Criterion Quiz
Travis Kelm
until now. But first, some history. Criterion recently sent a letter with
several questions to the members of its collector's club (which I'm sure may
of you are members). By correctly answering three of the four questions,
Criterion would credit your account with two bonus (you know the ones you
can save up to earn free Criterion discs). Suffice it too say, these
questions were rather hard. Thus I appeal you, friends, lovers, and laser
video afficianados, to help me with my quandry. So if you want to help a
poor young film buff earn enough credits so he can get the Criterion release
of _Man Bites Dog_, or if you just enjoy a good puzzle try to answer these
questions, and post or send me the answers (actually I only remember two of
them, and the actuall letter is miles away, thus feeling that speed is of
the essence (i.e. the deadline is fast approaching) I will type an
incomplete list and amend it later).
1) Which Criterion Collection films were directed by S. Kubrick?
2) A. Hitchcock is of course notorious for making appearences in his own
films. Name the films by Hitchcock in the Criterion Collection and briefly
describe his cameo in each of them.
Thanks for your support.
Gregory S Rogers
ANNOUNCEMENT
--------------------------------------------------------------------------
I am re-posting a tutorial which I wrote on Comb Filters. I have
been asked by a long list of net readers to repost this because they
missed it for one reason or another. Recently, while I was away for a
week I had 9 requests for this article.
Judging from the recent number of questions on this topic I think it
might be useful to have this second posting. I have also been asked to
post it to alt.video.laserdisc since some readers cannot access
rec.video (?). So it will appear in both forums this one time.
My apologies to those that read or saved it the first time. The only
changes are correction of some spelling errors.
--------------------------------------------------------------------------
Every week there are several postings in the video newsgroups asking
questions or providing "facts" about Comb Filters (Y/C Separators)
and/or S-Video (Y/C) vs Composite Video interconnection issues.
I think these are VERY important topics in getting the best performance
possible from a video system. Consequently, I have written a tutorial
reference on the topic. This tutorial is quite lengthy, over 2000 lines
total. Today, I have posted the tutorial in 4 parts to rec.video and
alt.video.laserdisc.
Y/C Separation Filters
by Greg Rogers
Contents
Part 1: Analog Y/C Separation Filters.
Part 2: Digital Y/C Separation Filters.
Part 3: Appendix 1: The Composite Video Signal
Part 4: Appendix 2: Y/C Separation Filter Theory
The main article is contained in Part 1 and Part 2. It describes the
various types of Y/C Separation Filters, how to identify them, their
benefits and their limitations. This portion of the tutorial is
non-technical and provides the basic information that a videophile should
know (and maybe a little more.) Part 2 depends on Part 1, they MUST be
read together for clarity.
Part 3 and Part 4 are Appendices. They are aimed at readers who want to
know more about how the Filters work (and why they have limitations).
They are more technical and aren't required to understand Parts 1 & 2.
There is some duplication of information between Parts 1/2 and 3/4. I
attempted to make Parts 3 and Part 4 readable as stand-alone documents.
Finally, this is copyrighted material. Please respect my request to NOT
duplicate or repost this information elsewhere.
Greg Rogers
Tektronix
Gregory S Rogers
**************************************************************************
This article is so long I've split it into several sections. The main
article is split into Part 1 and Part 2. Part 2 will not make any sense
unless you also read Part 1. Part 3 and Part 4 are Appendices. They are
considerably more technical and aren't required to understand Parts 1 & 2.
Y/C Separation Filters
by Greg Rogers
Contents
Part 1: Analog Y/C Separation Filters
Part 2: Digital Y/C Separation Filters
Part 3: Appendix 1: The Composite Video Signal
Part 4: Appendix 2: Y/C Separation Filter Theory
**************************************************************************
April 5, 1994 Copyright 1994 by Greg Rogers
Y/C Separation Filters
or
"Kookie, Kookie, Lend Me Your Comb"
This article presented in tribute
to Edd_B...@77.Sunset.Strip.com
Part 1: Analog Y/C Separation Filters
**************************************************************************
Part 1: Analog Y/C Separation Filters
Table of Contents
1.1 Y/C Separation Basics
1.1.1 Purpose of the Y/C Separation Article
1.1.2 Composite Video Signal Description
1.2 What Type of Products Have Y/C Separation Filters?
1.2.1 TV Monitors
1.2.2 LD Players
1.2.3 VCR's/Camcorders
1.3 Low Pass and Notch (or Band Pass) Filters
1.3.1 Description
1.3.2 Recognizing Notch/Low Pass Filters
1.4 Glass or CCD Line Comb Filters
1.4.1 Alternate Names
1.4.2 Description
1.4.3 Recognizing Glass or CCD Line Comb Filters
1.4.4 Benefits of Line Comb Filters
1.4.5 Limitations of Line Comb Filters
**************************************************************************
1.1 Y/C Separation Basics
1.1.1 Purpose of the Y/C Separation Article
The purpose of this article is to describe the different types of Y/C
(Luminance/Chrominance) Separation Filters used in TV monitors, VCR's,
LD Players, and other video components. They are often called comb
filters, although that isn't the only method of Y/C separation.
This subject is of importance for video products because there is a
widespread variation in the performance of different types of Y/C filters.
The differences are directly observable in the video artifacts that exist
after separation. Some artifacts occur because of incomplete separation
and others are side effects of the separation circuit itself.
This article is directed at video enthusiasts wishing to understand more
about this key element of their video products. It provides information
for making better informed buying decisions or to just satisfy the curiosity
about comb filters often expressed in this forum. It also provides means
for understanding how to choose the best interconnect strategy to take
full advantage of potential product performance capabilities.
It is important to make the best decision on whether to use the composite
video or the S-Video (more properly called the Y/C) signal interface when
connecting components. This decision should be made by determining which
component has the better Y/C separator. The interconnection should use
the better Y/C separator and bypass the inferior alternative. There should
be enough information in this article to identify the superior Y/C separators
either by name or by knowing what effects to look for in a comparison test.
This article is NOT intended to provide a detailed engineering level
understanding of how the various Y/C filter technologies are implemented.
However, Appendix 2 will describe in more technical detail how the various
solutions work and why they still have certain limitations.
1.1.2 Composite Video Signal Description
I will not describe the details of creating the composite signal here.
This is covered in the moderately more technical Appendix 1, which
accompanies this article. Read it for a more complete explanation.
Below is a simplified description of the composite video signal.
A composite video signal includes a luminance (brightness) signal and
a chrominance (color) signal. They are sometimes referred to as the
luma (Y) and chroma (C) signals for short. The C signal is formed by
combining two intermediate signals called the I & Q signals, which are
created from the original Red, Green, and Blue (RGB) light sensors of a
video camera or film scanner.
In a broadcast TV signal, over the air or delivered by a cable system,
a radio frequency (RF) signal is transmitted. This RF signal contains
both audio and video signals. The video signal is called the composite
video signal because it is the addition of the Y & C signals. The audio
signal won't be discussed in this article.
The video and narrow bandwidth audio signals are easily separated from
the RF signal because they don't overlap in the frequency spectrum. In
order to minimize the frequency spectrum used, the Y & C signals do
overlap. This permits more TV channels to be broadcast within the total
frequency spectrum available to broadcast television.
The problem created by overlapping the Y & C signal frequency spectra
is that it is very hard to separate them completely again. The original
designers of the U.S. NTSC television standard (in 1953) recognized this
limitation, but believed it was a good tradeoff to preserve frequency space
and to provide compatibility with the existing black and white TV system.
They very cleverly standardized on a system which interleaved the Y and C
frequencies as their spectra overlapped. The Y and C signals bunch up in
clusters around specific frequencies, fitting together like the teeth of
two combs. The signals alternate between clusters of Y frequencies and
clusters of C frequencies. A small section of the frequency distribution
is shown below.
| | | | | |
||| ||| ||| ||| ||| |||
||||| ||||| ||||| ||||| ||||| |||||
C Y C Y C Y
Fsc-Fh Fsc-Fh/2 Fsc Fsc+Fh/2 Fsc+Fh Fsc+3Fh/2
Fsc is approximately 3.58 MHz, the color carrier frequency.
Fh is approximately 15.734 KHz, the horizontal scan frequency.
During movement of certain images and patterns, the frequency clusters
can become smeared and widen out. In some cases the Y & C clusters will
overlap and can not be separated. But the interleaving works quite well
for most video images.
The original designers of the NTSC system expected technology to step up
to the separation problem and provide an economical means to separate the
Y/C signals much sooner than it has actually occurred. Comb filters date
back to much earlier than 1953, but they were too expensive to use.
It may be helpful in the following discussion, to know that the C signal
is modulated onto a 3.58 MHz (approx) color carrier and is limited to a
frequency band from approximately 2.1 MHz to 4.2 MHz. The Y signal is
spread from nearly DC to about 4.2 MHz. The smaller the picture detail,
the higher are the frequency components in the signal. It is also
helpful to know that a frame of video occurs about 30 times per second.
A frame of video consists of two fields of video, with a field rate of
about 60 Hz. The horizontal scan lines from one field are interlaced
with the scan lines in the next field. At each line in the field, the
C signal's phase is shifted by 180 degrees. In other words, the C signal
(but not the Y signal) is inverted at every other line. Similarly, in
each successive Frame, the C signal from the equivalent horizontal scan
line, is also inverted.
Again, please read Appendix 1, if you would like a more detailed explanation
of what the composite video signal is all about.
1.2 What Type of Products Have Y/C Separation Filters?
1.2.1. TV Monitors
TV monitors usually have several signal inputs. Broadcast RF inputs are
F connectors or antenna terminals. Composite Video is usually an RCA jack
but might be a BNC. S-Video, actually Y/C, uses a 4 pin DIN connector, or
possibly two BNC's. Some monitors may also have RGB connections.
Composite video, supplied directly or after recovery from the RF signal,
is then separated by a Y/C Separation Filter. The resulting C signal is
separated again into the I and Q signals by a process called demodulation.
