Ease of Integration & Perfor-

mance
 High clock speed (>250 MHz in
0.18um ASIC technologies)

DCT-FI  Low gate count

 Single clock cycle per sample
2D Forward and Inverse operation on both directions

Discrete Cosine  Low latency (89 cycles)

Transform Megafunction Design Quality

 Fully compliant with the JPEG
standard
The DCT-FI megafunction implements the combined 2D Forward/Inverse Cosine Trans-  Registered input and outputs
forms. Most of the image/video compression standards (JPEG, MPEGx, H.261, H.263,  Strictly positive edge triggered
DV etc) are based on the Discrete Cosine Transform (DCT). Able to operate over 8x8 fully synchronous design
and 16x16 blocks of samples/DCT coefficients, the DCT-FI covers the needs of hard-  Robust verification environment
ware image/video compression and decompression systems in the most efficient  No internal latches or tri-states,
manner. Possibly the fastest megafunction in the market, it is able to provide scan-ready design
processing rates up to 200 MSamples/sec in FPGA technologies and over 250 MSam-
Optional add-on Features
ples/sec in ASIC technologies. Furthermore, the megafunction allow the designers to
 Operation over 16x16 blocks of
perform area/quality trade-offs by adjusting the cosine coefficients and data-path preci-
samples/DCT coefficients
sion. Down-scaling in the frequency domain, as an optionally supported feature of the
 Down-scaling in the frequency
megafunction, allows reconstruction at various resolutions from the same input stream domain by run-time programm-
of coefficients. Finally the 2-4-8 DCT/IDCT transform, as this is specified in the DVC able integer factor (1 up to 8)
(DV) standard, can as well be optionally supported by the DCT-FI megafunction.  Programmable mode of opera-
Comprehensive documentation and a complete verification environment - including a tion (8-8 or 2-4-8)
bit-accurate model - help designers integrate and verify the megafunction. The DCT-FI
is designed for reuse in ASIC and FPGA implementations. The design is fully syn-
chronous with positive edge clocking and no internal tri-state buffers, and scan insertion
is straightforward.

Applications
The DCT-FI can be used in a variety of multimedia and image processing applications,
including:
 Office automation equipment (Multifunction printers, digital copiers etc)
 Digital cameras & camcorders
 Video production, video conference
 Display-projection systems
 Surveillance systems

Block Diagram

DCT-FI
dp_din Transpose dp_dout
SP SP
Memory
din
SP
dout
din_wen Stage 1 Stage 2
dp_waddr
dp_wen
dp_raddr

dp_ren

SP
clk
rst
clr
enable dout_wen
din_rdy dout_rdy
Control Unit sob
forward
eob

January 2008
Functional Description ple/DCT coefficient of an input block has been fed to the me-
gafunction.
The forward DCT (DCT) is a transform that converts a signal
into its constituent frequency components as represented by Implementation Results
a set of coefficients. The inverse DCT (IDCT) reconstructs
the original signal from its constituent DCT coefficients. A 2- DCT-FI reference designs have been evaluated in a variety
dimensional array of coefficients results by applying the DCT of technologies. The following sample Altera results are ob-
to 2-dimensional signals, such as images. The megafunction tained after area optimization during synthesis and place and
receives image samples or DCT coefficients and outputs route, while assuming that all megafunction I/Os are routed
DCT coefficients or image samples on a block by block basis, off-chip.
where each block has a size of either 8x8 or 16x16. The me- Altera Device Logic Frequency Special Features
gafunction implements the DCT or IDCT over the input blocks Apex 20KE
3,887 LEs 64 MHz 1 ESB
EP20K200E-1
by performing two 1-dimensional transforms, using row-
Apex-II
column decomposition, as defined by the following formulas: EP2A15-C7
3,889 LEs 102 MHz 1 ESB

DCT: Cyclone
2i  1 u 2 j  1 v
3,576 LEs 103 MHz 1 M4K
N 1 N 1 EP1C6-C6
2
Yuv  Cu C v   X ij cos cos  Stratix 1 M4K
N i 0 j 0 2N 2N EP1S10-C5
2,105 LEs 136 MHz
16 DSP blocks 9 bit
2 N 1  2 N 1
Cu   Cv  X ij cos
2i  1v  cos 2 j  1u Cyclone-II
2,231 LEs 151 MHz
1 M4K
 EP2C5-C6 16 DSP blocks 9 bit
N i 0  N j 0 2N  2N
Stratix-II 1 M4K
2,081 ALUTs 205 MHz
EP2S15-C3 16 DSP blocks 9 bit

IDCT: Support
N 1 N 1
2
X ij    Cu CvYuv cos
2i  1u cos 2 j  1v  The megafunction as delivered is warranted against defects
u 0 v 0 N 2N 2N for ninety days from purchase. Thirty days of phone and
N 1
2  N 1
2 2 j  1 v  cos 2i  1 u email technical support are included, starting with the first in-
 N
Cu 
N
CvYuv cos
2N

2N teraction. Additional maintenance and support options are
u 0  v 0  available.
where C  C  1 for
u v
u, v  0 and Cu  Cv  1 other-
Verification
2
wise, X ij are the image samples, Yuv are the DCT The megafunction has been verified through extensive simu-
lation and rigorous code coverage measurements. Being
coefficients.
embedded in numerous of products, the megafunction is sili-
The intermediate results being produced from the first 1- con proven in both FPGA and ASIC technologies.
dimensional transform are stored in the “Transpose Memory”.
The Transpose Memory is a dual ported RAM capable of Deliverables
storing an entire 8x8 or 16x16 block resulting from applying The megafunction is available in ASIC (synthesizable HDL)
the first stage of row decomposition. While the Transpose and FPGA (netlist) forms, and includes everything required
Memory is written in row-major order, the second stage of for successful implementation. The Altera version includes:
processing reads data from the Transpose Memory in a col-
 Post-synthesis EDIF netlist
umn-major order, effectively performing a transposition of the
intermediate results.  A bit-accurate model (BAM) of the megafunction including
support of custom test vector generation
The number of bits used for each intermediate result stored
in the Transpose Memory, as well as the number of bits used  Sophisticated self-checking Testbench (Verilog versions
to represent each of the cosine coefficients, is configurable at use Verilog 2001) supporting test vectors, expected re-
synthesis time. This allows the designers to perform their sults, and verification
own accuracy versus megafunction area tradeoffs. Further-  RTL and gate level (FPGAs) simulation scripts
more, the bit-width of both megafunction inputs and outputs
 Place and route scripts
is also configurable at synthesis time. It is noted that the de-
fault settings for these synthesis parameters, result to a  Comprehensive user documentation, including detailed
DCT/IDCT implementation that satisfy the accuracy criteria of specifications and a system integration guide
the JPEG standard.
The first DCT coefficient/image sample of a block will appear
at the output 89 clock cycles after the first image sam-
CAST, Inc. 11 Stonewall Court
Woodcliff Lake, NJ 07677 USA
tel 201-391-8300 fax 201-391-8694
Copyright © CAST, Inc. 2008, All Rights Reserved.
Contents subject to change without notice.
Trademarks are the property of their respective owners. This megafunction developed by the
multimedia experts at Alma Technologies S.A..