4:4:4 12-bit Uncompressed DVX100
 

Success!

After a month or two of slow progress, I managed to setup my test equipment and capture 4:4:4 12-bit uncompressed output straight from the DVX100 A/D converters.

My current test equipment is only capable of capturing a portion of a frame, but I wanted to try it out first to make sure it works. I have ordered the gear I need to perform uninterrupted capturing.

There are some side effects, for example, the signal out of the A/D converters is raw so all the corrections(WB,etc) done by the DVX have not been done yet and need to be done in post. This can be avoided by capturing the signal after it leaves the big chip that does all the color corrections, but the signal would then be 10-bit Y/C, whereas the signal I am getting now is a very fat RGB 12-bit.

It will be interesting to see what the difference is between the 12-bit RGB and the Y/C 10-bit signals.

I am going to attempt to put together a complete frame section by section and also post the image from tape for comparison.

Theoretically, this can be done with any camera but the DVX is a great choice because of the nice progressive scan CCD's. Once I have continuous capture running and tested, I plan to design a small portable interface(possibly firewire) that can be attached to the camera for portable use.

The RAW image is extremely fat...something like 38MB per second at 24fps...however this can be thinned down considerably.

More to come....

Juan

Juan:

Very interesting - love to hear and see more about how you are doing this and what results you get.

I think the DV format is the DVX100's biggest problem. What lossy or lossless codec are you thinking of using for processing the image?

He's going to have to find a RAW convertor first, or use the Y/C signal (probably the better option).

Right now, all I have is a set of 12-bit R,G,B values, which can be trivially padded to 16-bit or 24-bit.

There is more than one graphics format that works with interleaved RGB values. One example is BMP. A simple program can be written which creates a BMP from each frame simply by creating a standard header(since all frames are of equal size), and dumping the interleaved RGB data straight from the camera.

The independent frames can then be assembled into a video using any app like Shake for example.

The question here is what digital video format can handle this fat a signal...i could encode it in 4:2:2 raw and it would still be many times better than what comes out of the DVX100, and technically even the SDX900(ignoring lenses, CCD size). I am not sure if there is any format out there that handles 4:4:4 uncompressed, but I guess even if I do endup throwing out or averaging some of the samples it's still ~much~ better than DV.

If I had a choice, I would rather use the 12-bit RGB because this is a standard DVX100(non A) and has 10-bit Y/C output so I am loosing some precision. In the future, if I get my hands on a DVX100A, the Y/C 12-bit output will provide a signal that's 2/3 as fat as the RGB and probably just as good since it has no overlapping data.

Juan

What would be interesting to know is how many lines of resolution you can get off this vs. DV signal.

That is also interesting...and I could answer it if I had the data sheet for the CCD's which I do not have, but I am working on capturing the entire frame at once direct to a computer, so I should have the exact numbers soon. I assume there are dummy pixels around the image but as of now I am not sure how many effective lines the CCD reads out, without doing a rough estimate from the 440k figure.

I tried this experiment intially with a single-ccd sony camera. I did have the data sheet for the CCD so I wrote a program to de-mosaic the image, but the real interesting prospect would be to get a higher-end 1-CCD cam that also takes stills...i beleive those have high-res CCD's that readout the complete frame (in interlaced mode) and then is downsampled to DV frame-size.

The problem with doing this, is that since it was a 1-CCD approach, the data rate coming out of the single A/D converter and for a CCD that's almost 1MP in size is waaaay to fast for my equipment to capture. If I come up with a custom firewire solution then this can be done, but for now, i have to stick with 3-CCD systems which normally have a much lower sampling clock rate.

Juan

Forgive me for asking but what in the world is the point of this endeavor?

The signal captured and recorded by CCD has for more info i.e. bits of data than is recorded on tape. To record on tape, that data must be compressed and processed to the DV format, thus a signifigant amount of color and resolution goes out the window, lost forever.

