The final version can be found in Chapter 10
of the Doom9 Analog TV capture
guide.
Determining
the capture window of a capture card
Arachnotron, December 18th 2003
Introduction
I am assuming the reader is familiar with the concept of video lines,
resolution, aspect ratio and active picture area. If not, look at the
sites mentioned in the references. This is
just a quick guide on how to determine which part of a video line is
captured by a video capture device.
The official numbers, according to Rec. ITU-R BT.470-6, for the
active part of a video line are:
For PAL, a whole line takes 64 µs and the active part is 52.0 but
may lie between
51.7 and 52.0 µs.
For NTSC, a whole line takes 63.555 µs and the active part is
52.66 but may lie between 52.46 and
52.86 µs
(Actually, for NTSC there are a number of standards which all differ
slightly on the size of the active area and even on how to measure it.
But the number used in Rec. ITU-R BT.470-6 seems to be the one most
widely used when dealing with digital video formats)
Unfortunately, these are not the only 'standards' you can adhere to.
A capture device will typically sample the whole video line at a fixed
sampling rate, crop off all pixels which fall in the horizontal
blanking area and resize the remainder to whatever horizontal
resolution was requested by the user. The area's to be cropped are set
by the driver and mostly cannot be changed by the user. The
problem lies in that different manufacturers have different ideas about
how to interpret standards and how much of the original video
line can be cropped. And since it is the driver that determines the
capture window, even two identical devices can give different results
when used with different drivers.
The problem for you as the user is that the capture window influences
the aspect ratio of the picture. If two
devices capture at 720x480, but one of them uses a smaller capture
window than the other to generate those 720 pixels, the resulting
pictures will have different aspect ratio's. The picture below
illustrates this effect. It shows 4 strips that have been cropped from
720x576 captures
made using different devices and with a test
DVD as a source. The white crosses on the right originate from the
center of the test picture, and all strips have been aligned relative
to it.
. The four
strips are:
Original, cropped directly from a rip of the test DVD picture. A
DVD player outputs it's 720 pixels in 53.33 µs (see below)
Terratec Cameo Grabster, a device with a capture window of 53.33
µs
Hauppauge BT878a based card using the BTWincap drivers, capture
window 52.00 µs
Hauppauge CX23881 based card, Hauppauge drivers, capture window
is 51.56 µs
As the capture window becomes smaller, the pixels are devided over a
smaller area and the resulting picture is stretched out horizontally!
If
you want the aspect ratio to be correct on playback, you have to
correct for this effect. This page aims to give simple instructions on
how to
determine what capture window your card uses utilizing standard DVD
player as a signal source. Look in the Doom9 TV capture Guide
for more information on what to do once you know what it is.
Using the
analogue signal from a DVD player as a reference.
A DVD player operates at 13.5 MHz when it plays back a DVD. In other
words, it plays back
13500000 pixels in a second. This means that a single pixel in your
MPEG2 file is played back in exactly
1/13500000= 0.07407 µs or 74.07 ns. More information about the
way a DVD player plays back a MPEG2 file can be found here.
Note that this relationship is only
valid for a standard stand-alone DVD player. A multimedia-PC will do
some scaling of it's own when generating a TV picture and is unsuitable
as a reference for determining the capture window!
To calibrate your capture device, simply connect your DVD player to
your capture device. If possible, use an Y/C or S-video cable to do
this, but a composite video cable will do if neccesary.
Now play a DVD disk, capture a small clip en see how many of the
original DVD pixels made it into your capture file.
divide the number of DVD pixels by 13.5 and presto, you have your
capture window in µs. For example, 720 DVD pixels equal
720/13.5 = 53.33 µs.
Many DVD players will also play S-VCD's. A S-VCD pixel is
played
back at a lower rate than DVD's, 9 MHz. This means that a single
pixel is played back in exactly 1/9000000 =0.1111 µs or
111.1 ns. Because of this S-VCD pixels are 50% wider than DVD
pixels and the results will be slightly less accurate. Apart from that
the method works just the same as for a DVD, only now you divide
by 9 instead of 13.5. For example 480 S-VCD pixels equal
480 / 9 = 53.33 µs.
Because it can be difficult to determine exactly how many DVD pixels
made it into your capture using a ordinary DVD.
To help with this I created a test DVD for both PAL and NTSC.
