29C80F
2D Discrete Cosine Transform Circuit
Introduction
The 29C80F is a dedicated two-dimensional discrete Except for clock circuitry, the 29C80F is composed of 5
cosine transform circuit. The two-dimensional forward blocks :
transform (FDCT) or inverse transform (IDCT) is
performed on fixed 8 × 8 pixel or coefficient blocks (64
samples).
– 2 identical 1D DCT processors (line DCT and column
DCT).
– 1 memory for line/column transposition (between the 2
These blocks can be scanned from left to right, line by DCT processors).
line, or up to down, column by column, or in zig-zag order
for coefficient blocks only. If the input matrix is scanned
line by line and if the zig-zag order is not used, then the
output matrix will appear column by column
(line/column transposition between inputs and outputs).
– 1 memory for zig-zag scanning.
– 1 clipping operator (following the column DCT
processor).
All the internal sequencers are reset then started by a
pipelined signal, BLKIN (Block Input), and stopped after
a fixed number of cycles. The BLKOUT (Block Output)
signal indicates the beginning of a block on the output
data bus DO[0..11].
For FDCT, the input bus receives pixels coded with 8 or
9 bit in two’s complement format and the 29C80F output
coefficients are coded with 12 bit in two’s complement
format..
The latent period (the time between input data and its
For IDCT, the input bus receives coefficients coded with corresponding output result) is 128 CLK cycles
12 bit in two’s complement format and the 29C80F output (regardless of zig-zag scanning selection). The 29C80F
pixels are coded with 8 or 9 bit in two’s complement has been designed to process contiguous blocks.
format.
However, it is possible to introduce a gap period between
two blocks and/or to mix FDCT/IDCT by respecting
some recommendations.
A parallel architecture and a DCT based on modified
CHEN algorithm are used allowing high precision
compatible with the CCITT H261 requirements for
accuracy and fast operation up to 20 Mpixels/s.
The 29C80F is designed to cover a wide range of real time
DCT coding/decoding applications up to 20 MSamples/s.
Input Data Bus(1)
Output Data Bus(1)
(2)
Pixels 8-bit : DI [11..4]
FDCT
Coefficients : DO[11..0]
(3)
Pixels 9-bit : DI[11..3]
(4)
(5)
Pixels 8-bit : DI[11..4]
Pixels 9-bit : DI[11..3]
IDCT
Coefficients : DO[11..0]
Notes : 1. Data coded with 2’s complement
2. DI[3..0] must be tied to VIL
4. DO[3..0] forced to VOL
5. DO[2..0] forced to VOL
3. DI[2..0] must be tied to VIL
Features
D Forward and inverse 8 × 8 data transform
D DC to 20 MHz pixel rate (20 MHz clock)
D 9 bit two’s complement pixel format
D 8 bit two’s complement pixel format with optimised
accuracy
D 12 bit two’s complement coefficient format
D Fully compliant with CCITT H261 accuracy
D Selectable zig-zag scanning for coefficient blocks
D Full parallel architecture
D Radiation tolerant for space application
D Fully synchronous interface
D 128 block cycles latency
D Power down mode
D Tristate control output
D TTL compatible inputs and outputs
D Single 5 V ± 10 % power supply
D 44 pin MQFPJ
D Advanced 0.8 µm CMOS technology
MATRA MHS
65
Rev. D (25 Mar.97)