
Philips Semiconductors
Video Out
File: evo.fm5, modified 7/24/99
PRELIMINARY INFORMATION
7-19
7.14.4
Latency and Bandwidth Requirements
In order to avoid Hardware Bandwidth Errors the internal
set accordingly to the latency requirements of VO unit. In
the following all the numbers are given assuming data for
a new video line (Y, U, V and OL (overlay) planar mem-
ory format stored in memory) is aligned to 64 bytes. In
other words, it means {OL,Y,U,V}_OFFSET fields are a
multiple of 64 bytes. Otherwise internal VO arbitration for
OL, Y, U and V requests is changed, and following laten-
cies are not guaranteed. VO uses internal 64 byte buff-
ers.
Latency requirements for VO in image mode 4:2:2 or
4:2:0 co-sited or interspersed without up scaling and
overlay disabled, is expressed as:
During 128 VO clock cycles, VO block requires to
have 2 requests acked ([2 Ys, one U and one V]/2).
If VO clock is 27 MHz then VO has to get two requests
(128 bytes) from SDRAM in 128/.027 = 4740 ns.
Bandwidth (in bytes) requirement per video line within
the active image is:
B1x = [ceil(W/64) + ceil[(W/2)/64]*2 + 4] * 64
ceil(X) function is the least integral value greater than or
equal to X and W is the IMAGE_WIDTH field value.
In the same modes but with overlay enabled the latency
becomes:
During the rst 64 VO clock cycles at least one
request must be acked (the OL data).
During 128 VO clock cycles, VO unit requires to
have 4 requests acked ([4 OLs, 2 Ys, 1 V and 1 U]/2).
If VO clock runs at 54 MHz then VO has to get the first
request from SDRAM in 64/.054 = 1185 ns and average
a bandwidth/latency of 4 requests in 128/.054 = 2370 ns.
Bandwidth (in bytes) per video line within the active im-
age becomes:
B1x,OL = B1x + [ceil(W*2/64) + 4]*64
When VO mode is set to image mode with 2x up scaling
mode the latency requirements are multiplied by a factor
of 2 (i.e. instead of, for example, 1 request per 64 VO
clock cycles, the latency becomes 1 request per 128 VO
clock cycles). Bandwidth is roughly divided by 2:
B2x = (ceil[(W/2)/64] + ceil[((W/2)/2)/64]*2 + 6) * 64
B2x,OL = B2x + (ceil[((W*2)/2)/64] + 4)*64
Latency for raw mode or message passing mode is:
During 64 VO clock cycles, VO block requires to get 1
request from SDRAM.
If VO clock runs at 38 MHz then the latency is 64/.038 =
1684 ns and bandwidth is 38 MB/s.
7.15
DDS AND PLL FILTER DETAILS
The PLL filter serves to reduce the phase jitter of the
DDS synthesizer output. It can also be used to multiply
the DDS output frequency by 2x. The DDS and PLL filter
together provide a high quality, accurately programma-
ble output video clock. The complete system is sketched
is set in the ‘11’ position, and the PLL system is disabled.
To start the system, the following steps are needed:
Assign a DDS frequency (this starts the DDS). Allow
for at least 31 DSPCPU cycles for the DDS fre-
quency setting to take effect.
Choose a value for PLL_S, PLL_T (for 8-40 MHz
operation, a value of 1 for division by 2 is recom-
mended).
Choose a value for CLOCK_SELECT (for 8-80 MHz
operation, CLOCK_SELECT=00 is recommended).
Assign a VO_CTL word containing the above
choices. The rst assignment with CLOCK_SELECT
unequal 11 enables the PLL system. Allow for max.
50 microseconds to achieve lock.
Once the PLL is locked, small changes to the DDS fre-
quency are allowed, and the VO_CLK output will
smoothly track the frequency change.
Note that most consumer electronics equipment impos-
es
very high precision requirements on the value of the
color burst frequency. A video encoder derives the color
burst frequency from VO_CLK. In the case of changing
the VO_CLK frequency to software phase lock to a mas-
ter reference, special care is required to keep the color
burst signal frequency within a tolerance of some 50
00
01
10
11
Square-Wave DDS
FREQUENCY
VCO
8 - 90 MHz
VO_CLK
VO_CLK Internal
(to Frame Timing Gen.)
CLKOUT
3
× CPU Clock
0
3
Loop
Filter
Phase
Detect
PLL_S
div T+1
PLL_T
CLOCK_SELECT
div S+1
Figure 7-27. PLL filter block diagram