e005020.pdf

GENERAL INTEREST

Basics and Chipsets

Staging a DiY Stand-Alone MP3 Player

By Prof. F. P. Volpe and P. Elsesser

Traditionally, an MP3

player requires a semi-

conductor memory to

store MP3 data. MP3 players

capable of reading data

directly from a CD and with-

out ‘help’ from a PC are few and far between. That may change, however,

with the home-brew MP3 player described in this short series of articles.

Elektor Electronics

5/2000

MP3

GENERAL INTEREST

Strong Tone Signal

will cover ‘hands-on’ matters like the con-

struction of the MP3 decoder board, the

ATAPI support by a microcontroller and the

link to a CD-ROM drive.

Region where

Weaker Signals

are Masked

Why compression?

Frequency

000044 - 11

A digital audio signal typically consists of

samples with a size of 16 bits. According to

the sampling theorem, at a given bandwidth,

a sample has to be taken at a rate that equals

at least twice the frequency of the pro-

gramme material. If CD quality is required in

respect of bandwidth (44.1 kHz), then a data

rate of about 1.35 Mbit/s is needed to convey

an audio signal. In other words, one minute

of music requires about 10 Mbytes of data to

be conveyed or stored on a data carrier.

Unfortunately, that is not practicable even

with the huge capacity of today’s hard disks.

Obviously, the resultant transmission

(=download) times using media like the

Internet, Internet-Radio or Music On Demand

systems would be very long and therefore

prohibitive. In this case, the only solution is

to find a way to reduce the immense size of

the relevant data, the ‘real-world’ aim being

to convey a stereo music signal across an

ISDN link with a capacity of 64-kbit/s per

channel. In view of the required compression

rate of 1:12, that would seem to be possible

only if losses are accepted. Over a relatively

short period, a system called MPEG-1 Layer 3

has established itself as the de facto standard

for audio transmission via the Internet. MP3

employs compression algorithms that take

the real-life response of the human ear into

account. The resultant quality of reproduced

sound is so good that even trained experts

are unable to hear the difference between a

copy and its original. MP3 is also clearly

superior as compared to other MPEG Layers

such as the Digital Compact Cassette (MP1),

Digital Audio Broadcasting, the Video-CD

(MP2) and simpler systems like CELP,

Figure 1. A frequency range is masked (obliterated) by a loud signal.

Music programme material dis-

guised as MP3 files have become

big business on the Internet. How-

ever, before you can actually play a

music-CD with your own selection

of MP3 titles you typically have to

provide a link between your stereo

and your PC, by way of the sound-

card. Arguably, a stand-alone MP3

player with an internal standard

CD-ROM drive and a hardware MP3

decoder, ready for connection to

your stereo, represents a much bet-

ter solution because it avoids the

tedious process of having to ‘boot’

your PC every time you want to

play some music. Also, the often

objectionable noise level added by

the PC, and its very presence in the

living room, should be considered.

Such an MP3 player will be

described in the July/August 2000

issue of Elektor Electronics , and the

present article is intended as an

introduction to it. This month we

will describe the outlines and basics

of MP3 compression, as well as

some frequently seen chip sets for

MP3 decoding. We will also gaze

into the crystal ball for a bit in rela-

tion to future audio compression

technologies.

The follow-up article, to be pub-

lished in our annual Summer Cir-

cuits issue (i.e., July/August 2000),

120

120 phon (loudness level)

110

100

-Law

or ADPCM.

The essential point about MP3 is that the

system is based on a psycho-acoustic model

whose elementary structure will be dis-

cussed below. A far more extensive tutorial

about MPEG and audio compression tech-

niques is available in the form of a down-

loadable file (mpeg-tutorial.pdf) from the

Download Area on the Elektor Electronics

website at

http://www.elektor-electronics.co.uk

(document reproduced courtesy IEEE).

Fortunately, this document does not rely on

higher mathematics to explain the underlying

principles!

Minimum audible Field (MAF)

100

1000

10,000

Frequency (Hz)

000044 - 12

Figure 2. The human ear is particularly sensitive to speech signals.

