MPEG Digital Audio Coding

1
MPEG Digital Audio Coding
Setting the Standard for High-Quality Audio
Compression

• IEEE Signal Processing Magazine
• Sept, 1997
• Presented by Dan Li
• Oct, 2000

2
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

3
Introduction

• Audio coding implies
• ?fs, ? amp resolution, ? dynamic range, ?
  variations in PSD,
• stereophonic, multichannel, ?Q expectations...
• MPEG for audio coding
• MPEG-1 (IS 11172)
• stereophonic audio coding (SAC) at hi fs
  (32/44.1/48KHz)
• supports 256 Kbps 2-channel audio
• MPEG-2 (IS 13818)
• SAC at low fs (16/22.05/24KHz), plus multi- (2-5
  full BW) channel coding
• supports 64 Kbps audio

4
Bit Rate Reduction 1

• PCM bit rate
• Significant overhead ? the total bit rate
• e.g. For CD, an overhead of 2.91Mbps leads to a
  total bit rate of 4.32 Mbps!

5
Bit Rate Reduction 2

• Why reduction?
• to reduce transmission cost
• to provide cost-efficient storage
• to support transmission over low-capacity
  channels or variable-rate coding
• How to design?
• Basic requirements for hi-Q reconstruction
• robustness to variations in spectra and levels
• robustness to channel bit error packet losses
• ? complexity, ? power consumption
• Some techniques
• linear prediction, subband coding, transform
  coding, entropy coding...

6
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

7
Perceptual Coding 1

• Simultaneous masking
• A low-level signal can be masked by a
  simultaneously occurring stronger signal if they
  are close in frequency
• Masking threshold varies w/ time, depends on
  sound pressure level, the frequency of the
  masker, characters of the masker and maskee.
• SNR(m), SMR and NMR(m) (m-bit quantization)
• NMR(m)SMR-SNR(m)
• NMR(m)lt0?SMRltSNR(m) ?quantization noise level
  below masking threshold ?noise is masked (i.e.
  not audible!)
• Temporal masking
• A stronger signal can mask the weaker one even if
  the maskee precedes the masker.
• SMR- signal to mask ratio

8
Perceptual Coding 2
Mux
9
Frequency-Domain Coding

• Redundancy and irrelevancy are exploited to
  reduce bit rate
• Filterbanks-based encoder
• Transform coding (TC)
• DFT, DCT, MDCT (sliding window DCT)
• Quantize the transform coefficients for
  transmission
• Decoder reconstructs the signal in time domain
• Subband coding (SBC)
• subband
• Perfect reconstruction
• Hybrid filterbanks-based coding (HC)

10
Window Switching

• Motivation
• To reduce pre-echoes which exist in
  freq-domain coding. Pre-echoes typically occur
  when a long block is employed which contains
  nonstationary periods of the signal.
• Solution
• - switch between a bigger block size and a
  smaller one (e.g. N1024 and N64)
• - smaller blocks are only used during
  nonstationary periods to control pre-echo
  artifacts

11
Dynamic Bit Allocation

• Adaptive transform coding (ATC)
• Compute a DCT-based short-term spectral envelope
  by using DCT transform coefficients
• Estimate the short-term spectrum and calculate
  the optimum number of bits for each transform
  coefficient.
• Code and transmit the parameters of this spectrum
• Comments
• share many features with many recent frequency
  domain coders
• shortcomings block boundary effects,
  pre-echoes...

12
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

13
The Basics

• 3-layer structure (I, II, III)
• Hierarchy of layers higher layer incorporates
  function blocks of lower layer
• Layer I low complexity, good for consumer
  recording
• Layer II high efficiency with medium
  complexity, good for professional recording
  and for broadcast
• Layer III high complexity and high
  efficiency, suitable for very low bit-rates
  application as transmission through 64 Kbps
  channels
• Layer I II use SBC Layer III uses HC (SBCTC)
• of bits for each subband is determined on a
  block-by-block basis
• Operating modes
• Mono, stereo, dual w/ two separate channels,
  joint stereo,
• Specific to Layer III middle (LR) and side
  (L-R) channels

14
Layers I II 1
15
Layers I II 2

• Similar structure for Layer I II. Layer II has
  reduced redundancy, finer quantization, thus
  better performance
• Filterbanks
• 32 subband filters with the same bandwidth
• Impulse response of the BPF of subband k

