|
|
All the news now and in your digital future
OT: 60i vs 30psf, or "Why does my footage look like butt?"
Posted by Jeff Harrell
|
OT: 60i vs 30psf, or "Why does my footage look like butt?" March 22, 2010 06:30PM |
Once again, something's come up a few times lately on the forum, buried down in other threads, and I thought I'd drop a one-time informational post in here for everybody.
Often the question arises, "Why does my 30p footage look all strobey and jittery? I thought progressive was supposed to be better?"
The answer is that what everybody calls "30p" really ought more properly be called "30psf." Actually, it really ought to be called "29.97psf," but that's just too dang much typin'.
The "psf" in "30psf" stands for "progressive segmented frame." It's easiest to define this piece of jargon if we work backwards, and define what it's not.
Progressive-scan video, also sometimes called "frame-based" video, is video in which every frame is a single raster of pixels, so-and-so many pixels high by so-and-so many pixels wide. Other than the pixels themselves, there's no substructure to a progressive-scan frame. It's a complete unit, indivisible unless you want to talk about individual pixels. Each frame is notionally a single exposure of the entire sensor. (In reality, there's a slight delay between when the top row of pixels is scanned and when the bottom row is scanned; this is the cause of the much-maligned "rolling shutter" problem, also known as "jellocam."
The most notable examples of progressive-scan recording are 24p, as in 1080p24 for example, or 60p, as in 720p60. In point of fact, we're about to learn that these are the only significant examples of progressive-scan recording in the world. There are actually no other common progressive-scan formats. (Emphasis on the "common," please, for the pedants out there. There are others, but we're keeping things simple here.)
Interlaced video, on the other hand, has no frames. Instead, it's a sequence of fields. Each field is one exposure of half of the rows of pixels on the sensor. In interlaced video, we call a row of pixels a "scan line" for historical reasons. Digital video doesn't really distinguish between rows of pixels and columns of pixels, much, but we in TV do.
As everybody knows, NTSC runs at a base rate that we call 60 Hz, accepting that that's shorthand for an odd number that rounds off to about 59.94. In PAListan, the base rate is 50 Hz ? and they really mean 50. PAL people don't truck with oddball fractions.
For purposes of discussion here, I'm using the 59.94 base rate. If you live in a PAL country or just hate me, personally, feel free to mentally substitute "50" wherever you see "59.94."
Interlaced video is recorded like this:
At 1/59.94th of a second into the recording, we record the first field. In this example, I chose the odd field, meaning scan lines 1, 3, 5 and so on. This is the dominant field for HD video. SD video records the lower field first, but it really doesn't matter for purposes of illustration.
One 59.94th of a second later, the camera records the second field, which is the even field. That is, scan lines 2, 4, 6 et cetera.
Then at 3/59.94ths of a second, the camera records the third field, which means the odd scan lines again. And so on.
As you can see, the object in front of the camera (in this case, a pretty blue dot) is in a different position when each field is recorded. That's because the object is moving, and each field is recorded at a different instant of time, separated by 1/59.94th of a second.
When you play back interlaced footage ? and this is important ? the fields are drawn to the screen with the same timing with which they were recorded. In other words, the first field is drawn, then 1/59.94th of a second later the second field is drawn, then 1/59.94th later the third field, and so on in that fashion. It looks like this:
If this looks like a very slight motion blur, that's 'cause it kind of is. That's how your eye interprets it when you watch interlaced footage. That's what makes interlacing work, when played back properly. On a CRT television or a good LCD, each scan line will start to fade as soon as it's been drawn. Because the overall brightness of the picture is refreshed 60 times a second, there's no perceptible flicker. And because each field fades away before those scan lines are drawn over, interlaced video creates the illusion of very high-resolution smooth motion. It's kind of neat, to be honest.
But we're not here to talk about interlaced video. We're here to talk about 30psf ? though what I'm about to describe applies as well to 25psf, in PAL land. The relationship between 50i and 25psf is the same as the relationship between 60i and 30psf.
So I said progressive-scan video records the entire sensor at once (modulo read time), and also that frames of video in the progressive-scan domain are always treated as whole frames. Then I said that interlaced video records every other line of the sensor to create a sequence of fields which are played back with the same timing with which they were recorded.
Progressive-segmented-frame takes these two ideas and smooshes 'em up together.
Originally, PsF (as it's known) recording was invented to deal with film-transferred-to-video. It was discovered ? I think by some guys at Sony, though I'm not positive about that ? that if you take a PAL VTR and retime it slightly from 50 Hz to 48 Hz, you can record film on it as if the film were interlaced video. That is, each frame of film becomes a pair of fields, and those fields are written to tape and played back as if they were ordinary video, albeit running at a slower base rate.
