DVD Benchmark - Part 5 - Progressive Scan DVD - October, 2000
Don Munsil and
It has long been our feeling that one can learn much more about the quality of a DVD player by watching lousy DVD movie material than you can learn by watching great material. Great material tends to look, well, great, on just about every player, and we find ourselves squinting and striving to tease out the tiny differences in sharpness and resolution. But as we will see, there are really substantial differences in how well players handle badly encoded DVDs. There are plenty of DVD player reviews that will tell you how terrific a player looks when playing "The Fifth Element" or "Starship Troopers", but really, that’s like testing a car by only driving it on a perfectly smooth, level road. We’re here to change all that. We’re going off-road and showing you how well these players handle the rocks and potholes. This article is all about problems you might encounter with progressive DVD players, when you will encounter them, and what those problems look like. We took most of the currently available mass-market progressive DVD players, and a few expensive boutique players, along with some PC DVD software, and tested them with the most difficult DVDs we could find. Few players came through the gauntlet unscathed.
It almost goes without saying that all of these players will deliver a very nice image when fed good material (with a few exceptions, which we’ll cover later). There are differences between the players’ basic image quality, but for the most part, you’re going to notice very minor video quality differences when the players are firing on all cylinders. When the players are dealing with a crummy DVD, though, that’s when things get interesting. It should be noted in advance, that probably 90%, maybe 95%, of major Hollywood releases are going to play back basically fine on all of these players. If you’re happy with those kinds of numbers, just get the player with the best video quality, or the features that are most important to you. But if you want the other 5-10% of your DVDs to also look good, or if you have a reasonable number of the smaller-name DVDs, which are more likely to have mastering problems, you’ll want to pay close attention to our de-interlacing performance tests.
We also want to warn you: if you have a progressive-scan DVD player that you are happy with, you probably should not read this report. With some of these artifacts, you are much better off not knowing they are there, because once you start to notice them, they’ll drive you nuts, and then you will inevitably need to replace your player. We’re quite serious here – it’s happened to us. Once someone points out a flaw in your player, it’s very difficult to ignore it. If you don’t have a progressive player, and are thinking about getting one, then this report is for you. Or if you have a progressive player, and are seeing artifacts that are bothering you, and you want to find out more about those artifacts, or find a player that doesn’t have them, then again, this is a report you will want to read. But the rest of you: don’t blame us if reading this article ruins your enjoyment of your expensive DVD player. We warned you. We are calling them as we see them. If you want a review that will tell you, "Oh yes, this is a reference quality player," for most players, there are plenty of A/V magazines out there to do that for you. We won't be doing that here.
Before we begin, Brian will explain how film is transferred to video.
An Explanation of Film-to-Video Frame Rate Conversion
To better understand the upcoming concepts, one must be armed with some basic knowledge of how film gets transferred to video, as well as the nature of interlaced versus progressive display. As such, the following information is not intended to be a definitive paper on the subject, but should serve as a good introduction for all.
The visuals and animations presented here, though large in file size, are key and will reward repeat viewing.
Motion pictures are comprised not of motion at all, but numerous stills shown in rapid succession. For the films we all watch at the theater, 24 frames are shown in one second (24 frames per second, or fps). The NTSC television system differs from film in this regard, making it complicated to show film on video.
Televisions create their image by drawing (scanning) lines of light on the CRT face, left to right, top to bottom, to produce a picture over the entire screen. The resultant 'frames' as we call them, that make up the picture are interlaced into two fields per frame: that is, the first field consists of all the odd lines of a single frame (1 through 525), and the second field consists of all the even lines of that frame (2 through 524). The result is that only half of the video's image is drawn at one time. A simulation of this is shown on the left. Field 1 is scanned, showing half the picture, and then Field 2, which fills in the rest of the picture. Traditional talk quotes NTSC television as having 30 frames per second (as opposed to film's 24), each being comprised of two interlaced fields. We're going to try and unlearn that in this article. We need to realize that, just as fields 1 and 2 make an image, so do 2 and 3, 3 and 4, 4 and 5, and so on. So we don't want to think of interlaced televisions in terms of frames but rather in terms of fields, interlaced fields, and 60 of them per second.
The principal drawbacks of an interlaced display are (A) visible line structure, (B) flicker caused by the rapid alternating of the fields, and most important, (C) artifacts such as 'feathering' (also referred to as 'combing' or the 'zipper effect') and 'line twitter'. Visual artifacts like these last two occur anytime the subject or the camera moves from field to field. The subject will be in one position for one field, and in another position for the next, resulting in jagged edges (feathering) or shimmering horizontals lines (twitter). This is why you can see jaggies even when film is converted to video and fields 1 and 2 make up frame 1 with interlacing. You are seeing the effects of field 2 of frame 1 and field 1 of frame 2 being interlaced.
The animation on the right shows an example of motion on an interlaced display trying to show a tomato moving from left to right. Each field shows the tomato a little farther to the right than the previous. Because the fields are interlaced, jagged vertical edges can't help but exist, except during for the last two fields (5 and 6) where the tomato is stationary. The further back you are from an interlaced display (or the smaller the display is), the less this and other artifacts are noticed. If you want to see the effect in real life, just stick your nose up to an interlaced TV. Focus in on an objects edge that is stationary and wait for it to move. You will notice this right away.
Where the tomato animation deals with horizontal movement, line twitter is brought about by vertical movement. White horizontal lines in a brick building are an excellent example. As the camera or the wall pans up or down, if it is at a certain rate, the lines will be visible in one field and not in the next, disappearing and reappearing. It looks like stars blinking in the sky, or "twittering" if you will. It is also possible to see line twitter when there is no motion. If fine detail exists that is less than 2 scan lines high, it will only show up in one field. As the fields are flashed by, you will see the line appear and then disappear.
The proceeding basic knowledge on interlacing is necessary to understand the transfer of film to video, because it is an important factor in what we end up seeing.
