TV Comb Filters
With the transition to digital video, comb filters are no longer an important part of video processing and display.
Looking back at television and history, it is amazing that such complicated technology was needed and used to transmit color TV pictures over the air.
A comb filter is needed for the TV to show fine picture detail from analog broadcasts, laserdisk, and other composite sources. It also reduces discolorations in fine picture detail and provides purer color overall. Good comb filters reduce discoloration along vertical edges but unfortunately some comb filters add discoloration along horizontal edges.
Last update 2/18/15
Comb Filter Subtopics
Other Video Topics
In a Nutshell:
The analog video signal prepared for broadcast contains two major parts commingled, the luminance (makes a black and white picture in full detail) and chrominance (coloration with not quite all the detail). This method is used instead of red, green, and blue sub-signals in order to get the best looking picture that can be transmitted in the limited bandwidth of the broadcast channel. Every analog TV receiver and analog video recording device must contain a filter to separate the luminance and color (Y and C) again. Less than perfect Y/C separators lose resolution -- horizontal, vertical, or both. Also there are artifacts such as rainbow swirls where thin stripes should be, and crawling dots where patches of different colors meet. The perfect Y/C separator never did not exist, although some 3D comb filters come close.
In a video picture, some parts (notably where fine detail is present) need more treatment than others. The more elaborate the comb filters, the greater the surface area of the screen will be improved and to a better degree for typical video subject matter..
We will use the term "Y/C separator" from time to time to refer to both comb filters and their cheaper notch and bandpass filter substitutes.
Comb filters do not improve reception of weak broadcasts from far away stations.
Comb filters do not improve (and also do not worsen) feathered edges from less than perfect progressive scan conversion (de-interlacing).
The comb filter takes analog composite video and outputs S-video.
The TV set's Y/C separator is bypassed (not needed or used) when you feed in an S-video, analog component video, or digital video signal. The advent of DVD and digital TV is making comb filters less important but they are still needed for the best quality reproduction from laserdisks, analog broadcasts, and other composite source material (e.g. using yellow RCA video in jacks)..
Kinds of Filters
How The Filters Work
List of Subtopics
Here is what the advertisers told you:
o The comb filter improves the horizontal resolution out to 400, 500, or even more lines.
o Without a comb filter you get at most 260 lines of horizontal resolution.
o The comb filter makes the colors purer.
Here is what they didn't tell you:
o The comb filter may lose vertical resolution.
o The comb filter may cause thin halos with crawling dots to appear above and below objects of contrasting colors.
o Some comb filters are a lot better than other, and also more expensive.
o A comb filter can be out of adjustment producing results worse than having no comb filter.
This is how it works:
o The comb filter takes scan lines two at a time (or three at a time) and commingles them and through the miracle of phase cancellation the fine horizontal detail or purer color information is extracted.
o The analog video signal, intentionally constructed with color information phase reversed on every other line, helps the comb filter achieve its goal.
o What to look for when shopping
o Ordinary comb filters
Two line analog or digital
Three line analog
o 3 Line Adaptive filters (2D; dynamic)
o 3D (Motion Adaptive) filters
o Adaptation math
o Comb filters you buy separately
o What is it made of?
o Where the name comes from
o Video disks and video tapes
o Comb filter terminology
o Video glossary
o Related web sites
In an analog TV or video signal, the information to make just a black and white picture (luminance, or brightness, data) and the information needed to turn that picture into a color picture (chrominance data) are generated separately by the camera circuits, and are handled separately in the TV receiver or monitor to produce the picture.
However for analog broadcasting or recording on laser disks, the color information and the luminance information have to share in the channel or bandwidth space allotted or available and are combined to be transmittable on one wire into what is called the composite video signal. In NTSC (a U.S. standard) video the color information (transposed onto higher frequencies) is superimposed on the luminance information representing fine horizontal detail (also using higher frequencies), while the luminance information representing broad objects (lower frequencies) is not so encumbered.
Every TV set must separate the luminance and color information in order to display the picture, and contains all of the necessary circuits to do so. All of today's digital TV sets (with tuner and channel selector) have these circuits also.
To reduce interference with the luminance information and to fit everything in the bandwidth allotment, fine horizontal detail color information is omitted. The human eye and brain hardly notice this deficiency at normal viewing distances. Also medium detail color information (40 to 110 line) is omitted for all except a few colors (oranges and blues), again taking advantage of the fact the human eye and brain are yet less sensitive to inaccurate fine detailing in the other colors.
While it is theoretically possible to do a very good job of separating the color and luminance information for TV reception, it is very difficult and expensive to this day. Today's comb filters are the best practical solution so far for this task. They are found in most TV sets. In the process of removing the color information so that the fine horizontal detail can be utilized, usually some vertical resolution is lost. Comb filters themselves today come in different levels of sophistication, effectiveness, quality, and cost.
The input to a comb filter is composite video after demodulation from a TV channel RF carrier. The output of a comb filter is S-Video.
Digital broadcasting (e.g. ATSC) also transmits the video signal as separate luminance information and color information. The information is formatted in a manner that simple inexpensive processing can separate the luminance and color information perfectly. Here there is no need for a component component called a comb filter.
Evaluating TV Set Comb Filters
Not counting cases of extreme skimping on quality, there are four categories of comb filter implementation:
1. No comb filter
(then a small price jump, small improvement with a ...)
2. Two and three line ordinary filter
(then a medium price jump, big improvement with a ...)
3. Three line adaptive (a.k.a. 2D; dynamic) filter
(big price jump, small improvement to a ...)
4. Motion adaptive (3D) filter
Comb filters process every scan line in turn.
The two line comb filter uses the previous scan line content to assist in processing the current scan line.
Three line comb filters use the previous scan line content and the following scan line content to assist in processing the current scan line.
Good 3D comb filters use the previous scan line, the following scan line, the corresponding scan line in the previous frame (525 lines back) and perhaps the corresponding scan line in the next frame (525 lines later) when processing the current scan line. These filters are more expensive because all the intervening scan line content has to be stored until its turn to be used.
Buying a TV Set
Today (2014) composite video and analog broadcasts are such a small part of video sources that hardly anyone shops for a TV set for the quality of its comb filter. Also, given the combination other features one may want, it would be difficult to find a model that also had a superb comb filter.
In years past, this writer's opinion has been that the TV set purchaser should go for at least the three line adaptive comb filter. (The exception is if the tuner and composite input will never be used, all program material including broadcasts comb filtered elsewhere coming in through the HDMI jacks, component video jacks and perhaps an S-video jack). Examining the picture is important. Within a category, comb filter quality varies.
DVD players connected via composite jacks, and cable TV analog channels can be used for picture evaluation. VCR material is not suitable or more difficult to use for evaluation due to limited resolution and likely prior comb filtering and re-compositing of the video signal.
The most obvious unwanted artifacts left behind by less than perfect comb filtering are dot crawl and rainbow swirls. It is hard to look at just one TV set and decide how much dot crawl is too much but it is easy to compare two TV sets and decide which is better than the other. Compare both moving and still subject matter.
Dot crawl may occur where contrasting colors meet. It is caused by small quantities of color information left behind in the luminance signal as a result of imperfect comb filtering. If the dots are not moving they are sometimes referred to as hanging dots. Cartoons with black outlines around color patches are not good subject material. Try to get a TV with no horizontal dot crawl and almost unnoticeable to no vertical dot crawl. Very few TV comb fitlers eliminate all the diagonal dot crawl but the less the better.
Rainbow swirls swamp out closely spaced vertical or steep diagonal lines such as in a pinstriped shirt or a far away picket fence. They are caused by luminance information left behind in the color signal and getting into the color circuits. No comb filters can get rid of rainbow swirls on moving objects completely without introducing faded color artifacts elsewhere.
Beware of a still life picture that looks excellent but when something starts moving it breaks out into a moderate to severe case of dot crawl or rainbow swirls worse than another TV set has.
Beware of greatly reduced rainbow swirls where diagonal pinstripes or even single diagonal lines of certain thicknesses are blurred out also.
A detailed checklist of how to evaluate a comb filter, including the use of the "A Video Standard" or "Video Essentials" laser disks may be had by clicking here (text subject to discontinuance) and scrolling to the bottom of that page.
Remember, no comb filter available today is perfect. You must use your own judgment as to picture quality versus price.
If you have a laser disk or DVD player, select a disk and make a note of scenes and frames you want to look at when going to the store to evaluate a new TV set. Test subjects and test data should include fine vertical detail, which means some objects that occupy just one scan line, or perhaps two. These objects should be of varying horizontal fineness as well. On any one scan line, fine horizontal details are dots, broad horizontal details are dashes. Text in small print makes a good test subject. If you bring a DVD, have the salesman set 16:9 mode on the player and also connect the player via the composite (yellow RCA) jacks so the TV set's comb filter is utilized. View the credits at the end of a movie where the letters are small and thin. The horizontal bars of capital T and E should be absolutely sharp. Use freeze frame and single step and examine the letters at the corners and sides of the screen as well as in the middle. For a "standard comb filter" it is OK for the dot of the lowercase I to be duller and elongated slightly upwards on the comb filter set even though it fits sharp and clear on one or two lines (but maybe wider sideways) on an older cheaper set you might have back at home. A standard VHS VCR does not have enough horizontal resolution to conduct this test. To evaluate dot crawl (sometimes the dots stand still) find scenes where there are large color patches juxtaposed. Use both still frame and normal playback. If you (still) have a laserdisk player, still frame plays the same frame in the same phase over and over contrasted with the normal reversed phase of every other frame; a 3D comb filter's third dimension is not usable for and should not come on during LD still frame.
Comb filters are sometimes sold separately, as "composite to S-video converters". Examples are the Faroudja Laboratories VP100 Video Enhancer and the Crystal Vision VPS-1 Video Processing System which are digital 2D adaptive models that are very good. Unfortunately they are no longer being manufactured although they are occasionally sold secondhand for about $200.on eBay as of this writing (2014). A few audio-visual receivers and switchers contain comb filters to allow the composite input jacks to cross feed into the S-video output jack.
If the "kind" of comb filter is not specified, you will have to assume it is the simplest, two line, variety with dot crawl and all the artifacts associated with that kind of filter.
Top of Page
How The Filters Work
Bandwidth, Frequencies, and Numbers
Audible sounds are in the range 20 to 20,000 Hz. If only a flute is playing, most of the sound is from 300 to 500 Hz. If a tuba is playing, its sound is mostly from 50 to 200 Hz. But the entire band of 20 to 20,000 Hz must be available in order to have faithful reproduction.
The video signal consists of a band of frequencies, where broad luminance detail is represented by low frequencies, fine luminance detail is represented by high frequencies, and the inverse is true for color information in the NTSC format.
The NTSC broadcast channel spans 6 MHz. For broadcast, "zero" is not really zero Hz. For example channel 5 has the actual frequency band from 72 MHz to 78 MHz with the "carrier frequency" at about 73.25 MHz; zero on our scale would stand for the carrier frequency. So the scale below goes from minus 1.25 to plus 4.75 rather than from 72 to 78 or from zero to six. Also, for NTSC broadcasts, additional redundant information appears below the carrier frequency as the "lower sideband" and space (bandwidth) must be reserved for it.
Luminance data may be found all up and down the scale. The finer the horizontal detail, the further away from zero the information needed to reproduce it is found. Each megahertz away from zero represents approximately 80 (79.5) more lines of horizontal resolution. If for example the picture showed nothing but 80 lines of resolution, there being just 40 black and 40 white vertical lines alternating and just filling a(largest perfect circle fitting in the) 4:3 screen, the information needed would all be at the 1 MHz position on the scale with very little above and below. For a typical real life picture, most of the information is in the zero to 2 range and very little is out in the 3 to 5 range. For a broadcast channel, everything is cut off at minus 1.25 and plus 4.2 MHz to avoid running into the sound band and the next channel, so resolution is restricted to about 330 lines.
Color data may occupy the range 2.1 to 4.2 MHz with the carrier at about 3.58 MHz. The finer the horizontal detail, the further from 3.58 MHz the information representing it is found. Again, since the preponderance of horizontal detail in a real life picture is coarse, most of the information is close to 3.58 MHz. Only orange and blue information is allowed to occupy the range 3.0 down to 2.1 providing medium detail color resolution. Due to cost cutting skimping, some program material does not preserve any medium color detail in which case no color information will be found in the 2.1 to 3.0 MHz range. Coarse detail for all colors is represented in the range 3.0 to 4.2 which, taken alone, allows about 48 lines of horizontal resolution since we measure from the carrier to one side only and then compute 80 lines per megahertz. Some processing methods including all home VCR's allow only carrier plus or minus at most 0.5 MHz for color data for a theoretical limit of 40 lines and a practical limit somewhat less. (The carrier is a number different from 3.58 MHz for VCR recording.)
For non-aerial transmission of video, the 4.2 MHz upper limit does not apply and luminance information goes up to 5.3 MHz for laser disks and 7.0 MHz for DVD. This writer is not sure whether color information is allowed to extend above 4.2 MHz to increase the color resolution. Even if color information did extend above 4.2 MHz, not all TV comb filters reach that high to capture it.
To summarize, the color resolution for NTSC composite video is the same as luminance resolution in the vertical direction but under 50 lines for most colors in the horizontal direction.
Top of Page
Ordinary Comb Filter.
TV Set "Without" a Comb Filter
If a composite video signal (with color) is fed into the TV monitor or receiver luminance circuits with no filtering at all, the color information is interpreted as spurious fine details and the result is a silk screened grainy appearance over the entire picture.
So some kind of filter is always needed and present. The cheapest solution is to use simple filters (notch, low pass, bandpass filters) that pass only the coarse and medium horizontal detail (lower 3 MHz or so) to the luminance circuits and pass the bulk of the color information still commingled with the fine luminance detail (3 to 4.2 MHz) to the color circuits. For most of the picture content there is not much luminance information up in the 3 to 4.2 MHz band to cause problems. But where there are fine details such as striped clothing, the high frequency luminance contamination going into the color circuits causes noticeable rainbow swirls to appear in the picture at that location. Medium detail color information cannot be used because there is too much luminance contamination.
This filtering limits horizontal resolution to about 260 lines. It may be noted that even in the early days of TV, most TV receivers had only about that much resolution due to limited frequency response in the circuitry in turn due to manufacturing cost cutting. This allowed the designers of the NTSC color video signal to get away with commingling the color information with the finest detailed luminance information.
A sharp horizontal color change, or not so coarse color detail, results in more color information well below 3.58 MHz (the color carrier) and some of it remains in the medium detail portion of the signal going to the luminance circuits. It produces an irregularity wherever a scan line changes color abruptly. Considering a vertical or diagonal boundary between two solid color patches, all of these irregularities have the collective appearance of a silk screened effect at just that part of the picture, with dots that can crawl along the color boundary as the picture is repainted 30 frames per second.
"Ordinary" Two Line Comb Filter
What the simplest comb filter actually does is mix each scan line with its predecessor (in terms of arrival time; temporally adjacent) and then take the average to draw on the screen. It so happens that wherever the two lines happen to be alike in visual content (which is most of the time) and the lines are mixed together in one way (added), the "filtering" is theoretically perfect. The color information disappears and the fine details of luminance are now alone and easily put into the picture. Where the two lines are the same and are mixed together in a different way (one subtracted from the other), the luminance information disappears and purer more accurate color is had.
The improvements over "no" comb filter are: revealing of finer horizontal detail overall, and some reduction of rainbow swirls. Improvement of fine detail is most prominent where details consist of upright dark and light lines.
Why the color information disappears when identical parts of two consecutive scan lines are added together is because every other line as transmitted has the color information intentionally phase-reversed. The color information then "cancels out" leaving (theoretically) clean luminance data. "Subtracting" a line from its predecessor gives clean color data by cancelling out the luminance data. Within a solid color area, fine light to dark detail will now show up. Since in a real life picture, any two consecutively transmitted scan lines are usually the same in content over most of their lengths, the adding or subtracting concept works well over most of the picture area. Of course, on those parts along the pair of lines that don't match, adding and subtracting do not cancel anything and instead produce weird results, "artifacts" as video experts call them. The comb filter "does not know" whether the first line or the second line is "more correct" so all that can be done is to use the result which is the evenly weighted mathematical average of the two lines.
The NTSC composite video signal was purposely designed with the color information phase reversed on every other line. Whether this format was invented back ca. 1953 with comb filters in mind (not readily and economically available at the time) or whether it was only because the picture even without a comb filter looked better is a subject still debated by video historians.
This simple comb filter does add discoloration that having "no" comb filter does not suffer from. At a horizontal color boundary (adjacent scan lines with different color content) the result of the filtering is a third color (and usually has crawling dots within it). The discoloration is two scan lines thick because the problem occurs twice per frame, first during the odd field and again during the even field. Fortunately at normal viewing distances and for most color combinations it is not too obtrusive (the human eye is less sensitive to errors in fine color detail).
Top of Page
Old Style Three Line Filter
Three Line 2D Filter
Theoretical Example of Behavior (two line filter)
For the effect on color, consider a simple picture, with a red square on top of a green square. In the middle of either square, consecutive horizontal scan lines are the same and the commingling and cancellation work quite fine. Where the squares meet, we pick up a pair of scan lines that differ profoundly, red versus green. What comes out in the picture is a yellow line. This is the two differing scan lines mixed together yielding a third result related in math but not to the picture. More correctly we see a yellow boundary two lines thick because the commingling of a red line and a green line happens twice per frame, first during the odd interlaced field and again during the even interlaced field. The yellow lines have a fine undulating dark-light pattern due to imperfect (in practice) removal of the color information from the luminance information. This is dot crawl. The undulations usually shift during the 30 frames per second painting of the picture. (With "no" comb filter there is a perfectly sharp red to green horizontal boundary, thus the yellow boundary is an artifact of comb filtering.)
(Lines as shown are consecutively transmitted, that is all odd or all even)
Results top row: Result of commingling the first two original lines.
Results bottom row: Result of commingling the last two original lines.
Lower right view exaggerates what you would see on the screen. For best results examine it while not wearing eyeglasses.
A -- Solid color.
B -- One color with fine detail in the horizontal direction only.
C -- Lines differ at this point.
D -- Single isolated dot, black on white.
E -- Horizontal color boundary together with fine detail in the horizontal direction only.
Rainbow swirls may be seen in parts C and D, their exact appearance may vary
The comb filter leaves behind much less luminance contamination in the color information and vice versa than the notch and bandpass filters used in TV sets without comb filters. Much of the improvement afforded by the two line comb filter is the purification of color and reduction of rainbow swirls in areas with fine upright lines, as opposed to the overall improvement in resolution. The discoloration of horizontal color boundaries greatly offsets the gain in horizontal resolution.
Top of Page
Three Line 2D (Adaptive) Filter.
Three Line Comb Filters (Non-digital, non-adaptive)
These were found in more expensive TV sets from the mid 1980's to the mid 1990's but were not too common.
Improvement over the two line filter consists of sharper transition from one color to another at sharp horizontal color boundaries, and less dot crawl.
The same commingling idea is used but here we might take the first scan line at half strength, the second temporally adjacent line at full strength, and the third line at half strength. Where all three lines are the same we still get excellent filtering, since the two half strength lines are phased one way and the full strength line between them is phased the other way.
Let's go back to the red square atop the green square. With our three line comb filter we have perfect red mixtures until we get down to the boundary between the squares. When we first encounter a green line we take it at half strength. The two lines before are a red line at full strength and the preceding red line at half strength. In this 75% red 25% green mixture the red overwhelms the green more or less. Then one line further down we pick up the next green line at half strength, use the previous green line at full strength, and the line behind is red line and taken at half strength. This time the green at 75% overwhelms the red at 25%.
Results top row: Result of commingling the first three original lines.
Results bottom row: Result of commingling the last three original lines.
A -- Solid color.
B -- One color with fine detail in the horizontal direction only.
C -- All three lines differ at this point.
D -- Single isolated dot, black on white.
E -- Horizontal color boundary together with fine detail in the horizontal direction only.
Again this process happens twice, first in the odd interlaced field and then in the even interlaced field. So the total number of finished scan lines derived from less than 100% red or less than 100% green in this example is four using the 3 line filter versus two using the 2 line filter. As you move a magnifying glass down over a horizontal color boundary you can see the battle zone begin sooner with some dot crawl between the two colored squares. But the switch to the next color should be more abrupt to better match the sharp transition in the original source. Some video experts claim that the improvement that this analog 3 line comb filter offers over the two line comb filter is minimal considering the added cost.
If neither the line above nor the line below matches at a given spot along their lengths, we again get bizarre results (artifacts) at that spot. Fortunately, in a typical picture, the number of places about the screen where this happens is much much smaller than places about the screen where lines taken two at a time are different.
Let's consider part of a picture containing a thin dash and the tiny dot. The first time the line with the dot goes through a three line comb filter, it is at half strength with a full strength empty line and a half strength empty line before it. We get a very faint ghost of a dot in the result. For the next threesome we grab the line below the dot at half strength, take the line containing the dot at full strength and take one more line behind at half strength. The mixed result shows the dot clearly: 50% of the mix has the dot and two 25% portions don't. Going further on, the next line is an empty line taken at half strength, the preceding empty line is at full strength, and then the line with the dot is at half strength. The result has the dot very faint. So the final picture has the dot reasonably clear and small and although there are now two ghosts instead of one, the ghosts are very faint. Again, because the low frequency (coarse horizontal details) portion of the signal should not be put through the comb filter, a wide object such as a horizontal dash should not be ghosted at all.
We are not sure whether there existed 3 line analog comb filters that had fast acting accompanying circuits to change the commingling mixture during the drawing of a scan line for further improvement (adaptiveness; see next section). The result is that a long horizontal color boundary would have irregularities only at the ends rather than all the way across. To a limited extent the mathematics needed can be done using analog circuits.
Top of Page
3D Comb Filters.
2D Three Line Adaptive Comb Filters
The big improvement that the 2D comb filter brings is near elimination of dot crawl.
"To get a sharply detailed, more film-like performance requires the color separation process to adapt its characteristics based upon the content of the scene." (A reviewer of Crystal Vision's VPS-1 digital comb filter)
The above statement is certainly impressive looking but does not convey any useful information to someone who does not know how comb filters work.
These are three line comb filters that improve the picture further by changing the mixture from left to right along the scan lines respectively, as opposed to the fixed 25%, 50%, 25% mixture mentioned above or the fixed 50-50 mixture of a simple two line filter. For example, where just the last two of the three lines are the same, the comb filter mixes them 0%, 50%, and 50% to get near perfect cancellation (filtering). Further along the lines the first two lines might be the same and the mixture is switched to 50%, 50%, 0%. There may be intermediate mixtures, too, used if the filter logic "was not sure" whether the line above or the line below was a better match. Also using intermediate mixtures for a gradual switch gives a better picture instead of a sudden switch from 0/50/50 to 50/50/0.
Dot crawl can be eliminated over most of the picture and be made almost unnoticeable in the most difficult places, diagonal color boundaries. Color boundaries almost as crisp as if the luminance and color were never mixed can be achieved for most of the content of the picture. Again, if neither the line above nor the line below matches, these comb filters also cannot improve that spot on the picture. An artifact will appear just as with the two line comb filter. Some brands of these filters even have a 0%, 100%, 0% "mixture" to choose from and a notch filter that is switched in and out automatically on a few occasions. These comb filters are called two-dimensional (2D) because the mixture is varied vertically by using or not using the predecessor and/or successor lines.and can also be changed numerous times going horizontally along the line.
Top of Page
3D Motion Adaptive Filter.
Three line 2D Comb Filter Behavior
This is hypothetical. Different brands of comb filters may behave slightly differently.
We are going to construct the second scan line. Lines are "temporally adjacent", or transmitted consecutively. After all the odd lines are comb filtered, the process is repeated for all the even lines when they arrive.
(revised 5/28/00 to be more typical as opposed to ideal)
Position A horizontally along the lines: Coarse luminance detail. First two lines are the same so the three are commingled 50%, 50%, 0%, the phase cancellation is theoretically perfect, and the result is the same as the second line which we want. For coarse horizontal detail, the luminance signal might have come from the raw line 2 via a low pass filter with nothing contributed by the comb filter circuits.
Position B horizontally: Fine luminance detail, first two lines are the same so the three are still commingled 50/50/0, the cancellation is again almost perfect, and the result matches the second line of the threesome.
Position C: The subject line does not match either the line above or the line below. No commingling mixture will clean all of the chroma out of the luminance signal or vice versa so an artifact will appear. Exactly what the comb filter does varies from one brand to another. One might use a 25/50/25 mixture, another might send 0/100/0 to the luminance circuits and 0/0/0 (nothing) to the color circuits. Weakening (attenuating) the color signal for this tiny moment seems like a good strategy here since it eliminates or reduces rainbow swirls. Also this situation only occurs where there is fine detail and there the human eye does not find the lack of color to be too obtrusive.
Position D: Color boundary. The last two lines are the same so the three are commingled 0/50/50, the cancellation is almost perfect, and the result matches the second line.
The overall result is elimination of most of the ghosting and dot crawl associated with non-adaptive filters, and fewer places about the screen where artifacts show up.
Top of Page
3D (Motion Adaptive) Comb Filters
The 3D comb filter can achieve essentially perfect Y/C separation, eliminating all dot crawl and rainbow swirls for "stationary" subject material, and perform at least as well as the 2D filter for the rest of the picture.
"...higher-end televisions will utilize "3D" filters, which cross-reference the particular "2D" technique being utilized against three fields to determine the best filter to use." (Reviewer for Crystal Vision VP1 digitasl comb filter)
Preceding is another example of a statement intended to impress the reader/buyer and which by itself does not convey any useful information to someone who does not know how comb filters work.
By "choosing the best filter" we mean choosing the commingling mixture that yields the best separation of luminance and color.
Another source of information that can be drawn upon to hopefully get a scan line that has the same content for correct cancellation is the next (and/or previous) frame. If there was no motion of the subject matter, the corresponding line in the next frame (525 lines away or about 1/30 second later) will have the same content as the line we are working on. Its color content phase is also intentionally reversed So the (commingled) sum of these lines is pure luminance and the difference is pure color. Of course, if there is subject motion at that spot the lines will differ and should not be commingled. The filter logic should sense that, forego the "third dimension" and go back to a method that 2D filters use. A good 3D comb filter contains within it a 2D comb filter. While there are elaborate processes for detecting motion, possibly involving three frames instead of two, all that has to be done is sense whether portions of the lines are the same or different. This process is not perfect and is made more difficult if the picture is noisy (snowy). The circuits would also have to store an entire frame in a rolling data buffer since a line could not be drawn on the picture tube until the corresponding line in the next frame has arrived. These comb filters are called three dimensional because they (1) use or not use the the preceding line and/or succeeding line depending on content (vertical), (2) vary the commingling formula going from left to right along a line (horizontal), and (3) use or not use the corresponding line in an adjacent frame for commingling (depth or time).
"Many televisions which employ 3D filters will suffer a severe degradation in picture quality whenever the picture moves at all, as that automatically defeats the 3D aspect. This then exposes [ed. note: what could be a mediocre] 2D process underneath." (Reviewer for Crystal Vision VP1 digital comb filter)
The buyer has to be careful that the manufacturer did not skimp on the quality of the first two dimensions when offering the third dimension in a comb filter. There will always be some loss of quality as the subject moves and the filter is forced to go back to two dimensions but sometimes the moving subject breaks out into a severe case of dot crawl. Comparing the picture with that from another nearby TV with a 2D or 3D comb filter is easier than looking at just one picture.
Since the human eye does not notice deficiencies in moving subject matter as easily as in still subjects, manufacturers might get away with skimping on the first two dimensions of 3D comb filters. You do have to look specifically for dot crawl and other such artifacts.
The proponents of 3D comb filters state correctly that a good one does better than any 2D filter, notably for (stationary) fine diagonal details, and that artifacts are less noticeable in moving subject matter compared with stationary subject matter.
The proponents of 2D comb filters claim reduced cost, that the re-appearance of rainbow swirls when a subject starts moving is more annoying than not eliminating them completely when the subject is still, and that the amount of the picture content (screen surface area) where the 3D filter excels is small.
Top of Page
Why Is It Called a Comb Filter?
Imagine an array of alternating parallel evenly spaced tall red lines and short black lines drawn on paper as the graphical representation of the video information, and looking through a comb (yes, that thing you use to primp your hair) to see just the short black lines.
A video signal is arranged just this way. The frequencies actually used for luminance data occur in evenly spaced bunches with spaces in between. The frequencies actually used for color data are also evenly spaced bunches and they fit perfectly in between. True, we have commingled the luminance and color information, but theoretically each can be separated out uncorrupted by the other most of the time. The filter necessary to separate luminance from color ideally passes just the desired bunches of frequencies and reject the others.
Physically What Is a Comb Filter?
Because the scan lines are transmitted one after another, we can mix two adjacent (in transmit order) lines by introducing time delay. One kind of comb filter (glass) contains a tiny loudspeaker that plays the video signal into a tiny glass chamber where a tiny microphone picks it up at the other end. Another is similar except that a thin glass fiber transmits the signal from the speaker to the microphone like a toy tin can and string telephone. Still others (CCD's, or charged coupled devices) use silicon chips with a chain of "charged cells" so as to make the signal travel slower as is passes through. The circuits are arranged so that the video signal goes through both the apparatus described above and through a straight wire. At the other end where the signal is commingled, when the current line is coming down the wire, the previous line is coming out of the delay apparatus. This happens for each line in the picture. There are manual adjustments to fine tune things so the phasing cancels as best as possible. For the three line comb filter, there are two delay apparatuses, so we have available for commingling one line delayed twice, the next line delayed once, and the third line not delayed.
"Digital" does not by itself imply lack of dot crawl; some "two line digital comb filters" do exist. "Digital", assuming truthful usage, simply means that the video signal is chopped up into slices that can be regarded as pixels, possibly stored in a buffer (computer memory), and then processed using computer circuits.
Top of Page
Simple Two Line Filter.
Loss of Vertical Resolution; Commingling of Line Content
Almost all technical descriptions of comb filters state that "The comb filter takes advantage of the fact that generally any two (temporally) consecutive scan lines are nearly the same." All comb filters found in commercial video products commingle the content of consecutively transmitted scan lines. Because of this the comb filter must cause some percentage loss of vertical resolution. The underlying secret to minimize net loss of vertical resolution, as done by the more expensive adaptive comb filters, is to do most of the commingling of lines or portions of lines where they have the same visual content. That is, do the commingling in places where the subject matter possesses no vertical detail; even a 100% loss of "nothing" equals no net loss!
Commingling of scan lines is done because the only easy and inexpensive way to extract just the luminance information or just the color information is to make the other suffer phase cancellation. Indeed the color information is intentionally phase reversed on every other line and the process works extremely well on those parts of a line that are the same as the preceding line.
By "mixed together" or "commingling" we mean little by little from left to right. An infinitesimally tiny portion at the far left of one line is mixed with the corresponding tiny portion at the far left of the line above, the next tiny portion of the subject line mixed with the next tiny portion of the previous line, and so on. This is the electronic equivalent of drawing each line twice, first on top of its predecessor and then on the empty space below. (Actual drawing of the lines twice in this fashion will not work.)
Even the simplest and earliest (two line) comb filters got a jump in reducing loss of vertical resolution by not "combing" the lower video frequencies (coarse horizontal details) where there is no color contamination to begin with. One brand of early comb filter combed everything but used a feedback loop (trademarked "Vertical Detail Enhancement") to undo the vertical resolution loss amongst the coarse details. However adaptive circuits evaluate the scan line in such small chunks that coarse information cannot be identified as coarse.
In a spot where there is fine horizontal detail but the overall color is the same, non-adaptive filters leave the top and bottom endpoints fuzzy instead of crisp. Where coarse horizontal detail bypasses the filter, we should not have blurred tops and bottoms of wide objects.
Color is a different story because the bulk of the color information is right in the middle of the frequency band to be combed. With the simplest (two line) comb filter there will always be a two line thick discoloration artifact at horizontal color boundaries that wasn't in the original picture. This effectively cuts the vertical color resolution to a third of what it would be without any comb filter. The three line analog comb filter will reduce the discoloration. The adaptive comb filters can almost completely eliminate this boundary discoloration and also eliminate dot crawl except where the colored object is just one or two lines high. It is hard to say what the overall vertical resolution is after comb filtering because it is different all over the screen depending on the subject matter.