5/2000

Elektor Electronics

GENERAL INTEREST

Psycho-acoustics

100

The science of Psycho-Acoustics studies the

behaviour of the human hearing system in

relation to processing of acoustic information

in the brain. Psycho-acoustic principles have

been extensively used in the development of

MP3 and indeed many other compression

techniques.

The audible spectrum may be thought of

as consisting of 26 frequency bands.

The frequency range below 500 Hz is sub-

divided into five bands of 100 Hz each.

Above this range, the bandwidth is about

1/5th of the centre frequency. In the human

hearing system, soft sounds become less dis-

tinct and even inaudible when loud, discrete

sound levels occur within these ‘critical’

bands. As an aside, you should note that the

frequency resolving capacity of the human

hearing system is much more accurate than

the critical bands.

The above phenomenon is a condition to

allow masking of a soft sound by a loud

sound which occurs at a nearby frequency

and/or instant. The spectral masking range of

an individual loud sound is shown in Fig-

ure 1 . The masking (or, if you like, obliterat-

ing) sound may occur simultaneously with

the soft sound or even after it.

The actual effect of the masking depends

not only on the spectral and time-related

arrangement of the masking and masked

sound, but also on its tonality . Noise is far

easier to mask than a discrete frequency.

Conversely, a discrete frequency is a much

better mask than noise.

100

10k

000044 - 13

20k

f (Hz)

Figure 3. Three masks are applied to raise the co-hearing threshold.

the signal quality. In cases where no

compression is required, this is even

switched off completely.

process, a filter bank divides the

audible spectrum into 32 sub-bands

of equal width (750 Hz at a sam-

pling rate of 48 kHz). Next, 32-times

sub-sampling is applied to each

sub-band of the input signal, result-

ing in 32 PCM input data in a sub-

band sample. The filter bands equal

the previously mentioned critical

frequency bands of the human hear-

ing system, with three differences.

Firstly, the constant width of the

sub-bands is unlike that of the

human example, secondly, the filter

bank and its complementary func-

tion in the decoder are not loss-free,

and thirdly, an individual frequency

may affect the output signals

because of a certain overlap of the

sub-bands. Fortunately, the errors

Functional blocks

The elementary architecture of an

MPEG audio encoder is illustrated

in Figure 4 . Decompression is the

reverse operation or encoding. To be

able to apply the psycho-acoustic

model to the digital audio input sig-

nal (a PCM datastream arriving at

768 kBit/s), the signal first has to be

transposed to the frequency

domain. A fast Fourier transforma-

tion with 1024 coefficients is

employed as part of the computa-

tion of the psycho-acoustic model.

Fully synchronous with this

The starting point for a psycho-acoustic

model is the frequency-dependent charac-

teristic of the human hearing system. The

curves in Figure 2 illustrate that sounds

within the frequency range of human

speech are perceived with greater resolu-

tion than very high or very low sounds. The

lower, dashed, curve represent the absolute

hearing threshold , below which no sounds

are perceived by most of us. A masking

operation is applied to raise the absolute

threshold to the so-called co-hearing

threshold , as it appears three times as

dashed lines in Figure 3 for three different

masking signals.

A psycho-acoustic model analyses the

audio signal and employs complex algorithms

to compute the usability of masking sounds

in the relevant frequency range. The closer

the model gets to reality, the higher the com-

pression rate that can be achieved at a given

quality level of the output signal. However,

the rules of the model are ‘relaxed’ to an

extent required by the transmission rate and

MDCT

Huffmann

Encoding

Audio Signal

(PCM 768 kbit/s)

MDCT

Non-linear

Quantisation

and

Bitrate Control

Bitstream-

formatting

and

Error

Correction

Encoded

Bitstream

Filter Bank

(32 sub-bands)

559

560

Aux. Info

Encoding

MDCT

576

FFT

Psycho-

acoustic

Model

000044 - 14

Figure 4. Block diagram of an encoder to MPEG 2.5 Layer III.

Elektor Electronics

5/2000

GENERAL INTEREST

Audio

introduced by the filter bank are small

enough to be inaudible (<0.07 dB).

