29.97 and 23.98 SMPTE timecode -- The definitive explanation

[jacobfarron]

Let me start this by saying I have Pro Tools for video, film, and multimedia / Dialog Editing for Motion Pictures / Sound for Digital Video, and many other books. NONE have completely described why 23.98 lines up with 29.97, and the methodology behind it. And if they do, I have yet to comprehend it. I've read all the posts here, and the chapters dealing with timecode over and over and could not get it. Now I believe I do, it's all about the time reference! So, here's my entire explanation, compiled from various books and my own figurin':

Film speed is 60Hz – That is the length of one second for Film.
Film is shot at 24fps referenced to 60Hz.
24fps / 60Hz

Originally, black and white video was 30fps, referenced to 60Hz. Since it’s interlaced, each frame is composed of 2 fields, for a total of 60 fields.
30fps / 60Hz

NTSC color video was changed to 29.97fps, so there must be a new reference.
30fps / 60hz = 29.97fps / X (solve for X)
X = 59.94Hz
So, the new reference for NTSC is 59.94 HZ – This is the length of one second for NTSC, as well as the amount of fields per second.

Now, most video monitors will only display 29.97fps. To allow film to be played at this rate, first it must be referenced to video time instead of film time (NONE of my books described it in this way)
24fps / 60Hz = X / 59.94Hz (solve for X)
x = 23.976 (This is where the 23.98 number comes from)
This change in reference of time (60Hz verses 59.94Hz) is where the .1% slowdown comes from, NOT from a 2:3:2 pulldown.

Now, 23.976fps (aka 23.98fps) can be edited on a 23.98fps timeline with 23.98 timecode in Final Cut Pro, and edited in Pro Tools with a 23.98 session Time Code rate. HD video is often done this way as HD monitors can display 23.98fps, though I read film is usually edited in 29.97fps. So, film takes an additional step. After the .1% slowdown due to reference change, Film is generally pulled down via the 2:3:2 method which maps the first film frame to 2 fields of video, the next frame of film to 3 fields of video, and so on. Thereby fitting 23.98 progressive film frames on 29.97 interlaced video frames. There are duplicate fields when this happens. But, this allows Film to be displayed on a standard NTSC monitor.

Now, if an HD video camera wants the “film look”, they shoot 24P (which, unless it’s the Varicam or the like) is 23.98, and no speed adjustment is necessary before pulldown. If the camera shoots a true 24fps just like film, it must be slowed by .1% to simulate the effect of changing reference from 60hz to 59.94hz, to arrive at 23.98. This 23.98fps can either be left alone and played on HD monitors or computer monitors, or pulled down to 29.97fps, and played on any monitor – the pulldown only changes the frame count, not the speed. 29.97fps pulled from 23.98 has duplicate fields, whereas 23.98fps does not.

How audio is concerned:
When shooting HD 23.98fps, audio recorded separately (if not on the camera) is 48k at 29.97fps, because there is never a change in reference (59.94Hz). Thus, a .1% slowdown never needs to occur, even if the 23.98fps video is pulled down to 29.97fps – again, only the frame count changes.

For a film shoot, or TRUE 24fps, production audio is recorded at 48k and 30fps (60Hz reference aka pilot tone), when it’s slowed .1% it becomes 47,952Hz and it’s reference is now 59.94Hz, which matches NTSC–and will sync to both 23.98 AND 29.97 – because there is no time change between the two. Production audio can also be recorded at 48,048 and at 30fps, so that when pulled down .1% to a 59.94 reference, it plays back at 48k.


Therefore, audio can be interchanged from 23.98fps and 29.97fps video (keeping in mind audio must be slowed down if the 23.98 originated at 24fps film) because both are referenced to 59.94Hz. However, one may run into trouble having to autoassemble from production audio tapes that are in a different time code than the Pro Tools session.


So, if I've said something in error, please help me fix my understanding. If I've simply repeated what's already been said, show me where. And, if everything is wrong, Mod's just delete this post so everyone isn't as confused as I!

[hummer]

i have struggled with this as well, finding that most references do not adequately explain the intricacies.

i feel like a more simple explanation is this: (if i'm right)

24p is to 60i (30fps)
as
23.98p is to 59.94i (29.97fps)
via 2:3 pulldown.

meaning, you must slow the frame rate of 24p by .1% in order to successfully convert to an interlaced NTSC video framerate.

right?

here's my question...

i keep hearing that DVD is 24p, and that the player performs the pulldown for NTSC output. but it must actually be 23.98p that's coded onto the medium, right?

