MPEG Layer-3

An introduction to
MPEG Layer-3
K. Brandenburg and H. Popp
Fraunhofer Institut für Integrierte Schaltungen (IIS)

MPEG Layer-3, otherwise known as MP3, has generated a phenomenal

interest among Internet users, or at least among those who want to
download highly-compressed digital audio files at near-CD quality.

This article provides an introduction to the work of the MPEG group which
was, and still is, responsible for bringing this open (i.e. non-proprietary)
compression standard to the forefront of Internet audio downloads.

1. Introduction
The audio coding scheme MPEG Layer-3 will soon celebrate its 10th birthday, having
been standardized in 1991. In its first years, the scheme was mainly used within DSP-
based codecs for studio applications, allowing professionals to use ISDN phone lines as
cost-effective music links with high sound quality. In 1995, MPEG Layer-3 was selected
as the audio format for the digital satellite broadcasting system developed by World-
Space. This was its first step into the mass market. Its second step soon followed, due to
the use of the Internet for the electronic distribution of music. Here, the proliferation of
audio material – coded with MPEG Layer-3 (aka MP3) – has shown an exponential
growth since 1995. By early 1999, “.mp3” had become the most popular search term on
the Web (according to http://www.searchterms.com). In 1998, the “MPMAN” (by
Saehan Information Systems, South Korea) was the first portable MP3 player, pioneering
the road for numerous other manufacturers of consumer electronics.

“MP3” has been featured in many articles, mostly on the business pages of newspapers
and periodicals, due to its enormous impact on the recording industry. This article
explains the basic technology and some of the special features of MPEG Layer-3. It also
sheds some light on the factors which determine the quality of the coded audio signal.

EBU TECHNICAL REVIEW – June 2000 1 / 15

Popp and Brandenburg
 MPEG Layer-3

2. Why MPEG Layer-3?

MPEG Layer-3 emerged as the main tool for Internet audio delivery. Considering the
reasons, the following factors were definitely helpful.

Open standard

MPEG is defined as an open standard. The specification is available (for a fee) to every-
body. While there are a number of patents covering MPEG Audio encoding and decod-
ing, all patent-holders have declared that they will license the patents on fair and
reasonable terms to everybody. Public example source code is available to help imple-
menters to understand the text of the specification. As the format is well defined, no
problems with interoperability of equipment or software from different vendors have
been reported – except from some rare incomplete implementations.

Availability of encoders and decoders

DSP-based hardware and software encoders and decoders have been available for a
number of years – driven at first by the demand for professional use in broadcasting.

Supporting technologies

While audio compression is viewed as a main enabling technology, other evolving tech-
nologies contributed to the MP3 boom, such as:

the widespread use of computer sound cards;

computers becoming powerful enough to run software audio decoders and even
encoders in real-time;

fast Internet access for universities and businesses;

the availability of CD-ROM and CD-Audio writers.

In short, MPEG Layer-3 had the luck to be the right technology available at the right
time. In the meantime, research on perceptual audio coding progressed, and codecs with
better compression efficiency became available. Of these, MPEG-2 Advanced Audio
Coding (AAC) was developed as the successor of MPEG-1 Audio. Other – proprietary –
audio coding schemes were also introduced, claiming a higher performance than MP3.

EBU TECHNICAL REVIEW – June 2000 2 / 15

Popp and Brandenburg
 MPEG Layer-3

3. MPEG audio coding standards

MPEG, a working group formally named as ISO/IEC JTC1/SC29/ WG11, but mostly
known by its nickname, Moving Pictures Experts Group, was set up by the ISO/IEC
standardization body in 1988 to develop generic (i.e. useful for different applications)
standards for the coded representation of moving pictures, associated audio and their
combination. Since then, MPEG has undertaken the standardization of compression
techniques for video and audio. Originally, its main goal was video coding together with
audio coding for digital storage media. In the meantime, the MPEG audio coding stand-
ard found its way into many different applications, including

digital audio broadcasting (Eureka-147 DAB, WorldSpace, ARIB, DRM);

ISDN transmission for broadcast contribution and distribution purposes and for
commentary links;

archival storage within broadcasting;

sound for digital television (DVB, Video CD, ARIB);

internet streaming (Microsoft Netshow, Apple Quicktime);

portable audio devices (mpman, mplayer3, Rio, Lyra, YEPP and many more);

storage and exchange of music files on computers.

