VP520S

8

112-118

7, odd field

119-125

8, odd field

126-132

9, odd field

133-139

10, odd field

QCIF -> 625 line

0-6

1, even field

7-13

2, even field

14-20

3, even field

21-27

4, even field

28-34

1, odd field

35-41

2, odd field

42-48

3, odd field

49-55

4, odd field

HORIZONTAL FILTERS

Chrominance data is assumed to have already been

decimated down to half the horizontal sampling rate of the

luminance data, before it is applied to the VP520S. When

producing CIF data both luminance and chrominance are then

both decimated by two, when producing QCIF data they are

both decimated by four.

Simulations with actual video have shown that 8 tap CIF

filters and 16 tap QCIF filters give more than adequate

performance in the decimation mode. In the interpolation

mode these same simulations have shown the need for longer

filters in the luminance channel. The hardware thus supports

a 12 tap filter when interpolating luminance from CIF inputs,

but only 8 taps are provided for each chrominance channel.

Even longer filters are needed when QCIF data must be

interpolated, and the luminance channel is provided with 24

taps, and each chrominance channel with 16 taps.

Note that when interpolating by two the output rate is

double the input rate, but every other input will be conceptually

zero. Similarly when interpolating by four there are three

zero's between every data point, even though the output rate

is four times the input rate. Thus during any clock period only

one half or one quarter of the coefficients are actually in use,

and the computational burden is no greater than when doing

the equivalent decimation.

Since all the coefficients are not in use during any clock

cycle, it is convenient to refer to two smaller sets of coeffi-

cients. Thus the 12 tap CIF luminance filter, for example, can

be considered to have two sets of 6 coefficients, and the 24 tap

QCIF luminance filter to have four sets of 6 coefficients. The

addressing in the coefficient RAMs uses this concept of sets.

VERTICAL FILTERS

The vertical filters are designed to produce CIF with the

spatial relationship shown in Figure 7, and QCIF with the

spatial relationship shown in Figure 8. Original PAL or NTSC

video contains lines of coincident luminance and chromi-

nance, but the CIF specification requires that the decimated

chrominance information is shifted such that it lies mid way

between two luminance lines. This is achieved by choosing

the centre outputs from the filter which best fit the require-

ments. The filter outputs actually used by the device are

shown by the arrows in Figures 7 and 8, and are optimal when

the even field provides the original video.

It is assumed that one of the interlaced fields has been

discarded prior to the VP520S, and thus no further decimation

occurs when producing CIF luminance from PAL ( NTSC in

0

1

2

3

4

5

6

7

8

9

10

0

1

2

3

4

5

6

7

8

9

10

11

12

0

1

2

3

4

5

6

7

8

9

10

11

0

1

2

3

4

5

6

7

8

9

0

1

2

3

4

5

6

7

8

9

10

11

0

2

4

6

8

10

CIF LINES

LUM

CHROM

PAL LINES

EVEN

ODD

NTSC LINES

EVEN

ODD

REPEAT

REPEAT

REPEAT

Fig 7 : CIF Spatial Relationships