This breakthrough discovery was absolutely useless for actually watching this material. Play back a 24psf tape on a monitor that'll sync to 48 Hz and you'll see something truly appalling. But nobody used 24psf that way. Instead, 24psf was used to transport frames of film from the lab to the editing machine. If you hooked a 48 Hz PsF VTR up to your Avid, you could capture the fields from the videotape, then reassemble them in the computer into frames which would play back at 24 frames per second. This was great for editing film, 'cause you no longer had to deal with 3:2 pulldown.
Sometime later, somebody got the bright idea of applying the same principle to existing 50 and 60 Hz equipment. What if, they thought, we record both fields at the same time, like we would if we were doing a progressive-scan recording, but wrote them out to tape as if we were recording interlaced video?
And thus, 25psf and 30psf were born. But I guess for marketing reasons, or just due to natural suffix-erosion, the "sf" part got dropped, and those formats came to be known as 25p and 30p.
But these are not progressive-scan formats. This is really important to understand. They are progressive segmented frame formats, which makes them a thing apart and unique to themselves.
You record 30psf video just exactly like you're imagining: Read out the whole sensor, top to bottom, at once. You end up with two fields, both recorded at the same time, like this:
Most cameras ? I think all of them, but I can't swear to it ? do some filtering to minimize scan-line jitter when recording 25psf or 30psf. In other words, a slight vertical blur is performed to reduce the chance that any fine detail appears on just one of the two fields of any given "frame." But again, I can't swear that all cameras do this. Just all the ones I'm familiar with.
Note that absolutely nothing is recorded at 2/59.94ths and 4/59.94ths. If we were shooting interlaced, we'd get one field every 1/59.94th of a second, but since we're shooting psf, we get two fields every other 59.94th of a second. The subject in motion moves further, then, between each successive pair of fields. A nerd (hey, Strypes) would say a this point that the temporal resolution has been halved. The spatial resolution has also been somewhat reduced, because of that jitter filtering I mentioned.
But the biggest idiosyncrasy of 25psf or 30psf comes when we look at how it's played back on your television. Remember, your television thinks (for lack of a better word) you're looking at ordinary 50i or 60i material. So it plays it back as it that's just what it were:
You see that really funny thing going on at 3/59.94ths? We're getting half of the first frame, and half of the second frame. It's not until we get to 4/59.94ths that we see the whole second frame. If I drew the illustration wider, we'd see that every other frame is like this. In nerd-speak, these special intermediate frames that are built up of a field from one recorded frame and a field from another recorded frame are called "jitter frames." (If you're familiar with 3:2 pulldown, the BC and CD frames are your jitter frames.)
When you play back 25psf or 30psf material on a 50i or 60i monitor, half your frames are jitter frames.
This is why your footage looks "stroboscopic" or "jittery." Because it is.
Now, not all monitors (unfortunately) treat 50i and 60i material the same way. Some LCD televisions attempt to deinterlace interlaced material, so they can display whole frames. Some do a very poor job at this, and some are kind of okay. The newest televisions have this weirdo motion estimation feature that interpolates a lot more frames than are actually in your source material. It seems, from what I've heard, that nobody likes these televisions one damn bit, and since I don't own one, I can't guess how 25psf or 30psf material would fare on one.
On a computer, of course, 25psf or 30psf material is played back as if it were 25p or 30p. A computer doesn't know from interlacing; it just throws up pairs of fields at once. So in the Avid or in Final Cut, or when watching a Quicktime movie, 25psf or 30psf material looks just fine. Which is probably why people so often are shocked and dismayed to see their 25psf or 30psf stuff on television. Because TV works differently.
How do you avoid this stuttery, jittery look? Well, there's really only one way: open up your shutter. If you have a very long shutter (say, 1/30th, which would equate to 360° if it were possible to have a 360° leaf shutter), then the motion blur in your footage will reduce the perceptible jitter during playback. Of course, then you have two problems: lessened but still extant jitter, and also smeary, streaky looking footage. But if you have no choice, then it's at least something to do some test shots with.
Everybody is entitled to his or her own opinion of 25psf/30psf recording. I've got mine, and I make no secret of it. But whatever your opinion, it's important to understand the basics of what these formats are, and how they work under the hood, so you can make an informed choice about whether to use them or not.
Often the question arises, "Why does my 30p footage look all strobey and jittery? I thought progressive was supposed to be better?"
The answer is that what everybody calls "30p" really ought more properly be called "30psf." Actually, it really ought to be called "29.97psf," but that's just too dang much typin'.