The most widely-used audio compression formats are MPEG Audio Layer-2 and Layer-
3 (see below for definitions) and Dolby AC-3. A large number of systems currently
under development will use AAC.

MPEG-1

MPEG-1 is the name for the first phase of MPEG work, starting in 1988. This work was
finalized with the adoption of the ISO/IEC standard IS 11172 in late 1992. The audio
coding part of this standard (IS 11172-3) describes a generic coding system, designed to
fit the demands of many applications. MPEG-1 Audio consists of three operating modes

Abbreviations
AAC (MPEG-4) Advanced Audio Coding IEC International Electrotechnical Commission
ACTS Advanced Communications Technologies ISDN Integrated services digital network
and Services ISO International Organization for Standardiza-
ARIB Association of Radio Industries and tion
Businesses (Japan) JTC Joint Technical Committee
DAB Digital Audio Broadcasting MDCT Modified discrete cosine transform
DRM Digital Radio Mondiale MPEG Moving Picture Experts Group
DSP Digital signal processor / processing PEAQ Perceptual evaluation of audio quality
DVB Digital Video Broadcasting RACE R&D in Advanced Communications
HDTV High-definition television technologies in Europe

EBU TECHNICAL REVIEW – June 2000 3 / 15

Popp and Brandenburg
 MPEG Layer-3

called “Layers”, with increasing complexity and performance, named Layer-1, Layer-2
and Layer-3. Layer-3, with the highest complexity, was designed to provide the highest
sound quality at low bit-rates (around 128 kbit/s for a typical stereo signal).

MPEG-2

MPEG-2 denotes the second phase of MPEG. It introduced a lot of new concepts into
MPEG video coding, including support for interlaced video signals. The main applica-
tion area for MPEG-2 is digital television. The original MPEG-2 Audio standard (IS
13818-3) was finalized in 1994 and consisted of two extensions to MPEG-1 Audio:

Multichannel audio coding, including the 5.1 channel configuration well known
from cinema sound – this multichannel extension is done in a backward compati-
ble way, allowing MPEG-1 stereo decoders to reproduce a mixture of all available
channels.

Coding at lower sampling frequencies – this extension adds sampling frequencies
of 16 kHz, 22.05 kHz and 24 kHz to the MPEG-1 sampling frequencies of 32 kHz,
44.1 kHz and 48 kHz, improving the coding efficiency at very low bit-rates.

MPEG-2 AAC

In early 1994, verification tests showed that new coding algorithms, without backward
compatibility to MPEG-1, promised a significant improvement in coding efficiency. As
a result, a new work item was defined that finally led to the definition of a new MPEG
audio coding standard, MPEG-2 Advanced Audio Coding (AAC). The standard was
finalized in 1997 (IS 13818-7). AAC is a second-generation audio coding scheme for
generic coding of stereo and multichannel signals, supporting sampling frequencies from
8 kHz to 96 kHz and a number of audio channels ranging from 1 to 48.

MPEG-3

Originally, MPEG had a plan to define the video coding for HDTV applications in a fur-
ther phase, to be called MPEG-3. However, later on it turned out that the tools devel-
oped for MPEG-2 video coding would also address the HDTV requirements, and MPEG
gave up the plan to develop a special MPEG-3 standard. Sometimes MPEG Layer-3 (or
“MP3”) is misnamed MPEG-3.