This is mainly caused by the final MDCT

(Modified Discrete Cosine Transformation)

which divides each of these 32 sub-bands

into 18 sub-sub-bands. In the end we have

32×18 = 576 sub-bands of 42 Hz each (at a

sampling rate of 48 kHz).

In the next operation, noise allocation, the

ratio of the quantisation noise and the co-

hearing threshold is recovered from the dif-

ference of the signal-to-noise ratio (filter

bank) and the signal-to-mask ratio of the psy-

cho-acoustic model. The number of bits avail-

able in the output datastream is then deter-

mined (depending on the selected overall

data rate minus data on scaling factor, head-

ers and other auxiliary data). The encoder

varies the quantisation in a certain sequence,

weighs the spectral values, counts the number

of bits required for the output signal (the

number being further reduced by Huffmann

encoding). The outcome is used to compute

the permissible quantisation noise level. If,

after quantisation, bands are found with

unacceptably high distortion, the encoder

ampliﬁes these bands and so raises the size

of their quantisation stages. The operation is

repeated until distortion levels are below the

acceptable level. MP3 works with a variable

transmission rate and so creates a kind of

‘buffer’ for the duration of signals that require

a few bits only. The buffer is employed for

‘difficult’ encoding operations, at which the

data rate is fully exhausted.

The block diagram of the encoder is com-

pleted with a formatting unit that serves to

add the auxiliary data, and pack the lot into

frames ready for sending to the decoder.

CD-ROM

Audio Signal

Filter

ISO9660

left

ATAPI

I 2 C-Bus

AUX-In

LCD

PIC

16F874

MAS3507

DAC3550

Clock

Audio

Keyboard

Filter

right

000044 - 15

Figure 5. MP3 CD-player based on the MAS3507D decoder and DAC3550

digital-to-analogue converter, both from Micronas Intermetall. The setup is

controlled by a PIC17C756.

Audio

CD-ROM

Filter

ISO9660

left

LCD

ATAPI

Mic.

KS17C4000

DAC3550

Keyboard

Audio

Smart/media

Card

Filter

right

000044 - 16

Figure 6. MP3 CD player using the Samsung TL7231MD decoder. Besides

decoding MP3 data this chip also handles ADPCM-based audio signal compression.

From chipset to MP3 player

Audio

Different chipsets are currently available for

the purpose of decoding MP3 signals. In this

article, three options are discussed for realiz-

ing a stand-alone CD player for MP3 and

audio data, based on three different chip sets

from different manufacturers. The data source

is invariably a standard ATAPI-compatible

CD-ROM drive. A microcontroller is used to

look after the ATAPI protocol and supply the

data for the MP3 decoder. The same controller

also drives an LCD, and scans keys for navi-

gation within the CD directory structure. A

digital-to-analogue converter converts

decompressed data supplied by the MP3

decoder into plain audio signals.

At the time of writing, MP3 chipsets are

offered by three companies. Figure 5 shows

an MP3 player based on a chipset supplied

by Micronas. This player is controlled by a

PIC17C756 which reads the MP3 data from a

CD-ROM

Filter

ISO9660

left

ATAPI

I 2 C-Bus

LCD

PIC

17C756

STA013

CS4331

Audio

Keyboard

Filter

right

000044 - 17

Figure 7. MP3 CD player based on the STA013 decoder from STMicroelectronics

and the CS4331 DAC from Crystal Semiconductor.

5/2000

Elektor Electronics

GENERAL INTEREST

CD-ROM drive and transmits them serially to

a decoder chip type MAS3507D . You can

browse the CD contents and navigate by

means of a keyboard and a readout (LCD).

The display also indicates the ID-3 tag (data

information). Decoded data are applied to a

digital-to-analogue converter (DAC) type

DAC3550 , which supplies the analogue audio

signals. After cleaning in a third-order low-

pass filter, the audio signals are allowed to

leave the decoder.

Besides a digital input, the converter also

sports an analogue input. An I 2 C command is

used to select between these inputs. If the

microcontroller detects a music CD in the CD-

ROM drive, the DAC is ‘ordered’ to switch to

its analogue input.

Another chipset is shown in Figure 6 .

