Anyone shot much 35 Mpbs on the 700?
 

Hi there. Just got my new PDW-700 today (ex F350 owner) and absolutely delighted with it!
Has anyone compared the difference of the footage between shooting 4:2:2 at 50 mpbs and 4:2:0 at 35 mpbs. I'd be interested to see how much difference you really noticed. Shooting at 35Mpbs would certainly make things a little easier when it came to file sizes and editing. the 50 Mpbs files are pretty large to throw around - and you'd get a little more on a disc.

Cheers, Sparky

50 mb/s has been deemed a happy medium by many broadcasters, less than that and you'd maybe get questions asked. I think if it's material that's going to have any future archival use then stick with 50 mb/s, you'll be on much safer ground.
As for editing, AFAIK 50 mb/s XDCam works no problem - I used it on a Macbook Pro with no problem - but I'm no editor.
Steve

My main client has standardized on XDCAM HD 4:2:0 35mbps. They have a bunch of F350s, F70 decks and nothing to read 4:2:2 50mbps so for them, I always shoot at 35mbps. They tell me the F800 looks quite better than their house 350s, that is somewhat reassuring... On the 700, I don't think you will see much less compression artifacts when shooting at 50mbps vs. 35mbps because the additional 15mbps in the 50mbps format is most likely spent on the higher chroma sampling. Also, 50mbps is a full 1920 HD raster vs. 1440 for the 35mbps so, bits and bytes go in processing those extra pixels as well. Bottom line, the 700 and F800 look fantastic at both bit rates. For high end post and chroma key, I'd shoot at 50mbps without hesitation, for general acquisition and if you want to save on discs and/or wnat to enjoy more record time, go for the 35mbps.

Thierry.

My experience is similar to Thierry's. The 35Mb/s footage from a 700 looks very slightly softer than 50Mb/s due to the small resolution drop but there are no extra artifacts visible in first generation footage. The 50Mb/s footage does seem to hold up very slightly better in post production, but it's not a deal breaker.

Certainly the 700 looks a lot cleaner and less electronic than the PDW-F350 range, even at 35Mb/s. As always image quality starts at the front of the camera chain with the lens, then the sensors followed by codec and recording medium. Start with good clean pictures and everything else in the chain has an easier job to do.

Cool Mark,
This is the direction I'm heading into but it could be the 800 that I'll get next year.
I shoot an a daliy basis with my F350 and love it.
I would like to hear from you the challenges and new and exciting things that you have found shooting with the 700.

Cheers

I think this has been a really interesting thread. My clients loved the F350 footage I shot, so shooting with the same compression, but with a 2/3 inch camera with a fantastic sensor, could be a really good way forward. The majority of my work is high end corporate work, were I edit the footage as well, so this could be ideal. I got a bit of a shock when I first saw the file sizes of the 50 Mpbs footage! Thanks for your input.

Sparky

You say "same compression", don't forget that because of the extra pixels and 422 the 700 has the same compression at 50 mb/s as the 350 at 35 mb/s. With both at 35 mb/s the 700 is a fair bit more compressed.
Steve

Nothing wrong with the 350s. These are great and reliable cameras. I still have two of those and love their lightweight and simplicity. The 700 and F800 are definetly faster cameras. They also have better latitude, smoother hightlights handling and a very sharp sensor. Also, the 422 color space is better for high-end grading and post processing.

T.

There is no difference at 35Mb/s because both the PDW-700 and PDW-F350 record at 1440x1080 4:2:0. Compression ratio is approx 16:1.

When shooting at 1920x1080 4:2:2 50Mb/s with a PDW 700 or 1920x1080 4:2:0 with an EX or PMW-350 the compression ratio is around 20:1

I stand corrected, thanks Alister.
Steve

Not so - 50Mbs not only gives 4:2:2 sampling, but much lower overall compression as well, but the reasons aren't obvious, you can't use simple logic.

With the 4:2:2 sytem there are 2 chroma samples for every 2 luminance samples, against 1 chroma sample for 2 luminance samples for 4:2:0. Hence 4 total samples for 4:2:2 v 3 total samples for 4:2:0, so an increase in raw data rate of 33%. So, simple logic then goes on to say that 1.33x 35Mbs is about 47Mbs, or close enough to 50Mbs that it must be the same compression.

Unfortunately, that logic is flawed for two reasons.

Firstly, it assumes that chrominance samples are coded with equal weighting to luminance - and that's not true in practice. The 35Mbs aren't allocated roughly 24Mbs to luminance, 12Mbs to chrominance as may be expected. The bitrate allocation is biased towards coding luminance, much less to chrominance, so doubling the number of chroma samples doesn't need the extra bitrate (another 12Mbs) you'd initially expect.

Secondly, the extra samples have a high correlation to the original chroma samples of the 4:2:0 system - you're effectively interleaving an extra 960x540 matrix in with the original, and it's obviously going to have a lot of similarities. A clever coder will exploit that redundancy easily - hence not need anything like twice the bandwidth for twice the number of chroma samples.

Of the extra 15Mbs between the two systems, a relatively small percentage goes towards 4:2:2, the bulk goes to lowering overall compression.

My head is about to explode. But a lot has been explained :-)

See http://pro.sony.com/bbsccms/assets/f...mats_Guide.pdf there is a full table of compression ratios and sample sizes at the very bottom of the document.