MPEG-4

MPEG-4 intends to become the next major standard in the world of multimedia. The
first version was finished in late 1998 (IS 14496-3), and the second version at the end of

EBU TECHNICAL REVIEW – June 2000 4 / 15

Popp and Brandenburg
 MPEG Layer-3

1999. Unlike MPEG-1 and MPEG-2, the emphasis in MPEG-4 is on new functionalities
rather than better compression efficiency. Mobile as well as stationary user terminals,
database access, communications and new types of interactive services will be major
applications for MPEG-4. The new standard facilitates the growing interaction and over-
lap between the hitherto separate worlds of computing, electronic mass media (TV and
Radio) and telecommunications. MPEG-4 audio consists of a family of audio coding
algorithms – spanning the range from low bit-rate speech coding (down to 2 kbit/s) up to
high-quality audio coding at 64 kbit/s per channel and above. Generic audio coding at
medium to high bit-rates is done by AAC.

MPEG-7

Unlike MPEG-1, MPEG-2 and MPEG-4, MPEG-7 does not define compression algo-
rithms. MPEG-7 is a content representation standard for multimedia information search,
filtering, management and processing. MPEG-7 will be approved by July, 2001.

4. MPEG Layer-3 audio encoding

The following description of MPEG Layer-3 encoding focuses on the basic functions and
a number of details necessary to understand the implications of encoding options on the
sound quality. It is not meant to be a complete description of how to build an MPEG
Layer-3 encoder.

Flexibility

In order to be applicable to a number of very different application scenarios, MPEG

defined a data representation including a number of options.

Operating mode

MPEG-1 Audio works for both mono and stereo signals. A technique called joint stereo
coding can be used to achieve a more efficient combined coding of the left and right
channels of a stereophonic audio signal. Layer-3 allows both mid/side stereo coding and
intensity stereo coding. The latter is especially helpful for lower bit-rates, but bears the
risk of changing the sound image. The operating modes are:

single channel;

dual channel (two independent channels, for example containing different lan-
guage versions of the audio);

stereo (no joint stereo coding);

joint stereo.

EBU TECHNICAL REVIEW – June 2000 5 / 15

Popp and Brandenburg
 MPEG Layer-3

Sampling frequency

MPEG audio compression works on a number of different sampling frequencies.

MPEG-1 defines audio compression at 32 kHz, 44.1 kHz and 48 kHz. MPEG-2 extends
this to half the rates, i.e. 16 kHz, 22.05 and 24 kHz. “MPEG-2.5” is the name of a pro-
prietary extension to Layer-3, developed by Fraunhofer IIS, which introduces the sam-
pling frequencies 8 kHz, 11.05 kHz and 12 kHz.

Bit-rate

MPEG Audio does not just work at a fixed compression ratio. The selection of the bit-
rate of the compressed audio is, within some limits, completely left to the implementer or
operator of an MPEG audio coder. For Layer-3, the standard defines a range of bit-rates
from 8 kbit/s up to 320 kbit/s. Furthermore, Layer-3 decoders must support the switch-
ing of bit-rates from audio frame to audio frame. Combined with the bit reservoir tech-
nology, this allows both variable bit-rate coding and constant bit-rate coding at any fixed
value within the limits set by the standard.

Normative versus Informative

A very important property of the MPEG standards is the principle of minimizing the
amount of normative elements in the standard. In the case of MPEG Audio, this led to
the fact that only the data representation, i.e. the format of the compressed audio, and the
decoder are normative.

Decoder considerations

Even the decoder is not specified in a bit-exact fashion. Instead, formulae are given for
most parts of the algorithm, and compliance is defined by a maximum deviation of the
decoded signal from a reference decoder, implementing the formulae with double-preci-
sion arithmetic accuracy. This allows us to build decoders running both on floating-point
and fixed-point architectures. Depending on the skills of the implementers, fully-com-
pliant high-accuracy Layer-3 decoders can be constructed with down to 20-bit arithmetic
wordlength, without using double-precision calculations.

