PNX1301EH NXP Semiconductors, PNX1301EH Datasheet - Page 218

• Manufacturer Part Number: PNX1301EH
• Manufacturer: NXP Semiconductors

Specifications of PNX1301EH

• Lead Free Status / RoHS Status: Not Compliant

PNX1300/01/02/11 Data Book
output fractional X and Y subpixel offset is shown in
Figure
Output scaling calculation method
The output pixel distance in H and V in the ICP is calcu-
lated to high precision (16-bit fraction) even though the
output resolution is fixed at 1/32 of the input grid. Each
output pixel’s location relative to the input pixel grid is giv-
en by:
The X and Y locations may not be integer values, de-
pending on the scale factor. The resulting X and Y pixel
locations can be separated into an integer and a fraction-
al part. The integer part of the X and Y location selects
the pixel and line number closest to the output pixel, re-
spectively. The fractional part gives the fractional dis-
tance of the output pixel to the next X and Y input pixel
values. These fractional parts are the dX and dY values
shown in
The output pixel value can be calculated by interpolation
between the two input pixels or by 5-tap filtering using the
5 nearest pixels rather than the 2 nearest pixels. Interpo-
lation or filtering uses the fractional position values, ∆X
and ∆Y, to select the appropriate filter coefficients. In the
ICP, these values are limited to 5 bits for a resolution of
1/32, even though the actual position value has much
higher resolution. The ICP uses fractional values cen-
tered around the center pixel with a range of -16/32 to
+15/32.
To perform scaling, the X and Y locations of the output
pixel relative to the input pixel grid must be generated.
This includes both the integer part to locate the adjacent
pixels and the fractional part to choose the filter coeffi-
cients which generate the output value from the adjacent
pixels. This could be done by generating the output pixel
X and Y numbers and dividing each by its associated
scale factor. Since dividing is expensive in hardware and
14-8
Figure 14-9. ICP 1/32 output resolution
Output Pixels
Input Pixels
X location of output pixel = X0 of input line + output
Y location of output pixel = Y0 of input window
14-9.
Figure
dX
pixel number / X Scale Factor
+ output line number / Y scale factor
14-9.
1
PRELIMINARY SPECIFICATION
2
dY
time, the ICP effectively multiplies the X and Y pixel num-
bers by the inverse of the X and Y scaling factors, resp.
This is done by incrementing the X and Y input pixel
counters by X and Y increment values that are the in-
verse of the X and Y scale factors, resp. For output pixel
Xn, the inverse of the scale factor is added to the X input
location n times. This is equivalent to multiplying n by the
inverse of the scale factor.
The ICP uses a 16-bit integer and a16-bit fractional value
for the X and Y increment values. This allows a fractional
value resolution of 1/64K. Since the increment value will
be added 1024 times in a 1024-pixel line, any error in an
individual calculation will be multiplied by 1024. The high
resolution of the calculation prevents an accumulation of
error as you increment along the line.
Only the most significant 5 bits of the fractional value are
used by the filter coefficient RAMs. However, the X and
Y counters are incremented by the high-resolution X and
Y increment values. The result of this truncation is a
worst case error of approximately 1.5% in amplitude rel-
ative to arbitrary pixel output positions.
The error caused by discrete (1/32) resolution can be re-
duced to exactly zero if the output image size is adjusted
to have a repeating pattern that fits on these 1/32 bound-
aries. For zero error, this implies that the scaling factor
must be of the form of B/A, where B (the output pixel
count factor) is a sub-multiple of 32 [i.e. 1, 2, 4, 8, 16, 32],
and A (the input pixel count factor) is an integer deter-
mined by the nearest acceptable scale factor for a given
B. In the 640 to 1024 conversion case, the B/A ratio was
8/5, meeting this requirement.
The integer values, if accumulated, would be equal to the
total number of input pixels when scaling is complete.
The integer values for each pixel define the number of
pixels to read from memory and shift in to generate the
next output pixel. For example, a scaling factor of 1.0 will
result in one pixel shifted in for each output pixel gener-
ated. Upscaling will have integer increment values of
less than one. This means that the integer value will be
‘0’ for some pixels and ‘1’ for others. For example, up-
scaling by 2.0 will result in integer values of ‘1’ half the
time and ‘0’ for the other half, depending on the carry out
from the fractional increment.
Pixel shift bypassing for large down scaling
Down scaling will have integer increment values of great-
er than one. In this case, the integer value indicates the
number of pixels to read to obtain filter pixels for the next
output pixels. There are two ways to read and shift in the
pixels for down scaling: shift all and shift bypass. In the
shift all mode (the default mode) all five pixels are shifted
for each input value read and shifted in. Shift all mode
uses the five input pixels nearest the output pixel, inde-
pendent of scaling factor. In the shift bypass case, only
the last pixel is shifted in. For example, in a down scaling
of 10, nine pixels are read and the 10th pixel is shifted in
to the filter. Shift bypass mode is used for large down
scaling, i.e. down scaling factors of 2.0 or greater. The
shift bypass mode is selected by setting the GETB bit in
the parameter table. It uses input pixels that are nearest
the output pixel and those nearest each of the four output
Philips Semiconductors