Finally, the Y, I, and Q signals are decoded in a special matrix to
re-create, as closely as possible, the original RGB signals to drive the
CRT's red, green, and blue electron guns.
If the TV monitor is fed separate Y/C signals by using the Y/C input,
these signals bypass the monitors Y/C separation circuits altogether.
1.2.2 LD Players
The laserdisc stores its video signal in the composite video format.
Therefore, when the disc is played back the signal is initially composite
video. The LD player will have a composite output and may contain a Y/C
separation filter to provide a Y/C output.
It is a common mistake to think that because the LD stores its signal
in the composite video format that only the composite output should be
used. This seems logical since it avoids passing the signal though the
LD players Y/C separation filter. BUT, eventually the composite signal
will have to be separated by the monitor using the monitors Y/C separation
filter instead. Therefore, for the best performance, connect the LD player
to the monitor, in the way that uses the better Y/C separation filter. If
the LD player has the better filter, then use the Y/C output. If the
monitor has the better filter, then use the composite video output from
the LD player. This article will help identify which product has the
better filter.
1.2.3 VCR's/Camcorders
VCR's/Camcorders record video as separate Y & C signals, NOT composite
video. This applies to all consumer formats including VHS, S-VHS, 8mm,
Hi-8, etc. Prior to recording, the VCR must separate composite video
into Y/C video the same as a monitor. Therefore, the same LD player to
monitor rules, apply when connecting a video signal from a LD player to
a VCR.
Since VCR's record video as separate Y/C signals, the VCR's Y/C output
should always be used to connect to a monitor or another VCR for dubbing.
This avoids summing the Y/C signals in the output of the VCR and splitting
them again in the other product.
1.3 Low Pass and Notch (or Band Pass) Filters
1.3.1 Description
Early TV monitors and VCR's, and very inexpensive products today, use
simple analog filters for Y/C separation. The composite video is split
into two signals by filtering. In the Y path a low pass or notch filter
rejects anything above 2.5-3.0 MHz. The exact frequency varies between
products. This effectively filters most of the C signal out of the
composite signal, leaving the Y signal. Unfortunately, it also removes
the higher frequency Y signal components, above 2.5-3.0 MHz. This loss
of bandwidth reduces the horizontal resolution of the luminance signal
and fine details in the picture are lost.
However, if the C signal were not filtered out of the Y signal, it would
create a varying luminance signal. In particular, the color carrier
would create a checkerboard luminance appearance where ever patches of color
existed. Any C signal that isn't completely filtered out of the Y signal
will create this problem. This artifact of incomplete Y/C separation is
called cross-luminance.
In the second path, to create the C signal, a band pass filter passes only
frequencies from about 3.0 MHz to 4.2 MHz. Unfortunately, this doesn't
remove the luminance frequencies in the same range. These high frequency
luminance signals are in the C signal and therefore create unwanted color
patterns, or rainbows, in the areas of small detail. They are
particularly obvious in any areas containing fine closely spaced lines.
This unwanted color interference is called cross-color.
If the lower frequencies had not been filtered out, then the cross-color
would occur everywhere in the picture, not just in areas where fine detail
is present. This technique also reduces the available bandwidth of the
C signal, which reduces the already marginal color resolution. Fortunately,
the eye is far less sensitive to color resolution than luminance resolution,
but this still limits the quality of the picture.
1.3.2 Recognizing Notch/Low Pass Filters
The easiest way to determine if this type of Y/C separation is being used,
is to look at the multiburst test pattern on "A Video Standard", LD-101,
from Reference Recordings. To test the monitor, feed the signal from the
LD player to the monitor using the composite video connection. This
forces the monitor to use its Y/C separation circuit. If you wish to test
the LD player, then you must use the Y/C connection (S-Video) so that the
LD player uses its Y/C separation filter and not the monitors. Look at
frame 50816. If the monitor produces stray colors in the high frequency
bursts (narrow vertical lines) at the right side of the pattern, the Y/C
separation filter is probably of this type.
It is still possible to have minor cross-color effects even with the
more sophisticated filters described later, particularly on closely
spaced diagonal lines. However, none of the more sophisticated filters
should show cross-color using the multiburst pattern suggested here.
To test the VCR's input Y/C separation filter, connect the composite output
from the LD player to the VCR and observe the multiburst pattern on the
monitor using the Y/C output of the VCR. If you haven't got Y/C signal
outputs on your VCR, then you must use the composite output between the VCR
and the monitor. In this case, first test that the monitor doesn't show
cross color on this test pattern, as described above. (i.e. The monitor
doesn't have this type of Y/C filter.) Then, if cross-color occurs, it
must be the VCR's input Y/C separation filter at fault, not the monitor.
Note: It is very unlikely that any product that includes a Y/C output
or Y/C input, uses this type of Y/C separation filter. Today, they are
almost exclusively limited to very low cost products. I can't imagine
any LD players that would have this filter type. Interestingly,
professional broadcast monitors may have these type of filters available
which can be switched in for comparison purposes with one of the better
filters listed below.
1.4 Glass or CCD Line Comb Filters
1.4.1 Alternate Names
These are sometimes called simply Glass Comb Filters, CCD Comb Filters,
1-H [Glass or CCD] Comb Filters, or 2-H [Glass or CCD] Comb Filters.
The may be called a Line, 2-Line, or even 3-Line Comb Filter. They
are NEVER (correctly) called Digital or Logical Comb Filters. In
particular, a 3-Line Logical Comb Filter is a digital filter, described
later.
1.4.2 Description
These are the most common Y/C separation filters in use today. Most
VCR's, most TV monitors except the top end monitors, and mid priced
and lower LD players typically use these filters. The glass comb filters
and CCD comb filters perform in very nearly the same way, but the CCD
filters are newer technology. Products less than about 5-6 years old
will mostly have CCD comb filters and products more than 10 years old will
almost always have glass comb filters. In between, both types are common.
Appendix 2, contains an explanation on how a line comb filter works.
Read it if you want the theory explained. In brief, the line comb filter
operates by delaying the last, composite video, horizontal scan line and
comparing it to the current horizontal line. Adding the two lines
together, and adjusting the output gain, provides the Y signal.
Subtracting the current line from the delayed line and adjusting the
output gain, provides the C signal. These two operations are concurrent
and then the next line is processed with a delayed version of the current
line in the same manner. Therefore, the Y and C signals are separated a
line at a time as the field is scanned.
The process just described creates two filters which have frequency
responses that look like the teeth of a comb. The responses (teeth of
the combs) are interleaved just as the Y and C signal clusters are
interleaved. This allows the filters to separate the Y and C signals.
This is illustrated in the diagram below.
Comb Filter Frequency Response
100%|... ............ ............ ............
| .
| . . . . . .
Y | . . . . . .
| . . . . . .
0 |______ . __________________ . __________________ . _______________
Frequency
100%|............. ............ ............ ...
| . . . . . .
| . . . . . .
C | . . . . . .
| . . . . . .
0 |________________ . __________________ . __________________ . _____
Frequency
| | | | | |
||| ||| ||| ||| ||| |||
||||| ||||| ||||| ||||| ||||| |||||
C Y C Y C Y
(The filters actually look more like the top halves of sine waves, i.e. a
full wave rectified sine wave. This was as good as I could draw in ASCII.)
The type of filter just described is known as a 1-H line comb filter
since it uses a 1-horizontal scan line delay to process the signals.
Other more complex comb filters can be built using 2-horizontal scan line
delays and are called appropriately 2-H line comb filters. They have a
slightly different comb filter shape (more like a full sine wave shape)
but work in essentially the same manner. They are not often used because
of their higher cost vs marginal performance improvements.
The line comb filter can be implemented with two types of analog
components. These simple filters can also be easily implemented with
digital processing, but that alone, isn't likely to be done. With
digital processing further enhancements are possible as explained later.
The Glass Comb Filter converts the composite signal to an acoustic signal
using a piezo-electric transducer. The acoustic signal is bounced through a
special glass cavity and reconverted by another transducer to an electrical
signal exactly the time of one horizontal scan line later. This is
called a 1-H delay, for 1 horizontal line time delay. The current line
is summed in one operation and subtracted from the delayed signal in
a parallel operation. Since the signal output of the delay line is
continuous, the sum (Y) output and the difference (C) output of the glass
comb filter is a continuous signal.
The CCD (Charge Coupled Device) comb filter works in nearly the same
manner, but the signal is sampled, usually at 4 times the color carrier
frequency. The CCD looks internally like a long chain of capacitor cells
that can hold a charge directly related to the input signal size. The CCD
is an analog device, which samples the size of the signal at its input and
then with each successive sample, it transfers the charge of each previous
cell toward the output. By selecting the proper number of cells and the
appropriate sample rate, a 1-H delay can be achieved. The current composite
video line and the delayed line are summed and subtracted to create the Y
and C signals as above.
The fact that the signal is sampled instead of continuous produces no
limitations in performance since the sample rate is sufficiently higher
than the input frequencies. The sample rate must be higher than twice
the highest frequency component that will be sampled to avoid a problem
called aliasing. I won't discuss aliasing further, except to say that if
it were to occur, additional frequencies not actually in the original
signal would be generated, creating video artifacts.
(For the technically minded the sample rate is about 14.3 MHz, with a
Nyquist frequency of about 7.15 MHz. The highest video frequencies are
kept below the Nyquist frequency.)
1.4.3 Recognizing Glass or CCD Line Comb Filters
First determine that the product doesn't have just analog filters as
described in case 1. Next select the vertical color bar pattern on frame
50815 of AVS. This frame has vertical color bars on top and patches of
color underneath. Look at the areas of vertical color transitions
between the bars on top and the patches below. If the product uses a line
comb filter, the horizontal edges will have a pattern of alternating
light and dark squares on either side of the horizontal boundary. This
appears as a zipper or a checkerboard appearance along the vertical color
transition. It is an artifact of how the comb filter works when
processing the lines around the color transition. It is an example of
cross-luminance, because the color carrier is modulating the luminance
signal creating the light and dark alternating pattern. It is also
commonly called "hanging dots".