So, how do we space 24 film frames per second over 60 video fields per second? Simple math says that for every 4 film frames (1/6 sec.), we need 10 video fields (also 1/6 second). But we can't just double up the fields on every fourth film frame or we'd get a real 'stuttered' look. Instead, a process is used known as 3-2 pulldown to create 10 video fields from 4 film frames. They alternate between creating 3 fields and 2 fields from each film frame. Hence the name 3-2.
Consider now our flow chart of the 3-2 pulldown performed on four frames of this movie scene:
Note that anytime a field follows one made from a different film frame, there exist the possibility for anomalies in what we see, feathering being one example. Absolutely any differences between the two film frames that make up the video frame (the last field of one frame and the first field of the next frame), be it brightness, color, or especially motion, are going to result in some artifact as the two fields merge on screen. Even our little animated synthesis of the final interlaced product, which actually contains 10 interlaced pieces, shows evidence of such anomalies as the Millennium Falcon gets closer. Such is life.
As long as you are watching your movies on an ordinary interlaced display, there is not much more to tell you. What you see is pretty much what is shown as the interlaced content in the above illustration. But should you have the fortune to be using a progressive display TV, the following comes into play.
Progressive displays, such as high-performance CRT/LCD/DLP projectors and the new HDTV-ready TVs, can show progressive scanned images as opposed to interlaced. In order to show a progressive image, the CRT must scan at a higher rate, which is 2x the speed of NTSC. Because we are scanning at twice the speed, we can draw an entire frame in the same amount of time it took to draw a single field. We learned above that an interlaced display is really showing 60 fields per second. But with progressive, each "field" is now a complete picture including all scan lines, top to bottom, so we will now call it a frame, and we are showing 60 of those per second. (Of course, only 24 of those are unique if the source is film based) The benefits of a progressive display are no flicker, scan lines are much less visible (permitting closer seating to the display), and they have none of the artifacts we described for the interlaced display, as long as the source material is progressive in nature (film or a progressive video camera).
The term commonly used to restore the progressive image is de-interlacing, though you might also hear it called re-interleaving.
De-interlacing (or re-interleaving) involves assembling each pair of interlaced fields into one progressive frame (1/60 of a second long), and showing it at least twice to fill the one second time. The need for 60 flashes on the screen each second stems from a biological property called the Flicker Fusion Frequency, meaning how many flashes that we need to see each second so that we (our brains) fuse the image into one where we don't see a flicker.
For every film frame that had three fields made from it, the third field is a duplicate of the first, and (if the MPEG-2 compression is set right) won't even be stored on the DVD. Instead of encoding the duplicate fields, the DVD flags repeat_first_field and top_field_first are used to instruct the MPEG decoder on what to do.
The progressive output player will assemble 2 fields from each film frame and create a complete progressive one that looks just like the original film frame. You should now be thinking that the DVD will once again have 24 frames to show in one second. But the progressive display is still expecting 60 complete frames (one frame being made of two fields assembled together) a second. In order to space them out, the DVD player shows the complete frames in this order: 1, 1, 1, 2, 2, 3, 3, 3, 4, 4 and so on.
This form of display gives us a moving image very close to the original film. It has a tendency to "judder" a bit though, as every other film frame lasts 1/60 of a second longer than the previous one. Even our little synthesis of the final product, which actually contains 10 pieces, shows this judder. In the future, both the player and the display need to increase the display rate above 60 fields per second, to 72 per second. At that point, the fields would only last 1/72 of a second, permitting the player to show every film frame three times (24 x 3 = 72), eliminating the motion judder, and also helping us with the Flicker Fusion Frequency problem (60 flashes per second are just barely enough in a well lit viewing environment). This would look like: 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4 and so on. 72 fps will only work with film based sources, as it is a multiple of 24. It will not work well with video sources which are 30 fps.
The process above is really for progressive source material since each 'pair' of 2 interlaced fields come from a single film frame, so it's fairly simple. De-interlacing native NTSC interlaced video material is much more complicated. In such video material, each field is a unique image in time, and in order to be de-interlaced at an acceptable level, it requires getting into motion adaptive and motion compensation algorithms to overcome the inherent problems of the interlaced material. There is no best method, and the two mentioned are VERY expensive to implement. We will get around to explaining it when we have reviewed some more projectors and HDTVs.
- Brian Florian -
Why are progressive players better?
With all the hype flying around about progressive DVD players, many people have assumed that when they get their new progressive player home and plug it in, the difference will smack them in the face and it will be like watching a whole new film. We’re here to tell you that’s not so. If you don’t know what to look for, the difference between interlaced and progressive can be quite subtle. Not to say that there’s no improvement, but it’s not the kind of improvement that will knock most viewers off their feet. Once you understand the improvements, though, and know what to look for, we think most people will be hooked, and will no longer want to watch their DVDs in interlaced form.
The magnitude of the change you will see also depends on what you were watching before. If you switch from a regular interlaced TV to a progressive-scanned 480p picture, you should see a much smoother, more film-like picture, with much less obvious scan line structure, and more apparent vertical resolution. The difference should, in fact, be pretty obvious. But if, like most new high-end TVs, your TV has a built in de-interlacer (often called a “line doubler”), then the television has been converting your interlaced signal to 480p already, so the smoothness and lack of line structure are already there. So what can a progressive DVD player offer? Better de-interlacing, to start with. In other words, the de-interlacer in the DVD player is likely better than the one in your TV.
Most high-end TVs have a motion-adaptive de-interlacer, with no film mode, so they can’t recreate a perfect progressive film frame from film-originated sources. (The TV has its de-interlacer optimized just for video sources, such as news broadcasts.) But nearly all progressive DVD players, certainly all the ones we review here, have a de-interlacer with a film mode (remember, film sources are 24 frames per second, and video sources are 30 frames per second). The result is that when watching film-originated DVDs, you will get far less line twitter on thin horizontal and near-horizontal lines, you will keep full resolution on camera pans and zooms, and you will see less noise on high-detail areas. To see this effect, look at thin lines in the background, especially on slow camera movement. Look at high-detail areas of the picture, like trees and bushes, again, especially on slow camera moves. Once you see how sharp and clear those areas look with good film-mode de-interlacing, you won’t want to go back.
