NTSC Decoding Basics (Part 3)
by Steve Somers, Vice President of Engineering
Line Comb Filter Decoders
I'd like to thank all of you for your complimentary input on my articles and their value to you. Here in Part 3, we will move forward into the application of the comb filter in NTSC decoding. And, in case you have not read parts one and two, I want to reintroduce the basic NTSC decoding system diagram as Figure 1. The Y/C Separator portion continues to be the real focus of this series as this function is the most critical in the system.
Figure 1: Basic NTSC Decoding System
I previously covered operation of the Notch/Bandpass Filter. Recall that two simple, passive type filters (of course, there are active variants but they accomplish the same function) are used to perform simple separation of luma information and chroma information in the frequency domain. Since this type of filter cannot tell the difference between one frequency and another with respect to time, some luma information spills into the chroma channel and vice versa. Therefore, it's hardly an ideal system but it is low in cost and provides acceptable performance (perhaps up to 250 lines of horizontal resolution) for mass production purposes.
Since the beginning of professional television broadcast, the TV industry has looked for an affordable method to separate luma from chroma more efficiently. It took about twenty years after the introduction of color television before reasonably affordable comb filter methods were realized. And, for the most part, those methods were analog in their earliest implementations. Today, digital processing is the norm for comb filters and the technical intrigue continues as we inch closer to the full digital television age.
Regardless of the specific components used to implement the "combing process", we are going to look at the basic topologies currently in use and understand the relative advantages of each. Lowest in the food chain is the single line comb filter; next, the 2-line comb filter; and yes, the 3-line version; replaced by 2D adaptive methods; and, currently the sophisticated 3D motion-adaptive comb filter. All comb filter topologies share an important concept…picture information is about the same from line to line. Fundamentally, incoming information is stored in some way; then compared to successive information in such a way as to attempt cancellation (in the time domain) of luma information from the chroma channel and, vice versa. Crystal clear, right?
Figure 2: Single Line Comb Filter
OK, Form One or Two Lines
The earliest and simplest versions of a comb filter delayed one horizontal line of picture information and compared it to the current line passing through the system (see Figure 2). Adding the two lines together causes the Chroma to cancel itself thereby providing the Luma information. Why? In the NTSC system, the chroma burst reference and the chroma information is reversed by 180 degrees on alternate lines. Going back to the early days, this phase reversal effect caused the perceived luminance of the system to remain constant even though the color subcarrier is superimposed upon the luma information. [Remember that the color system was mandated to provide acceptable performance on monochrome receivers.] Conversely, subtracting the delayed line from the current line cancels the luma portion and provides the chroma signal. Take a look at the shape of the typical system responses in Figure 3 to see why we call it a "comb filter". Comparing the Y and C frequency components in the correct time relationship creates a filter system where very little luma or chroma cross over into the wrong processing channel. This greatly improves horizontal resolution. (Until recent years, comb filters were analog implementations using glass delay lines, or SAWs, (surface acoustic waves) or charge-coupled devices (CCDs) to obtain the relatively long delay for an entire horizontal line. Today, most designs are digital utilizing A/D and D/A conversion with memory chips providing the delays needed for digital processing of the signal to derive luma and chroma.)
Figure 3: Comb Filter Response
The action described above occurs continuously on a line-by-line basis thus providing a simultaneous flow of Y and C information. But, the "combing" action is limited to the immediate picture field being received. If the image is static, the combing is very near perfect. An obvious, perhaps objectionable artifact is the transition between different colors vertically through the picture. For example, moving from a red area downward to a green area creates a yellow line at the horizontal boundary between the two colors, since one red line would be delayed and added to the first green line. Worse yet, when motion is involved, the cancellation effect is imperfect and gives rise to various image artifacts.
Figure 4: 2-Line Comb Filter
Adding a second delay path to create a 2-line comb filter, Figure 4, can help reduce the problems with color boundaries. Here, two horizontal lines are delayed and added back to the incoming video information. Considering that a total of three lines of information are involved, the information after the first delay will become the "middle" line in the sequence. The incoming video is compared at a weighted value of one-fourth, the last delay weighted by one-fourth, and the first delay weighted by one-half. Therefore, at horizontal boundaries of colors, the transition from one color to another is softened. Perhaps the largest disadvantage to the 2-line comb filter is that vertical resolution suffers due to the constant weighted averaging of picture information. For this reason, other forms of adaptive comb filters are more popular for professional applications, but cost more.
If two lines are better than one, then three should be better yet, right? Yes, depending. In most cases, more resolution and color decoding improvement may be obtained. The processing format is similar to the 2-line comb, but may utilize additional averaging methods between line samples and typically provides more decisive control. Depending on implementation, additional vertical resolution may be lost.
All line combs suffer from "chroma mesh failure" when images contain vertical color changes. At the color transition, the image information is changing such that some chroma is not cancelled and is passed through as luma information, causing cross-luminance effects…or, the typical "hanging dots" at the transition point. Diagonal lines present another difficult situation as luma information is changing position from line to line. Samples of the luma do not match with prior or later samples and an artifact is created. The non-cancelled luma passes through into the chroma system and becomes the "rainbow effect" seen dancing about on areas where predominant luma detail having oblique angles occupies the image.
Comb When You Can
Most implementations of comb filters today are digital. Even when the internal design is a digital line comb, there is some intelligence usually included with these filters. Why? The digital comb filter must operate with precise timing in order to provide proper cancellation of luma and chroma. If the luma and chroma timing relationships are lost, the image quality is worse than when using a notch/bandpass filter.
So, most digital comb filters regularly make decisions on when to implement the comb and when to implement the notch/bandpass approach. This occurs most predominantly with standard VCRs. The signal timing from a VCR is not capable [without a timebase corrector] of providing the pixel-to-pixel accuracy required in a digital decoder. During VCR playback, a display's digital decoder detects the large signal phase errors and automatically switches back to the notch/bandpass mode of Y/C separation that was discussed in the last article. Considering that the realizable resolution from the typical VCR is not any better than the performance of a good notch/bandpass design, you will not see much difference in performance. Certainly, there is some exception for picture quality here with the use of S-video type recording and playback, but that application is not the most common for analog VCRs. For extremely stable sources such as direct broadcast or DVD, the comb filter will switch into operation.
Adapt or Die!
The next steps in comb filter technology lead to adaptive techniques such as 2D and 3D motion adaptive combing. Image motion is the bane of line-based comb filters. Adaptation allows the system further decision-making capability during adverse image content changes. Unfortunately, this topic will have to be the next, and final, episode in this series on NTSC decoding basics. Unlike the horizontal dots in a cheap comb filter, there's no intention to keep you hanging.
Return to top.