There will also appear a zipper effect along the vertical edges of the
color bars particularly the transition between the green and magenta
bars. If the horizontal edge does NOT have hanging dots, but the
vertical edge does have dots, then this is not a simple line comb filter.
Be thankful, and see the section on digital filters.
1.4.4 Benefits of Line Comb Filters
The line comb filter will NOT suffer from poor horizontal resolution since
the Y signal is not low pass filtered at 2.5-3.0 MHz. Therefore, it will
provide much better fine details in the picture than the previous low pass
filters. In addition, it will separate most of the higher frequency Y
signals out of the C signal. This will dramatically reduce the
cross-color effects discussed earlier. Again look at the multi-burst
pattern in frame 50816 of AVS. There should be virtually no colored
artifacts in the picture.
1.4.5 Limitations of Line Comb Filters
The line comb filter can not completely avoid cross-color effects
particularly when fine diagonal lines are present in the picture. AVS
doesn't have a very good test for this effect. The best example is to
look at frame 50828, and notice the concentric circles in the center of the
picture around the number 30. Notice the faint rainbows that spread out
from the center around the closely spaced circles. Single step ahead and
the color patterns will seem to shift and move. This again is an example
of cross-color, but it is better than in the previous simple filter case.
Another limitation of the line comb filter was already described as the
creation of hanging dots, or cross-luminance effects along the edge of
vertical color transitions. Again this is usually preferable to the
simple filter solution which throws away the luminance resolution.
End of Part 1: Analog Y/C Separation Filters
----------------------Copyright 1994 Greg Rogers-------------------------
------------------All Rights Under Copyright Reserved--------------------
Permission is granted for reproduction of this article, in part, ONLY for
follow-up articles in rec.video. No other reproduction, in whole or in
part is permitted without the permission of Greg Rogers.
-------------------------------------------------------------------------
**************************************************************************
Greg Rogers
Tektronix
Gregory S Rogers
***************************************************************************
This article is so long I've split it into several sections. The main
article is split into Part 1 and Part 2. Part 2 will not make any sense
unless you also read Part 1. Part 3 and Part 4 are Appendices. They are
considerably more technical and aren't required to understand Parts 1 & 2.
Y/C Separation Filters
by Greg Rogers
Contents
Part 1: Analog Y/C Separation Filters
Part 2: Digital Y/C Separation Filters
Part 3: Appendix 1: The Composite Video Signal
Part 4: Appendix 2: Y/C Separation Filter Theory
**************************************************************************
April 5, 1994 Copyright 1994 by Greg Rogers
Y/C Separation Filters
or
"Kookie, Kookie, Lend Me Your Comb"
This article presented in tribute
to Edd_B...@77.Sunset.Strip.com
Part 2: Digital Y/C Separation Filters
**************************************************************************
Part 2: Digital Y/C Separation Filters
Table of Contents
2.1 2-D Adaptive Y/C Separation Filters
2.1.1 Alternate Names
2.1.2 Description
2.1.3 Benefits of 2-D Adaptive Y/C Separation Filters
2.1.4 Limitations of 2-D Adaptive Y/C Separation Filters
2.1.5 Recognizing 2-D Adaptive Y/C Separation Filters
2.2 3-D Motion Adaptive Y/C Separation Filters
2.2.1 Alternate Names
2.2.2 Description
2.2.3 Benefits of 3-D Motion Adaptive Y/C Separation Filters
2.2.4 Limitations of 3-D Motion Adaptive Y/C Separation Filters
2.2.5 Recognizing 3-D Motion Adaptive Y/C Separation Filters
2.3 Y/C Separation Filter Summary
**************************************************************************
2.1 2-D Adaptive Y/C Separation Filters
2.1.1 Alternate Names
Manufacturers almost never refer to these filters by their technical name.
They may call them Digital Comb Filters, Digital Dynamic Comb Filters,
3-Line Logical Filters, or almost anything which includes 'Digital Filter'.
An exception are 3-D Digital Filters, a much more sophisticated design
which is presented in section 2.2.
For all practical purposes these filters must be implemented after analog
to digital conversion. Hence, any name which includes 'CCD' (which is
an analog sampling process, not digital) will NOT be one of these filters.
2.1.2 Description
2-D Adaptive Filters are intended to solve some of the problems inherent
in conventional analog line comb filters. The term "Adaptive" indicates
that the filter changes its algorithm in response to the video image.
Digital logic is applied to "evaluate" the image and switch to the
"best" applicable filter algorithm.
All digital filters must first convert the composite video to a digital
signal using an A/D converter. The most common type of 2-D filters,
implement with digital processing, a basic comb filter mode similar
to an analog CCD comb filter. Adaptive logic is then used to eliminate
the horizontal edge 'hanging dots' problem of analog line comb filters.
The hanging dots are caused by comb filtering two successive lines with
different color values at the same horizontal positions along the lines.
The basic comb filter mode will fail to separate the Y/C signals correctly.
The color carrier crosses over into the luminance signal and causes the
hanging dot appearance. This is explained in detail in Appendix 2.
In the most common type of 2-D Adaptive Y/C Separation Filter, the adaptive
logic ensures that this will not occur during vertical transitions between
two colors. The logic examines three successive horizontal lines
simultaneously. At a vertical color transition, either the first or last
two lines will be the same color as the scan steps down the screen. First,
the top two lines will be the same color and the logic directs them to the
comb filter. When the scan moves down another line, the last two (of three)
lines will have the new color and be directed to the comb filter. Hence,
two lines with different colors will not be input to the comb filter at a
boundary. The Y/C signals are fully separated and the hanging dots are
eliminated.
Finally, in this implementation, whenever there is no vertical correlation or
if all three lines are correlated, the basic comb filter is selected, which
performs essentially the same algorithm as a CCD line comb filter.
2.1.3 Benefits of 2-D Adaptive Y/C Separation Filters
From the proceeding description, the 2-D Adaptive Y/C Separation Filter
works much like a CCD comb filter. However, on the horizontal borders
between vertical color transitions, cross-luminance dots are eliminated.
This works nearly perfect for the very sharp color transitions found on
the SMPTE color bar test pattern on frame 50815 of AVS. It doesn't work
quite as effectively on the gradual color transitions found in real world
video. Nevertheless, it is a rather significant and obvious improvement
when viewing normal video material.
The problem of dots on the vertical edges of the color bars is not
improved with the common implementation described above. The Mitsubishi
'601' and '701' series monitors contain an even more sophisticated type of
2-D Adaptive Y/C Separation Filter. Mitsubishi calls it a Digital
Dynamic Comb Filter. This filter incorporates three different filtering
algorithms with logic to pick the "best" filter depending on evaluation
of the image. One mode eliminates cross-luminance on vertical edges and
another mode fixes the horizontal edges. The third mode is used for the
uncorrelated portion of the image.
The Mitsubishi 2-D Adaptive Filter has been widely praised for its almost
perfect Y/C separation on the SMPTE color bar test pattern. However,
during normal video, I prefer the performance of the more common type 2-D
filter on SOME video material. Personally, this would create a difficult
choice if all other monitor performance parameters were equal. I would
suggest closely examining still frames 13710-13740 on AVS with both filter
types. Look for a cross hatched dot pattern on the lemons & limes and the
oranges in the background.
This is also a good scene for observing the differences between CCD comb
filters and the common 2-D Adaptive Filter designs. You should notice how
much worse the Y/C separation is with the CCD comb filters. Notice the
hanging dots on the horizontal edges of the display counter with the CCD
comb filter. Unfortunately, AVS has very few real world video scenes that
have fine details useful for comparison purposes.
Another significant benefit of all 2-D (and 3-D) filters is that the
processing is done digitally. This improves many limitations of the analog
technology used in glass and CCD line comb filters. Lower noise, better
gain and phase matching, improved bandwidth, better sum and difference
processing, etc. results from generating the line delays and processing
the signals digitally. All of these factors contribute to an improved
video image.
2.1.4 Limitations of 2-D Adaptive Y/C Separation Filters
These filters share the same limitations as the CCD line comb filters
except that the cross luminance problems of horizontal hanging dots
is fixed on vertical color transitions in all 2-D implementations.
In addition, the Mitsubishi enhanced 2-D Adaptive Y/C Separation Filter
also fixes the problem of cross-luminance dots on the vertical edges of
horizontal color transitions.
Other improvements of the basic comb filter limitations occur due to the
digital signal processing vs analog processing as discussed above.
2.1.5 Recognizing 2-D Adaptive Y/C Separation Filters
From the proceeding description, it should be obvious that frame 50815
from AVS can be used to examine the vertical color transitions at the
base of the color bars. If the filter hasn't already been proven to be
one of the previous types, and the hanging dots are absent, then the filter
is of the 2-D adaptive variety (or even better, a 3-D filter described
next). If the dots along the vertical edges of the color bars are also
absent, then the filter is of the Mitsubishi enhanced 2-D adaptive variety
or is a 3-D filter.
If the filter passes the 2-D or enhanced 2-D test, look at frame 50828
next. Look for the rainbow pattern around the concentric circles at the
middle of the test pattern. They are probably less prominent than the CCD
comb filter (due to the more accurate digital processing) but they should
still be present in the 2-D adaptive filters. If they are gone, the
product may have a 3-D filter.
The 2-D filters have other artifacts that are harder to describe and AVS
has few suitable frames for detecting them. For instance, the common
2-D adaptive filter (not the Mitsubishi) cannot properly separate Y/C
signals for a single horizontal line of a different color in a field. Go
back over the description of the adaptive filter process and this should
be clear.
Neither of the 2-D filter types can properly separate complex patterns with
fine details which lack sufficient vertical or horizontal correlation.
Examining different still patterns, with high frequency details at different
angles in the image, will reveal these limitations of the 2-D adaptive filter
process. Eliminating the remaining picture artifacts from still pictures
requires an even more sophisticated 3-D filter.