[jacobfarron]

Yes. And to further my post - my editor notes that film cameras were only referenced to 60hz around the time TV began appearing. Which doesn't change any of the technical details I believe. I've done the calculations over and over, and I believe my post is accurate. I also pulled out the manual to our Varicam, and it can reference either 59.94hz, or 60hz. So, everything seems to be confirming what I've found. I felt like I'd discovered gold when I finally realized that the pulldown itself does not change the speed of the video, the change of reference changes speed. I know a lot of the pro's on here probably have known that forever - but I'm just finally getting it.

And to Mr Purcell, your book finally gave me enough information to fully comprehend all of the numbers. Some of my other books simply said "23.98 has a special relationship with 29.97"... I can't accept that! So thanks for a great book.

[D Pinder]

You seem to have got it. The bottom line:

24 and 30 fps = film speed
29.97 and 23.976 = video speed

[jacobfarron]

hummer, please note that many times editors use 60i to mean 59.94. Just like they say they're shooting 30drop, when we all know it's 29.97DF.

[starcrash13]

Originally Posted by hummer

i keep hearing that DVD is 24p, and that the player performs the pulldown for NTSC output. but it must actually be 23.98p that's coded onto the medium, right?

Correct. DVDs are encoded at 23.976 and the DVD player adds the 3:2 pulldown to display NTSC (29.97).

[starcrash13]

Originally Posted by D Pinder

You seem to have got it. The bottom line:

24 and 30 fps = film speed
29.97 and 23.976 = video speed

Exactly. This is the way I understand it.

[Geert van den Berg]

I don't have enough experience with american frame rates and HD, I did do a reverse telecine on a feature we had to dub though.

If all the info appears to be correct, maybe this thread should be a sticky? Handy for anyone in doubt.

[MixinMonkey]

Originally Posted by starcrash13

Correct. DVDs are encoded at 23.976 and the DVD player adds the 3:2 pulldown to display NTSC (29.97).

Well, you're close.
If it's an HD disc like Blu-ray (or the almost completely hosed HD-DVD) then it's usually running 23.98 for films.

But, for regular standard definition region 1 DVD's.. these discs are mastered, encoded & authored in NTSC @ 29.97.
(NDFTC or DFTC depending on the studio), not 23.98.
So, the disc comes packaged to you already pulled down -0.1% @ NTSC speed from authoring/duplication.

other regions besides N. America & Japan would of course be PAL @ 25 frame EBU, pulled UP approx 4%.. ugh... ...sometimes...
but that's a whole 'nother sticky right there.

[starcrash13]

Originally Posted by MixinMonkey

But, for regular standard definition region 1 DVD's.. these discs are mastered, encoded & authored in NTSC @ 29.97.
(NDFTC or DFTC depending on the studio), not 23.98.
So, the disc comes packaged to you already pulled down -0.1% @ NTSC speed from authoring/duplication.

Standard DVD's can be encoded using either frame rate. I should have been more specific. From what I understand, most current major studio DVDs are 23.976 and the 3:2 pulldown happens at the DVD player for standard NTSC television. However, on an HDTV with a progressive scan display, no conversion needs to be made and the DVD plays actual 24p (well, 23.976 really).

Also, you either misspoke or you are mistaken about the pulldown from 24p to 29.97 being "-0.1%". A pulldown of .1% is a change in speed used to go from 24 to 23.976. In film being telecined to video, a .1% pulldown is followed by a 3:2 pulldown. Going from 23.976 to NTSC requires a 3:2 pulldown which is not a speed change, but rather an addition of repeated video fields to go from 24 frames to 60 interlaced video fields.

Sorry to be so nit-picky but, after all, this is "THE definitive explanation of 29.97 and 23.98" frame rates, right?

Not that Wikipedia is the most reliable source in the world, but there is a pretty good article on 24p that supports my understanding of how this stuff works.

24p - Wikipedia, the free encyclopedia

[jacobfarron]

Thank you starcrash! I'm just as picky.

[charles maynes]

You are missing a very serious reality with Red and HD for theatrical "film" releases-


We are crossing a threshold where "film" as we know it is simply going away-

In the past and present (at least for Steven Spielberg) film is shot on film at 60hz reference- production audio is at 60hz reference- All is pulled down to .99 speed for post, then taken back to 1.0 speed in the theatre, with music being .01% sharp to drive the perfect pitch folks crazy-


the new paradigm is that 23.98 IS 1.0 speed on set and the film speed adjustment can be done as a after mix process- In many ways this is easier because the telecine process is technically unneeded.

it still is not making the transition easy though-

[jacobfarron]

I've been hearing about this. The company where I work has been dealing with Red a lot. We have a great post workflow... but as I understand it, no one is posting at the full 4k resolution yet.

[robcap]

Great thread!

