Wearable Computer System for HD Capture

[Chris Ames]

Originally Posted by Dale Backus

I read that they're going to be integrating Flash Disks (up to 512 Gig has been announced) more and more into consumer products such as small form factor laptops and things... ... ...if this is a problem, they better figure it out soon....

It is a problem and they have solved it already in the Solid State Disks. The problem with USB flash drives is that a sector must be erased before new data can be written to it. And the conservative estimate for the number of erase cycles for flash memory is 100,000 cycles.

This is no big deal if you are just copying a few files to a USB thumb drive here and there to transport from one PC to another. But if you are using the devices as a scratch disk or as virtual memory... (and if his entire XP install is running off it, then the page file/virtual memory file is running off it as well) ... then you significantly decrease the life of the flash drive.

From what I've read the 32gb SSDs have onboard RAM that handles constantly changing data without writing it to the more sensitive flash memory. Also the drives have onboard software that monitors how many times each sector has been used and decommissions/retires the sectors once they have lived a good life, but before they physically kick the bucket. Meanwhile the OS is oblivious that this is happening under the covers.

USB flash thumb drives do not have these features. They will just up and perish one day without warning. Hopefully not during a crucial moment.

It is still an awesome idea, but like I said before. It's a good idea to keep a backup installation on a spare flash disk to minimize the impact.

Chris
http://dewde.com/videos

[Kevin Kondra]

Hello Everyone,

Thanks for the interest - I'm in a small town in Manitoba on vacation for three weeks so my internet access is sporadic. I've seen some things that need answering though:

The OS is on a 4GB IDE flash drive, not USB drive as some people have indicated. This drive should have reasonably intelligent circuitry to randomize write locations and prolong the life of the drive. The form factor is tiny so it fits right into the IDE pin socket (or so it should but the USB risers get in the way).

I'll try out Cineform NeoHD, maybe even in the next few days as I'm going to do some filming here with a new generator.

The cpu, T7200 with 667 MHz RAM, has enough power to record 60i without dropping frames using BM MJPEG. The data rate seems to be 13 MB/s. What most people may not know is that BM supplies a second MJPEG codec that isn't used with their recording app and has selectable quality values. Upping the bit rate to 20MB/s makes a noticeable difference in visual quality. They number used is similar to JPEG compression quality numbers, where 100 is supposed to be lossless, and 80 is default. 20 MB/s is 92 on the scale.

The batteries I'm using are 60 Watt-hour units. The power specs for the motherboard show 19 volts, and some current rating which is the max possible. The max rating is with all USB ports sucking a full 500 mA each, ethernet, compact PCI, PCIe and cooling fans all running with full current draw. Fortunately my setup does not use this much power. Running out of juice in the middle of a shot sucks though, so I got a new 118 watt-hour unit.

Like Serge wrote, a mesh bag will let air through. There is a cooling fan in the case. It doesn't seem to get hot at all. My backpack has the case strapped on the outside for maximum cooling.

OK, have to wait for more info later.

[Kevin Kondra]

I have video to share, does anyone have any great ideas on how to share it?

[Wayne Morellini]

Hi

You won't be able to do cineform on these, yet, but look up pico itx, and nano-itx motherboards. They start at around 7cm*5.4cm. If an PCIex1 interface is ever available, you could it with Intensity card, and power supply disk and battery along side the motherboard next to the card in an much smaller case.

I think we will see an Expresscard version of the Intensity some day, which would fit an wider range of these Nano/PICO ITX boards.

Ultra Mobile PC tablets might be available in 5inch + forms, and if there was full Expresscard. and Intensity available in Expresscard, you have something the size of an external display.

There is an company, Zero technologies I think. Designing an wireless HDMI transfer solution using UWB (wireless version of USB). They use Analogue Devices Jpeg2000 (wavelet like cineform) codec chips to compress. If anybody makes an USB version that can feed 220mb/s (dual codec chip) 4:2:2/4:4:4 to computer then it would make an great solution.


An friend did an white paper proposal fro an HDSDI standard to transfer an raw bayer video. While David is here, I would suggest that it might be beneficial for cineform, and others it would benefit, to do an proposal for an raw bayer/single chip format for HDMI and it's derivatives (UID and Display port) basically using existing greyscale, with an designated pixel format. Most cameras are now, or moving towards bayer/signal chip, and you could fit far better images compressed in the same bandwidth, particularly HDTV/cable in CineformRAW. 4:2:0 color is dated.

[Chris Ames]

Originally Posted by Kevin Kondra

Hello Everyone,
The OS is on a 4GB IDE flash drive, not USB drive as some people have indicated. This drive should have reasonably intelligent circuitry to randomize write locations and prolong the life of the drive. The form factor is tiny so it fits right into the IDE pin socket (or so it should but the USB risers get in the way).

Very cool. I read that some of the upper end flash drives do have firmware that intelligently distributes write/erase cells in order to maximize life. Also, the fact that it is high capacity helps a good bit especially if you don't use all of it. The bigger the pool of unused cells, the better. Because it reduces the frequency in which each individual cell is used, thus prolonging its life.