Although i'm not enough of electronics/video engineer as Juan obviously is, the point would be to use the DV to capture a signal in a higher format. E.G., using an 720p MPEG-2 based HD format, it would be fascinating to see how the DVX100 or 100A would compare to the JVC-HD CAM. I think it would be signigantly better image, both resolution and latitude.

If Juan can come with a system to get 720p 24p stream from the DVX100, he might become a millionaire overnight.

Or even see if there's enough info to process a true 1080p signal.

Or several other RGB based formats that he outlined.

Read up on the net or here about the Thompson Viper (and recent similar) cameras - it uses this basic idea to get a better image than possible with HD cameras.

The main goal of this endeavor departs from the fact that with any DV camera, the signal captured is of MUCH higher quality and contains much more information than the DV signal that is actually recorded. This is also true with all but the highest end cameras. The Thomson GrassValley Viper is the ultimate example of this concept. This camera simply outputs the raw 4:4:4 uncompressed stream and capture/processing is done on a separate computer unit.

I don't know the exact details but the rough estimate is that 80% of the data is thrown away in the 4:1:1 quantizing and DV compression process. This experiment attempt to get the clean signal with no degradation.

After all, you PAID for the CCD's and the electronics. Why shouldn't you be able to get what you paid for?

It is relatively simple for manufacturers to include some interface which allows the end user to get this quality...the main reason is that they do not want to outshine their high-end products.

If someone doesn't beleive me, all you have to do is take a look at the DVX100 service manual...it has full board layouts, completely labeled, except for the side of the board that one would need to carry out the experiment i'm doing.

Although i'm enthusiastic about this i wouldn't go as far as saying that I will gain monetary anything like Stephen says, but the encouragement is appreciated :)

It does rise an interesting question...how many people would actually be willing to void the warranty on their DVX's to get this quality output from their cameras? It would be invaluable for transfers to film and green-screens among other things.

Juan

I see.

So, in summary Juan, your goal is to turn a $3,500 camera (in this case the Panasonic DVX100) into the equivalent of a $100,000++ Thompson Viper by intercepting the raw imaging signals from the 1/3" CCD block before the camera processes them?

I seem to recall someone attempting a similar maneuver with the XL1s a year or so ago here. If you can find the thread it might offer you some pointers.

OK, well thanks for that clarification and good luck with your project.

Juan:

Based on the number people dissassembling and altering their camera with home-grown mini35's, I would a pretty reasonable amount.

Of course, it would depend how much skill required and/or someone could to the board work for them. Also, if an reasonably elegant solution were devised the retain some measure of portability, durability for location shooting and recording time of at least 30 minutes.

The DVX100 already has great mindshare and the only cams to enter that discussion are the JVC HD cams (but it's a bear to shoot with and the camera and codec leave alot to deserved) and the XL1s with mini35, but it cost's a ton and ain't exactly portable either.

So, if you take those two cams as the primary target, I think you would grab a lot of interest.

I personally think that 24fps 720p is the holy grail of indie filmmakers. More rez than PAL, none of the 25 Mb/s DV crap, no 25 fps pal problems, etc. etc.

Wait, I'm really confused. How are you going to get 720p out of the DVX's CCDs? Is there that much info even being captured on the CCDs?

I'm pretty sure that I can't get a full 720p frame out of the CCD's on the DVX.

However, the CCD's ~do~ have more pixel sensors than there are actual pixels per DV 720x480 frame, so the raw output will have a larger frame size than 720x480(and 720x576), but once again I don't think it reaches a 720p frame. For example, i beleive the DVX's CCD's have 440k pixels, while a 1024x720 frame is over 700k pixels.

The largest advantage will probably be the fact that it will have FULL color information and no compression artifacts.

I'm going back to the lab tomorrow, I will try to get an exact figure for raw frame size.

Juan

And more lattitude?

I doubt it.

I estimate that the color will be better not only because it is full color sampling for each pixel, but since it is 12-bit and not 10-bit. The data will be more precise but the white and black will be at the same values.

There might be something about the quantizing and compression that I am overlooking, which limits the effective latitude of the DV output, but i'm not sure.