You can
download them in the download
area. You need RAR to unpack
the files.
The test DVD contains a 10 s clip in which a single frame is repeated.
The DVD is set to play in a loop. Below you can see the NTSC version of
the test frame. Click it to see it full-size. They contain a
ready made VIDEO_TS folder which can be
burned to video-DVD using Nero burning Rom or any other DVD burning
program.
If you don't own a DVD burner, a S-VCD compliant MPEG2 file can be
found here
also. Use your favorite SVCD authoring program to make a SVCD out of
it. The SVCD MPEG file contains 20 seconds of the test image shown
below (click to see it full size)
To be able to analyze the
results with one-pixel accuracy it is grayscale. Both DVD's and capture
devices distribute color values across two or more neighbouring
pixels,
which makes a color picture difficult to use accurately. The picture contains a cross
made up of a sequence of horizontal
and a sequence of vertical perpendicular lines. The lines are spaced
exactly 20 pixels apart. Every 40 pixels, the pixel number is written
next to a line. The DVD pictures are exactly 720x480 and 720x576
pixels in size, the SVCD pictures 480x480 and 480x576.
Capturing and how to interpret your results
Connect the DVD player using a good quality S-Video lead.
Composite also works, but gives less detail.
Capture at a high horizontal resolution, preferably 720x480/576
or higher, using a non-lossy codec like huffyuv if possible. If your
capture application won't let you do that, try VirtualVCR which can be
obtained here .
Select a single clean frame using the offset buttons
Save the frame to a .bmp file, by selecting 'File', 'Save Image Sequence' from the menu
and setting the output format to 'Windows
BMP'
Open the BMP file in Microsoft paint. (Start, Programs,
accessories)
Zoom in to 800% and enable the grid (view, zoom , Custom...., 800% and view, zoom, show grid. ) The
picture below shows sections of a PAL capture made using a BT878 based
card and the BTWincap drivers v5.3.6.1 at 720x576.
The
picture below shows sections of a PAL capture made using a BT878 based
card
and the BTWincap drivers v5.3.6.1 at 720x576. Click to see the
whole captured frame.
Search for the first visible vertical line on the left. Note the
DVD pixel number (the numbers written above the lines), i.e. 20, 40, 60
etc.The vertical line is probably composed out of 2 or more gray lines
with varying shades of gray. look for the line with the lightest shade
and note the number of pixels up to but not including it. If
the vertical line contains two lines of exactly
the same shade of gray, note the number of pixels to the left of the
first of those two lines and add 0.5. In the illustration above the
line on the leftmost position originates from DVD vertical line number
20. The vertical line with the
lightest shade of gray in the capture has 12 pixels leading up to it as
indicated by the red arrow.
Do the same for the last vertical line that is still in your
capture window on the right. In the
illustration above the line on the rightmost position originates
from DVD vertical line number
700. The one pixel vertical
line with
the lightest shade of gray in this line has 10 pixels to the right of it, as
indicated by the red arrow.
Calculate the number of capture pixels between the two. This
number is 720 - (pixels to the left) - (pixels to the right) . In the
example this is 720 - 12 - 10 =
698 pixels.
Calculate the number of DVD pixels between the lines from the
numbers printed above/below the lines. This is (Right DVD
pixel number -
left number) +1 . In the example
this is 700 - 20 + 1 =681DVD
pixels.
From this you can calculate the ratio of DVD pixels per captured
pixel. This is (DVD
pixels from step 11) / ( capture pixels from step 10). In the example this number
is 681 / 698 = 0.9756 DVD pixels
per captured pixel.
Now
multiply this number by the horizontal resolution of your capture. In
this example: 0.9756 * 720 = 702.4 DVD pixels in the capture window
Divide this number by
13.5 and you have the window in µs. In the example: 702.4 / 13.5 = 52.03 µs
Note: you might see some black
vertical bars on the left and right sides of your capture. This is
normal. Some DVD players crop a few pixels from the sides. In the
capture these show up as black. These black pixels still count, your
capture is still
720x480 or 720x576 and the receipe above still works.
The procedure when using a SVCD is identical, only at step 14 you
divide by 9
instead of 13.5.
Why are the white lines in my capture so blurred?