2.2 3-D Motion Adaptive Y/C Separation Filters
2.2.1 Alternate Names
There have been limited implementations of this advanced technology so few
names have surfaced. Look for names which include, 3-D, Motion Adaptive,
or Inter-Frame as part of Y/C Separation or Comb Filters.
2.2.2 Description
This type of Y/C separation filter is again an adaptive filter which
varies its algorithm depending on the content of the video. The big
difference from the 2-D filters, is that the 3-D filters use the same
horizontal line information from two different frames to provide the
inputs to the comb filter for Y/C separation. If no movement occurs
between frames, then the two lines represent exactly the same image and
the Y/C separation can be essentially complete. However, if movement has
occurred in the image between frames, then this inter-frame combing will
not work at all. Hence, the 3-D filter is adaptive, and chooses to use
inter-frame combing ONLY in the absence of motion.
3-D filters must compare differences between video frames to determine the
presence of motion. There are several different schemes for looking at the
differences in video between frames to decide if motion occurs in the image.
If the digital logic determines that there is NO motion, then the Y/C
separation is done by feeding the same horizontal lines in two successive
frames to the comb filter. See Appendix 2 for more technical details.
Since the same line is used from two successive fields, this frame comb
filter avoids the problems of the line comb filter since it only processes
correlated signals. Hence the 3-D filter can do a nearly perfect job of
separating the Y/C information in a motionless picture.
If motion exists in the picture, it means that the corresponding lines in
successive frames are uncorrelated (have different Y/C content). In this
case, using the inter-frame comb filter creates completely erroneous
information. Whenever the adaptive logic determines that motion exists
between frames, it switches over to processing the composite video using a
2-D intra-field (Adaptive Line) comb filter as described in section 2.1.
Hence, the 3-D Y/C Separation Filter performs just as well as the 2-D
Adaptive Filters when motion is present between frames, and performs with
almost perfect separation where ever the video is still between frames.
The 3-D Adaptive Separation Filters are considerably more expensive than
the 2-D variety because several entire fields of video have to be stored
instead of just two lines of video in the 2-D case. The video memory and
the more complex motion sensing logic, along with the associated digital
filters, quickly drive up the price of the 3-D filters. There have been
very few of these advanced filters implemented so far in consumer TV
monitors.
Improved Definition TV's (IDTV's) that included line doublers, were
likely to include some form of 3-D Y/C Separation Filter. The necessary
frame memories and motion sensing logic were required to properly implement
the line doubling process. Separate line doubler units, are likely to
include 3-D filters for the same reason. The discontinued Mitsubishi 35X7
($8000) was reported to have a 3-D filter but I have been unable to verify
this by observation.
I am not aware of any other CURRENT conventional consumer TV monitors that
have 3-D filters. I am just beginning to survey the available monitors for
this feature, so this may be incorrect. Toshiba has announced a 56" 16:9
projection system that reportedly will have a 3-D filter. Toshiba has
developed a version of a 3-D Adaptive Y/C Separation Filter that minimizes
the necessary video memory using techniques which are beyond the scope of
this article to explain. Therefore, this is a likely candidate for
inclusion in this new TV monitor.
In general, with the reduction in video memory costs and the widespread
application of new application specific IC's (ASIC's), I expect to see
many new high end TV monitors begin to appear with 3-D Adaptive Y/C
Separation Filters. This represents the current state-of-the-art in NTSC
Y/C separation.
2.2.3 Benefits of 3-D Motion Adaptive Y/C Separation Filters
From the description above, it should be clear that these filters are
capable of virtually perfect Y/C separation on stationary or still images.
Cross luminance and cross color should be completely absent in the still
images. In addition, the vertical resolution should also appear improved
over the 2-D and analog comb filters, since the luminance averaging between
lines in a field is completely eliminated.
2.2.4 Limitations of 3-D Motion Adaptive Y/C Separation Filters
The only theoretical limitation of these filters is the inability to
prevent cross color and cross luminance effects with moving images. In
this respect they perform the same as 2-D Adaptive Filters, to which they
default during moving images. In image areas of sufficient movement, the
Y and C signal spectra is smeared until their components can actually occupy
the same frequencies. If that happens, complete Y/C separation becomes
impossible. A solution to this problem requires pre-filtering the source
material in the Y and C signal domains, before creating the composite video
signal at the source. This interesting topic is beyond the scope of this
article.
2.2.5 Recognizing 3-D Motion Adaptive Y/C Separation Filters
All of the preceding frame references for AVS can be examined.
Theoretically, no indications of cross-luminance or cross-color should be
found in any of the still frames. In particular, the cross-color
(rainbow) effects at frame 50828 should be completely eliminated for the
first time.
===========================================================================
2.3 Y/C Separation Filter Summary
Technology What to Look For
Analog Low Pass Limited horizontal (and color) resolution.
and Notch Filters Cross-color effects when observing higher
frequencies in the multiburst pattern
(Frame 50816) from AVS.
Analog Line Comb Filters Problems listed above are corrected. Hanging
Glass or CCD dots appear on the horizontal borders between
vertical color transitions at the base of the
color bars, AVS Frame 50815. Dots along the
vertical edge of the color bars are also
present.
2-D Adaptive Y/C The "common" version of this technology will
Separation Filter correct the examples above except for the
dots on the vertical edges of the color bars.
Cross-color effects will still be seen on
closely spaced diagonal lines (Frame 50828).
Cross-luminance dots in the background of
Frames 13710-13740 will be improved over
the analog line comb filter types.
Enhanced Version This version of the 2-D Adaptive Y/C Filter
will also correct the dots on the vertical
edges of the color bars. However, note the
cross-luminance dots in the background of
Frames 13710-13740 compared to the "common"
2-D filter.
3-D Adaptive Motion No cross-luminance or cross-color effects
Y/C Separation Filter should be seen on any stationary pictures.
============================================================================
End of Part 2: Digital Y/C Separation Filters
----------------------Copyright 1994 Greg Rogers-------------------------
------------------All Rights Under Copyright Reserved--------------------
Permission is granted for reproduction of this article, in part, ONLY for
follow-up articles in rec.video. No other reproduction, in whole or in
part is permitted without the permission of Greg Rogers.
-------------------------------------------------------------------------
*************************************************************************
Greg Rogers
Tektronix
Gregory S Rogers
**************************************************************************
This article is so long I've split it into several sections. The main
article is split into Part 1 and Part 2. Part 2 will not make any sense
unless you also read Part 1. Part 3 and Part 4 are Appendices. They are
considerably more technical and aren't required to understand Parts 1 & 2.
Y/C Separation Filters
by Greg Rogers
Contents
Part 1: Analog Y/C Separation Filters
Part 2: Digital Y/C Separation Filters
Part 3: Appendix 1: The Composite Video Signal
Part 4: Appendix 2: Y/C Separation Filter Theory
**************************************************************************
April 5, 1994 Copyright 1994 by Greg Rogers
Y/C Separation Filters
or
"Kookie, Kookie, Lend Me Your Comb"
This article presented in tribute
to Edd_B...@77.Sunset.Strip.com
Appendix 1:The Composite Video Signal
**************************************************************************
Appendix 1:The Composite Video Signal
A.1 What's in the Composite Video Signal?
Start with the original Red, Green, and Blue signal components from the
video camera. Gamma Correction (a linearity correction involving cameras
and CRT displays) can be ignored in this discussion since it doesn't affect
the Y/C separation process. The color components are combined to form a
luminance (brightness) signal, Y, that maintains compatibility with the
original monochrome TV system.
Y = 0.30R + 0.59G + 0.11B
where R,G,B are the red, green and blue (Gamma corrected) video signals.
The coefficients chosen in the Y equation are related to the eye's
sensitivity to the RGB colors. In broadcast video, the Y signal has a full
video bandwidth of 4.2 MHz. Laserdiscs, some VCR's, and some Camcorders
have considerably larger bandwidths (5.5 Mhz or more) because they are not
forced to include the sound on a 4.5 MHz carrier. The Y signal has a
higher bandwidth than the I and Q signals discussed below. Therefore, it
provides the fine detail in the picture.
Since there are three unknowns (R,G,B) that must be recovered in the monitor
to drive the display, we need three equations involving those variables.
Two new signals are formed in addition to Y:
I = 0.74(R-Y) - 0.27(B-Y) = 0.60R - 0.28G - 0.32B
Q = 0.48(R-Y) + 0.41(B-Y) = 0.21R - 0.52G + 0.31B
Values rounded to two decimal places.
These two signals are then bandwidth limited. This means they can not
provide the same level of resolution detail as the Y signal. The I signal
is bandwidth limited to about 1.5 MHz and the Q signal to about 0.6 MHz.
The reason they are different is a function of the eyes acuity to the
particular colors represented by the signals. In the mid-resolution region
the color perceptiveness of the eye is best in the orange-red/blue-green
colors represented on the I axis. Therefore, the I signal gets the higher
bandwidth. Additional discussion of this topic is beyond the scope of the
Y/C separation issue, and will be saved for another article.
The I & Q signals are then applied to separate balanced modulators. The
modulators are fed carriers at 3.58 MHz that are 90 degrees out of phase.
This produces two double sideband suppressed-carrier AM signals. The signals
are then added together to form the resultant quadrature amplitude modulated
chrominance signal, C. The C signal is then low pass filtered at about
4.2 MHz to limit the upper sideband that would otherwise extend to about
5.1 MHz. Hence, the I signal is a vestigial sideband signal.
Finally, the Y and C signals along with various sync pulses are summed.
Since the color sub-carrier was suppressed, 8-9 cycles of the 3.58 MHz
sub-carrier is sent with each horizontal sync pulse. This allows the
carrier to be re-inserted with the proper frequency and phase when the
C signal is eventually demodulated and I & Q are recovered. The resultant
summed Y, C, color burst and sync pulses form the Composite Video signal.
A1.2 So what's the Problem? How does it affect Image Quality?