If you don’t really see any difference, then perhaps your TV has a better de-interlacer than most, and includes a film mode. Then you will see very subtle differences, perhaps no difference at all, switching between progressive and interlaced output on your DVD player. The de-interlacer in the TV may be just as good as the one in the DVD player. Even in this case, though, the progressive player has an advantage: better resolution. When the television de-interlaces the analog signal, it has to digitize the signal, send it through a de-interlacing chip, and convert it back to analog to feed to the CRTs. That process inevitably loses some resolution. It might not be much. You might feel like you can live with it. But it would be a good idea to get a copy of the Avia DVD and look at the resolution pattern on it to see exactly how much you are losing. About 10 TVL (TV Lines) of loss is good, 15 is average, and 20 or more is bad. With the progressive DVD player, you should be able to get all 540 TVL. But if your TV has a good film-mode de-interlacer, and you feel like you can live with the resolution loss, then save yourself the money and get a nice interlaced player.
Sometimes you see folks commenting on the improved blacks and color saturation of their progressive players. This is a mirage. Progressive players are not supposed to improve black level or color saturation. The reason it may look different is that the standard black level on a progressive player is usually different from the standard black level on an interlaced player, for technical reasons having to do with conflicting TV standards. And when black levels go down, the saturation of colors on the screen inevitably goes up, because white is being removed from the color. Once you calibrate your display with Avia or Video Essentials, the color, contrast, and black level should look exactly the same with interlaced vs. progressive. The only advantage of a progressive DVD player is in the lack of interlace artifacts. Of course, if your progressive DVD player is a lot better than your old interlaced player, it may look better for a variety of reasons. But it still shouldn’t change your black level or color saturation.
How progressive players work
How the information is stored on disc
It’s important to understand at the outset that DVDs are designed for interlaced displays. There’s a persistent myth that DVDs are inherently progressive, and all a DVD player needs to do to display a progressive signal is to grab the progressive frames off the disc and show them. This is not exactly true. First of all, a significant amount of DVD content was never progressive to begin with. Anything shot with a video camera, which includes many concerts, most supplementary documentaries, and many TV shows, is inherently interlaced. Only content that was originally shot on film, or with a progressive TV camera, or created in a computer, is progressive from the get-go. But even for such content, there is no requirement that it be stored on the DVD progressively.
DVDs are based on MPEG-2 encoding, which allows for either progressive or interlaced sequences. However, no one ever uses progressive sequences on DVD, because the players are specifically designed for interlaced output. Interestingly, while the sequences (i.e., the films and videos) are never stored progressive, there’s nothing wrong with using individual progressive frames. This may sound like a semantic distinction, but it’s not. If the sequence is progressive, then all sorts of rules kick into place, which ensure that the material stays progressive from start to finish. Whereas, if the sequence is interlaced, then there are many fewer rules, and no requirement to use progressive frames. You can mix and match interlaced fields and progressive frames to your heart’s content, as long as each second of MPEG-2 data contains 60 fields, no more, no less. (PAL DVDs have 50 fields per second.) The progressive frames, when they are used, are purely for compression efficiency, but the video is still interlaced as far as the MPEG decoder is concerned.
In interlaced sequences, the encoder can either keep the fields separate, or combine them together into one frame, whichever is best for compression purposes. There is a flag on each picture stored in the MPEG-2 stream called “picture_structure” that can be either “frame” for a full 720x480 pixel frame, or “top field” or “bottom field” for a single 720x240 field. (We’ll learn about top and bottom fields later.) And it is allowed, but again not required, to set a flag called “progressive_frame” as a hint to the decoder that the fields in that frame were taken from the same frame of film. This allows for better pause and slow motion modes, and better down-conversion of 16x9 images for 4x3 displays. But this is again, purely optional. The content will play fine whether the data is structured as fields or frames, and whether the flag is present or not.
In fact, the encoder is allowed to combine fields that are not from the same film frame together, as often that produces better compression, even for inherently interlaced video. In such cases, the encoder is not supposed to set the progressive_frame flag, but again, if it does happen to get set, it will make no difference for normal playback on an interlaced display. And since interlaced displays are the only thing DVD was designed for, sloppiness with flags is more common than you’d think. What we are about to tell you is that DVD manufacturers have not been paying attention to what they were supposed to, when programming the DVDs, such that progressive scan DVD players are now having a hard time delivering good progressive scan signals to the TV when those discs are played. Time to wake up. Progressive players and TVs are here. What now . . . yet more re-releases of DVD movies?
The flags on the disc, and the structure of the frames, are purely hints. Interlaced video can be stored on the disc as frames, and progressive frames can be broken into fields. The progressive_frame flag can be there or not. It doesn’t make any difference for interlaced playback. As we will see, though, it can make a big difference for progressive playback.
As Brian explained above, when film is transferred to video, 24 frames per second of film must be converted to 60 fields (30 frames) per second of video. The way this is accomplished is to show the first frame of film for 3 fields, then show the second frame of film for 2 fields, then the third frame of film for 3 fields, and so forth. This sequence, or “cadence” of 3, 2, 3, 2, 3, 2 is what is called 3-2 pulldown. (Sometimes this is written as 3/2 or 3:2, but this looks like a ratio, so we prefer it written as “3-2.”)
There are two extra flags available in MPEG-2 to make it easier to create a disc that has 3-2 pulldown. These two flags are called “repeat_first_field” and “top_field_first.” A frame in the MPEG stream can have “repeat_first_field” set to “true,” and that tells the decoder to generate 3 fields from this frame, rather than 2. The decoder plays the first field, then the second field, then the first field again, thereby making the 3-field section of the 3-2 pulldown. The next frame would generally have “repeat_first_field” set to “false,” so the decoder will generate 2 fields. Because fields have to alternate between even (bottom) and odd (top), the “top_field_first” flag tells the encoder which of the two fields in the frame should be sent out first. The “top” field is the odd numbered scan lines: 1, 3, 5…, while the “bottom” field is the even numbered scan lines: 2, 4, 6… If the top field is first on a 3-field frame, the decoder will output top, bottom, top. The next field needs to be a bottom field, so the next frame will have top_field_first set to “false.”
