
521 - 33 - 04
4
DEVICE DESCRIPTION
The GF9103 is composed of five main sections:
1.4:2:2 De-Multiplexer
2.FIR Filtering and Setup Insertion
3.Color Space Conversion
4.Digital SIN X/X Correction
5.Output Processing
4:2:2 DE-MULTIPLEXER
The de-multiplexer accepts data multiplexed in a SMPTE
125M compliant format from the
SI
9..0
input data port.
SI
9
is
the Most Significant Bit and
SI
0
is the Least Significant Bit.
The input data stream is assumed to be a multiplexed
stream of C
B
Y C
R
[Y] C
B
..., where the three words C
B
Y C
R
refer to cosited samples and where [Y] refers to an isolated
Luminance sample. When operating the GF9103 with 8 bit
input data,
SI
9..2
should be used to present data to the
device and
SI
1..0
should be set low.
At least once during a power cycle, the GF9103 must be
synchronized to the incoming data stream. The GF9103 is
synchronized by holding SYNC high on clock period N and
low on clock period N+1 when the first C
B
sample is
presented to the
SI
9..0
inputs. SYNC may be held low until
re-synchronization is desired, or it may be toggled at every
occurrence of a C
B
sample. Refer to the timing diagram in
Figure 9 for required operation of the SYNC control signal.
The internal de-multiplexer will de-multiplex all data in the
input data stream including any ancillary, EDH,VITC, and
EAV/SAV ... signals that may be present. Since this data is
passed directly to the interpolation filters in the same way
that active video would be, it is recommended that such
data be replaced with appropriate blanking levels prior to
entering the GF9103.
The output of the 4:2:2 de-multiplexer consists of three 10
bit channels of YC
B
C
R
data. All three channels are then fed
to their respective interpolation filter.
INTERPOLATION FILTERS
Within the interpolation stage, the Luminance data is over-
sampled by a factor of two and the C
B
and C
R
data is over-
sampled by a factor of four so that the 4:2:2 data is
converted to 8:8:8 data. By over-sampling the 4:2:2 data to
8:8:8 data, the size, cost and complexity of the analog
reconstruction filters following Digital to Analog converters
are reduced.
The Luminance data is over-sampled by a linear phase FIR
filter providing 0.0 dB DC gain, +0.038/-0.025 pass- band
ripple [0.0 s to 0.21 s], 6 dB attenuation at s/4, and 47
dB stopband attenuation [0.30 s to 0.50 s]. Figure 3 and
Figure 4 present the frequency response of the Luminance
interpolation filter.
The C
B
and C
R
data is over-sampled by a linear phase FIR
filter providing 0.0 dB DC gain, passband ripple of +0.2 dB/-
0.2 dB [0.0 to 0.07 s], 6 dB attenuation at s/8 and a
stopband attenuation of 28 dB [ 0.17s to 0.50 s].
Figure 5 and Figure 6 present the frequency response of the
C
B
and C
R
interpolation filters.
Following the interpolation process, a DC offset may be
introduced into the Luminance channel. Setup insertion is
enabled and disabled by the SETUP control signal. While
SETUP is high, the Luminance data is scaled by a factor of
+947/1024 and an offset of +71 (decimal) is added. While
SETUP is low, no scaling or offset is applied and the data
passes through the stage unmodified. The timing diagram
in Figure 10 demonstrates the operation of the SETUP
control signal.
COLOR SPACE CONVERSION
Two operating modes exist for the color space converter
section. These two modes are controlled by the
SELECT_MATRIX control signal. While SELECT_MATRIX is
low, the de-matrixing 3 x 3 multiplier is bypassed so that
over-sampled Y C
B
C
R
data is passed through the stage
unmodified. While SELECT_MATRIX is high, the 3 x 3
multiplier implements the following color space conversion:
SIN X/X CORRECTION
While BYPASS is high, SIN X/X correction is enabled on
each of the three output channels. SIN X/X correction is
implemented by passing the data through a FIR filter with
the frequency response shown in Figure 7. While BYPASS is
low, the FIR filter is bypassed and each channel is passed
directly to the output processing section. Total latency
through the device is 22 clock cycles when BYPASS is low
and 24 clock cycles when BYPASS is high.
OUTPUT PROCESSING
Output data may be rounded to 10 or 8 bit accuracy.
RND10/8
should be set high for 10 bit output rounding and
set low for 8 bit output rounding. Rounding to 8 bit accuracy
is accomplished by adding a rounding bit to SO
1
and then
zeroing both SO
0
and SO
1
.
C
B
and C
R
data may be output as signed (two’s
complement) or unsigned (offset binary) data depending on
the state of the CONVERT control signal. When CONVERT
is set high, the C
B
and C
R
channels are output as signed
(two’s complement) data. When CONVERT is set low, C
B
and C
R
are output as unsigned (offset binary) data,
obtained by inverting the sign bit of the two's complement
number. When operating in RGB output mode, the
CONVERT pin is over-ridden and RGB data is always
output as unsigned (offset binary) data.
G
B
R
1
1
1
-689/2048
3548/2048
0
-1430/2048
0
2807/2048
Y
C
B
C
R
=