The "psf" in "30psf" stands for "progressive segmented frame." It's easiest to define this piece of jargon if we work backwards, and define what it's not.
Progressive-scan video, also sometimes called "frame-based" video, is video in which every frame is a single raster of pixels, so-and-so many pixels high by so-and-so many pixels wide. Other than the pixels themselves, there's no substructure to a progressive-scan frame. It's a complete unit, indivisible unless you want to talk about individual pixels. Each frame is notionally a single exposure of the entire sensor. (In reality, there's a slight delay between when the top row of pixels is scanned and when the bottom row is scanned; this is the cause of the much-maligned "rolling shutter" problem, also known as "jellocam."
The most notable examples of progressive-scan recording are 24p, as in 1080p24 for example, or 60p, as in 720p60. In point of fact, we're about to learn that these are the only significant examples of progressive-scan recording in the world. There are actually no other common progressive-scan formats. (Emphasis on the "common," please, for the pedants out there. There are others, but we're keeping things simple here.)
Interlaced video, on the other hand, has no frames. Instead, it's a sequence of fields. Each field is one exposure of half of the rows of pixels on the sensor. In interlaced video, we call a row of pixels a "scan line" for historical reasons. Digital video doesn't really distinguish between rows of pixels and columns of pixels, much, but we in TV do.
As everybody knows, NTSC runs at a base rate that we call 60 Hz, accepting that that's shorthand for an odd number that rounds off to about 59.94. In PAListan, the base rate is 50 Hz ? and they really mean 50. PAL people don't truck with oddball fractions.
For purposes of discussion here, I'm using the 59.94 base rate. If you live in a PAL country or just hate me, personally, feel free to mentally substitute "50" wherever you see "59.94."
Interlaced video is recorded like this:
At 1/59.94th of a second into the recording, we record the first field. In this example, I chose the odd field, meaning scan lines 1, 3, 5 and so on. This is the dominant field for HD video. SD video records the lower field first, but it really doesn't matter for purposes of illustration.
One 59.94th of a second later, the camera records the second field, which is the even field. That is, scan lines 2, 4, 6 et cetera.
Then at 3/59.94ths of a second, the camera records the third field, which means the odd scan lines again. And so on.
As you can see, the object in front of the camera (in this case, a pretty blue dot) is in a different position when each field is recorded. That's because the object is moving, and each field is recorded at a different instant of time, separated by 1/59.94th of a second.
When you play back interlaced footage ? and this is important ? the fields are drawn to the screen with the same timing with which they were recorded. In other words, the first field is drawn, then 1/59.94th of a second later the second field is drawn, then 1/59.94th later the third field, and so on in that fashion. It looks like this:
If this looks like a very slight motion blur, that's 'cause it kind of is. That's how your eye interprets it when you watch interlaced footage. That's what makes interlacing work, when played back properly. On a CRT television or a good LCD, each scan line will start to fade as soon as it's been drawn. Because the overall brightness of the picture is refreshed 60 times a second, there's no perceptible flicker. And because each field fades away before those scan lines are drawn over, interlaced video creates the illusion of very high-resolution smooth motion. It's kind of neat, to be honest.
But we're not here to talk about interlaced video. We're here to talk about 30psf ? though what I'm about to describe applies as well to 25psf, in PAL land. The relationship between 50i and 25psf is the same as the relationship between 60i and 30psf.
So I said progressive-scan video records the entire sensor at once (modulo read time), and also that frames of video in the progressive-scan domain are always treated as whole frames. Then I said that interlaced video records every other line of the sensor to create a sequence of fields which are played back with the same timing with which they were recorded.
Progressive-segmented-frame takes these two ideas and smooshes 'em up together.
Originally, PsF (as it's known) recording was invented to deal with film-transferred-to-video. It was discovered ? I think by some guys at Sony, though I'm not positive about that ? that if you take a PAL VTR and retime it slightly from 50 Hz to 48 Hz, you can record film on it as if the film were interlaced video. That is, each frame of film becomes a pair of fields, and those fields are written to tape and played back as if they were ordinary video, albeit running at a slower base rate.
This breakthrough discovery was absolutely useless for actually watching this material. Play back a 24psf tape on a monitor that'll sync to 48 Hz and you'll see something truly appalling. But nobody used 24psf that way. Instead, 24psf was used to transport frames of film from the lab to the editing machine. If you hooked a 48 Hz PsF VTR up to your Avid, you could capture the fields from the videotape, then reassemble them in the computer into frames which would play back at 24 frames per second. This was great for editing film, 'cause you no longer had to deal with 3:2 pulldown.