Again, it’s important to note that there is no requirement to follow this 3-2 sequence. The DVD decoder doesn’t care. It just follows the flags as written on the disc. As long as there are 60 fields in each second of video, everything is fine.
Here are some examples of what legal flag sequences look like. Imagine that we have a sequence of 4 film frames that we want to convert to video and store as an MPEG-2 stream.. We need to turn those 4 frames into 10 fields (3 + 2 + 3 + 2).
First, the most common way, using 4 MPEG pictures and all the flags:
MPEG picture Film Frame Picture_structure Progressive_frame Repeat_first_field Top_field_first 1 1 Frame True True True 2 2 Frame True False False 3 3 Frame True True False 4 4 Frame True False True
However, it would be perfectly acceptable to encode that same sequence of film like this, using 10 MPEG pictures:
MPEG picture Film Frame Picture_structure Progressive_frame Repeat_first_field Top_field_first 1 1 Top Field False False False 2 1 Bottom Field False False False 3 1 Top Field False False False 4 2 Bottom Field False False False 5 2 Top Field False False False 6 3 Bottom Field False False False 7 3 Top Field False False False 8 3 Bottom Field False False False 9 4 Top Field False False False 10 4 Bottom Field False False False
Or like this, using 5 MPEG pictures (MPEG pictures 2 and 3 contain fields from two different film frames):
MPEG picture Film Frame Picture_structure Progressive_frame Repeat_first_field Top_field_first 1 1 Frame False False True 2 1 & 2 Frame False False True 3 2 & 3 Frame False False True 4 3 Frame False False True 5 4 Frame False False True
These are all real examples, actually used on real DVDs. And there are dozens of other legal variations. In each case, exactly the same sequence of fields will be produced at the decoder output, even though the flags and the number of pictures actually stored on the disc will be different. The compression factor will be best with the first variation, which is the only reason it’s the most popular. (It’s certainly not for the purpose of making progressive DVD players work better.)
Film-mode de-interlacing ("Re-Interleaving")
To display a perfect progressive image from a film-sourced DVD, the player needs to figure out which fields in the MPEG stream go together to make each film frame. In theory, the progressive_frame flag should tell the player that the frames on the disc were originally from a film, and will go together, but as we’ve mentioned, that flag is not always set correctly.
So what most players do is use a standard MPEG-2 decoder to generate digital interlaced video, then feed that video to a de-interlacing chip. The chip makes decisions constantly about whether the video was originally from film by looking for repeated fields. In the standard 3-2 cadence, the 1st and 3rd fields are identical. If the de-interlacing chip sees a constant stream of 5-field sequences in which the 1st and 3rd fields are identical, it switches to film-mode de-interlacing.
Once it’s in film mode, the de-interlacer just combines fields 1 and 2 to make one progressive image, outputs that for 3 progressive frames, then combines fields 4 and 5 to make another progressive image, and outputs that for two frames. Then it repeats the process with the next 5 fields. The player is still outputting frames in a 3-2 pattern, but it’s creating 60 full progressive frames per second instead of 60 fields per second. Once the chip is in film mode, the de-interlacing algorithm is incredibly simple, and the complete film frame is recreated without loss or compromise.
The most common, and most distracting, artifact one encounters in film mode happens when the de-interlacer blithely combines together two fields that weren't meant to go together, usually because the 3-2 sequence is interrupted and the de-interlacer doesn't adapt quickly enough. When this happens, the odd numbered lines of the image are from one moment in time, and the even numbered lines are from a different moment. If something in the image is moving, it looks like there are spiky lines sticking out from the sides of the object like the tines of a comb. Hence the effect is usually called combing, though it is also sometimes referred to as feathering or zippering. Click here for a picture of what it looks like.
Sometimes, the cadence doesn’t stay regular. For various reasons (detailed later), the 3-2 cadence may break from time to time, or perhaps the video was never sourced originally from film. Documentaries, concerts, and made-for-TV material often is shot on video cameras, and then there is no good way to create perfect progressive frames. Video cameras capture 60 separate fields per second, and each one is separated in time, so moving objects are in a different position in each field.
In any case, if the de-interlacer doesn’t see a 3-2 film cadence, it must switch to video-mode de-interlacing. Here the algorithms get much tougher. There are two very simple techniques, neither of which is very good, and a host of progressively more complex algorithms. We’ll divide them into five large categories, roughly in order of complexity:
Single-field interpolation (or “Bob”)
This just involves taking each field and scaling it to a full frame. The missing lines between each of the scan lines in the field are filled in with interpolated data from the lines above and below. Done badly, the screen looks blocky and pixellated. Even done well, the image looks very soft, as image resolution is unavoidably lost. In addition, thin horizontal lines will tend to “twitter” as the camera moves. These thin lines will fall on just one field of the frame, so they will appear and disappear as the player alternates between the odd fields and even fields. This is the most basic de-interlacing algorithm, and the one that almost every de-interlacer falls back on when nothing else will work. Click here to see how single-field interpolation affects the image.
Field combining (or “Weave”)
In this technique, each pair of two consecutive fields is merged together to form a frame. This generally only works well if there is very little or no movement between the two fields, such as is the case with a single frame of film. If there is movement between fields, the image will have combing, which is very distracting. Hence very few de-interlacers use this as a primary algorithm.
This is a technique used by most software PC DVD players. Because software decoders generally don’t have enough horsepower to do motion adaptive de-interlacing and MPEG-2 decoding at the same time, the players use shortcut techniques to get reasonable-looking results. Most commonly, they weave together pairs of fields that are stored together as MPEG frames, and soften the image slightly in the vertical direction so any combs that result will look more like double images than combs. This causes loss of vertical resolution and bizarre-looking jitter on movement and pans. Once you notice it, it can become difficult to watch. See below for pictures of what this technique looks like.
This is a whole class of algorithms that attempt to switch between different ways of de-interlacing depending on whether an area of the screen appears to be still or moving. If an area is still, the algorithm uses the image data from two fields and weaves them together, but for moving areas, the algorithm just interpolates (bobs) the current field. This preserves resolution on the still sections of the screen, where the viewer is most likely to notice it, and reduces combing on the moving sections of the screen, at the expense of resolution. Done well, this looks very good. Most good de-interlacers use some form of motion-adaptive algorithm.
This is something one generally only finds on very, very expensive de-interlacing solutions, and we mention it here for completeness. This involves doing elaborate image analysis to identify the moving areas of the image, and weaving together the same image from two fields, with individual areas shifted to compensate for the movement. It involves a lot of processing power, and is not found on any DVD player we know of.
What does "true" progressive mean?
Just about every DVD player manufacturer claims, in one way or another, that their player is the only “true” progressive player on the market, and claims that other solutions use some kind of primitive line-doubler. This is, not to put too fine a point on it, absurd. With the possible exception of some very low-cost progressive players that we didn’t review, all progressive players are capable of outputting the entire film frame, without compromise. They are all “true” progressive players. Whether the player reads the progressive frame directly off the disc, or recreates it with a de-interlacer in the digital domain, the end result is the same. What varies between the players is their video performance, and the ability to handle material that wasn’t encoded the “standard” way.
Progressive players vs. external de-interlacers
Most progressive players contain the same de-interlacing chips used in external de-interlacers like the Crystal Image, the Focus Enhancements Quadscan, and the Silicon Image iScan. The major advantage the progressive player has over the external de-interlacer is that in the player, the de-interlacing can be done to the video in the digital domain, using the digital video directly out of the MPEG decoder, without any intervening analog conversions. An external decoder must use the analog signal from the player, re-digitize it to feed into the de-interlacing chip, then convert it again to analog to feed the display. This extra set of A/D and D/A conversions inevitably reduces the display resolution and increases video noise. If it's done well, the image degradation is relatively minor. But all else being equal, the progressive DVD player will always produce a better 480p image than an interlaced DVD player connected to an external de-interlacer with the same chipset.
If the display device requires an external scaler for optimal display quality, because it doesn't display a good image with a 480p signal, then there's no real advantage to using a progressive DVD player. As long as the de-interlacing chip in the external scaler is of similar quality to the one in the DVD player, the final video quality should be essentially the same whether one feeds the scaler 480p or 480i. In addition, most external scalers don't accept 480p inputs, making the decision moot. What would be best is to either build the scaling into the player or add a digital output to the player, and a digital input to the external de-interlacer/scaler. We review one add-on product that adds scaling and de-interlacing to an interlaced player, and while there were some issues with that specific product, the idea is sound.
A look at some common chipsets
Genesis gmVLX1A-X & gmAFMC
This is by far the most common de-interlacing chipset used in progressive DVD players. The gmVLX1A-X is the actual de-interlacer, and has both a film mode and a video mode, as well as a “graphics” mode which is just a simple weave of every pair of fields. The main chip does not, however, have any way of analyzing the video to determine whether it should be in film mode or not, and that’s where the gmAFMC comes in. It uses data (provided by the gmVLX1A-X) to figure out if there is a 3-2 pulldown sequence coming in, and switches the main chip on the fly between video and film modes.
Not all players use both chips. Some players use their own strategy for deciding when to be in film mode and when to be in video mode, so they forgo the gmAFMC. We’ll mention that in the individual player reviews. But most players use the two chips as a set.
The video mode on the Genesis is not motion-adaptive as we have defined the term. It uses "vertical-temporal filtering" which appears to us to be a slightly more advanced version of the vertical filtering we mentioned above, with the current field providing more of the input to the finished frame than the next or previous fields. It switches to a simple weave algorithm when the image is still or near-still.
The Genesis chipset offers lots of useful options, including scaling 4x3 input to 16x9 output, and scaling any input signal to a wide range of output resolutions. Most DVD players don’t use these features – they are mainly designed for the projector and stand-alone scaler market.
The main Achilles heel of the Genesis chipset is that it doesn’t buffer more than one field, so it can’t look ahead to see cadence breaks coming in advance. Accordingly, the Genesis is almost guaranteed to comb on 3-2 pattern breaks. By the time it knows the cadence has broken, it has already sent out at least one bad frame. In addition, the Genesis is not very good at identifying or handling 2-2 cadence, which comes up more than you might think.
The players that use the Genesis chipset don’t all have exactly the same de-interlacing performance. There are tradeoffs that can be made by adjusting the sensitivity of the gmAFMC chip to cadence mismatches, and that causes some players to do better than others on certain kinds of material. Often, though, on other material the advantage is flip-flopped. There is no perfect setting.
Silicon Image Sil503 (DVDO DV103)
(DVDO recently was purchased by Silicon Image, so we'll refer to current chips by their Silicon Image names, and older chips by the DVDO names.) This chip (like its earlier incarnations, the DV101 and DV102/Sil502) is a single-chip solution that does both de-interlacing and mode detection. It does not scale the signal to different resolutions, but the latest revisions of the chip can do 4x3 to 16x9 conversion.
The Silicon Image chip buffers 4 fields at all times, which it uses for cadence analysis as well as motion analysis in video mode. As a result, it is much better than the Genesis at handling cadence problems, and combs very little. Its video-mode de-interlacing algorithm is motion-adaptive, and is substantially better than the one on the Genesis. It handles 2-2 cadence with no trouble. Overall, it’s one of the best performing de-interlacers on the market, comparable only to much more expensive products from Faroudja, and a few others.
Unfortunately, for reasons unknown, the Silicon Image chip is so far only available in very expensive boutique players, and in their own stand-alone line-doubler. The stand-alone iScan de-interlacing product is nice, but suffers from the inevitable loss of resolution that extra A/D and D/A conversions cause. We fervently wish that more manufacturers would start using this chip, and after you read the player reports, you will too.
National Semiconductor NDV8501
This is National’s new DVD-on-a-chip design, and combines MPEG decoding, de-interlacing, and video DACs all on one chip. Since everything is on one chip, it’s easy for the de-interlacing logic to take advantage of the flags in the MPEG stream, and in fact the chip does make use of the flags almost exclusively. This works well when the encoding and flags are fairly standard, but video-sourced material and material with non-standard encoding or incorrect flags are not handled as well.
The National’s video de-interlacing algorithm is not motion-adaptive, and uses either single-field interpolation or vertical filtering, depending on user selection and possibly the amount of change from field to field. This is a fairly watchable solution under many circumstances, but if there is a lot of motion in the frame, it starts to look very odd. It also looks stuttery during pans and zooms.
What can go wrong
Flag reading vs. cadence reading
Progressive DVD players can be divided into two basic types: flag-reading and cadence reading. Flag-reading players use the flags embedded in the MPEG stream to make decisions about what de-interlacing algorithm to use. Cadence-reading players ignore the flags, and analyze the content of the frames as they come out of the MPEG decoder to figure out which algorithms to use.
Flag-reading players have two major difficulties: they are often tripped up by errors in the MPEG stream, where the flags are not correct, and they drop to video mode if the encoding is non-standard, even on good film material. They generally have no way of doing 3-2 pulldown detection other than looking for the alternating "repeat_first_field" flag, so if it's missing, the film mode won't kick in, and you end up seeing the video-mode de-interlacing. This isn't unwatchable, but the main reason to get a progressive DVD player is to get film-mode de-interlacing. You can get video-mode de-interlacing from the de-interlacer in the TV. In general, we marked a test "fail" if the player went to video mode on material that was sourced from film.
Cadence-reading players are not tripped up by non-standard or missing flags, which is good, but they are often tripped up by glitches in the cadence. If the 3-2 sequence is interrupted, often the cadence-reading players will stay in film mode for a handful of frames, and the result on screen is combing, which is very distracting. Similarly, once a cadence-reading player is in video mode, it typically takes a few moments to re-synchronize to the cadence and return to film mode. During that time, the player is using the inferior video-mode de-interlacing when it perhaps doesn't need to. Our tests will tell you how long it took the player to return to film mode after being forced into video mode by a bad cadence break.
Alternating progressive flag
On some material sourced originally from film, the progressive_frame flag is on for one frame, then off for the next frame, then on for the next frame, and so on. This causes no problems for an interlaced display, but progressive decoders that use the flags religiously will switch every other frame between film-mode decoding and video-mode decoding, with results that range from almost unnoticeable, to very annoying. "Titanic" and "Austin Powers" are examples of discs with this problem. On the WHQL disc (see below for more info on WHQL), the Film Alternate 2 sequence has this problem.
No progressive flag
Some discs have no progressive_frame flag set on any pictures on the disc, and don’t use repeat_first_field. They just dump the film transfer with its inherent 3-2 pulldown onto the disc without any special flags, as though it was video. When this is done, 2 out of every 5 frames on the disc contain fields from two different frames of film (see the third MPEG flags table at the top of the article for a diagram). Again, a player that uses the flags exclusively will treat this material as video, and you lose the special film mode de-interlacing that makes a progressive player look so good. In the worst case, the player will combine fields that are stored in one MPEG frame, but came from two different film frames, causing combing. "Cirque Du Soleil: Quidam" and the trailer on the "Galaxy Quest" disc are two examples of this problem.
Progressive flag on non-progressive material
This is a rare problem, but one that does crop up. This is when material that came originally from video, or film that has not been transferred correctly, has the progressive_frame flag set on every picture in the MPEG stream, even though some or all of the fields come from different points in time. This particular problem causes players that read the flags to comb badly (see below for pictures of combing). The "Galaxy Quest" main menu intro, "The Big Lebowski" making-of documentary, and the "Apollo 13" making-of documentary all have this problem, as does the WHQL Mixed Mode Alternate 1 sequence.
3-3 and 2-2 sequences
This usually happens when two MPEG pictures in a row have the repeat_first_field flag set the same, either true or false. If they are both true, you get a 3-3 sequence. If they are both false, you get a 2-2 sequence. Players that look exclusively at the cadence can be tripped up by both of these problems, if they continue to stay in film mode for a few frames even after they detect a cadence problem. This problem is sometimes caused by something like seamless branching, where the end of one segment ends with a 3 frame, and the next segment begins with a 3 frame. Most encoders won’t catch a problem like that. Also, some DVD encoders will do odd things to the flags right around a chapter stop, like switching to a 2-2 cadence, or dropping the progressive-frame flag, or both. "The Big Lebowski" making-of documentary and WHQL Chapter Stops sequences have examples of both of these cadence issues. (Note: an encoder is used in manufacturing the DVD, while the decoder is used in the DVD player.)
On certain kinds of material, mainly documentaries about film and making-of or behind-the-scenes videos, sections of video are interspersed with sections of film. Players that use cadence matching may see the film cadence and settle into the pattern, only to have a problem when the pattern abruptly ends. Almost all making-of documentaries demonstrate this problem, and in fact many cadence-reading players comb on all such supplements. "The Big Lebowski" making-of video demonstrates this kind of problem, but in general all making-of documentaries will have the same set of issues.
Shot on film, edited on video
A fairly large amount of material these days is shot on film to get a film “look,” then transferred to video for editing and other post-production, as editing and post are much cheaper for video. Almost all episodic TV, music videos, and made-for-TV movies are done this way. This is a torture test for cadence-reading de-interlacers, as each scene will have the 3-2 cadence internally, but edits will more often than not break it, since the makers didn’t care about keeping the film cadence intact. In our test suite, we used "More Tales of the City", which is a textbook example of this kind of material. However, you will find that most music videos and made-for-TV material will have the same problem.
This is really a special case of the above problem, where a documentary will pull together segments of film and edit them together on video. This often happens in trailers and ads designed for showing on TV, and in making-of documentaries, where segments from different parts of the film are edited together for effect. Again, the 3-2 cadence will generally break on more the half the cuts.
