Full-Screen Video Over 28.8k: The Claims Continue 459
gwernol writes "Over at Screen Daily they are claiming that an Australian company has demonstrated a high quality, full-screen video-on-demand service that is delivered over a 28.8k modem. They claim this will 'eliminate the need for broadband.' If this is true, then they'll change the world. Of course, the basic technology has been around for a while, see this article from 1998 or this one from earlier this year. I remain extremely sceptical. If this is real, why won't they allow proper independent testing? But it is interesting that they're getting funding. Could this be the last great Internet scam?"
Several readers also pointed out this brief report at imdb.com as well. We've mentioned this before, but the news here is the reportedly successful demo. It would be a lot easier to swallow if he'd let people test it independently, but video-over-28.8 sure is tantalizing.
The last? It was one of the first and best (Score:4, Interesting)
The above is all that in necessary to say on this subject, but due to the postercomment compression filter, I have to add this meaningless paragraph.
Re:Well if its full screen over 28.8 (Score:2, Interesting)
What is this FULL-SCREEN video ?? 320x200 ? 640x480 ? true NTSC @ 29.97 FPS ? DVD resolution ? HDTV ?
Or is it 1/4 tv resolution zoomed to fit the screen ? with 1/2 the fps ?
Maybe all they did was improove the zoom, iterpolation and anti-aliasing algorithms in the player. So they send a crappy video and it ends up looking ok.
Anyway its all hot air until we get some technical data.
Decoding, not compression (Score:2, Interesting)
That said, assuming they have the compression, nobody probably has a cpu for decoding it.
Sounds like WEB technologies (Score:3, Interesting)
In fact someone came up with a mathematical statement that said the only way their claims would hold water was if they just gave out 64 bit serial numbers and stored the data somewhere else. Not to different from what we call Freenet now.
Needless to say these guys ended up going under after the investors figured out they were not only full of it, but 10 lbs of it in a 5 lbs bag.
Re:MP3... (Score:1, Interesting)
Re:Claim is not unreasonable... (Score:2, Interesting)
Ok, what is "Scientfic Image Data"? Pictures of planets?
What is "12-bit dynamic range"?
At 30 fps, 0.33 MB per frame, that's 10 MB of image data per second. Compressed 1000 to one, you're only talking about 10 kilobytes per second.
Ok, what is your source resolution and color depth? How did you come to
Even assuming you could get that down to 10K, a 28.8K modem runs at about 2.8K a second. It would take you 3.5 seconds to download those 30 frames. That would bring your frame rate down to 8.5FPS. This doesn't even include Audio.
If you're willing to suffer with less dynamic range around spike bits of data, it's not unreasonable to think that another factor of four could come out of that...
So now you are talking about a 4000:1 compression ratio? Sign me up! The highest I've read about is between 10:1 and 20:1 compression for MPEG4!
Even if you had a typo and meant 100:1 then another factor of four would put the compression ration at 400:1. That is hardly realistic.
Re:MP3... (Score:3, Interesting)
For example, take a 10 second clip of 640x480 24-bit, RGB, 29.97 fps video (no audio). The math sez its:
640 x 480 x 3 x 29.97 x 10 = 263.41 MB (approx).
Yet 10 seconds of 10 Mbits MPEG-2 video (very high quality) takes up just 10 Megabytes of space. That's a compression ratio of over 26:1!
Over a 28.8kbps modem over the internet we are looking at about 2.6kbps of data (headers and other overhead removed). This means the above 263 MB video is supposed to compress down to less than (don't forget about the sound!) 26 k. That's a compression ratio of 10374:1!
I can believe a leap of 10x, *maybe* 50x. But a leap of 400x is just something I have to try on my own terms before I believe it.
Is it actually DIGITAL compression? (Score:2, Interesting)
Just because he's using a modem doesn't mean that he's actually transmitting digital data over the phone line. What sort of video compression can be achieved when you don't need (or get) bit-perfect transmission, but rather encode video properties directly in the analog signal? Errors then show up as slight inconsistencies from the original color or position - but on motion video, this would be irrelevant.
The compression would still need the common video codec functionallity to remove redundancy, and send the changed areas more frequently than static images, but if the modem link mapped QAM data directly to position and color signals, it might just be possible to paint a fairly high quality picture.
For that matter, some fractal compression techniques are quite tolerant of minor errors in their probability and/or mapping factors - combine this with sending color information as analog data, and now you might be able to have a link that is unidirectional (the whole audio bandwidth can be dedicated to the video stream without need for a reverse channel) and error tolerant (no re-transmit on error or dropouts due to transient line noise).
Maybe it isn't a scam.
Re:Uh, no (Score:2, Interesting)
Actually, our eyes don't have a fixed fps as so many of you nerdlings tend to think. There IS a limit to how rapid changes we are able to see, but they are very dependent on brightness. We have problems seeing dark changes that happen in tenths of a second, but noone will miss a bright flash even if it lasts 200ths of a second.
In normal lighting, 10-12fps is not even in the
same ballpark as our vision. 75 fps is more like it.
Re:Claim is not unreasonable... (Score:2, Interesting)
This is a common mistake that people make. Someone designs a compression scheme that works really well for specific cases and thinks that it will work in the general case. Hell, I once designed a custom lossless scheme for handling certain classes of bitmaps that beat lzw by a factor 5:1, but I guarentee you if you applied it to bitmaps that we were not interested in, it would have been very unimpressive. I suspect the same can be said for the wavelets your group was using.
Re:Lets do the math... (Score:1, Interesting)
Let's take 640x480x16. It's around 614K. We try
fractal compression approach here, so we divided
a picture into 8x8 blocks with x,y,offs,scale,rot coords.
We end with 9600 blocks.
An arithmetic coder allows us to achieve
415Kbit=9600*(log2(640)+log2(480)+16+7+2) or 52K
bytes per frame. For 320x240x16 this going down
to 6.2K (320/8*240/8*(log2(320)+log2(240)+16+7+2)).
This is about twice as big as 2.8K, yes? And we get
nasty side effects because of big blocks (4x4 is far
better for small resolution pictures).
I've seen (a long time ago) a paper on 3D fractal
compression. Let's see. The bit count for this scheme
will be
BC=X*Y*Z/(N^3)*(log2(X)+log2(Y)+log2(Z)+O+S+R),
where X,Y,Z - sizes on X,Y and time axis, O is
bit count for offset (16), S is bit count for scale
factor (7) and R is bit count needed to encode
rotations (to swap or not to swap direction around
appropriate axis, which are three).
For 320x240x24 (24 frames of 320x240 pictures) and
N=4 we get BC=1.2 Mbits. But for 3D compression
bigger (than 2D compression) N size does not introduce
same nasty looking artifacts. As far as I can remember,
even 16x16x16 block looks pretty well. So let's choose
N=8 and we get (don't take your breathe) - 157Kbits.
You still breathe? Ok, for 1 second and N=16 we get
an estimate BC=21Kbit, well within 28800. For four
seconds encoded we get an estimate throughput
of 22000 bits per second. For 4 sec 640x480 N=16 we
get throughput of 91Kbit/sec.
The other side is the memory requirements. We will
hold RGB (or YCrCb) in three separate bytes. We have
to double our buffer because we have scaled and
original version of picture. So we get X*Y*Z*3*2 bytes
for each compressed block. It's 44M for 240x320x4sec.
Two links:
http://inls.ucsd.edu/y/Fractals/ - for uninitiated;
includes reference to "Three-Dimensional fractal video
coding" paper.
http://www.cse.sc.edu/~culik/ - Karel Culik invents
Weighted Finite Automata transform, which is more
efficient than fractal-based, but uses similar approach.
This page includes links to several WFA software
systems to experiment with.