4:4:4 10bit single CMOS HD project

[Steve Nordhauser]

CCD course:
Please be sure to notice that Wayne's CCD course is just that - about CCDs. Similarly, the information on our web site is mostly about CMOS sensors. There is plenty of overlap but some differences.

[Anders Holck Petersen]

<<<-- The standard black for NTSC is 7.5 IRE and for PAL it is 16 IRE, ( I suggest always use 16) and they work very well..
Remember that your video master goes thru a standard gamma correction on the film recorder to be transfered to film, and if the video original isn't captured within TV Standards limits (From 7.5 or 16 to 235 IRE) -->>>


This info is not correct.

The IRE scale is for NTSC only and the legal values goes from 0 to 100 IRE.

All analog American NTSC signals should have additional "setup" which is a blacklevel correction that puts all blacks at 7.5 IRE.

Analog PAL uses Volts instead, and the legal scale goes from 0 to 0.7 V.

Both scales are only relevant in the Analog domain eg. after the DA conversion.

ALL digital video signals both PAL NTSC and HD, has a legal range of 16 to 235 if we assume a 8 bit depth. Levels 0-15 is for subblacks used by old DSK keying, and 236-255 is for additional latitude in the highlights.

In NTSC a level of 7.5 IRE is digitized at 16 and 100 IRE is digitized at 235.

So internally a PAL signal or a ntsc signal with or without setup should have its blacks at 16 and be whiteclipped at 235. when in the digital domain.

[Juan M. M. Fiebelkorn]

as you like :)

I made a big mistake ,though putting IRE after 235 instead of saying 100 IRE.

[Jason Rodriguez]

If you can mention it, who did you send your footage too?

Good places I'd recommend are Technicolor NY, Efilm, I/O Film, Cinesite, and Duart. I'm sure there are more good places, but I've found these to be the most knowledgeable or experienced with tape-to-film transfers.

[Jason Rodriguez]

BTW guys, new info:

With the Altasens chip, you can NOT do 720p AND keep the 2/3" optical format-you must to windowing, reducing the optical format to 1/2". Now I don't think this is a deal-breaker by a long shot, but this is just something I thought you'd all need to know now, and not get surprised with later.

I'm thinking then for my setup I'll use the 2.5" 36GB Seagate Savvio's (10K RPM U320 drives). I'll stripe four of them together on one SCSI channel, and that should give me enough bandwidth for 40fps 1920x1080 at 12bits. Hopefully Rob can incorporate 12-bits linear to 10-bits Log so that we can easily do slo-motion with this drive setup, because then we're only recording 10-bits to the hard-drives, and that'll get me another 8fps up to 48fps before I hit what I think will be the 124MB/s barrier on a U160 interface (that's as fast as I can find a MiniPCI card right now since they are typically 32-bit 33Mhz). I'd like to get 40fps at least though, I think that should be enough slow-motion for 1920x1080. Either that or I'm thinking that 1/2" optical format won't be so bad, but I'd like to stay away from that if I can (plus I'm not interested in the lower resolution of 720p, but I do like how high a speed it will give me, so I'm weighing the choices-but either way, four of these drives striped together should give me plenty of bandwidth).

[Wayne Morellini]

I also remember an Australian 7 Network tech teaching us about the super white/black values being reserved in Pal because they interfered with certain emergency radio freequencies, and also causing the TV station transmission towers to kick in extra generators and cooling.

Jason-
You can probably solve the slow motion problem by using a fast 64bit frame grabber to save the short sequence to 4 or 8GB of memory and then saving it to disk betrween shots.

[Rob Scott]

Jason Rodriguez wrote:
With the Altasens chip, you can NOT do 720p AND keep the 2/3" optical format-you must to windowing, reducing the optical format to 1/2".

Actually, it appears that as long as we support the oddball Bayer pattern, we can keep the 2/3" size.

before I hit what I think will be the 124MB/s barrier on a U160 interface (that's as fast as I can find a MiniPCI card right now since they are typically 32-bit 33Mhz).

I'm not an expert of these systems (yet), but if you're using 32-bit/33 MHz PCI, you're going to have a problem with bandwidth. The frame grabber card is going to use a good chunk of your PCI bandwidth, so if you have the hard drives running over PCI as well, you will need to use 64-bit/66 MHz, or a split bus, or both. (Somebody please correct me if I'm wrong. :-)

You can probably solve the slow motion problem by using a fast 64bit frame grabber to save the short sequence to 4 or 8GB of memory and then saving it to disk betrween shots.

That's a good idea too. Elsewhere someone suggested doing this in software, so of course you'd need a motherboard that would hold that much RAM, a 64-bit CPU and a 64-bit OS -- but 64-bit Windows is not available yet (again, unless I'm mistaken).