Chroma upsampling error (Streaky saturated colors)
On many progressive players we looked at, there were strange streaks running through the edges of saturated color fields, especially bright red. In other places, it looked like the red areas were blocky or had stair-stepped edges. It turns out that this is not a de-interlacing problem at all, but it shows up much more clearly when the video is converted to progressive.
DVDs store luma, which is the black and white signal, and chroma, the color signal, separately. The chroma signal has half the resolution in both directions, so for a 720x480 frame, the chroma information is only 360x240. Since the eye is much more sensitive to luma changes, the much lower chroma resolution looks fine. However, at some point the MPEG decoder on the DVD player must convert the 360x240 chroma signal to a 720x480 (or 360x480) signal so it can be sent on to the video DACs. This conversion is being done incorrectly on a surprisingly wide variety of DVD players. The short version is that there is a bug in the 4:2:0 to 4:2:2 conversion that happens inside of the MPEG decoder. We have informed the manufacturers about this bug, and await their response.
Different DVD players show the problem to varying degrees, and different DVDs will show this problem more than others. However, we were able to look directly at the chroma channel without the luma channel, which makes it easy to see what the player is really doing with the chroma. And even though the problem is most visible on saturated reds, it’s actually there on every part of the screen. The problem can make normally smooth edges look stairstepped or streaky. In the worst cases it looks like there are black horizontal lines running through bright-colored objects.
What’s amazing about this problem is that it’s been right under everyone’s nose for years. Lots of different DVD players have the problem, and it’s not limited to progressive players. We have been able to see the problem even on interlaced displays. We can only speculate as to why it has gone largely unnoticed until now. For one thing, it’s harder to see on an interlaced display. For another, the eye is less sensitive to chroma (which is why DVDs can get away with one quarter the resolution for the chroma channel). Several popular de-interlacers hide the problem, like Sony’s DRC and the Faroudja units, probably by smoothing the chroma channel (which is not fixing the problem, it’s just papering over it). And last, unless you know what the picture is supposed to look like, it would be easy to assume the artifact is in the DVD. Since most people have only one DVD player, they can’t switch back and forth and really see the differences.
Click here to see a picture of what this artifact looks like.
What it looks like when things go wrong
Look at the picture below, and note the way there are spiky lines, like the tines of a comb, on the sides of the moving person in this frame. This is because the first field (the odd numbered scan lines) is from one frame of film, and the second field (the even numbered scan lines) is from the next frame of film. The image has been zoomed 200% so you can see the details.
This picture (below) is the same frames used for the “combing” demo above, but with the vertical filtering method turned on. You can see that the combs have been converted into something more like a double image. This makes the combed frames less distracting, but it’s still visible.
Below, on the left, is a full film frame, with both fields combined together (weave) to make a complete progressive frame. The right image is just the first field of the frame, with the missing scan lines interpolated. Note the loss of resolution and the increased stair stepping on the diagonal lines. Again, the images are zoomed 200%
Weave Single-field interpolation
Chroma Upsampling Error (Streaky Saturated Colors)
Here is an image, below, with the correct decoding on the left, and the “bad” decoding on the right. Look particularly at the blue around the word "Toy". See the horizontal streaks? Note that the “bad” example is not necessarily what you will see on all DVD players. Some players may look better or worse, though the nature of the problem and the places it appears will be the same.
These next two are the same images with the luma removed, so you can see clearly the artifacts in the chroma channels. Again, notice the horizontal streaks. This bug is in a lot of players now, due to a problem with the MPEG decoder algorithms. We hope that this report will stimulate a fix.
Evaluation Equipment and Process
We looked at all of the progressive DVD players using a Sony VW10HT LCD projector, which was selected because of its ability to handle a wide variety of different signal types and its very high resolution display. The projector’s one major weakness is a lack of a very dark black level, but that didn’t affect the tests we were making. We used custom-made Canare LV-61S cables for most players, and a custom VGA to 5 RCA cable for the PC DVD players and the iScan Plus.
Players were scored on a sheet that had a pass/fail designation for each of the tests, plus areas for entering numerical scores for certain tests, and notes for anything unusual or interesting. All tests were witnessed by at least two people. For looking at the chroma separate from the luma, we used an iScan Plus V2, because it produces H and V sync signals even with Y’Pb’Pr’ output, and the 10HT is able to sync to external H and V sync even in Component mode, so we were able to remove the Luma cable and still have a stable picture.
What we looked at
WHQL DVD Test Annex 2.0
This disc is produced by Microsoft, specifically by the Windows Hardware Quality Lab. It’s designed for testing PC-based DVD players. But, it makes a great test disc for finding problems with freestanding DVD players. Since PC DVD players are all inherently progressive, much of the test material on WHQL is for testing de-interlacing and other progressive issues. It’s by far the most revealing single disc available for testing how well a player handles various real-world “bad” material.
This is the most basic film-based test imaginable. The flags are correct, and the 3-2 cadence is correct. If a player can’t handle this test, it can’t handle film-based DVDs correctly.
A player passes this test if it stays in film mode for the entire test.
This is the same piece of material as Film 1, but encoded with a different MPEG-2 encoder. In this version, the progressive_frame flag toggles off and on, every frame. It’s set to “true” for the first frame, then “false” for the next frame, and so forth. Players that ignore the flags and just look at the cadence will handle this just as well as Film 1. Players that do look at the flags may trip up and comb or fall into video mode.
Again, to pass this test the player needs to stay in film mode for the whole sequence.
Mixed Mode 1
This is a sequence that switches every few seconds between film-based (24 frames, encoded with 3-2 pulldown) material and video-based (60 individual fields). The flags are all correctly encoded, and the cadence is correct for the film-based sections. A player passes this test if it doesn’t comb when the material switches from film to video, and if it takes no more than 3 frames to switch back to film mode when the cadence resumes.
For players that didn’t pass the test, we estimated the number of frames (progressive video frames, 60 per second) it took to get back into film mode from video mode.