You will see that the compression ratios for EX (21:1) and 422 (20:1) are almost the same, only a 5% difference which I wouldn't call "much lower", rather a tiny bit lower. Simple logic can sometimes be correct and I stand by my statement that both are "around 20:1".

That's not true. The bit rate allocation is not biased to luminance as you suggest as the chroma is already being subsampled prior to encoding compared to the luma, to further subsample would mean the signal is no longer truly 4:2:0.

Again, further subsampling through coding would mean the signal cannot be claimed to be 4:2:2. The compression ratio remains the same, thus the bandwidth doubles. But the end result is a cleaner decode is possible for the same bit rate/compression ratio as you have more samples to work with. It's the reduced subsampling that may give the 4:2:2 cameras the edge over multiple generations.

So I stand by what I said, the compression ratios for the 4:2:2 cameras and EX cameras are pretty much the same. The ratio difference of 5% is very small indeed and in most situations it would be impossible to see, the contents of the image will make more difference to the final picture than the tiny ratio difference. This is one of the reasons why I believe that in the future the BBC will accept 35Mb/s 4:2:0 for HD broadcast. The BBC's argument that only 422 at 50Mb/s is robust enough to go through the entire production chain doesn't really hold water. Yes, 422 with it's extra chroma samples should give slightly fewer chroma artifacts but I think the reality is that the difference is so small that in the real world it will make little if any difference, especially as HD in the UK is currently broadcast at 1440x1080 4:2:0.

The interesting thing is the 16:1 ratio of the older 1440x1080 XDCAM HD cameras. This lower compression ratio suggests that this mode (which is also available on the new EX1R and PMW-350) may be more robust in post production. However I think that the subsampling (1440 as opposed to 1920) may erode any advantage that the lower compression ratio offers.

Mine too! Interesting though.
Steve

I think that's to read too much into that statement. It's quite true that the compression ratios are very similar in terms of simply "data in:data out", but wrong to assume that compression ratio uniquely defines quality, even within a single basic codec family. Most crucially, the same compression ratio will give different qualities for different image sizes, everything else equal.
You're misunderstanding what I'm saying. The chroma signals aren't further sub-sampled - they just get allocated a lower bitrate than may be expected.

Let's imagine a single frame, and let's say we want to code this as three separate frames corresponding to Y, U, and V, in Photoshop, say. The Y frame will have a resolution of 1920x1080, and each of the U, V frames will have a resolution of 960x540, if we assume 4:2:0. Let's further assume we have 150kB for the whole process. "Simple logic" would mean that we assign that data proportionate to no of pixels - 100kB to Y (2megapixels) and 25 kB each to U,V (0.5megapixels). But we don't need to do that - we could allocate 125kB to the Y frame, and 12.5kB each to the U,V frames. So in Photoshop terms, we code the Y frame with a quality of (say) 7, and the U,V frames with a quality setting of (say) 2.

Which is very similar to what happens with XDCAM.

Now, we can extend that to say we now have not 150kB to play with, but 200kB - 1.33x as much. We are going 4:2:2 - so our U,V frames each become 960x1080 or 1megapixel. Even if we then double the data allocated to each (to 25kB), that still allows us now 150kB in place of 125kB for the Y frame.

Try an experiment in Photoshop. Open a frame, then compress as a JPEG with a quality setting of 5. I've just done that, with a 1067x800 pixel image, and get a file size of 199kB. Now rescale vertically, to give a 1067x400 (squashed) image and do the same thing, same quality setting. "Simple logic" will suggest a file size of 100kB, won't it? In practice, when I do it, I get 118kB - about 20% more than theory would seem to suggest - so what's going on?

It's roughly what I was referring to when I said "Secondly, the extra samples have a high correlation to the original chroma samples of the 4:2:0 system ....." You don't need twice the number of bits to code twice the number of pixels *FOR THE SAME BASIC IMAGE*. You can take advantage of correlation, and it's the same for 4:2:0 to 4:2:2.

In my hypothetical example above, you don't need 2x12.5kB for the same quality - you can get away with less for the same compression quality. So you might be able to divert even more of the 200kB away to lessen the compression in the luminance image.
It's not really the BBCs argument, but rather that of the EBU after some very extensive and scientific trials. I don't think they actually rule 35Mbs 4:2:0 out, but rather only give a full thumbs up for general acquisition to 50Mbs. It's quite conceivable that broadcasters will decide that (if they go the Sony route) 50Mbs is de rigour for higher end programming, but 35Mbs is fine for news, sport etc - we'll have to wait and see.

It may be a comparable situation to the last ten years, when 4:2:2 and Digibeta were seen as most desirable - but DSR500s with 4:2:0 and DVCAM were seen as quite good enough for some types of programming. Actually had some advantages - cost and ease of integration with non-linear field editing.
The EBU recommendations are based on multiple generations (with pixel and frame shifts) and passage through the broadcast chain with decompressions and recompressions. Hopefully, I've convinced you now that the differences are much more than just more chroma samples (and yes, much of the benefit to be lost for final 4:2:0 transmission coding) but significant improvements in lower luminance compression.

At least you can put something like a nanoFlash on a PMW350 to get full approval. You can't put an external 2/3" chipset on an HPX301.