Encoder considerations

Encoding of MPEG Audio is completely left to the implementer of the standard. As a

helpful guide-line, the ISO standards contain the description of example encoders.
While these descriptions were derived from the original encoders used for verification
tests, a lot of experience and knowledge is necessary to implement good-quality MPEG

EBU TECHNICAL REVIEW – June 2000 6 / 15

Popp and Brandenburg
 MPEG Layer-3

audio encoders. The amount of investment necessary to engineer a high-quality MPEG

audio encoder has kept the number of independently-developed encoder implementa-
tions very low.

5. MPEG Layer-3 – the algorithm

The following paragraphs Quantized

samples
describe the Layer-3 encod- Digital
Analysis Quantization Encoding of Encoded
ing algorithm along with the audio
filterbank & coding bitstream bitstream
signal
basic blocks of a perceptual
encoder. More details about Perceptual
Layer-3 can be found in [1] model
and [2]. Fig. 1 shows the
block diagram of a typical Figure 1
MPEG Layer-3 encoder. Typical MPEG Layer-3 encoder

Filterbank

The filterbank used in MPEG Layer-3 belongs to the class of hybrid filterbanks. It is
built by cascading two different kinds of filterbanks, first a polyphase filterbank (as used
in Layer-1 and Layer-2) and second, a Modified Discrete Cosine Transform (MDCT) fil-
terbank. The polyphase filterbank fulfils the purpose of making Layer-3 more similar to
Layer-1 and Layer-2. The subdivision of each polyphase frequency band into 18 finer
sub-bands increases the potential for redundancy removal, leading to better coding effi-
ciency for tonal signals. As a further positive result of the higher frequency resolution,
the error signal can be better controlled, allowing a finer tracking of the masking thresh-
old. The filterbank can be switched to a lower frequency resolution to avoid pre-echoes
(see below).

Perceptual model

The perceptual model mainly determines the quality of a given encoder implementation.
Since the original informative part in the standard was written, a lot of additional work
has gone into this part of the encoder . The perceptual model either uses a separate filter-
bank as described in [1] or combines the calculation of energy values (for the masking
calculations) and the main filterbank. The output of the perceptual model consists of val-
ues for the masking threshold or the allowed noise for each coder partition. In Layer-3,
these coder partitions are roughly equivalent to the critical bands of human hearing. If
the quantization noise can be kept below the masking threshold for each coder partition,
then the compression result should be indistinguishable from the original signal.

EBU TECHNICAL REVIEW – June 2000 7 / 15

Popp and Brandenburg
 MPEG Layer-3

Quantization and coding

A system of two nested iteration loops is the common solution for quantization and cod-
ing in a Layer-3 encoder. Quantization is done via a power-law quantizer. In this way,
larger values are automatically coded with less accuracy, and some noise shaping is
already built into the quantization process. The quantized values are coded by Huffman
coding. To adapt the coding process to different local statistics of the music signals, the
optimum Huffman table is selected from a number of choices. The Huffman coding
works on pairs and, in the case of very small numbers to be coded, in quadruples. To get
even better adaption to signal statistics, different Huffman code tables can be selected for
different parts of the spectrum. Since Huffman coding is basically a variable code length
method and because noise shaping has to be done to keep the quantization noise below
the masking threshold, a global gain value (which determines the quantization step size)
and scalefactors (which determine the noise-shaping factors for each scalefactor band)
are applied before actual quantization. The process to find the optimum gain and scale-
factors for a given block, bit-rate and output from the perceptual model is usually done
by two nested iteration loops in an analysis-by-synthesis way:

Inner iteration loop (rate loop)

The Huffman code tables assign shorter code words to (more frequent) smaller
quantized values. If the number of bits resulting from the coding operation
exceeds the number of bits available to code a given block of data, this can be
corrected by adjusting the global gain to result in a larger quantization step size,
leading to smaller quantized values. This operation is repeated with different
quantization step sizes until the resulting bit demand for Huffman coding is
small enough. The loop is called rate loop because it modifies the overall coder
rate until it is small enough.