Sometime later, somebody got the bright idea of applying the same principle to existing 50 and 60 Hz equipment. What if, they thought, we record both fields at the same time, like we would if we were doing a progressive-scan recording, but wrote them out to tape as if we were recording interlaced video?
And thus, 25psf and 30psf were born. But I guess for marketing reasons, or just due to natural suffix-erosion, the "sf" part got dropped, and those formats came to be known as 25p and 30p.
But these are not progressive-scan formats. This is really important to understand. They are progressive segmented frame formats, which makes them a thing apart and unique to themselves.
You record 30psf video just exactly like you're imagining: Read out the whole sensor, top to bottom, at once. You end up with two fields, both recorded at the same time, like this:
Most cameras ? I think all of them, but I can't swear to it ? do some filtering to minimize scan-line jitter when recording 25psf or 30psf. In other words, a slight vertical blur is performed to reduce the chance that any fine detail appears on just one of the two fields of any given "frame." But again, I can't swear that all cameras do this. Just all the ones I'm familiar with.
Note that absolutely nothing is recorded at 2/59.94ths and 4/59.94ths. If we were shooting interlaced, we'd get one field every 1/59.94th of a second, but since we're shooting psf, we get two fields every other 59.94th of a second. The subject in motion moves further, then, between each successive pair of fields. A nerd (hey, Strypes) would say a this point that the temporal resolution has been halved. The spatial resolution has also been somewhat reduced, because of that jitter filtering I mentioned.
But the biggest idiosyncrasy of 25psf or 30psf comes when we look at how it's played back on your television. Remember, your television thinks (for lack of a better word) you're looking at ordinary 50i or 60i material. So it plays it back as it that's just what it were:
You see that really funny thing going on at 3/59.94ths? We're getting half of the first frame, and half of the second frame. It's not until we get to 4/59.94ths that we see the whole second frame. If I drew the illustration wider, we'd see that every other frame is like this. In nerd-speak, these special intermediate frames that are built up of a field from one recorded frame and a field from another recorded frame are called "jitter frames." (If you're familiar with 3:2 pulldown, the BC and CD frames are your jitter frames.)
When you play back 25psf or 30psf material on a 50i or 60i monitor, half your frames are jitter frames.
This is why your footage looks "stroboscopic" or "jittery." Because it is.
Now, not all monitors (unfortunately) treat 50i and 60i material the same way. Some LCD televisions attempt to deinterlace interlaced material, so they can display whole frames. Some do a very poor job at this, and some are kind of okay. The newest televisions have this weirdo motion estimation feature that interpolates a lot more frames than are actually in your source material. It seems, from what I've heard, that nobody likes these televisions one damn bit, and since I don't own one, I can't guess how 25psf or 30psf material would fare on one.
On a computer, of course, 25psf or 30psf material is played back as if it were 25p or 30p. A computer doesn't know from interlacing; it just throws up pairs of fields at once. So in the Avid or in Final Cut, or when watching a Quicktime movie, 25psf or 30psf material looks just fine. Which is probably why people so often are shocked and dismayed to see their 25psf or 30psf stuff on television. Because TV works differently.
How do you avoid this stuttery, jittery look? Well, there's really only one way: open up your shutter. If you have a very long shutter (say, 1/30th, which would equate to 360° if it were possible to have a 360° leaf shutter), then the motion blur in your footage will reduce the perceptible jitter during playback. Of course, then you have two problems: lessened but still extant jitter, and also smeary, streaky looking footage. But if you have no choice, then it's at least something to do some test shots with.
Everybody is entitled to his or her own opinion of 25psf/30psf recording. I've got mine, and I make no secret of it. But whatever your opinion, it's important to understand the basics of what these formats are, and how they work under the hood, so you can make an informed choice about whether to use them or not.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 22, 2010 06:41PM |
Jeff, this is an ARTICLE, not a post. Why not talk to Ken Stone about getting these published on his site. Then I'll start linking to you!
And you might make some $$ too! People pay for stuff like this dude...
www.shanerosseditor.com
Listen to THE EDIT BAY Podcast on iTunes
[itunes.apple.com]
And you might make some $$ too! People pay for stuff like this dude...
www.shanerosseditor.com
Listen to THE EDIT BAY Podcast on iTunes
[itunes.apple.com]
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 22, 2010 06:46PM |
Meh. It's all the same, as far as Google is concerned. I just like sharing what little bits and pieces of trivia I've been able to pick up along the way.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 22, 2010 07:28PM |
Nice, Jeff. I always saw PsF as something you get whenever you float a progressive frame across a video system (eg. some SDI signals) for playback on analog equipment, and once that frame gets divided and stored as fields, you get PsF. There are some differences due to the way it is stored, eg. when you send a PsF signal into a monitor that can only accept a progressive signal (eg. the HP Dreamcolor), it won't work like it should.