When you capture the DVD signal from the test DVD you will probably
notice that instead of a nice sharp single pixel wide vertical line you
get a several pixels wide line made up out of single pixel wide lines
with varying shades of grey. This has nothing to do with the
quality of your capture card. This 'blurring' is caused by the way a
DVD player plays back a single pixel. The picture below shows what
happens when a DVD player processes a single gray line with a white
pixel in it:
The row of squares on top represent some of the pixels in the original
line. Below that is a graph representing the analogue signal resulting
from it. It was taken from an actual oscilloscope picture made from the
signal of my DVD player. When the signal goes up, the color is white.
Down is black; anything in between is a shade of gray. The peak
resulting from one pixel is actually almost three DVD pixels wide, with
oscillations at the side. When you capture a signal like this using for
example a
BT878 based card with BTWincap drivers at 720x576, the pixels resulting
from this will look like the row of pixels below. But by looking for
the lightest
pixel in your capture you can still locate where the top of the peak is
and where the original white pixel must have been.
A special case: BT878 based capture cards
TV cards using the Brooktree/Conexant BT878 chip have been around for a
long time. As a result there are a number of specialized tools for this
chip floating around on the internet. One of them, BTtool, enables you
to read back how the chip is being programmed by the driver and the
capture application. To function, it needs a driver, BT878.sys to be
installed. Both can be found in the download
area.
After installation of the driver, a reboot is required, even if the
installer does not request it.
To understand the values you use below, a short description of how
BT878 cards (and for that matter CX2388x based cards ) operate is in
order. The BT878 is a programmable chip, and it is programmed by it's
drivers. The drivers do this by setting registers, the digital
equivalent of switches and nobs on the chip.
First, the card digitizes the whole scanline, 64 µs, with a
sample rate of 17.73447 MHz. This yield 64 * 17.73447 = 1135
samples per complete line (for PAL; for NTSC the number is
63.5555 * 14.31818 = 910 pixels ) . These 1135 samples are reduced to
the required number of samples per line by interpolation. In other
words, the original samples are sampled again, but at a lower rate
resulting in the desired reduced number of pixels per line. The amount
of reduction is set by a scaled sampling rate. The HScale register
contains the ratio between the unscaled and the scaled sample ratio.
Because a whole line contains non-picture information on the left
(horizontal blanking) and for a short stretch on the end (Front Porch)
the resulting pixels have to be cropped to obtain the required number
of pixels. the HDelay register determines how many of the scaled pixels
are to be cropped on the left. The HActive register determines hor many
pixels after that are used. Any pixel left after HDelay and HActive is
ignored.
More on how the scaler in a capture device operates can be found here
. More details on the BT878 registers can be found in the
DataSheet.
You can determine the capture window of a BT878 based device as follows:
Warning!
The BTtool and driver can make your system unstable while they are
running. Please do not run any other applications while testing to
prevent data-loss !
Do not start BTtool yet!
Open your capture application, set the capture resolution and
enable
the preview.
Start BTtool and write down the following hexadecimal
numbers
which can be found in the upper left corner of the BTtool screen:
Crop
Vactive Lo
HDelay low
HActive Lo
Hscale Hi
HScale Lo
Select your TV system and fill in the numbers in the calculator
provided below.
Press the Compute button
Warning:
Do not change your capture settings while BTtool is running! First
close BTtool, make your changes, and restart BTtool afterwards. Example: The picture shows the BTtool
readings for the PAL 720x576
capture on a BT878
based card with BTWincap drivers from the example used in the previous
section. These values have already been set as defaults in the
calculator below. Press Compute to see the results. In our example , the HScale
setting results in a scaled sampling rate of 13.849 MHz. This results in 13.849 *
64 = 886 scaled pixels
per line. These lines of pixels still
contain
the horizontal blanking information before the active video area
begins. So HDelay sets the number of pixels at the start which must be
cropped, 143 in the example.
The final register is HActive, which sets the number of
pixels
following HDelay which hopefully contain the picture.
HActive
will
normally be the number of horizontal pixels you chose for the capture,
and
in this case is 720. Once
these
720 pixels are in, the final 886-143-720 = 23 pixels are skipped.
From this information, the capture window can be determined:
The capture window now is 720 /
13.849 = 51.99 µs.
This equals 51.99 * 13.5 = 701.9 DVD
pixels
This
calculator is an adaptation of a calculator programmed by TREVLAC
. Click here to see the original version
as he made it.