Outer iteration loop (noise control loop)

To shape the quantization noise according to the masking threshold, scalefactors
are applied to each scalefactor band. The systems starts with a default factor of
1.0 for each band. If the quantization noise in a given band is found to exceed
the masking threshold (allowed noise) as supplied by the perceptual model, the
scalefactor for this band is adjusted to reduce the quantization noise. Since
achieving a smaller quantization noise requires a larger number of quantization
steps and thus a higher bit-rate, the rate adjustment loop has to be repeated every
time new scalefactors are used. In other words, the rate loop is nested within the
noise control loop. The outer (noise control) loop is executed until the actual
noise (computed from the difference of the original spectral values minus the
quantized spectral values) is below the masking threshold for every scalefactor
band (i.e. critical band).

While the inner iteration loop always converges (if necessary, by setting the quantization
step size large enough to zero out all spectral values), this is not true for the combination
of both iteration loops. If the perceptual model requires quantization step sizes so small
that the rate loop always has to increase them to enable coding at the required bit-rate,

EBU TECHNICAL REVIEW – June 2000 8 / 15

Popp and Brandenburg
 MPEG Layer-3

both can go on forever. To avoid this situation, several conditions can be checked to stop
the iterations more early. However, for fast encoding and good coding results, such a
condition should be avoided. This is one reason why an MPEG Layer-3 encoder usually
needs tuning of the parameter sets of the perceptual model for each bit-rate.

6. Quality Considerations

As explained above, the pure compliance of an encoder with an MPEG audio standard
does not guarantee any quality of the compressed music. Audio quality differs between
different items, depending on basic parameters including, of course, the bit-rate of the
compressed audio and the sophistication of different encoders, even if they work with the
same set of basic parameters. To gain more insight into the level of quality possible with
MP3 and AAC, let us first have a look at typical artefacts associated with perceptual
audio coders.

Common types of artefacts

Unlike analogue hi-fi equipment or FM broadcasting, perceptual encoders exhibit sound

deficiencies when run at too low bit-rates or with the wrong parameters. These so-called
“artefacts” are in most cases different from usual noise or distortion signals. Perceptual
audio coding schemes such as MPEG Layer-3 introduce an error signal that can be
described as a time-varying error at certain frequencies, which is not constrained to the
harmonics of the music signal. The resulting music signal may sound:

distorted, but not like harmonic distortions;

noisy, but with the noise introduced only in a certain frequency range;

rough, with the roughness often being very objectionable as the error may change
its characteristics about every 24 ms.

Loss of bandwidth

If an encoder does not find a way to encode a block of music data with the required fidel-
ity within the limits of the available bit-rate, it “runs out of bits”. This may lead to the
deletion of some frequency lines, typically affecting the high-frequency content. Com-
pared to a constant bandwidth reduction, such an effect becomes more objectionable if
the effective bandwidth changes frame-by-frame (e.g. every 24 ms).

EBU TECHNICAL REVIEW – June 2000 9 / 15

Popp and Brandenburg
 MPEG Layer-3

Pre-echoes

Pre-echoes are very common artefacts, in the case of perceptual audio coding schemes
using high-frequency resolution. The name “pre-echo”, although somewhat misleading,
nicely describes the artefact, which is a noise signal occurring even before the music
event that causes such noise.

To understand the origin of Bitstream Decoding of Inverse Synthesis Audio

pre-echoes, let us consider in bitstream quantization filterbank out

the decoder of a perceptual

coding system (see Fig. 2). Figure 2
The reconstructed fre- Decoder of a perceptual coding system
quency lines are combined
by the synthesis filterbank, consisting of a modulation matrix and a synthesis window.
The quantization error introduced by the encoder can be seen as a signal added to the
original frequency lines, with a length in time that is equal to the length of the synthesis
window. Thus, reconstruction errors are spread over the full window length. If the
music signal contains a sudden increase in signal energy (e.g. a castanet attack), the
quantization error is increased as well. If such an attack occurs well within the analysis
window, its error signal will be spread within the full synthesis window, preceding the
actual cause for its existence in time. If such a pre-noise signal extends beyond the pre-
masking period of the human ear, it becomes audible and is called “pre-echo”.