[Jason Rodriguez]

Well I'm not really sure it's mission critical that we keep the 2/3" optical format. Like I said, that's really not a deal-breaker. Also an odd bayer pattern is going to cause some weird artifacting, especially when you have two greens right next to each other and then a red or blue right next to each other and back to green. Now I saw Wayne's post on the technical discussion board, and it does seem like it might work, but I'm worried about new weird artifacting patterns and other associated artifacts. For instance, variable number of gradients needs nine pixels total, or a 3x3 block to work correctly-with the sub-pixel mode we only have 2x2 blocks-so more advanced algorithms are going to get messed up. I think it's just best that we get a really fast drive array in as small a package as we can. Again, we have new 2.5" 10K drives from Seagate (Savvio), and they can sustain around 53MB/s.

Also Rob, the board that I'm using does have a split bus, since it's using 64-bit PCI-X for one PCI slot, and then the other slot is 32-bit MiniPCI. The 36GB models are around $350 right now, so four of those striped together should do a nice job of keeping around 120MB/s the whole way through. And when we want 720p, we simply window down to that format and crank that sucker as far as we can push!

Again, something else to consider is that most video lenses, especially those on DV cameras, don't open that far f-stop wise because of the prism needed for 3-chips (like maybe f2.8). You can get 16mm lenses though that open to f1.3, so the depth of field problem becomes mitigated at that point-that's amost 2 1/3 stops wider than f2.8. The only "problem" you might have is with super wide-angles, but that's about it IMHO.

But again, if I can get this thing up to 120-124MB/s drive-speed wise, then we should get around 40fps at 1920x1080, especially with 10-bit log conversions from 12-bit linear.

BTW, a motherboard that holds 8GB of RAM is HUGE-I'm not going to be carrying that around on my shoulder.

[Jason Rodriguez]

Okay, I looked at lenses some more, and it appears as though you can get primes that are around 5.9-6mm for 16mm, which translates to around 22-25mm when thinking of it in 35mm terms when used with a 1/2" chip format. So no real need to get worried about using 1/2" chips with 720p windowing.

[Rob Scott]

Jason, since you have a split bus, you should be in good shape. I wanted to make sure you didn't have any nasty surprises later on, but I believe you're on top of things.

You're probably right about artifacting with the oddball Bayer pattern. To get a good image, someone would probably have to create a custom algorithm.

One more thought: Those 10K drives sound nice and fast, but with four you're only going to have 144GB - enough for about 1.5 hours at 720p and 45 minutes at 1080p (10 bits, 24 fps).

[Jason Rodriguez]

45 minutes should be plenty for one magazine. That's like one tape. And using USB 2.0, we can offload 144GB fairly quickly, like 50 minutes max (hmm, that does seem like a long time).

BTW Rob, if your software could make a new folder for each capture, like "scene-01-shot-001-take-001" "scene-01-shot-001-take-002", etc., that would speed up the transfer process quicker because we could keep the USB drive hooked up all the time, and just offload the takes we want when there's a break in shooting (without searching through a bunch of file sequences-THAT would be very confusing and take forever).

[Rob Scott]

Jason Rodriguez wrote:
45 minutes should be plenty for one magazine ... offload [in] 50 minutes

If the drives were removable, that would be nice. I'm not sure how practical that is though.

BTW Rob, if your software could make a new folder for each capture, like "scene-01-shot-001-take-001" "scene-01-shot-001-take-002"

Without promising anything :-), I'll keep that in mind. How would you suggest handling setting the scene/shot/take? (In the UI, I mean.)

[Jason Rodriguez]

I liked the way that S-Two did it with their D.MAG digital disk recorder.

At the setup of the new "reel" it would ask you what the reel number was and what the scene number was (if you were continuing from the same reel as before). So in a reel (for us it would be our hard-drives), it would then add a folder for the scene number that you had entered, and then with each stop/start, it would make a folder for the take.

Each time you erased the hard-drives, it would set-up a new "reel" number (if you wanted), or you could keep the reel number from before. Then it would save the files out using the name format of r001_s001_t001_XXXXXXX.dpx. Of course they were using DPX files (we'd be using IHD files), but that seemed to be a very easy way to identify what file was what, and it was embedded nicely inside a reel-scene-take folder system. Wasn't hard to setup at all on the user end, and it helped a lot in keeping track of what when where, even when files got seperated from their respective folders (or when you were importing files into a program, they all didn't have the same number sequence). BTW, the seven digit number sequence was actually timecode. I'm not really sure that's necessary for us though, but seven padded numbers does give us a lot of frames for recording, if one decides to fill up their hard-drives.

If the drives were removable, that would be nice. I'm not sure how practical that is though.

I think I'm going to up the capacity then using the 73GB hard-drives. That'd be like having "two" magazines if I went with the 36GB HD's. 1.5 hours of 1080/24p is a lot.

[Obin Olson]

I have new scene/take files every time you hit "record" they all stay in the folder you make when you start shooting

[Jason Rodriguez]

Obin,

Do you think it's possible though to have the information for each scene/take in a new folder hierarchy of scene->take? Then you're not scrolling through a million files trying to find the right scene/take number for the files you want to copy over.