Here, the Samsung TL7231 is used as the

MP3 decoder. Because this chip integrates a

digital-to-analogue converter, the number of

components is drastically reduced. In addi-

tion to the MP3 decoder function, the

TL7231MD also has an ADPCM Codec. Using

this chip it is possible to store audio signals

in compressed form, and play them back

again.

If PCB space is at a premium, the solution

is to employ the STA013 from STMicroelec-

tronics ( Figure 7 ). This chip comes in a SO-28

case. The CS4331 digital-to-analogue con-

verter from Crystal Semiconductor comes in

a SO-8 enclosure. It should be noted that this

MP3 decoder wants a conﬁguration ﬁle after

every reset. This file is supplied by the ST7

central controller in the system.

Compression and the future

A number of other audio compression methods exist besides MPEG Layer III.

These are partly well established, partly ‘being designed’. At least for the near

future MP3 seems to have won the battle for dominance on the Internet. How-

ever, Sony’s AC-3 system has pushed MPEG aside in the DVD market.

MPEG-2

allows ‘low’ sampling frequencies like 16 kHz, 22.05 kHz and 24 kHz besides the

more usual 32 kHz, 44.1 kHz and 48 kHz standards. It also supports 5.1-Sur-

round as well as multi-language broadcasts. Otherwise downwards compatible

with MPEG-1.

MPEG-2.5

again halves the sampling rates, allowing bit rates up to 8 kbit/s to be achieved.

Otherwise this standard is downwards compatible with MPEG-1. The Micronas

chipset used in the Stand-Alone MP3 Player is designed for MPEG-2.5.

MPEG-2-AAC (Advanced Audio Coding)

is not downwards compatible. It allows sampling frequencies of 32 kHz, 44.1 kHz

and 48 kHz as well as single-channel, dual-channel 5.1-Surround and multi-lan-

guage broadcasts. Provision is made for error detection/correction. AAC is

increasingly used by broadcast stations and, interestingly, for a new satellite radio

station covering the southern hemisphere. In Japan, HDTV and all broadcast sys-

tems are to employ AAC.

MPEG-4

allows acoustic events in a room to be described. This is of particular interest to

multimedia applications because a ‘speed control’ enables different media to be

synchronized.

ATRAC

is used in MiniDisc systems and supplies a data rate of 140 kbit/s/channel. The

main difference with MPEG-1 is that the ﬁlter bank uses a different method to

resolve the spectrum. In higher frequency bands, the frequency resolution is

reduced and the time-resolution is increased. (the ‘higher’ ﬁlter sections operate

faster).

(000044-1)

Note:

The construction of a Stand-Alone MP3 Player will

be discussed in the July/August 2000 issue of Elek-

tor Electronics.

Dolby AC-3

is MPEG’s biggest competitor and extensively applied with DVD and HDTV. AC-3

supports various formats including 5.1-Surround. The bitrate may lie between

32 kbit/s and 640 kbit/s. The quality is about the same as MPEG-2, however it is

below that of AAC with 5.1-Surround.

MS Audio

Impossible compression system devised by Microsoft.

Literature

- Micronas Intermetall: MAS 3507D MPEG 1/2

Layer 2/3 Audio Decoder. Preliminary Data

Sheet, Edition Oct. 21, 1998.

- STMicroelectronics: AN1090 STA013

MPEG 2.5 Layer III Source Decoder.

Application Note, March 1999.

- Micronas Intermetall: DAC 3550 Stereo Audio

DAC. Preliminary Data Sheet, Edition April 23,

1999.

- Draft ISO/IEC 9660: Information tech-

nology – Volume and ﬁle structure of

CD-ROM for information interchange.

- Samsung Semiconductor: TL7231MD Full Layer-

III ISO/IEC 11172-3 Audio Decoder. Data Sheet.

- A Tutorial on MPEG/Audio Compres-

sion, published in IEEE Multimedia

Journal, Summer 1995

- STMicroelectronics: STA013/STA013T MPEG

2.5 Layer III Audio Decoder. Data Sheet, Sep-

tember 1999.

Elektor Electronics

5/2000

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: