NTSC Decoding Basics (Part 4)
by Steve Somers, Vice President of Engineering
Adaptive Comb Filter Decoders
This is the last installment of our series on NTSC decoding basics, and Y/C separation in particular. Meanwhile the quest for high quality decoders continues on into the sunset of NTSC. The emphasis here is on the first step in the process known as Y/C, luma/chroma, separation since it does represent the most crucial step in realizing the full potential of the transmission medium. The earlier installments covered NTSC signal creation, notch/bandpass filtering, and line comb filters.
Implementation of high-speed digital processing and low cost memory components facilitated many improvements in Y/C separation. These strides in decoding performance involve decision-making, called adaptation, based upon image content. Moreover, having more image memory available means that video processing can now take place beyond the original 2D boundaries (i.e. the information within a picture field). Processing decisions now extend into the temporal realm, or from picture frame to picture frame. This means more intelligence in Y/C separation as subject matter moves across the screen from frame to frame.
2D or not 2D?
Since conventional (line type) comb filters have problems handling diagonal lines and vertical color changes, it seems the first order of business would be to tackle these tricky situations. Remember that when processing diagonal lines, succeeding luma, Y, information is shifted in time and does not properly cancel line-to-line or field-to-field. This results in Y information being interpreted as chroma, C, information. The outcome is "cross color" or the rainbow effect in the region of the diagonal lines. The vertical color changes, which do not match in time at the transition point, result in chroma information being interpreted as Y information. Hence, you see the "hanging dots" at the color boundary.
Figure 1 — Two-line comb filter
Now, suppose we take the two-line comb design discussed previously (see Figure 1). This design is comparing data between two stored lines and the incoming data. It utilizes averaging to soften the transition between colors. If we add a system called a correlator, data between the three lines can be tested before any actual averaging or summation takes place (see Figure 2).
Figure 2 — Correlation detector
Here, if there is significant correlation of data value between line 1 and line 2, then CF= 0 and a difference between line 2 and line 1 is used. But, if line 2 and line 3 correlate to a higher degree, CF=1 and the difference between line 2 and line 3 is used. If there is no correlation at all, then CF= 0 and the difference between line 2 and line 1 is used. This type adaptive system works only within a picture field and is called intrafield filtering. Additional logic is usually added to detect whether the system sampling of data is stable enough to perform proper combing. If not, the system switches back to the notch/bandpass filtering method. This is quite common when decoding video from the typical VHS VCR.
The term "2D" indicates that the filter implements detection of both horizontal transitions (along horizontal lines) and vertical transitions (between horizontal lines) within the picture field of interest. The term adaptive carries wide meaning in that comb filter manufacturers create unique methods that enable the filter to make better decisions as to the process algorithms to use. Adaptive processing can lead to increased noise or graininess in the image. To combat this, a system function may be added called coring. The coring function modifies (or outputs the equivalent of zeroes) the data values near a transition so as to remove the random artifacts. This, in conjunction with a contouring circuit, adjusts the values to provide a much more pleasing image transition.
So, the 2D adaptive filter attempts to eliminate the hanging dots at color transitions. If a situation occurs where there are different colors on three successive lines, the filter fails and artifacts appear. The problem will be most noticeable on specific, abrupt color changes and not so obvious on gradual color transitions. On diagonal lines, the 2D adaptive filter is less effective. There is typically not much luma correlation from one line to another. Although some averaging occurs, this filter type is only marginally better than a regular line comb.
Cure for motion sickness?
3D motion adaptive comb filters represent the most sophisticated comb technology available. While still pricey compared to previously discussed filter types, they can provide near perfect separation of Y and C for still frame images. In 3D comb filtering, picture information is taken and compared to information in successive frames (called inter-frame filtering) as opposed to 2D filters which process data taken from successive lines within a field or frame (intra-frame filtering).
For still images, the picture data from one frame to the next is essentially identical. Since it has a high degree of correlation, making comparisons (similar in topology to the line comb examples) provides near perfect output of correlated Y and C. In this case, diagonal lines and color transitions can be matched very well. This is the key advantage over the 2D filter.
However, if there is picture movement or color changes between frames, the 3D system will produce noticeable errors. Here is where the motion adaptation comes into play. The 3D filter compares the data of several video frames to determine correlation (still image) or lack of it (motion present). If motion or serious color changes are detected, the system switches to 2D adaptive filter operation. Under this condition the 3D filter may perform no better than the 2D filter. Different implementations of the filter will use different algorithms for the determination of motion or color changes. This motion detection is not a trivial pursuit and will separate the good 3D comb filter from the not-so-good one.
You, too, can become an expert
After all the foregoing, what does it mean to you? How can you tell one filter from another? That is probably the most important question on your mind. Further, it's one thing to recognize the type of comb filter design and quite another to determine if it's performing correctly. Let's see if I can provide you with "something to walk away with". Refer to Table 1 for a compilation of features and problems with each of the topologies discussed in this series on Y/C separation in the NTSC decoding process. While you may not find that memorizing the various features is any advantage to you, pay particular attention to those features that help identify the comb filter type of most interest to you. You never know when and where you'll see a good comb filter.
||Low luma bandwidth; High amounts of cross-color and cross-luma.||Soft edge definition; gross amounts of dot crawl around text and at vertical color boundaries; regions with close spaced lines go gray.||Multi-burst: look for gray band at 3.58 MHz band region; no detail at higher bands.
Color Bars: heavy dot crawl on vertical transitions; poor green/magenta transition.
Resolution Wedge: cross-color in wedge along with rapid loss of res near the wedge end.
||Increased horizontal resolution; hanging dots at color boundaries; vertical dot crawl; severe artifacts at vertical color transitions.||Can better see close spaced lines although lots of cross-color; false colors at vertical color transitions; better BW than notch.||Multi-burst: improved hor res; look for more clarity of 3 - 4.2 MHz bands.
Color Bars: dot crawl on vertical edges; abrupt color errors between upper & lower bar colors.
Resolution Wedge: some res improvement, but lots of cross-color.
||Increased horizontal res, but less vertical res; better color transitions; hanging dots; dot crawl.||Can better see close spaced lines although some cross-color; improved color transitions; better BW than notch; lower vertical res; hanging dots on vertical color transitions.||Multi-burst: good horizontal res; clarity of all bursts.
Color Bars: improved vertical color transitions; hanging dots along the vertical color transitions.
Resolution Wedge: lower vertical res, still see cross-color.
||Nearly no hanging dots; good hor bandwidth; improved cross-color performance.||Fails to eliminate dots in regions of high detail; nearly no dots at base of text or other color transitions.||Multi-burst: good horizontal res and clarity of all bursts.
Color Bars: near elimination of hanging dots; cleaner color transitions; some vertical dot crawl.
Resolution Wedge: some cross-color; some loss of vertical res; good horizontal res & correlation.
||Near perfect on still images; switches to 2D during motion; improved vertical res.||Best H & V BW with no dot crawl or hanging dots; possible artifacts around moving objects.||Multi-burst: good horizontal res and clarity of all bursts.
Color Bars: virtually no dots vertical or horizontal; smooth color transitions; good green/magenta transition.
Resolution Wedge: when stationary, full res w/o cross-color artifacts… when moving, same as 2D.
|* Color Bars refers to SMPTE Color Bars because of their complementary color transition region below each bar.|