A low-cost architecture design with efficient data arrangement and memory configuration for MPEG-2 audio decoder

A Low-Cost, Architecture Design with Efficient Data Arrangement

and Memory Configuration for

MPEG-2 Audio Decoder

Tsung-Han Tsai, Liang- Gee Chen,Sheng-Chieh Huang,Hao-Chieh Chang,

Departmenit

Electrical Engineering, National Taiwan University, Taiwan, R.

0.

C.

Email: hanovideo. ee.ntu. edu.tw

ABSTRACT

The paper deslcribes a low-cost MPEG-2 audio decoder with a modified fast algorithm for decoding. In the modified decoding scheme, the computation amount of the bottleneck module can be reduced into one-forths of the original one. Also, the ma-

jor memory storage only requires half size of the s-

tandard synthesis subband filterbank. The decoder is developed for the approaches of simplicity and low-cost architectui:e design, with the techniques of intelligent d a t a arrangement and memory configu- ration.

extension of the existing standard t o lower sampling frequencies and lower bit rates. Besides, backward compatibility is the key aspects t o ensure the exist- ing two channel decoders will still be able t o decode compatible stereo information from five multichan- nel signals. This implies the provision of compati- bility matrices, using adequate inverse matrix coeffi- cients. Figure 1 describes the operation of MPEG-2 audio codec and the coding relationship among the five audio channels (L, R, C, LS, RS), the two basic channels (TO, T1) and the three extended channels (T2, T3, T4).

1. Introduction 3. The Modified Fast Algorithm for Decoding

The I S 0 MPEG,-2 audio standard has develope- d a world-wide standard audio coding algorith- m , which can significant reduce the requirements of transmission ba:ndwidth and data storage with low distortion. W.ith the recent advances in VL-

SI and ATM networking technology, the low-cost MPEG-2 audio decoder in real-time system be- comes more and more essential for multimedia ap- plications. This paper presents a low-cost MPEG-2 audio decoder, which is capable of decoding MPEG- 2 standard multichannel audio bitstream for Layer I

and I1 with a modi:fied decoding scheme. The mod- ified decoding scheme takes the advantages of low- cost computation and low memory requirements.

This

t a arrangement and memory configuration for low complexity and high efficiency applications.

2. MPEG-2 Audio Coding

The MPEG-2 audio coding standard is an exten- sion of MPEG-1. Emphasis of the new activity is

The MPEG-2 decoding flow chart is shown in Figure 2. Also, within the synthesis subband fil-

ter, the inverse Modified Discrete Cosine Transform (IMDCT), inverse Pseudo Quadrature Mirror Filter (IPQMF) will be further realized as shown in Figure

3. According to the computation power analysis for

MPEG-2 audio decoding in Table 1, the most com- putation load highly depends on the realization of IMDCT module.

First, the original inverse MDCT of a sequence

S ( k ) is defined as following [l]: (16

+

+

c

(1)

i

...

Taking the advantage of the symmetric proper- ties, the inverse MDCT standard function of Equa- tion (1) can be leaded into a new formula with a

on multichannel and multilingual audio and on w - @ u c t i o n of computation amount as following:

divided into five individual memory banks. Each bank matches an audio channel data. The bank can

*

contains 16 audio samples.

(2) According t o the proposed algorithm and the memory configuration, only 512 clocks is needed for the computation of the IMDCT transform. Also, the IPQMF takes 512 clocks for a cycle. These make

modules with high efficiency and as shown in Fig-

with

can be reduced Only to 512 word- two blocks. In the meantime, the IPQMF module blocks. The

for IPQMF can be realized by

Figure 7. These imply two pointers address the s-

tarting and ending blocks to realize a circular buffer access for the IPQMF shifting.

15 (32 + i ) ( 2 k

_{' I T ] *}

+

v;

...

k

The proposed modified

requires about the pipeline processing with IMDCT and IPQMF 1/4 amount of multiplier-accumulate computation

quired size for the RAM buffer in which the QMF CT are written into synthesis window

data

s per channel, instead of the original size of 1024

words per channel. Table 2 shows the comparisons memory data

rithms for the computation complexity and memo- ry requirement

Of the Is' suggestion method' Moreover' the re- ure 6. In each cycle, the data processed from IMD-

reads the data from buffer with

between the Original and proposed decoding algo- the dedicated address generator and illustrated in

4. A r c h i t e c t u r e D e s i g n

Figure 4 describes the overall architecture dia- gram of our proposed design. It includes 5 prima- ry modules called: preprocessor, inverse quantiza- tion, multichannel processor, IMDCT module and

er information from audio bit-stream (DATAIN), ciency design is realized. then extracts multichannel processing modes and

some control informations t o store in control reg- ister. Inverse quantization module processes the inverse quantization function to get the quantized samples. These samples are fed into multichannel

processor t o reconstruct five channel samples. The

synthesis subband filterbank, viewed as the most computation load, can be performed in two pipeline stage of IMDCT and IPQMF for high throughput consideration. These two modules mainly contain

a multiplier-accumulator (MAC) respectively. Fi-

6. Conclusions

We have described a low-cost MPEG-2 audio de- coder. Based on our modified fast algorithm for de- coding and efficient data arrangement and memory

IPQMF module* The preprocessor interprets head- configuration, the goal for low-cost and high effi-

R e f e r e n c e s

MPEG, "ISO CD 11172-3: coding of moving pictures and associated audio for digital storage media at up to about 1.5 Mb/s", Nov 1991.

[2] MPEG, "IS0 CD 13818-3: coding of moving pictures and associated audio for digital storage media at up to about 1.5 Mb/s", Nov 1994. [ 1 ~

nally the output interface performs the parallel-to-

serial operation t o get the five PCM audio channel L'

f

c

Lb

samples with some word select signals (WORDSEL) L5

and the associated PCM clock signal (PCMCLK). R1 n*

M P E G Z C i C o D f R MPIG-2 ENCODER

5 . Efficient Data A r r a n g e m e n t and Memory C o n f i g u r a t i o n

The synthesis window buffer takes the important role in synthesis subband process. Thus we take the efficient memory configuration for synthesis window buffer as shown in Figure 5. This buffer can k - 6 6

Figure 1: Backward Compatible for MPEG-2 Audio Codec

Table 1: Required Computation Power (MOPS) for MPEG-2 Decoding Function Degrouping Requantization Rescalization

D

I s i f i c a t i o n Required Processing Power

(

)

0.88 1.44 0.96 3.28 0.576 1.44 2.016 61.44 19.22 81.36 86.656 I& SYNTHESIS SUBBAND FILTER

kMC

MULTI-CHANNEL SYNTHESIS SUBBAND

c-

-

DECODER FILTER

1

5

BASIC CHANNEL EXTENDED CHANNEL

DECODING OF BIT ALLOCATION

c

c

-

k

z

j

Figure 2: MPEG-2 Decoding Flow Chart

Input 32 New Subband Samples

I I I for]

-

1

Figure 3: Synthesis Subband Flow Chart

I

I

Bank0 Bank1 Bank2 Bank3 Bcnk4

Buffer

'.

=

..

Item Original Proposed Ratio MAC transform 1 4 Per 2048 512 Buffer Size IPQMF Per channel

Figure 5: Memory configuration for synthesis win-

dow buffer 1 2

-

u

Figure 7: IPQMF Memory Data Access per Audio Channel

Table 2: Comparisons between the Original and Proposed Decoding Algorithms for the Computa- tion Complexity and Memory Requirement

Figure 6: Pipeline processing for IMDCT and IPQMF