There are a number of techniques to avoid audible pre-echoes, including variable bit-rate
coding or a local increase in the bit-rate to reduce the amplitude of the pre-echo. In gen-
eral, these artefacts belong to the “most difficult to avoid” category.

Roughness, “double-speak”

Especially at low bit-rates and low sampling frequencies, there is a mismatch between
time resolution of the coder and the time structure of some signals. This effect is most
noticeable on speech signals and when listening via headphones. As a single voice tends
to sound like it has been recorded twice and then overlaid, this effect is sometimes called
“double-speak”.

7. Not all encoders are created equal

The MPEG standards do not prescribe the implementation of the audio encoder. In an
extreme case, one could completely avoid implementing the perceptual model, decide
not to use the scalefactors (and therefore the outer iteration loop), and do a very simple
inner iteration loop. Such an encoder would be very fast (potentially much faster than
any current encoder products), would be compliant with the standard, would even pro-
duce a nice audio quality for some signals, but would sound very bad with a large selec-

EBU TECHNICAL REVIEW – June 2000 10 / 15

Popp and Brandenburg
 MPEG Layer-3

tion of music. While such a project would be easy to implement, it is much more
difficult to build an encoder that offers very high audio quality across all types of music
and even with the most exotic test items. In MPEG, testing had always aimed to verify
sufficient encoder performance in worst-case scenarios. Nonetheless, the current MP3
encoders show remarkable differences in their ability to produce, in a consistent way,
high sound quality at low bit-rates.

7.1. How to measure sound quality

Measuring the sound quality of perceptual audio codecs has developed into an art of its
own, over the last ten years. Basically, there are three methods: Listening tests, simple
objective measurement methods and perceptual measurement techniques.

Listening tests

To date, large-scale and well-controlled listening tests are still the only method available
for comparing the performance of different coding algorithms and different encoders.
With input from a number of broadcasters and the MPEG audio group, the ITU-R has
developed a very elaborate set of rules for listening tests. These tests aim to stress the
encoders under worst-case conditions, i.e. the testers try to find sound material which is
most difficult to encode, and then evaluate the performance of the encoders under test for
this material. This procedure is based on the observation that, in a lot of cases, coding
artefacts become audible and even objectionable only after an extensive training period.
Since the use of equipment based on audio compression technology (such as portable
audio players) itself constitutes extensive training, everybody can become an expert lis-
tener over a period of time. Therefore, right from the beginning, encoders should be
tuned better to satisfy the quality requirements of expert listeners.

Simple objective measurement techniques

Over and over again, people tried to get a measure of encoder quality by looking at
parameters such as the signal-to-noise-ratio or bandwidth of the decoded signal. As the
basic paradigm of perceptual audio coders relies on improving the subjective quality –
by shaping the quantization noise over frequency (and time), leading to an SNR which is
lower than is possible without noise shaping – these measurements defy the whole pur-
pose of perceptual coding. As explained below, to rely on the bandwidth of the encoded
signal does not show a very good understanding of the subject. Another approach is to
look at the codec output for certain test signal inputs, such as transients or multi-tone sig-
nals. While the results of such a test may tell the expert a lot about the codec under test,
it is very dangerous to rely solely on such results.

EBU TECHNICAL REVIEW – June 2000 11 / 15

Popp and Brandenburg
 MPEG Layer-3

Digital audio Distortion Coded

signal (PCM) control loop audio signal
Sub-band Line
31 575 Non-uniform
192 kbit/s
(768 kbit/s) quantization ..
Filterbank
MDCT rate control
Huffman .
32 sub-bands encoding Bitstream 32 kbit/s
loop
0 0 formatting
Coding of CRC-Check
Window
switching side-information

