to analog converters) by more than 6dB. This
means that the peak amplitude of the actual waveform might be more than twice as high
in amplitude as the highest sample value. This is only likely to happen when music is
heavily compressed however, and most music will practically require less than 6dB of
headroom above the highest sample value to ensure accurate reproduction.

July 2003 Page 7

In the end the mastering engineer will be left with two choices if they want the music to
be in compliance. They will either need to compress or limit the music less, so that the
RMS value of the music is lower, lowering the ‘perceived’ level of the music in order to
keep it legal, meaning that the music will take advantage of the increase in headroom to
have more dynamic range. The other solution is that they can continue to compress the
music and limit it as hard, but will have to turn it down to give room for the peaks that
occur above the sample levels. Either way, the result of compliance will require a
lowering of music that ends up in delivery formats to a return to the legal range.

It is worth noting that Sony’s new SACD format includes measures that prevent the
music from ever clipping in the way described. Mastering engineers who work on SACD
releases have observed the notion that heavily compressing the audio inevitably results in
the need to ‘turn down’ the overall level on the disk. Left with the choice of compressing
the disk and turning it down, or simply leaving it the opportunity to ‘breathe’ with some
headroom, most mastering engineers are mastering to SACD disks differently than they
have been to DVD’s and CDs. Many professionals in the audio industry are claiming that
audio that has been remastered for the SACD format ‘breathes more,’ ‘has more life,’ and
‘doesn’t have the digital harshness’ of the CD counterparts.

July 2003 Page 8

Conclusion

The mastering or mixing engineer that first starts using an oversampled peak meter
capable of representing the audio waveforms may at first be frustrated that it is difficult
to get their final result as loud as they could otherwise. This is only partially true.

Of course, since many popular music CD’s have been clipping consumer digital to analog
converters, accommodating those systems will inevitably require lowering the level in
some method or another, resulting in a quieter final product in some capacity, although
likely by only a few decibels. Where this is not necessarily true, however, is that the
PCM system and the CD both allow for the representation of illegal waveforms such that
it is not a requirement to lower the level just because the mastering engineer is endowed
with tools that show him that he is allowing a distorted result to be reproduced. The red
book format for CDs and the DVD specs both allow for this illegal content, and the
mastering engineer is still allowed to put out releases that meet the spec while allowing
consumers’ players to distort. At least with an oversampled peak meter the engineer will
be able to know that the music is clipping, by how much, and where. With this
knowledge the engineer can then decide with complete information whether or not to
accommodate the legal range of digital audio on a PCM sampled system.

The consumer has continued to complain that CDs and DVD’s sound ‘harsh,’ the
mastering engineers have argued that peak meters continue to be inaccurate, and
everyone continues to demand better sounding mixes. Utilizing an oversampled peak
meter in the digital audio studio that represents the reconstruction filters in digital to
analog converters is the first step toward an improvement in audio quality in music
releases.

July 2003 Page 9

References and further reading

Aldrich, Nika. Digital Audio Explained For the Audio Engineer. San Francisco:
Backbeat Books, 2004.

Banquer, Dan, Dick Pierce, Herbie Robinson, et al. "Intersample Peaking." Pro Audio
Mailing List. 21 December, 2002 - 31 December, 2002.

Nielsen, Soren and Thomas Lund. "Level Control in Digital Mastering." Preprint 5019,
107th AES Convention. Denmark, 1999.

Nielsen, Soren and Thomas Lund. "0dBFS+ Levels in Digital Mastering." TC

Electronic: Risskov, Denmark. 17 July, 2003.
<http://www.tcelectronic.com/media/Level_paper_AES109.pdf>.

Nyquist, Henry. "Certain Topics in Telegraph Transmission Theory." Transactions of

the AIEE. Vol. 47 (April 1928): 617-644.

Shannon, Claude E. "Communication in the Presence of Noise." Proceedings of the IRE.

Vol. 37 (January 1949): 10-21.

July 2003 Page 10