Chapter Stops 1
This is a sequence with 4 chapter stops encoded in the middle of it. The actual existence of a chapter stop on the disc does not itself cause a problem, but placing a chapter stop in the disc mastering software causes many MPEG encoders to do strange things for a few frames before and after the chapter stop. In this case, the encoder stops encoding repeat_first_field and progressive_frame entirely for about 8 frames before and 4 frames after the chapter stop, causing both cadence and flag problems. There are many discs, including "Titanic", that have exactly this problem on almost every chapter stop.
To pass this test, a player has to stay in film mode for all four chapter stops. If the player freezes on a bad frame at the very end of the section, while waiting for the menu to load, that was not counted as a problem. In cases where the player didn’t pass, we estimated the number of frames it took to return to film mode.
Chapter Stops 2
This has a smaller problem, where the encoder drops the progressive_frame and repeat_first_field flags for somewhere between 0 and 3 frames after every chapter stop. Passing conditions are the same as the first Chapter Stops test.
Video Essentials: Snell & Wilcox Zone Plate
This sequence is great for looking at whether the de-interlacing algorithm is motion-adaptive, and if so, how good it is at isolating moving areas from stationary ones. There is a moving circular pattern in the center of the test, which is moving with a film cadence during the first half of the test, and a video cadence in the second half. However, we were watching the thin almost-horizontal lines on the left side of the image, as these are completely motionless, and should therefore stay exactly the same throughout the test. If the lines flickered, softened, or in any way changed during the test, it shows that the de-interlacer is not good at isolating stationary areas of the screen from moving areas, and you will lose resolution in the background whenever something is moving in the foreground.
To pass this test, the player had to reproduce the left-hand thinnest line pattern with full resolution, with no change throughout the test. Again, a change on the very last frame before the chapter stop was not counted as a problem.
Big Lebowski Making-of Featurette
We just looked at the opening montage of film clips, up until the first shot of the Coen brothers talking, and counted the number of times it combed. This is a perfect example of a film montage in which a number of segments of film that have the 3-2 cadence have been edited together, causing cadence breaks on most of the cuts. The progressive_frame flag is on for the whole sequence, even though some of the frames should not have it turned on, so this sequence combs even on players that read the flags.
To pass this test, the player has to get through the whole opening without combing. For the ones that failed, we counted the number of times we saw a combed frame.
Galaxy Quest Menu
This is the opening flyby of the spaceship that leads into the main menu of "Galaxy Quest". It is film, and was created with 3-2 pulldown, but then transferred onto the DVD without the repeat_first_field flag (see our third example of legal flag sequences), so 2 out of every 5 frames are not correct progressive frames. However, the progressive_frame flag has been turned on for every frame, which confuses many players that read the flags off the disc.
To pass this test, the player has to go into film mode, and not comb.
Galaxy Quest Trailer
This is the theatrical trailer, in the “special features” section of the disc. It is, like the menu lead-in mentioned above, transferred to disc without using the repeat_first_field flag, so 2 of every 5 frames are not properly progressive (see our example above). In this case, the progressive_frame flag is off for the whole running time, so it makes a good contrast to the first example from this disc. They both have the same cadence, but one has progressive_frame set to “true,” and one has it set to “false.” A good player will ignore the flags and treat both the same.
To pass this test, the player has to go into film mode, and not comb.
More Tales of the City
This whole miniseries was shot originally on film, probably because of the film “look,” then transferred to video and edited on video. For this reason, it has a cadence break on most of the cuts, and hence combs all the time on many players.
To pass this test, the player has to go into film mode, and not comb. Minor drops from film mode here and there on cuts were not considered a problem.
Titanic Chapter 7->8
The whole of "Titanic" (like "Austin Powers 1" and the first release of "Terminator 2") has alternating progressive_frame flags throughout the film. This is the same issue that WHQL Film 2 demonstrates. In addition, many of the chapter stops in the film have the same strange cadence and flag breakdowns that are evident in WHQL Chapter Stops 1. To test both of these issues, we went to the very end of chapter 7 and watched the transition to chapter 8. The beginning of Chapter 8 of "Titanic" is a great place to look for de-interlacing artifacts, as you have a bunch of thin diagonal lines (the railings, the deck boards), and a very slow camera move, which is a really hard scene to handle well for even a motion-adaptive de-interlacer. So if the player drops to video mode here, it’s very evident.
A passing grade here requires that the player have no more than a one or two frame drop to video mode on the chapter break, and no combing.
Apollo 13 Making-of
This making-of featurette is a rarity – video marked progressive. The whole thing, practically, is video, but the progressive_frame flag is set to “true” the whole time. On players that read the flags, this can cause severe combing throughout.
To pass this test, the player must go into video mode, and not comb.
Toy Story Chapter 4
This is a great place to see the chroma upsampling problem (streaky saturated colors). There are lots of bright red objects in this scene, including the top of the Tinkertoy canister, the top of the microphone Woody is holding, the fire engine, the hats on the little firemen, the red top of the notepad Woody is using, the Operation box, and the letters on the Mouse Trap box. Once you know what to look for, you can see if a player has the problem within 3 seconds of the beginning of this scene, but we watched the first minute or so to get an idea of how bad the problem was.
For us to give a passing grade, we had to see no evidence of the chroma upsampling problem in the first minute of chapter 4. We also tried to gather a subjective impression of how bad the streakiness was.
Here are links to reviews we have done using the above tests:
First Progressive DVD Shootout, Dec. 2000
- Don Munsil -
Dolby Logo Copyright Dolby Technologies
"Return of the Jedi" Copyright 1983, Lucasfilm
WHQL Images Copyright 1999, Microsoft Corporation
"Toy Story" Images Copyright 1995, Pixar and Disney
"Titanic" Images Copyright 1997, Paramount/Fox
"Galaxy Quest" Copyright 1999, Dreamworks
Go to Part 3 - Functionality Part 4 - Usability Part 5 - Progressive Scan
© Copyright 2000 Secrets of Home Theater & High Fidelity
Return to Table of Contents for this Issue.