FFT Psycho-acoustic
1024 points model
External
control

Figure 3
An MPEG Layer-3 encoder

Perceptual measurement techniques

For 15 years, there has been a lot of research into applying psycho-acoustic modelling to
the prediction of sound quality and the audibility of certain artefacts. While the state of
the art is not yet sufficient to make large-scale and well-prepared listening tests obsolete,
perceptual measurement techniques have progressed to the point where they are a very
useful supplement to listening tests and, in some cases, are already replacing them. The
ITU-R Task Group 10/4 worked on the standardization of perceptual measurement tech-
niques and finally produced a Recommendation for a system called PEAQ (Perceptual
Evaluation of Audio Quality). This Recommendation defines a multi-mode system
based on the collaborative efforts of all the leading laboratories working on perceptual
measurement techniques.

7.2. The bandwidth myth

Reports about encoder testing often mention the bandwidth of the compressed audio sig-
nal. In a lot of cases this is due to misunderstandings about human hearing on the one
hand and encoding strategies on the other hand.

Hearing at high frequencies

It is certainly true that a large number of (especially young) subjects are perfectly able to
hear single sounds at frequencies up to and sometimes well above 20 kHz. However,
contrary to popular belief, the authors are not aware of any scientific experiment show-
ing beyond doubt that there is any listener (trained or not) who may detect the difference
between a (complex) music signal with content up to 20 kHz, and the same signal with a
bandlimit of around 16 kHz. There are some hints that a few listeners may have such
capabilities, but the full scientific proof has not yet been given. Therefore, it is a reason-

EBU TECHNICAL REVIEW – June 2000 12 / 15

Popp and Brandenburg
 MPEG Layer-3

able strategy to limit the frequency response of an MP3 encoder to 16 kHz. Due to the
brick-wall characteristic of the filters used in the codec, this can be done easily.

Please note, however, that this is not a general rule which can be applied to other types of
audio equipment (in particular, analogue). Typical audio equipment has to support much
higher frequencies in order to have the required perfectly flat frequency response up to
16 kHz: any deviation from the ideal straight line below the frequency cut-off point is
very audible.

Encoding strategies

While a loss of audio HF response will produce a coding artefact, it does not necessarily
mean that an encoder which produces a higher audio bandwidth will sound any better.
There is, in fact, a basic trade-off depending on how the available bits are used. If they
are used to improve the frequency response, they are no longer available to produce a
clean sound at lower frequencies. Leaving this trade-off to the encoder often leads to a
poor audio signal, with the high frequency cut-off point varying from frame to frame.
According to the current state of the art, it is best to introduce a fixed bandwidth limita-
tion.

Technically, MP3 can reproduce signal content up to the limit given by the actual sam-
pling frequency. If there is a comparison, at the same bit-rate, between an encoder with a
fixed limited frequency response, and another encoder with a much larger bandwidth,
experience tells us that in most cases the encoder with the lower bandwidth produces bet-
ter sounding compressed audio. However, there is a limit to this statement: at low bit-
rates (64 kbit/s for stereo and lower), the question of the best trade-off in terms of band-
width versus cleanness of sound is a hotly-contested question of taste. We have found
that even trained listeners sometimes completely disagree about the bandwidth a given
encoder should be run at.

7.3. Tuning for different bit-rates

As explained above, the double iteration loops do not converge if there is a mismatch
between the coding requirements as given by the perceptual model and the bit-rate avail-
able to code a block of music. To avoid this situation, it is wise to set the parameters in
the psycho-acoustic model in a way that the iteration loops will normally converge. This
may require settings which lead to audible differences, but the final coding result is still
better than the one from a perceptual model set to avoid any audible differences, com-
bined with coding loops which do not converge in a sensible way. To achieve this bal-
ance between requirements from the perceptual model and the available bit-rate, the
coding parameters have to be readjusted if the encoder is run at different bit-rates. Such
tuning procedures are responsible for a large part of the effort being put towards develop-
ing an MP3 encoder.