16
Layers I II 3

• What of quantization bits to use?
• dynamic bit allocation
• A uniform midtread quantizer to meet the
  requirements of both the bit-rate and the masking
• Iteratively determine the bits that minimizes
  the NMR (so that the quantization noise is
  sufficiently below masking threshold)
• If SMRlt0 for several subbands (usually high
  bands), the number of assigned bits 0 since
  within these subbands signals are masked thus not
  to be transmitted
• see Coding Examples p. 67

17
Layer III 1
Rate and distortion control lope
Scaler Quantizer
Mux
Hoffman Coding
Analysis Filterbank
MDCT w/ Dynamic Windowing
Digital Channel
Coding of Side Info
FFT
Masking Threshold
PCM input
De- mux
Dequan- tizer Descaler
Synthesis Filterbank
Hoffman Decoding
Inverse MDCT w/ Dynamic Windowing
PCM output
Decoding of Side Info
18
Layer III 2

• New features
• Switched hybrid filterbank
• SBC?32 SB each SB signal ? block-wise MDCT
  transform ? High frequency resolution
• Adpative/dynamic block/window size switching
  6-pt to control pre-echoes, 18-pt to get hi freq
  resolution
• Hoffman coding
• efficiently represent quantizer indices.
• Analysis by synthesis method
• Best audio quality with Layer III in the case of
  wideband audio coding!

19
Frame and Multiplex Structure

• Frame structure packetization
• In one packet 4-byte header 184-byte payload
• Each frame can be decoded independently from
  previous frames
• Length of frames not fixed due to many factors
• Multiplex structure
• A packet structure for multiplexing audio, video
  and ancillary data in one stream

20
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

21
Multichannel Coding

• p/q multichannel loudspeaker configuration
• p front channels and q back channels
• 3/2 system SL, L, C, R, SR, ? surround-sound
  field
• 5.1 system 3/2 system low-freq enhancement
  channel (subwoofer)
• MPEG-2 multichannel audio-coding
• Standard 1 MPEG-1 compatible
• Standard 2 Not compatible with MPEG-1 Advanced
  Audio Coding (AAC)

22
MPEG-2 Audio Coding 1

• Compatibility of MPEG-2 with MPEG-1
• Forward and Backward compatibility
• Backward compatibility matrix

A
L0
LS L C R RS
A
MPEG-1 Encoder
Mux
1-H
1-Anc data
1-PL
R0
T3
MPEG-1 Frame
MPEG-2 Ext. Encoder
T4
T5
2-Anc
1-H
2-H
2-PL
1-PL
MPEG-2 Encoder
MPEG-2 Frame
23
MPEG-2 Audio Coding 2

• Advanced Audio Coding (AAC)
• Supports up to 46 channels for various
  multichannel loudspeakers configurations
  (including mono, stereo and 5.1 system)
• High quality at low bit rates
• Key features
• High resolution filterbanks, adaptive
  prediction techniques, quantization and noiseless
  coding
• 3 profiles of AAC
• Main profile Highest quality
• Low complexity profile No prediction and noise
  shaping
• Sampling-rate-scaleable profile lowest
  complexity

24
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

25
MPEG-4

• Proposals for multimedia application
• Standardized configuration supports various
  applications
• Higher compression rate
• Verification model core coders
• New functionalities to meet different needs

26
What will be covered?

• Key technologies in audio coding
• MPEG-1 standard
• MPEG-2 standard
• MPEG-4 standard
• Typical applications for MPEG audio compression
• Only standard for audio is to be discussed.

27
Applications

• Digital broadcast and transmission systems
• cable-TV networks, satellite-based digital
  broadcasting, HDTV system, cellular mobile radio
  networks, LAN, etc.
• Digital storage
• DVD (support 5.1 system), Digital Compact
  Cassette,
• MP3 (121 compression ratio!, near CD quality)

28
Websites of Interest

• MPEG in general
• http//www.cselt.it/mpeg/
• http//garuda.imag.fr/MPEG4/
• http//www.mpeg.org
• MPEG audio
• http//www.tnt.uni-hannover.de/project/mpeg/audio/
  
• http//viswiz.gmd.de/DVP/Public/deliv/deliv.211/mp
  eg/audio001.htm
• MPEG coding
• http//www.apocalypse.org/pub/u/howitt/sbc.tutoria
  l.html
• http//www.cs.tut.fi/ypsilon/80545/Codinglinks.ht
  ml
• MPEG tools
• http//bmrc.berkeley.edu/frame/research/mpeg/