Building a Digital Video Capture System - Part II


So, if you've read Building a Digital Video Capture System - Part I you should have a general idea about the issues involved with building your own digital video capture setup. In this part we'll get down to some system specifics and then take a look at a handful of capture systems out there and how they stack up.

But first, I got a lot of questions about how big a video stream really is based on what I wrote in Part I. I'll stick by my guns on this one. Raw, uncompressed video is about 21 MB/s. Here's the math:

720x486x29.97fps = 10487102.4 pixels per second.

Since broadcast video is stored using a YUV 4:2:2 color scheme the luminance values are stored for every pixel while the color values are stored for every other pixel. This equates to 32 bits for every two pixels or, for the sake of simplicity, 16 bits per pixel (1/3 more efficient than a computer's 24 bit 'true-color' RGB scheme actually). So...

10487102.4 * 16 = 335587276.8 bits/s
335587276.8 / 8 = 20,974,204.8 Bytes per second

Modulated RF video signals transmitted from your local TV station are about 25 MB/s (there's a lot of extra stuff in those RF signals such as audio, SAP, VIB signals, a bunch of other reference signals, and space reserved by the FCC for test signals, EBS, and even military signals).

DV video is 25Mbits/s because it's compressed (and don't let anyone tell you it isn't compressed or that it is a lossless CODEC). For starters DV only contains 1/4 the color information (YUV 4:1:1 for NTSC and YUV 4:2:0 for PAL) and most DV camcorders do some pretty severe clipping of the blacks (so if you plan to shoot a lot of moody, dark sequences you might want to avoid DV). At its heart, DV uses a DCT (discrete cosine transform) compression algorithm, which is lossy and the conversion from RGB to YUV also introduces quantization errors. They used to say, in the capture industry, that a capture rate of 5-6 MB/s was the magic number because a 4:1 compression ratio was about as far as most broadcasters were willing to degrade an image.

I also got a number of emails that questioned my statements about 2 GB limits in file systems. According to a Pinnacle technician FAT-16 has a limit of 2 GB, FAT-32 has a limit of 4 GB, and NTFS has a limit of 4 TB. The Pinnacle manuals state, "A limitation in the FAT 32 file system of Windows 98 and Widows ME allows you to capture only files with a maximum size of 3.9 GB." According to the PYRO ProDV manuals the limit is 4 GB. The PYRO manuals also state that the DVTools software they bundle has a file size limit of 2 GB. One reader, Brick Ekesten wrote, "The 2 GB limit is not imposed by FAT formatting but instead by the values used by Microsoft's AVI stream wrapper. It was assumed originally to have been the FAT-16 limitation but it has been possible to format a drive to use larger than 2 GB partitions but the AVI stream wrapper has not changed. There are still capture cards which are limited to 2 GB file sizes despite the fact that people are using FAT-32 or NTFS because the AVI wrapper has some hard coded limits on file size." Apparently Apple also fell into the same trap since the Mac OS (below 7.5) and up until the latest releases, QuickTime also had a 2 GB limit. So, whether the problem really exists or not, whether it's in the file system or the software, whether it's 2 GB, 3.9 GB, 4 GB, or 4 TB most capture card manufacturers seem to think there is some kind of limit and most of them offer some sort of workaround.