Loss of Horizontal Resolution (10/99)
It is more correct to say that a Y/C separator reclaims or does not reclaim horizontal detail as opposed to lose it. The best 3D comb filters extract (reclaim) all the horizontal resolution present in the original composite video signal for stationary subject matter and over most of the picture (diagonal details being the weak spots) for moving subject matter. It may be noted that color (chrominance) resolution is limited to about 120 lines for certain color combinations and as low as 40 for others. Some comb filters reclaim all of this, others limit color resolution to around 40 lines for all colors. This latter depends on the quality as opposed to the kind of filter.
Video Disk and Video Tape Player Considerations
If you are feeding an S-video, RGB, or component video signal into the TV, the TV comb filter is skipped. But the incoming video signal may have gone through a comb filter elsewhere. A laser disk is recorded with composite video, that is, with the color already superimposed on the luminance information. If the LD player has an S-video output, it must have a built in comb filter to separate the luminance from the color. Evaluate the comb filter in your LD player by using an S-video cable to connect the equipment. Evaluate the comb filter in the TV set by making the connection with a standard video cable which bypasses the player's comb filter. Keep the connection that you feel gives a better picture. Evaluate the comb filter in an S-VHS VCR while viewing a broadcast received on its built in tuner (or an incoming composite video signal), and using S-video output.. We currently recommend always using the S-video output of an S-VHS VCR for playing tapes; use in addition a composite connection if the TV allows you to select it and if your TV's comb filter is better. DVD players do not have comb filters because the disks are recorded with the luminance and color information already separated. In your home theater iit is far better to keep DVD luminance and color separated through use of the S-video (Y/C) connection or the component (not composite) video connection. Whereas the regular Video In/Out (composite) connections combine the luminance and color, forcing the comb filter in your TV set to do the last honors of separating them again.
Special note regarding S-VHS VCR's and inexpensive composite to S-video converters -- A few models use a notch filter rather than a comb filter. There is luminance resolution in the "310 to 400 line range" but intermediate resolution in the "260 to 310 line range" is missing, "notched out". You will see "laser disk quality" from a test tape but a wedge test pattern will show a blurred out section in the middle. Analog broadcasts will still show the 260 line limit on horizontal resolution since there is no content in the 310 to 400 line range. We have tested two NTSC to VGA converters with TV tuners, both of which that exhibit this behavior.
Top of Page
Comb Filter Adjustments
The delay elements in a comb filter must produce the correct delay in order for the filter to perform properly. Take, for example, an approximately 4 megahertz signal, which is within the frequency band that is "combed" and can very well be fine luminance information or coarse color information. For 4 MHz, one cycle takes 250 nanoseconds, and if the timing is off by 125 nanoseconds, phasing is the reverse of what it should be. Information that should cancel is instead doubled in strength, and vice versa. This would produce totally incorrect results.All the frequencies from 2.1 MHz to 4.2 MHz must cancel or not cancel together. Using a colored wedge test pattern, if the wedge is alternately blurred and clear in numerous places, that signifies that the comb filter delay is still not correctly tuned. Also the delay must be exactly one line's draw time so that the lines' contents are not skewed as they are commingled from left to right. Otherwise lines will not cancel properly even if they are the same. More dot crawl that becomes even greater with added horizontal detail will be seen. One source reports that two line comb filters with very little dot crawl do exist; loose tolerances during manufacture are what results in proliferous dot crawl.
Analog comb filters usually have a delay adjustment.
Digital comb filters (usually) don't need a delay adjustment because they (should) synchronize themselves to the start (back porch) of each scan line and also they process each scan line the same way. This prevents phase errors assuming the incoming video signal is constructed uniformly and correctly.
All comb filters, digital ones included, can give less than perfect results when the incoming video signal has time base errors, that is, the time for each scan line is not absolutely constant.
Top of Page
Choosing the Best Filtering
How does the comb filter know when two lines are the same?
First off, non-adaptive comb filters neither know nor care. They simply use a fixed commingling formula producing two outputs, one for the luminance circuits and the other for the color circuits.
Adaptation requires math. Adaptive comb filters contain a computer (computer chip). As we mentioned earlier, if two scan lines are the same at a given spot and are added together, the color (chrominance) disappears. If one is subtracted from the other, the luminance disappears.
Scan line commingling goes on continuously in the first part of the comb filter that contains the delay elements. For example the first part of a hypothetical three line 2D filter might deliver seven outputs representing waveform amplitude for each small chunk of the scan line we are about to put on the screen.
1. The raw first line
2. The raw second line
3. The raw third line
4. The sum of the first two lines
5. The difference between the first two lines
6. The sum of the last two lines
7. The difference of the last two lines
Actually two sets of seven outputs are needed, one representing lower frequencies or coarse horizontal details, and the other set representing fine details.
The "small chunk" of fine detail (pixel) is the white spot in the middle of the diagram above. The green pixels represent the corresponding portions of the scan lines before and after. With scan lines divided into pieces this small, each pixel is represented digitally by a single number that stands for the amplitude if you graphed the video signal waveform. A separate set of the above seven outputs is computed and remembered for each of several three-high columns of pixels before and after (shown in red). Several consecutive sets of these outputs must be utilized to come up with the final result for the one pixel in the middle. Then the comb filter moves on to the next pixel to the right. A "small chunk" of coarse detail would be represented by, say, seven pixels as suggested by all seven pixels in a row in the diagram above.
According to a representative of Camelot Technologies, their Crystal Vision VPS-1 comb filter divides each scan line into about 1500 pixels. At video frequencies, each pixel represents less than one half of a cycle worth of a laserdisk video signal waveform representing the finest horizontal detail.
A zero value for any of the seven fine detail outputs means nothing all by itself. The comb filter does not know the luminance or color content of any of these outputs. This is where the seven coarse detail outputs come in. The coarse detail outputs represent luminance only, there is no color information at low video frequencies. So if the raw first line and the raw second line are the same for coarse detail (#5 is zero), we might assume (not foolproof) that scan lines 1 and 2 are the same.
The seven fine detail outputs contribute to the final output. If we assume that the first and second lines have the same content then #4 is selected as the luminance and #5 is selected as the color which is the 50/50/0 mixture of the original three lines.
If we find that coarse detail #2 and #3 are the same, then we use #6 as the luminance final output and #7 as the color. This is the 0/50/50 mixture.
If all three lines are the same in terms of coarse detail, the best bet is to use a 25/50/25 mixture as the final output since finding the coarse detail the same does not guarantee that everything is the same. If everything was the same, the 25/50/25 mixture gives the same result as a 0/50/50 or a 50/50/0 mixture.
If all three lines are different it would be up to the filter designer to decide whether the final output should be uncombed line 2 or a 25/50/25 mixture of lines 1, 2, and 3 respectively.
For typical video subject matter the above processing will yield pure luminance and pure color over most of the surface of the screen.
If all three original lines are different in a spot on the screen, then #8 will not match #1 and also #10 will not match #2. Here the behavior of the comb filter differs depending on the make and model, and is likely a trade secret. For example the output may be the raw second line (#2) with notch filtering applied. If nothing at all (0/0/0 "mixture") is delivered as color for that spot on the screen, there will be no rainbow swirl as that spot on the screen is rendered in black and white.
Can a two line adaptive comb filter exist? Yes, this writer has seen advertisements mentioning a "two line digital comb filter". It might have just the preliminary outputs and intermediate computations #1, #2, #4, and #5. If the two lines had the same content, the "combed" preliminary outputs #4 for luminance and #5 for color become the final output. If not, a proprietary formula, for example #2 notch filtered, is used for the final output. However three line adaptive comb filters are sufficiently inexpensive these days (5/00) that a two line adaptive comb filter is hardly worth manufacturing. Note that "digital" is also incorrectly used to describe simple non-adaptive comb filters containing a chip with "charged coupled devices", or a series of electronic cells, as delay elements.
Top of Page
Comb Filter Terminology
1D, 2D, 3D -- A 1D, or one dimensional, comb filter has only the vertical dimension, commingling information from the subject line and the line above and possibly the line below using a constant mixing formula.. A 2D comb filter has the vertical dimension of using consecutive scan lines in the mixture and the horizontal dimension of varying the mixture from left to right along each scan line. A 3D comb filter has the two dimensions just mentioned plus the third dimension of depth (or time) by using or not using information in the next video frame. "D" and "lines" are not the same.
Lines -- Refers to the number of scan lines taken into consideration (for commingling) at any given time; for 1D comb filters it is either two or three; for 2D filters it is always three. The 3D filter uses three lines from the subject frame plus one line from the previous frame and possibly one line from the succeeding frame..
1H, 2H -- Two line comb filters are also called "1H" because they contain elements or circuits that produce one horizontal lines' worth of time delay. Three line comb filters are sometimes called "2H" because their internal circuits or elements have two lines' of time delay.
Adaptive, Motion Adaptive -- Just "adaptive" means a 2D comb filter; it "adapts" according to the picture content by varying the commingling of the scan lines under consideration as the scanning goes from left to right across the screen. "Motion adaptive" means a 3D comb filter. It adapts by using or not using information from the next video frame depending on subject motion and also has 2D behavior. Adaptive and motion adaptive comb filters are digital, using computer chip circuits. Non-adaptive filters, which include some 2H and all 1H filters, are one dimensional (1D)
Correlating -- Determining that parts of two or more scan lines have the same content.
Glass -- This comb filter uses a glass or ceramic element with a hollow or chamber inside to generate the delay needed to commingle adjacent scan lines.
CCD, or Charged Coupled Device -- This comb filter uses a circuit with hundreds of tiny electronic cells that together produce the delay needed to commingle adjacent scan lines.
Cross Color -- Luminance information contaminating the color signal, resulting in rainbow swirls.
Cross Luminance -- Color information contaminating the luminance signal, resulting in dot crawl.
Dynamic -- Adaptive.
Temporally adjacent -- Refers to scan lines (or rows of pixels) of interlaced video that are transmitted consecutively and therefore arrive consecutively in time. A pair of such lines are either both odd or both even.
Spatially adjacent -- Refers to scan lines of interlaced video that are painted next to each other on the screen; one is odd and the other is even. Any two such lines are about 1/60 second apart (for NTSC) in time.
Y/C Separator -- A more generic term that encompasses video comb filters and the cheaper notch and bandpass filters used in their place.
Other Links (subject to change or discontinuance)
Stiill more on comb filters (Greg Rogers)
More on video signal formats
Camelot Technologies (Crystal Vision VPS-1 Video Processing System with comb filter, $650.)
Faroudja Laboratories (VP-100 Video Enhancer with comb filter, $550.)
Kramer Electronics (2D adaptive comb filter, $300.)
Tributaries (C2S Composite to S-video cable with notch filter (?) suitable for use with VCR; $100.; also C2S Plus (probably) two line comb filter, $300.)
Note: A powered (non-passive) composite to S-video converter cannot be turned around and used for S-video to composite conversion. Some passive (non-powered) composite to S-video converters are reversible, some are not. S-video to composite converters, also usually non-reversible, do exist.
Other television and video topics
Go to topics on unrelated subjects
Return to top of page
(c) Copyright 1998-2009, Allan W. Jayne, Jr.
If you would like to contribute an idea for our web page, please send us an e-mail. Sorry, but due to the volume of e-mail we cannot reply personally to all inquiries.