Chris

[Kevin Kondra]

So I have good news and bad news. The bad news is that the computer does not have enough processing power to encode to Blackmagic's MJPEG format in real time. I didn't find this out for a while because I thought my 3:2 pulldown removal process was the culprit. It turns out the system was gracefully dropping frames, and since the codec isn't totally multithreaded, CPU usage only showed 56%. Unfortunately 33% of frames were dropped so the video was ugly and pulldown removal wasn't ever going to work.

The good news is that I have been using the trial of Cineform's NeoHD and it works great. CPU usage was up as high as 99% (but never 100%) and no frames are dropped. This is using the medium quality setting. Real-time pulldown removal simplifies my workflow greatly - take the removable disk from the portable computer to the workstation and start using the capture files directly in a project (or copy them locally first).

Other news, the system over all works very well. With the new battery I've done about 15 minutes of recording and 30 minutes of other use with the battery still showing >80% charge.

Samples next.

[Kevin Kondra]

So here are two samples, same shot recorded simultaneously on tape and computer. Computer recording was with Cineform's NeoHD, medium quality setting. Tape recording was HDV.

Both files were encoded to an XVid format to keep the file sizes reasonable. Encoding settings were identical except I added a 16:9 aspect ratio flag for the HDV based file. HuffYUV was used as an intermediate codec for processing the HDV file, the Cineform file was encoded directly in one step.

www.hydraclime.com/Cineform_Sample_3_XVid.avi

www.hydraclime.com/HDV_Sample_3_XVid.avi

If anyone is interested, I can provide other types of shots (any requests?) or both of these in HuffYUV versions for an accurate comparison.

[Richard Leadbetter]

CineForm would really benefit from the 2.33GHz Core 2 CPU. That said, what type of video are you acquiring? 1080p/24? 1080i/30? If the latter, are you doing the pulldown on the fly?

[Serge Victorovich]

Originally Posted by Kevin Kondra

So I have good news and bad news. The bad news is that the computer does not have enough processing power to encode to Blackmagic's MJPEG format in real time. I didn't find this out for a while because I thought my 3:2 pulldown removal process was the culprit. It turns out the system was gracefully dropping frames, and since the codec isn't totally multithreaded, CPU usage only showed 56%. Unfortunately 33% of frames were dropped so the video was ugly and pulldown removal wasn't ever going to work.

The good news is that I have been using the trial of Cineform's NeoHD and it works great. CPU usage was up as high as 99% (but never 100%) and no frames are dropped. This is using the medium quality setting. Real-time pulldown removal simplifies my workflow greatly - take the removable disk from the portable computer to the workstation and start using the capture files directly in a project (or copy them locally first).

Other news, the system over all works very well. With the new battery I've done about 15 minutes of recording and 30 minutes of other use with the battery still showing >80% charge.

Samples next.

Kevin, with yours hardware possible the best choice for injest is NeoHDV at 1440x1080p23,976 in 8bit. CPU usage can be less than 90% w/o dropped frames. And if you than compare 1920x1080(NeoHD 10 bit) and 1440x1080(NeoHDV 8bit) visually no difference, because 1920x1080 is upscaled on-the-fly 1440x1080 (need more computation). Also Intensity has only 8 bit video processing. You can always make upgrade from NeoHDV to NeoHD/2K :)

HuffYUV was used as an intermediate codec for processing the HDV file

Why not Cineform? Use HDlink Of NeoHD for transcode HDV to Cineform with same setting as used for injest.

[Richard Leadbetter]

My guess is that he's trying to retain as much quality as possible. By using huffyuv he's losing NO detail whatsoever.

The only problem is that by encoding to XviD afterwards, the purity of that workflow is badly compromised in trying to show the quality difference.

My advice would be to extract stills from your streams for comparison and upload them here in the .PNG format. Then we can see the difference. Try to take some shots with plenty of motion or fine detail as that's where you should see a huge gulf between the direct CineForm and the HDV capture.

[Kevin Kondra]

Here's some comparison shots. In the first one you can see the detail in the bricks and the pine trees in the background.

www.hydraclime.com/Cineform10000.png
www.hydraclime.com/HDV10000.png


The second shot show better colours in the foreground bushes.

www.hydraclime.com/Cineform20000.png
www.hydraclime.com/HDV20000.png

[Andrew Plumb]

Kevin, assuming this is the board you're using, have you max'd out the RAM across both SODIMM slots?

Sometimes you can squeeze a bit more bandwidth out of the number crunching if you use two smaller RAM modules instead of a single larger...

[Kevin Kondra]

Serge,

I don't understand why I would want to capture the HDMI/1080i signal with NeoHDV? Was there ever a consensus on how the HV20 processes the video signal internally to HDMI and HDV (is HDMI output just an upconverted HDV signal?)

[David Newman]

Kevin,

If you PC can handle it, run some CineForm captures at higher quality. Medium quality was not designed for live HDMI captures, while it looks better than HDV it can look better still.

[Kevin Kondra]

Andrew,

I'm using the AOpen motherboard without onboard RAID, similar to the one you linked. Both RAM slots are in use. The computer won't go any faster unless I overclock it, which I'll probably do next. Faster mobile CPUs are available, but I don't want to upgrade yet.