So knowing all this, what's the problem? The problem is that the C signal
takes up some of the same frequency space as the Y signal. The C signals
frequency range is from about 2.1 MHz to about 4.2 MHz, and the Y signals
frequency range is from 0 to 4.2 MHz. Somehow these signals must be
separated at the monitor so that the original R,G,B signals can be recovered
by solving the three equations. This decoding must start by separating the
color signal, C, and the luminance signal, Y, as completely as possible.
Any C signal remaining in Y will modulate the luminance and create
cross-luminance, a dot structure ("hanging dots"). This is caused by the
3.58 MHz color carrier getting into the luminance signal. Conversely,
any fine structure high frequency luminance information remaining in the C
signal will create color "beat patterns" on the edges of objects and in
fine patterns like the stripes on a shirt. This is called "cross color".
An early solution to the problem was to put a low pass filter on the
composite signal and filter out the color signal above about 2.5 MHz to
"recover" the Y signal. The reduced bandwidth of the Y signal drastically
limited the resolution in the picture. A band pass filter was used to
recover the C signal but it was still contaminated by the high frequency
luminance crosstalk and suffered cross color effects.
A1.3 Frequency Interleaving
Fortunately, we can take advantage of another property of the composite
video. The video signal is periodic in nature as a result of being
interrupted by the horizontal scanning and blanking process, and the
vertical blanking process. Therefore, it becomes sampled data. The
trick is to pick the horizontal and vertical scanning rates and the color
sub-carrier frequency in particular harmonic relationships. The color
sub-carrier frequency, Fsc, is not exactly 3.58 MHz, but is actually
3.579545 MHz. This corresponds to the 455th harmonic of the horizontal
scanning frequency, fH, divided by two.
fH = 15,734.26 Hz
Fsc = 455 x fH/2 = 3.579545 MHz
Just to complete all the critical relationships, since there are 525 lines
in a video frame and the frame consists of two interlaced fields, there
are 262.5 lines in a field. Therefore, the vertical field rate is:
fV = fH / 262.5 = 59.94 Hz.
Since there are two fields in a frame, the frame rate is fV/2 = 29.97 Hz.
Since the video signal is periodic in nature as described above, the
spectral distribution of the video frequencies are bunched together in
clusters. The Fourier analysis of a static video signal shows the
energy spectrum is concentrated in clusters separated by 15.734 KHz, the
horizontal scan rate. Each cluster has sidebands with 59.94 and 29.97 Hz
spacing. Hence, the luminance signal which has a bandwidth of 4.2 MHz does
not have a continuous distribution of energy across that band, but rather
exists as clusters of energy, each separated by 15.734 KHz. These clusters
are themselves not very wide, therefore most of the space between the
15.734 KHz harmonics is empty.
The chrominance signal is also periodic in nature, since it appears on
each horizontal scan and is interrupted by the blanking process. Therefore,
the chrominance signal will also cluster at 15.734 KHz intervals, spread
across its bandwidth from about 2.1 MHz to 4.2 MHz. So by picking the
color subcarrier at an odd harmonic (455) of fH/2, which we showed above
is 3.579545 MHz, the energy of the chroma signal clusters will be centered
exactly between the luminance signal clusters. Hence, the Y and C signals
can occupy the same frequency space by this process known as frequency
interleaving.
So at last we come to the purpose of the comb filter. The comb filter
has a frequency response that has nulls at periodic frequency intervals.
At the center frequency between each null, the comb filter passes full
amplitude signals. By tuning the comb filter to be periodic at the same
15.734 KHz intervals it can pass the Y signal while rejecting the C signal,
or visa versa. The relationship of the comb filter frequency response and
the interleaved Y/C signals is illustrated below.
Comb Filter Frequency Response
100%|... ............ ............ ............
| .
| . . . . . .
Y | . . . . . .
| . . . . . .
0 |______ . __________________ . __________________ . _______________
Frequency
100%|............. ............ ............ ...
| . . . . . .
| . . . . . .
C | . . . . . .
| . . . . . .
0 |________________ . __________________ . __________________ . _____
Frequency
| | | | | |
||| ||| ||| ||| ||| |||
||||| ||||| ||||| ||||| ||||| |||||
C Y C Y C Y
Fsc-Fh Fsc-Fh/2 Fsc Fsc+Fh/2 Fsc+Fh Fsc+3Fh/2
Fh*453/2 Fh*454/2 Fh*455/2 Fh*456/2 Fh*457/2 Fh*458/2
(The filters actually look more like the top halves of sinewaves, i.e. a
full wave rectified sinewave. This was as good as I could draw in ASCII.)
The performance of comb filters on different video signals can be analyzed
in the frequency domain using 3-D Fourier analysis techniques. However,
the frequency domain approach is far too complex for discussion here.
Instead, in appendix 2, I will introduce a simple but effective model of the
composite video signal that will be adequate to illustrate the operation and
limitations of the Y/C separation filters discussed in the main article.
End Appendix 1
----------------------Copyright 1994 Greg Rogers-------------------------
------------------All Rights Under Copyright Reserved--------------------
Permission is granted for reproduction of this article, in part, ONLY for
follow-up articles in rec.video. No other reproduction, in whole or in
part is permitted without the permission of Greg Rogers.
-------------------------------------------------------------------------
**************************************************************************
Greg Rogers
Tektronix
Gregory S Rogers
**************************************************************************
This article is so long I've split it into several sections. The main
article is split into Part 1 and Part 2. Part 2 will not make any sense
unless you also read Part 1. Part 3 and Part 4 are Appendices. They are
considerably more technical and aren't required to understand Parts 1 & 2.
Y/C Separation Filters
by Greg Rogers
Contents
Part 1: Analog Y/C Separation Filters
Part 2: Digital Y/C Separation Filters
Part 3: Appendix 1: The Composite Video Signal
Part 4: Appendix 2: Y/C Separation Filter Theory
**************************************************************************
April 5, 1994 Copyright 1994 by Greg Rogers
Y/C Separation Filters
or
"Kookie, Kookie, Lend Me Your Comb"
This article presented in tribute
to Edd_B...@77.Sunset.Strip.com
Appendix 2: Y/C Filter Separation Theory
**************************************************************************
Appendix 2: Y/C Filter Separation Theory
Table of Contents
A2.1 Y/C Separation
A2.1.1 The Y/C Separation Objective
A2.1.2 The General Line Comb Filter Model
A2.1.3 The Line Comb Filter - Frequency Domain Response
A2.1.4 A Simplified Model
A2.1.5 The Comb Filter Revisited - The Simplified Model
A2.2 Analog Y/C Separation Filters
A2.2.1 Example 1: Solid Color with Vertically Correlated
Luminance Pattern
A2.2.2 Example 2: Fine Horizontal Lines
A2.2.3 Example 3: Diagonal Lines
A2.2.4 Example 4: Hanging Dots
A2.3 Digital 2-D Adaptive Y/C Separation Filters
A2.3.1 Digital Y/C Separation Theory
A2.3.2 2-D Adaptive Y/C Separation Filter
A2.3.3 Example 4: Vertical Color Transition -
2-D Adaptive Filter
A2.3.4 Example 5: Single Line Color Transition
A2.4 Digital 3-D Motion Adaptive Y/C Separation Filters
A2.4.1 The Inter-Frame Y/C Separator
A2.4.2 Example 6 - Fine Vertical Color and Luminance
Transitions
A2.4.3 3-D Motion Adaptive Y/C Separation Filter Topologies
A2.4.4 Motion Detector
****************************************************************************
Appendix 2: Y/C Filter Separation Theory
A2.1 Y/C Separation
A2.1.1 The Y/C Separation Objective
The objective of Y/C Separation Filters is to separate the interleaved
luminance (Y) and chrominance (C) signals in the composite (Y+C) video
signal. As explained in Appendix 1, these signals are interleaved in
clusters, with the Y clusters centered around even multiples of Fh/2
(where Fh is the horizontal scan rate frequency, 15.734 KHz), and the
C clusters centered around the odd multiples of Fh/2. Fh is chosen so
that the color sub-carrier frequency, Fsc, is at the 455th multiple of
Fh/2, approximately 3.58 MHz. A SMALL segment of the frequency spectrum
around Fsc is indicated below. The actual bandwidths of the signal
spectra are discussed in Appendix 1.
| | | | | |
||| ||| ||| ||| ||| |||
||||| ||||| ||||| ||||| ||||| |||||
C Y C Y C Y
Fsc-Fh Fsc-Fh/2 Fsc Fsc+Fh/2 Fsc+Fh Fsc+3Fh/2
Fh*453/2 Fh*454/2 Fh*455/2 Fh*456/2 Fh*457/2 Fh*458/2
A2.1.2 The General Line Comb Filter Model
The general block diagram for a 1-H line comb filters is shown below.
-----------------> Diff ---> 1/2 --> C
| |
| |
Composite -------> 1-H Delay ----->
Signal (Y+C) | |
| |
-----------------> Sum ---> 1/2 --> Y
1-H is the time for one horizontal line to be scanned =
1/(horizontal scan frequency) = 1/Fh = 1/15.734 KHz = 63.6 uS.
The current line is summed together with the previous line to
generate the Y signal, or subtracted from the previous line to
generate the C signal.
The 1-H delay can be generated by an acoustical glass delay line, by an
analog sampling delay line (CCD), or by digitizing the waveform and
delaying it in digital memory. The fundamental comb filter mechanism
is the same in each case. For the glass delay line, the output is a
continuous waveform, and for the two sampling methods, the output is a
sampled waveform, usually at a sample rate of 4 * Fsc.
A2.1.3 The Line Comb Filter - Frequency Domain Response
The frequency response at the comb filter outputs is a result of the 1-H
time delay compared to the period (1/frequency) of the signal at the input.
When the input frequency's period is an exact multiple of the time delay
(the frequency is equivalent to an even multiple of Fh/2), the input signal
(non-delayed) appears at the Sum or Difference nodes exactly in phase with
the delayed signal. In this case, the output of the Sum node is twice the
magnitude of the original signal. The output of the Difference node is
zero. Therefore,
At f = n * Fh/2 = n * 15.734/2 KHz for n = 2,4,6,...
C output = 0
Y output = |(Y+C)| = Y
since the C signal is 0 at even multiples of Fh/2
Similarly, when the input frequency's period is an exact odd multiple of
1/2 the time delay, it will be 180 degrees out of phase with the delayed
previous line. The output of the Difference amp will be twice the original
signal and the output of the Sum amp will be zero.
At f = n * Fh/2 = n * 15.734/2 KHz for n = 1,3,5,7,...
Y output = 0
C output = |(Y+C)| = C
since the Y signal is 0 at odd multiples of Fh/2
At frequencies in between the zero output frequencies, the response of both
the Y & C outputs resembles the positive half of a sine wave. Overall the
response looks like the teeth of a comb, hence the name comb filter. The
response is plotted (as well as I can on an ASCII terminal) below.
Comb Filter Frequency Domain Response
1 |... ............ ............ ............
| . . . . . .
| . . . . . .
Y | . . . . . .
| . . . . . .
0 |______ . __________________ . __________________ . _______________
1 |............. ............ ............ ...
| . . . . . .
| . . . . . .
C | . . . . . .
| . . . . . .
0 |________________ . __________________ . __________________ . _____
| | | | | |
||| ||| ||| ||| ||| |||
||||| ||||| ||||| ||||| ||||| |||||
C Y C Y C Y
Fsc-Fh Fsc-Fh/2 Fsc Fsc+Fh/2 Fsc+Fh Fsc+3Fh/2
Fh*453/2 Fh*454/2 Fh*455/2 Fh*456/2 Fh*457/2 Fh*458/2
Other types of line comb filters that use two or more 1-H time delay's are
possible. They differ in the shape of the frequency response between the
same maxima and minima frequencies. For instance, a 2-H line comb filter
will have a frequency response that is shaped like a full sine wave (not
like a full-wave rectified sinewave, as the 1-H filter appears). These
other shapes have various trade-offs in the rejection of unwanted signals.
However, 2-H filters are very seldom used in practice, and their
characteristics won't be discussed further here.
Note: Although 2-H line comb filters (almost always created with analog
sampling CCD's) use three video lines, they are NOT the same thing as the
2-D Adaptive Y/C Separation Filters, which also use three video lines. The
2-H filters use all three lines at all times, with no adaptive intelligence.
The 2-D filters, which are much more sophisticated, and always digital, will
be discussed later.
A2.1.4 A Simplified Composite Video Analysis Model
It is possible (and very useful) to analyze the effects of different
Y/C separation filters on complex video signals in the frequency domain.
However, this becomes too complex mathematically to describe here.
Frequency domain analysis requires the use of Fourier Transforms in three
dimensions. Graphs must be plotted in 3-D spatiotemporal frequency space.
When Fourier Transforms are introduced into an analysis in even one
dimension, let alone three, it may mean "something wonderful is about to
happen" ["2010"] for an engineer or mathematician. But it probably means
something incomprehensible is about to be discussed to everyone else. So
an alternative, conceptually simple model will be used below, to illustrate
a few of the Y/C separation filter behaviors discussed in the main article.
However, be advised that because of its simplicity, the model is limited
in the effects it can describe. More complex issues still require Mr.
Fourier's work.
The composite signal is the summation of the Y and C signals. The C signal
changes phase by 180 degrees at the start of each successive scan line.
This is a consequence of the fact that the frequency of the color sub-
carrier is chosen to be an odd multiple of 1/2 the horizontal scan line
frequency (Fsc = 455 * Fh/2). At the end of each scan line, the color
sub-carrier has been through 227.5 periods and starts the next line 1/2
period out of phase with the previous line. Therefore, for every sample
point directly above or below the sample on the current line, the C signal
phase is different by 180 degrees. i.e. The C signal is inverted between
lines.
Since there are 525 lines (an odd number) in a (2-field) frame of video,
the C signals on corresponding lines in alternate frames, are also exactly
180 degrees out of phase.
The C signal has the same frequency as the color sub-carrier. Since the
sample rate is usually chosen as 4*Fsc, sample points are separated by 90
degrees on the C signal. Therefore, the phase of the C signal also changes
180 degrees at every other sample point on each horizontal video line.
The following chart illustrates the relationships above. If the video
image is stationary for several scan lines duration or for several frames,
then a simple but useful model of the composite video signal is:
Within a field At line n Composite Video
of frame m OR between frames Same horiz position
Line n-1 Line n, Frame m-1 Y - C
Line n Line n, Frame m Y + C
Line n+1 Line n, Frame m+1 Y - C
************************************************************************
Signal Notation
In the remainder of the article I will indicate the phase angle of the
chroma signal, C, by putting the phase angle in parenthesis.
i.e. a chroma signal of magnitude C, but phase angle 90 degrees is
indicated as C(90). C(180) = -C.
When two signals of equal magnitude but opposite phase are added, the
signals cancel to 0. Hence, C(0) + C(180) = 0. This is exactly the same
as C + -C = 0. Also C(90) + C(-90) = 0.
*****************************************************************************
Three lines of a small section (8 sample points horizontally) of composite
video are shown below. The example section is a solid color with constant
luminance.
L1 = line 1 Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90)
L2 = line 2 Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90)
L3 = line 3 Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90)
The alternating phase of the C signal can be seen both between lines and
between every other point on each horizontal line.
A2.1.5 The Comb Filter Revisited - The Simplified Model
It is simple to explain the operation of the basic comb filter using this
model. If the same video image (same chroma and luminance) is found at the
same horizontal points on two successive lines, then the composite signal
on the first line is Y+C, and the composite signal directly below it on
the next line, of the same field, is Y-C.
To extract the Y information for these two points, their composite sampled
values are added. To extract the C information, their sampled values are
subtracted.
To extract Y, Sample(line1) + Sample(line2) = (Y+C) + (Y-C) = 2*Y
To extract C, Sample(line1) - Sample(line2) = (Y+C) - (Y-C) = 2*C
To create an entire comb filtered Y/C line, this process is repeated,
stepping point by point across the width of the two input lines.
Next we look at several basic video images.
A2.2 Analog Y/C Separation Filters
A2.2.1 Example 1: Solid Color with Vertically Correlated Luminance Pattern
Consider horizontal scan lines which contain a single color and some
luminance pattern, Y = Y(h), which repeats on each line. Therefore, the
luminance value is the same at each position, h, on each line. The
luminance pattern exists across the horizontal width of the screen
and can create vertical lines or some other pattern which is uniform
vertically. It is said to be correlated in the vertical direction.
Next, comb filter three lines to separate the Y/C components of the lines.
The 1-H line comb filter performs the the sum and difference operations
from above, on each vertical pair of points, two lines at a time.
Looking at a horizontal section of 8 samples on the three lines:
L1 = line 1 Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90)
L2 = line 2 Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90)
L3 = line 3 Y+C Y+C(+90) Y-C Y+C(-90) Y+C Y+C(+90) Y-C Y+C(-90)
Y1 = (L1+L2)/2 = Y Y Y Y Y Y Y Y
Y2 = (L2+L3)/2 = Y Y Y Y Y Y Y Y
C1 = (L1-L2)/2 = C C(+90) -C C(-90) C C(+90) -C C(-90)
C2 = (L2-L3)/2 = -C C(-90) C C(+90) -C C(-90) C C(+90)
The Y and C signals have been totally separated by the comb filter even
though they were originally added together to create the composite signal.
Remember that the Y signal is actually Y(h) on EVERY line. Vertical lines
on the display are separated according to the Y(h) horizontal frequency
(this is called a spatial frequency). The Y signal can have a high
(temporal) frequency content which extends through and above the C signal
bandwidth. (The Y/C frequency clusters stay separated regardless of the
horizontal frequency of the lines. This is true even if the spatial
frequency is equal to the color sub-carrier frequency. This is where Mr.
Fourier comes into play, but it is easier to see why this works using the
simplified model.)
The comb filter effectively eliminates any crosstalk of the Y signal into
the C signal and prevents cross-color effects from occurring. Notice that
the resulting C signal maintains the necessary phase inversions between
lines. For this simple case the separation is perfect.
A2.2.2 Example 2: Fine Horizontal Lines
In the rest of the examples, I will omit every other point to simplify
the diagrams. This doesn't create any loss of generality.
Next consider the effect of the luminance signal changing from line
to line. If the luminance signal changes from 100% white level to black
level, this will represent the highest vertical frequency content, at
full amplitude, that is possible in a single field.
L1 = line 1 Ya+C Ya-C Ya+C
L2 = line 2 Yb-C Yb+C Yb-C
L3 = line 3 Ya+C Ya-C Ya+C
Y1 = (L1+L2)/2 = (Ya+Yb)/2 (Ya+Yb)/2 (Ya+Yb)/2
Y2 = (L2+L3)/2 = (Ya+Yb)/2 (Ya+Yb)/2 (Ya+Yb)/2
Notice that the comb filter has averaged the vertical information. This
creates an effective loss of vertical luminance resolution. Next, notice
what happens to the color information shown below.
C1 = (L1-L2)/2 = Yc+C Yc-C Yc+C
C2 = (L2-L3)/2 = -Yc-C -Yc+C -Yc-C
where Yc = (Ya-Yb)/2
In this example the color information could be contaminated by the
leakage of Yc into the color signal. However, Yc is a constant value on
each line, unlike C, which is a modulated 3.58 MHz carrier. In order to
eliminate this type of low frequency luminance crosstalk, the C signal can
be high pass filtered after the comb filter. Since the high pass filter
can be at a quite low frequency, no loss of chroma resolution results.
This example shows that finely spaced horizontal lines do not create
cross-color problems.
A2.2.3 Example 3: Diagonal Lines
The preceding two examples showed that cross-color problems are not
created from finely spaced lines as long as they are vertical or
horizontal lines. This example looks at what happens when they become
diagonal lines.
Consider diagonal lines which are at 45 degrees. The Y value on a given
line alternates between two luminance values, which may be full black and
100% white. But it is not necessary that the luminance values cover such
an extreme.
L1 = line 1 Ya+C Yb-C Ya+C
L2 = line 2 Yb-C Ya+C Yb-C
L3 = line 3 Ya+C Yb-C Ya+C
Y1 = (L1+L2)/2 = (Ya+Yb)/2 (Ya+Yb)/2 (Ya+Yb)/2
Y2 = (L2+L3)/2 = (Ya+Yb)/2 (Ya+Yb)/2 (Ya+Yb)/2
Again in this example the Y values are averaged between lines.
C1 = (L1-L2)/2 = Yc+C -Yc-C Yc+C
C2 = (L2-L3)/2 = -Yc-C Yc+C -Yc-C
where Yc = (Ya-Yb)/2
The C signal is now modulated by the the difference between the Y values.
Notice that the Yc component alternates phase between samples and between
lines, making it indistinguishable from a legitimate color signal
component. This creates the familiar cross-color effects associated with
finely spaced diagonal lines. When the lines are not exactly at a 45
degree relationship with the sample rate, the modulation depth varies
along the lines and the cross-color takes on its characteristic rainbow
appearance.
This shows that a line comb filter cannot eliminate this all too common
Y/C separation artifact. This shows up often in real world video,
usually in the fine stripes in a suit or shirt collar. As the object
moves, the angle of the lines change and the rainbow pattern varies with
the movement. But it also shows up in static images with fine details
that are not horizontal or vertical lines.
A2.2.4 Example 4: Hanging Dots
Next, lets look at the problem which occurs when there is a vertical
color change. This is the problem of hanging dots which occurs on the
horizontal boundary line between two color areas.
L1 = line 1 Ya+C Ya-C Ya+C
L2 = line 2 Ya-C Ya+C Ya-C
L3 = line 3 Yb+C' Yb-C' Yb+C'
L4 = line 4 Yb-C' Yb+C' Yb-C'
Y1 = (L1+L2)/2 = Ya Ya Ya
Y2 = (L2+L3)/2 = Y1+C1 Y1-C1 Y1+C1
Y3 = (L3+L4)/2 = Yb Yb Yb
where C1 = (C'-C)/2
Y1 = (Ya+Yb)/2
Notice that the luminance signal is modulated by the color sub-carrier
at a magnitude determined by the difference between the two colors.
Therefore, different colors will modulate the intensity of the luminance
differently. This modulation creates a "dot" pattern on the horizontal
transition line between the two colors. This can be clearly seen on AVS
using the color bar pattern at frame 50815. This is an example of
cross-luminance.
C1 = (L1-L2)/2 = C -C C
C2 = (L2-L3)/2 = Y2+C2 Y2-C2 Y2+C2
C3 = (L3-L4)/2 = C' -C' C'
where Y2 = (Ya-Yb)2
C2 = -(C+C')/2
Again in this example the Y2 term is filtered out after the comb filter
by a high pass filter. The C2 term is an average of the two colors which
causes a narrow color smear at the transition border.
This example will be re-visited below to see how the 2-D Adaptive Filter
fixes the problem of the hanging dots.
A2.3 Digital 2-D Adaptive Y/C Separation Filters
A2.3.1 Digital Y/C Separation Theory
These filters are implemented by first converting the composite video
signal to a digital signal by inputting it to an A/D converter. The A/D
converters are almost always 8 bit converters and digitize at 14.31818
Mega-samples per second. This is four times the color carrier frequency,
usually written as 4*fsc. (Remember that fsc is about 3.58 MHz). It is
necessary to sample at a rate at least twice the highest frequency
component of the video signal to avoid aliasing. Since the highest
frequency components of the composite signal in standard broadcast video
are limited to 4.2 MHz, and LD's may reach about 5.5 MHz or so, 4*fsc is
an adequate sample rate.
After Y/C separation, the Y and C signals must be re-converted to analog.
In LD players, which output both the separated Y/C and the composite
video, the composite video may also be a re-converted analog signal
instead of the original composite signal as stored on the LD. Why would
the composite signal output go through the A/D and D/A conversions steps?
Simply because the LD player can process the digital version of the
composite video for noise reduction using sophisticated DSP techniques.
This processing can be very effective as in the Pioneer CLD-97. In this
case, the original analog composite signal will be discarded and the
digitally processed composite signal will be used. Each of the digitized
signals are re-converted using D/A converters (again almost always 8 bit).
Note: There has been much made of the fact that processing the separate
Y/C signals and recombining them in the Pioneer CLD-95 caused an
undesirable delay between the Y/C components in the composite signal
output. It should be noted that this processing all occurred on analog
signals, after separation by an analog CCD comb filter. It has given the
whole concept of processing the separate Y/C components and recombining
them a bad name. The irony of this is that the CLD-95 did very little
processing (analog Y noise reduction) in this mode.
The concept of Y/C processing, as it has become known, was a flaw in analog
implementation of the CLD-95. Always use the CLD-95's Y/C outputs.
Incidentally, the CLD-95 had a unique very sophisticated two stage analog CCD
comb filter. It does not follow the model presented here at all. It was
an attempt to solve some of the analog CCD comb filter problems by analog
processing. Although a spirited attempt in pre-digital times it will not
be seen again.
Digital processing done in the CLD-97, does not create the same problem and
provides superior digital adaptive field noise reduction on the separate
Y/C signal components. The superb noise performance of the CLD-97
demonstrates the desirability of this technique. The signals are
separated, processed, and recombined entirely in the digital domain.
General Digital Y/C Separation Model
<--Analog ---> <----------- Digital ----------> <--- Analog --->
Composite ---> A/D ---> Y/C ---> Y ---> D/A ---> Y
Video | Sep ---> C ---> D/A ---> C
(Y + C) | & ---> Composite ---> D/A ---> Composite
| Optional Video Video
| Noise
| Reduction OR
|
|------------------------------------------------> Composite
Video
A2.3.2 2-D Adaptive Y/C Separation Filter
Before explaining the 2-D filter, it should be noted that it is possible
to implement a simple 1-H line comb filter in the digital domain, just
like it is done in the analog domain. In fact, it is easier and would be
more accurate in the digital domain. A 1-H line delay is created by
storing the previous digitized line in digital memory and then performing
the desired sum or difference between the lines using a digital adder or
subtractor.
Once the composite signal is digitized, any further gain, offset, noise,
or frequency response problems inherent in analog comb filters can be
avoided by using digital signal processing. However, I am not aware of
any products that simply emulate the 1-H or 2-H analog comb filters
digitally. This is because there are major advantages available by
implementing additional signal processing in the digital domain. This
is called adaptive processing and is described next.
The 2-D filter works very similar to a line comb filter except that it
solves one of their more annoying problems. When there is a vertical
transition in color across a horizontal boundary line in the video,
the line comb filter suffers from "hanging dots" as explained above.
The luminance signal at the horizontal boundary is modulated by the
color carrier. This contamination of the luminance signal creates a
pattern of alternating light and dark "dots" along the horizontal line.
A block diagram of a 2-D Adaptive Y/C Separation Filter is shown below.
2-D Adaptive Y/C Separation Filter
Composite ----- 1-H -------- 1-H -------> L1
Video | Delay | Delay
| |
| -----------------> L2
|
-----------------------------> L3
L1 ------ - Diff -------
+ |
| ------> 1-K --->
L2 ----------- Sum -- 1/2 --------> C
| ------> K ---> |
+ | | |
L3 ------ - Diff ------- | -
| L2 --> + Diff ---> Y
L1 -----> Vertical |
L2 -----> Correlation (K) -----
L3 -----> Detector
In this circuit, the Vertical Correlation Detector determines the
degree of vertical correlation, K. If L1/L2 are highly correlated,
then K=0 and L1/L2 are input to the comb filter. If L2/L3 are
highly correlated, then K=1 and L2/L3 are used. If no correlation
exists then K is set equal to 0 and L1/L2 are used by default.
Some 2-D filters calculate K to be a fraction based on the relative
correlation of the two line pairs. In that case, a proportional
combination of the line pairs is used for generating C and then Y.
In the examples below, only the simple case of K = 0 or 1 is used.
A2.3.3 Example 4: Vertical Color Transition - 2-D Adaptive Filter
The diagram below is repeated from the "hanging dots" example above.
Again, notice the resulting luminance line Y2 which is modulated by C1.
Analog Line Comb Filter
L1 = line 1 Ya+C Ya-C Ya+C
L2 = line 2 Ya-C Ya+C Ya-C
L3 = line 3 Yb+C' Yb-C' Yb+C'
L4 = line 4 Yb-C' Yb+C' Yb-C'
Y1 = (L1+L2)/2 = Ya Ya Ya
Y2 = (L2+L3)/2 = Y1+C1 Y1-C1 Y1+C1
Y3 = (L3+L4)/2 = Yb Yb Yb
where C1 = (C'-C)/2
Y1 = (Ya+Yb)/2
Now look at how the 2-D Adaptive Filter fixes this problem.
2-D Adaptive Filter
L1 = line 1 Ya+C Ya-C Ya+C
L2 = line 2 Ya-C Ya+C Ya-C
L3 = line 3 Yb+C' Yb-C' Yb+C'
L4 = line 4 Yb-C' Yb+C' Yb-C'
To generate Y1, the adaptive logic compares L1, L2, and L3 for correlation.
L1 and L2 are correlated and are input to the comb filter (remember that C
changes sign each line even though the color is the same).
Y1 = (L1+L2)/2 = Ya Ya Ya
To generate Y2, the adaptive logic compares L2, L3, and L4 for correlation.
L3 and L4 are correlated and are input to the comb filter.
Y2 = (L3+L4)/2 = Yb Yb Yb
Using the adaptive logic, the vertical transition between colors is made
without crosstalk of the color signal into Y.
A2.3.4 Example 5: Single Line Color Transition
Although the 2-D adaptive filter solved the line comb filter's problem of
horizontal hanging dots on vertical color transitions between two patches
of color, it is not a complete cure for all stationary images. It is still
possible for stationary images to contain fine detail and transitions that
can not be correctly separated by the 2-D adaptive filter. One very simple
example is given below just to illustrate the potential for misbehavior.
In this example, a single scan line has a different color in the middle of
another color. This may seem to be a very thin line, but its only one field
of the frame. The other field could have the same single line. Together the
two interlaced fields would form a frame with a horizontal band, two scan
lines in height, a more typical video image.
L1 = line 1 Ya+C Ya-C Ya+C
L2 = line 2 Ya-C Ya+C Ya-C
L3 = line 3 Yb+C' Yb-C' Yb+C'
L4 = line 4 Yb-C Yb+C Yb-C
L5 = line 5 Yb+C Yb-C Yb+C
Applying the adaptive logic, the comb filter would see the following
inputs in succession:
Y1 = (L1+L2)/2 = Ya Ya Ya
*Y2 = (L2+L3)/2 = Ya+Cx Ya-Cx Ya+Cx
Y3 = (L4+L5)/2 = Yb Yb Yb
Where Cx = (C'-C)/2
* = No correlation of successive lines found, comb
filter uses first two of three lines compared.
In this example the single line color change creates a line of hanging
dots, where there is no correlation between two successive video lines.
In other common video situations, the color changes gradually over a
vertical area of an object due to shading, etc. In this case the C
component may change line by line over several lines. The hanging dots
(cross luminance) will be generated over this entire area. Other more
complex stationary images will also cause cross-luminance and complementary
cross-color problems.
Also bear in mind, that the cross luminance (dots) problem is easy to
see on frozen images with a LD player, as alternating light and dark
dots. However, on stationary images, in normal play mode, the alternating
C phase of the frames and lines, tend to be averaged out by the eye. In
some cases the cross luminance effect just appears to add noise to the
picture especially at the edges of objects, and in other cases the dots
appear to crawl around on the edges of objects. Either way the effects
of cross luminance degrade the video image. But it may be easy to blame
the problems on noise or poor resolution by mistake.
A2.4 Digital 3-D Motion Adaptive Y/C Separation Filters
The above examples show that the 2-D Adaptive Y/C separation filters can
be more effective than line comb filters in some common video image
situations. However, in real video material (beyond test patterns and
simple horizontal and vertical line dominated images), the images may not
have sufficient correlated lines to permit the comb filter to perfectly
separate the Y/C signals. The more complex and finely detailed the image,
the more unwanted artifacts that will remain, even when using a 2-D filter.
A2.4.1 The Inter-Frame Y/C Separator
The 3-D filter can solve the limitations of the 2-D filter, in the absence
of motion, by comb filtering the same scan line from two successive frames.
This is technically known as Inter-Frame Y/C Separation. As explained
previously, the phase of the C signal shifts by 180 degrees between the
same scan lines in alternate FRAMES. This is analogous to the C signal
phase shift between successive lines within a single field. Therefore,
as long as there is no movement in the image between frames, the comb
filter inputs receive the same signal except for the C inversion. This
leads to near perfect Y/C separation for stationary pictures. However,
if there is any movement between frames, then the comb filter outputs
totally erroneous Y/C signals.
The Inter-Frame Y/C Separation is the same principle as the basic line
comb filter, except that the lines being processed are the same lines from
successive frames, instead of successive scan lines from the same field.
The delay element in the Inter-Frame Y/C Separator is a complete frame,
instead of a single line in the Line Comb Filter. This requires a digital
memory that holds 525 lines instead of a single line of video. Additional
frame memory is required for the motion detector, to be discussed shortly.
Therefore, the cost of a 3D Filter is much higher than 2D or Line Comb
Filters. The block diagram of a simple Inter-Frame Separation Filter is
shown below.
Simple Inter-Frame Y/C Separator
-----------------> Diff ---> 1/2 --> C
| - |
| |
Composite -------> Frame Delay --->
Signal (Y+C) | |
| |
-----------------> Sum ---> 1/2 --> Y
In this circuit, the Frame Delay stores two fields of
the composite video. The previous frame is subtracted
from the current frame, line by line, sample by sample,
to generate the C signal and summed to create a Y signal.
A2.4.2 Example 6 - Fine Vertical Color and Luminance Transitions
This is similar to example 5, but shows that even the general case of
color changes and luminance changes between each line can be separated
correctly by the 3-D filter. Shown below are 3 lines from a pattern
that remains stationary between two frames.
Frame 1 (F1)
L1 = line 1 Ya+Ca Ya-Ca Ya+Ca
L2 = line 2 Yb-Cb Yb+Cb Yb-Cb
L3 = line 3 Yc+Cc Yc-Cc Yc+Cc
Frame 2 (F2)
L1 = line 1 Ya-Ca Ya+Ca Ya-Ca
L2 = line 2 Yb+Cb Yb-Cb Yb+Cb
L3 = line 3 Yc-Cc Yc+Cc Yc-Cc
Y1 = L1(F1)+L1(F2)/2 Ya Ya Ya
Y2 = L2(F1)+L2(F2)/2 Yb Yb Yb
Y3 = L3(F1)+L3(F2)/2 Yc Yc Yc
C1 = L1(F2)-L1(F1)/2 -Ca Ca -Ca
C2 = L2(F2)-L2(F1)/2 Cb -Cb Cb
C3 = L3(F2)-L3(F1)/2 -Cc Cc -Cc
In this rather general example it can be seen that the Inter-Frame Y/C
Separator completely separates the Y and C signals even though they are
changing each line. However, remember this only works in the absence of
motion. Otherwise, frame 2 will have a different image than frame 1,
and the result would be totally incorrect. It is much better in that
case to revert to a 2-D filter, which may be able to find areas of the
picture which have vertically correlated lines within a single field.
This is known as the Intra-Field Y/C Separator in the 3-D Adaptive Filter.
Several other topologies are possible for the basic Inter-Frame Y/C
Separation Filter but won't be shown here. Some versions use two Frame
Delays (Memories) in a method similar to a 2-H Line comb filter.
A2.4.3 3-D Motion Adaptive Y/C Separation Filter Topologies
An overall block diagram of one type of 3D Motion Adaptive Y/C Separation
Filter is shown below:
3-D Motion Adaptive Y/C Separation Filter Type 1
----> Inter-Frame ---> C ---
| Y/C Separator |
| No \
| \
Composite ------> Motion Detector ------> OR ------------> C
Video | / |
| Yes / |
| | -
|---> Intra-Field ---> C --- Diff ----> Y
| Y/C Separator +
| (2-D Adaptive Filter) |
| |
--------------------------------------
In this topology, the Y/C separators only generate
the C signal outputs. The Y signal is created by
subtracting the selected C signal from the composite (Y+C).
The motion detector selects the desired C signal.
The Intra-Field Y/C Separator is usually a 2-D Adaptive
Y/C Separator as described earlier.
Very closely related, but with a few more degrees of design freedom for
tailoring the filter characteristics, is the filter shown below:
3-D Motion Adaptive Y/C Separation Filter Type 2
----> Inter-Frame ---> C ---------------
| Y/C Separator ---> Y ---- |
| | No |
| | Motion |
| \ \
| \ \
Composite ------> Motion ---------------> OR --> Y OR --> C
Video | Detector / /
| / Yes /
| | Motion |
| | |
----> Intra-Field ---> Y ---- |
Y/C Separator ---> C ---------------
(2-D Adaptive Filter)
In this topology, the motion detector selects between
the Inter-Frame for stationary images, and the Intra-Field
for moving images, and uses both the Y and C signals
from the Y/C Separation Filter selected.
Another alternative is to use the motion detector to create a signal, K,
which is proportional to the amount of motion sensed. The block diagram
below explains this case.
3-D Motion Adaptive Y/C Separation Filter Type 3
----> Inter-Frame ---> C ---------------
| Y/C Separator ---> Y ---- |
| | |
| K K
| | |
|
Composite ------> Motion --> K Sum --> Y Sum --> C
Video | Detector
| | |
| 1-K 1-K
| | |
----> Intra-Field ---> Y ---- |
Y/C Separator ---> C ---------------
(2-D Adaptive Filter)
In this topology, the motion detector determines the
degree of motion, K, and then the Y and C signals are
generated by using proportional amounts of the Inter-
Frame and Intra-Field Y/C Separator outputs.
If the motion detector makes only a binary, yes/no
judgements, then K=1 or 0 only. The Sum nodes are
then simple switches. In this case, the Type 3
Filter is equivalent to the Type 2 Filter.
A2.4.4 Motion Detector
It is very important to correctly detect motion in the picture, so that
the Inter-Frame Y/C Separator is only applied for stationary images. In
fact, if the motion detector makes an error in judging motion, it should
only be to use the Intra-Field Y/C Separator unnecessarily on stationary
images, but never the other way around. The motion detectors are
designed with this bias in mind.
This is a very complex part of the 3-D Filter, more complex than the Y/C
separators themselves. The basic concept is to compare frame to frame
changes in the lower frequencies of the Y signal, which can be obtained
by simply low pass filtering the composite signal. And also to compare
changes in the C signal (and higher Y frequencies) after a two frame delay.
It is necessary to use a two frame delay for the C signal since the phase
of the C signal is inverted on alternate frames. The higher Y frequencies
can be subtracted out by using the output of the 2-D filter.
A very simple block diagram of a motion detector is shown below:
Motion Detector
Composite --- LPF ----- Frame -- Diff --------------------
Video | (Y lf) | Delay | |
Digitized | | | Motion -->
| -------------- Logic
| |
-- BPF --- Diff --- Frame --- Frame --- Diff --
(C+Y hf) | | Delay Delay |
| | |
Y hf -------------------------
(from 2-D filter)
Motion detectors vary a great deal in different implementations both
to improve performance and to reduce cost. The large amount of memory
required in the frame delays increase the cost of the 3-D filters.
Topologies that sub-sample the reduced bandwidth signals above, have
been proposed that would reduce the total memory to a single conventional
frame memory size (about 2 Mbits).
End of Appendix 2