EBU TECHNICAL REVIEW – June 2000 13 / 15

Popp and Brandenburg
 MPEG Layer-3

Karlheinz Brandenburg received M.S. (Diplom) degrees in Electrical Engi-

neering (1980) and in Mathematics (1982) from Erlangen University, Ger-
many. In 1989, he gained a Ph.D. in Electrical Engineering, also from
Erlangen University, for work on digital audiocoding and perceptual meas-
urement techniques. From 1989 to 1990, he was with AT&T Bell Laborato-
ries in Murray Hill, NJ, USA, where he worked on the ASPEC perceptual
coding technique and on the definition of the ISO/IEC MPEG Layer-3 sys-
tem. In 1990, he returned to Erlangen University to continue his research
on audio coding and to teach a course on digital audio technology.

In 1993 Dr Brandenburg became head of the Audio/Multimedia department at the Fraunhofer

Institute for Integrated Circuits (FhG-IIS). Since 2000, he has been teaching at Ilmenau Techni-
cal University and has become the director of the new Fraunhofer Group for Electronic Media
Technologies in Ilmenau, Germany.

Karlheinz Brandenburg has received three awards from the AES for his work on perceptual
audio coding and psycho-acoustics. He is a member of the technical committee on Audio and
Electro-acoustics of the IEEE Signal Processing Society. He has worked within the MPEG-Audio
committee since its beginnings in 1988. In recent years, he worked on MPEG-2 Advanced
Audio Coding (standardized in 1997) and helped to organize the work for MPEG-4 Audio. He
is a member of SDMI (the Secure Digital Music Initiative) and currently chairs the AES Standards
Committee working group AESSC-06-04 on Internet Audio Delivery Systems. He has been
granted 25 patents and has several more pending.

Harald Popp was born in Erlangen, Germany, in 1956. In 1981, he received an M.S. (Diplom)
in Electrical Engineering from Erlangen University. From 1982 to 1984, he continued his work
at Erlangen University and carried out a technology transfer project for
advanced cable fault location. In 1984, he joined the Fraunhofer Institut
Integrierte Schaltungen (IIS) at its inauguration and worked as a board-level
hardware designer in various industrial projects. From 1987, he was
responsible for the real-time audio coding systems of the IIS (LC-ATC, OCF,
ASPEC, MPEG-Layer-3).

Today, Mr Popp is head of the Studio Department, focusing on effective

DSP-based real-time implementations of audio and video coding schemes.

8. Conclusions
By using an encoder with good performance, both MPEG Layer-3 and MPEG-2
Advanced Audio Coding (AAC) can significantly compress music signals, while still
maintaining CD or near-CD quality. Between the two systems, Layer-3 – with somewhat
lower complexity – is the system of choice for current near-CD quality applications.
AAC is its designated successor, providing near-CD quality at even larger compression
rates, and enabling higher quality encoding and playback up to high definition audio at
96 kHz sampling rate.

EBU TECHNICAL REVIEW – June 2000 14 / 15

Popp and Brandenburg
 MPEG Layer-3

Acknowledgements
The authors would like to thank all their colleagues at Fraunhofer IIS and MPEG Audio
for all the wonderful collaborative work over the last 11 years, since the start of MPEG.
Special thanks are due to Jürgen Herre, Martin Dietz, Oliver Kunz and Jürgen Koller for
their helpful suggestions. It should be noted that part of the audio coding work at Fraun-
hofer IIS-A has been supported by the Bavarian Ministry for Economy, Transportation
and Technology, and also the European Commission (within the RACE and ACTS pro-
grammes)

Bibliography
[1] K. Brandenburg, G. Stoll: ISO-MPEG-1 Audio: A Generic Standard for Coding
of High Quality Digital Audio
“Collected Papers on Digital Audio Bit-Rate Reduction (Gilchrist and Grewin),
AES, 1996.

[2] M. Dietz, H. Popp, K. Brandenburg and R. Friedrich: Audio Compression for Net-
work Transmission
Journal of the AES, Vol. 44, No. 1-2, 1996.

EBU TECHNICAL REVIEW – June 2000 15 / 15

Popp and Brandenburg