One wrench, at least to me. How does drop frame fit into the mix. ie. Isn't 23.98 a non drop format, but 29.97 can either be DF or ND - of course video for broadcast is DF where DF just accounts for the .03 difference if I understand this correctly.

My question revolves around a 23.98 ND project converted to a 29.97 DF project. If I look at the counters of both projects across a 22 minute show, the difference adds up to a couple of seconds due to the 2 frames occasionally not counted (but the video not physically omitted.) I have done this on our video system by burning in the 23.98 TC and pulling it out to NTSC 29.97DF and superimposing the new TC. Some posters above have mentioned that the frame rate would match - mine is close, but not dead on - unless I misunderstood the post...

Roberto

[charles maynes]

Originally Posted by robcap

Great thread!

One wrench, at least to me. How does drop frame fit into the mix. ie. Isn't 23.98 a non drop format, but 29.97 can either be DF or ND - of course video for broadcast is DF where DF just accounts for the .03 difference if I understand this correctly.

My question revolves around a 23.98 ND project converted to a 29.97 DF project. If I look at the counters of both projects across a 22 minute show, the difference adds up to a couple of seconds due to the 2 frames occasionally not counted (but the video not physically omitted.) I have done this on our video system by burning in the 23.98 TC and pulling it out to NTSC 29.97DF and superimposing the new TC. Some posters above have mentioned that the frame rate would match - mine is close, but not dead on - unless I misunderstood the post...

Roberto

that is a very good question- and you will not likely find a good answer for it here-

since dropframe is a TV only format- you might check with NTSC to see what the SOP is for that production workflow...

[minister]

23.976 does not have a Drop Frame option. 23.976 cannot be used for true Time of Day. The same 1000/1001 ratio applies and the duration of one hour of 23.976 Fps Time Code will actually be 1 hour and 3.6 seconds and the time code at this frame rate, will read 00:01:03:14.

When you perform the 2:3 pulldown to get to 29.97, the two numbers will drift apart if you try to go to 29.97 Drop. To do it right, you need the right devices and know-how.

[Remoteness]

I copied this thread over to this forum because I thought the advice could be very enlightenmenting to us Remotesters.

I hope you folks agree.

[hociman]

It should be noted that 23.98 (or 23.976, depending on what manufacturer it is) has no drop-frame option because of the integer number of frames.

At 29.97 non-drop timecode, there are 107,892 frames in an hour. At 29.97, you lose .03 frames per second in your count, which equates to 108 lost frames in an hour. If you drop frames on every minute except the 10th, you drop 2 frames * 54 minutes = 108 frames. The difference in the number of frames over the course of an hour between non-drop and drop frame timecode (108) is an whole number which makes the impementation of dropped frames practical in that it can represent clock time.

At 24fps timecode, there are 86,400 frames in an hour. At 23.98(.976) timecode, there are 86,313.6 frames in an hour. The difference here is 87 frames (rounding down) or 88 frames (rounding up). The point is there is rounding involved, so there is no whole number relationship, thus negating the ability to drop frames and get an accurate clock time on the content. If anyone says they are working at 24 drop, they are not correct.

[charles maynes]

Originally Posted by hociman

It should be noted that 23.98 (or 23.976, depending on what manufacturer it is) has no drop-frame option because of the integer number of frames.

At 29.97 non-drop timecode, there are 107,892 frames in an hour. At 29.97, you lose .03 frames per second in your count, which equates to 108 lost frames in an hour. If you drop frames on every minute except the 10th, you drop 2 frames * 54 minutes = 108 frames. The difference in the number of frames over the course of an hour between non-drop and drop frame timecode (108) is an whole number which makes the impementation of dropped frames practical in that it can represent clock time.

At 24fps timecode, there are 86,400 frames in an hour. At 23.98(.976) timecode, there are 86,313.6 frames in an hour. The difference here is 87 frames (rounding down) or 88 frames (rounding up). The point is there is rounding involved, so there is no whole number relationship, thus negating the ability to drop frames and get an accurate clock time on the content. If anyone says they are working at 24 drop, they are not correct.

J-

Since this is a digital format, wouldn't frames be arbitrary? There is not a real "frame" in digital, because it is a continuous capture.
Since dropframe was something introduced by the necessity of converting film to video, and the inability of the NTSC format to actually run those frames against the same reference as film, the drop frame counting method was a convenience for TV production to time out their programs to match real-time. The penalty of having the material play slow was seen as a necessary compromise.

This is an academic point btw...


cm

[rhumphries]

Since this is a digital format, wouldn't frames be arbitrary? There is not a real "frame" in digital, because it is a continuous capture.

For the audio yes, but not for the picture that it's is in sync with-- that is still one discrete field after another.