Another difference would be the way the signal is compressed (eg. DCT based compression which is used by almost every digital video format), compression on progressive frames is easier and more efficient as the frame is split into blocks of 16 and 8. Interlaced DCT based compression is less efficient, as it requires the frame to be split fields and compressed to prevent smearing across fields which contain different temporal samples (eg. the reason why P-jpg compression should never be used on an interlaced signal).
One thing that is nice about having stuff like this on a forum, is that different people can contribute, and it's different from reading an article (eg. I still can't figure out why Charles Poynton mentioned that the CRT gamma response is "desirable", if we work to correct it right after it comes off the CCDs).
www.strypesinpost.com
Another difference would be the way the signal is compressed (eg. DCT based compression which is used by almost every digital video format), compression on progressive frames is easier and more efficient as the frame is split into blocks of 16 and 8. Interlaced DCT based compression is less efficient, as it requires the frame to be split fields and compressed to prevent smearing across fields which contain different temporal samples (eg. the reason why P-jpg compression should never be used on an interlaced signal).
One thing that is nice about having stuff like this on a forum, is that different people can contribute, and it's different from reading an article (eg. I still can't figure out why Charles Poynton mentioned that the CRT gamma response is "desirable", if we work to correct it right after it comes off the CCDs).
www.strypesinpost.com
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 01:22PM |
Didn't you just get a new job Jeff? How do you find the time for these dissertations??
When life gives you dilemmas...make dilemmanade.
When life gives you dilemmas...make dilemmanade.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 01:38PM |
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 03:56PM |
Jeff. This is an article. Can lafcpug publish it? Shane doesn't seem to know we have articles here too
Michael Horton
-------------------
Michael Horton
-------------------
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 04:07PM |
Quote
Shane doesn't seem to know we have articles here too
lol
When life gives you dilemmas...make dilemmanade.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 04:40PM |
Thanks for the explanation, Graeme. I never saw gamma take on that role in video.
www.strypesinpost.com
www.strypesinpost.com
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 04:49PM |
When Jeff and Graeme have dueling dissertations in the same thread...I get a little nosebleed.
When life gives you dilemmas...make dilemmanade.
When life gives you dilemmas...make dilemmanade.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 04:59PM |
Pretty sure Graeme's reply was meant to go into the gamma a bit down the page, rather than this one. But of course he's right. Gamma is to video kind of, if you squint as log is to film. Apply a transform function lets you get away with less luma precision because you're putting more bits where they do more good. But thinking about it in those terms always makes my head hurt. I'm a 16-bit kind of guy.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 05:17PM |
Quote
Remember that in PAL land (so that we can ignore the issues of 3:2 pulldown), film being shot at 25fps would be telecined as such to 50i video, effectively psf by any other name.
I'm always amused when I hear the process of converting progressive-scan to PsF described as "2:2 pulldown." Because it really means you do nothing, but it's really quite specific about exactly what type of nothing you do.
Quote
Strypes mentioned gamma above in reference to Poynton, so was answering on that aspect.
Oh, whoops. I totally missed that, and got my threads crossed. Pardon the pun.
Quote
How does material recorded at 59.94fps get recorded and played back?
It depends too heavily on your monitor to generalize. My television, for example, is a CRT, so it resyncs to 720p60 naturally. It shows me 60 frames per second, progressive-scan. Well. 59.94. I actually don't know how the average LCD handles it, or even if there is an average LCD in that respect.
Quote
How do these displays play back 29.97fps?
That's kind of the point. There is no conventional broadcast format that records 29.97 frames per second. The 30psf format still records 59.94 fields per second, just like ordinary 60i does, the major difference being that those fields are read out of the sensor at the same time instead of 1/59.94th apart.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 07:10PM |
Is anybodies else's head about to explode?
Michael Horton
-------------------
Michael Horton
-------------------
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 23, 2010 10:54PM |
Yeah Mike...uh-oh....(CUT TO: GRAFIXJOE SHAKING VIOLENTLY, then...)
When life gives you dilemmas...make dilemmanade.
When life gives you dilemmas...make dilemmanade.
|
Re: OT: 60i vs 30psf, or "Why does my footage look like butt?" March 24, 2010 11:30AM |
Yeah, boy!! You spin me right-round!!
When life gives you dilemmas...make dilemmanade.
When life gives you dilemmas...make dilemmanade.
Sorry, only registered users may post in this forum.