This 23.98 vs 29.97DF is a real issue that we need to be aware of. For instance: I recently came across a show that that was shot at 23.98. The offline was also at 23.98, but the online is at 29.97DF. The first problem was that all the lengths of the acts were off because 23.98 doesn't match the clock on the wall (video editor's problem-- not audio's!)

Additionally, when we say that 23.98 TC matches 29.97 TC on the seconds, but not the frames-- this is true, but if the audio guy is working on 29.97DF and the offline editor is working on 23.98, that doesn't apply. The timecodes will drift with respect to each other by about 3.3 seconds over an hour. That makes manually syncing a shot change a real nightmare.

-Richard

[Eric L]

Originally Posted by charles maynes

Since this is a digital format, wouldn't frames be arbitrary? There is not a real "frame" in digital, because it is a continuous capture.

I think we should start referring to it as a "scan interval" as opposed to a frame. With all the confusion over HD fields and frames. 1080i vs 1080p or 59.94 vs 29.97. It is important to find a common language here, because 59.94 can now refer to both field rate in a 1080i project, or frame rate in a 720p project. Is it a "frame" or is it a scan interval?

[robcap]

The offline was also at 23.98, but the online is at 29.97DF

Richard: This situation you have described is exactly what is happening to us on a continuing basis. We have animation and live action coming to us in this manner. Our solution, for now, is to have the editors all stay at 23.98 - even the online edits at this rate until the final output. (The editors do the math and bring their show to time in 23.98 - no problems!) In audio post, we import the 23.98 OMF into a 29.97df project. We ask DV video of the project with burnin, and use our Fairlight Pyxis to pull out the video to 29.97df. We are then able to look at both TC - the 23.98 burn-in from the original, as well as the burn-in displayed by the Fairlight Pyxis. Therefore when referring to specific TC during reviews, we can always refer to the correct TC without doing any conversions. And yes, we are off by a little over a second across a 22 minute show.

ROberto

[hociman]

Originally Posted by Eric L

59.94 can now refer to both field rate in a 1080i project, or frame rate in a 720p project.

But are there really 59.94 frames per second in a 720p project? I could be wrong, but I don't think so. The last D-5 HD tape I worked with that was 720p/59.94 had a timecode track of 29.97. I've never heard of 59.94 fps timecode, but wouldn't SMPTE have found a way to make that possible if there were truly 59.94 fps?

I don't wish to hijack the thread, but if this is going to be a sticky, its important that this be cleared up, even if I am wrong on this point.

[Eric L]

Yes, that is what I was taught about 720p 59.94 D5 as well. But in another thread started by Jacob, (2 sync issues)we discussed how PT now has a 59.94 timeline and that some editors in FCP are cutting on a 59.94 fps timeline.

I, myself, have no experience with either of these and can not figure out why one would work this way. I had never heard of it until this thread. I think this is some of the confusion I'm referring to. Mine and others.

I have spent more time than I should have trying to research this on the internet and there is so much conflicting info, it gets tiring.

[jacobfarron]

Eric - in my other thread, even though my timecode burn for Quicktime ref was in 59.94 timecode, the DVCProHD tape I laid back to actually had 29.97 timecode on it.. go figure. Now I've started wondering... what if the tape I needed to lay back to actually had 59.94 timecode on it? Pro Tools wouldn't lock to that. So I'd have to give my editor a bounce and have him do it. We're going to set up some tests later in the week to see how this goes and if it's something we need to have a workflow solution for.

rhumphries - Thanks for bringing this up. I haven't come across that yet. Can this problem be avoided by never using DF when pulling down from 23.98 to 29.97?

[rhumphries]

Originally Posted by jacobfarron

rhumphries - Thanks for bringing this up. I haven't come across that yet. Can this problem be avoided by never using DF when pulling down from 23.98 to 29.97?

Yes, but shows don't get broadcast in 29.97NDF or in 23.976, only 29.97DF, so if you are doing broadcast work then you have to reconcile the two frame rates at some point.

-Richard

[jacobfarron]

As luck would have it... I'm coming up on this issue this week. We have HDCam masters at 23.98, and are transfering to 29.97 NDF for post (thankfully) and will never be in drop. Our JH-3 play deck can do the pulldown (we do it with the deck instead of FCP) with either DF or NDF (so those of you using this deck be careful) - but a nice feature is it will display the original timecode and the pulldown timecode burn at the same time. I don't think our HDW-500 can do that.

[huub]

But, for ALL framerates, audio is recorded(on location) at exactly 48000hz right?

[charles maynes]

Originally Posted by huub

But, for ALL framerates, audio is recorded(on location) at exactly 48000hz right?

not in my experience....

[huub]

Please explain?

Thanks,
huub