What is Sony's XAVC Format?

[Eric C. Petrie]

Over the last 2 years Sony has slowly began to roll out a new format, XAVC. What began with the F5 & F55 cinema cameras is now a format that is used in multiple ENG cameras, DSLRs, broadcast decks, and more. It’s clear that Sony intends XAVC to be the do-it-all format for the next generation. But what is it?

The core technology of XAVC is h.264 mpeg-4 AVC, the same technology that is used in AVCHD cameras and DSLRs. But while AVCHD typically runs between 24 and 35mbps and is 8-bit 4:2:0 Long-GOP, XAVC aims much much higher. The reason XAVC was invented was Sony needed a next-generation format to be able to deliver 4K content with. It needed to be high quality in order to appeal to the commercial and cinema world. One important aspect was moving up to 10-bit recording. Sony’s also not a big fan of licensing existing codecs. So, in 2012 when they were getting ready to launch the F5 & F55 they wanted to have their own high quality deliverable native editable format that could encompass 4K workflow. Hence XAVC was born.

There are a few current flavors, or implementations, of XAVC. The first is full quality XAVC. This is a 1080p, 100mbps(at 30p, 200mbps at 60p), 4:2:2, 10-bit, Intra-Frame format. It’s extremely robust. It can also be 4K, 300mbps(at 30p, 600mbps at 60p), 4:2:2, 10-bit, Intra-Frame format.

There is another flavor that is very similar but it is Long-GOP. This is a 1080p, 50mbps(at 30p, 100mbps at 60p), 4:2:2, 10-bit, format. So you still retain much of the detail but you lose motion information. It can also be 4K, 150mbps(at 30p, 300mbps at 60p), 4:2:2, 10-bit. This flavor of XAVC hasn’t been put into use yet but it is expected to be made available as a firmware update to current XAVC hardware during 2014............

Read more What is Sony’s XAVC? | News from the NW Digital Cinema Dealer

epetrie@provideoandtape.com

[Noa Put]

You ask a question and then provide the answer yourself? That's not how a forum works :)
Excellent info, there is so much changing these days it's hard to keep up, your post makes it much more transparant.

[Eric C. Petrie]

The questions in the tag grabs peoples attention. Plus it makes it pop up on Google search more readily when people try to find an answer. Thanks for taking the time to check out the post. I'm glad it was helpful.

[Christopher Young]

Sony's whole XAVC / XAVC-S overview was pretty well explained in their white paper release of 2103 called

XAVC_SpecificationOverview_Rev2_130913.

This revision 2 release from Sept last I believe is the latest version. Quite an interesting read. See attached PDF

Chris Young
CYV Productions
Sydney.

[Andrew Smith]

I've only just started making my way through the white paper, and it's interesting to note two things already.

1. Apparently TV industry types don't see the mpeg2-based codecs as being a suitable replacement for their current XDCAM disc / HDCAM tape formats. They want to move over to an mp4 wrapper/codec ... hence the need for XAVC.

2. Sony are not publishing the write speeds of the XQD cards in their table on page 3. The other media listed are the technology variants of the SxS format.

Andrew

[Ron Evans]

XAVC is an MXF wrapper and XAVC-S an mp4 wrapper. Sony have a range of XQD cards to cover the lower speed 8 bit 420 XAVC-S format as well as the XAVC 10bit 422 variants. XQD S series are up to 180MBps write speed and the slower N series are 125MBps which I use in my FDR-AX1. Write speed are on the Sony WEB site for each of these cards.32GB High Speed XQD Memory Card - Memory Cards & Adapters Sony Store - Sony Business Store

Ron Evans

[Greg Boston]

Originally Posted by Andrew Smith

1. Apparently TV industry types don't see the mpeg2-based codecs as being a suitable replacement for their current XDCAM disc / HDCAM tape formats. They want to move over to an mp4 wrapper/codec ... hence the need for XAVC.

A bit ironic since XDCAM is an MPEG2 based codec.

-gb-

[Andrew Smith]

That's what I was thinking.

I suspect the real story is that they don't want to move to a new codec infrastructure and still remain with the same quality as before. If you're going to invest in upgrading your systems, you might as well improve your vision quality once again (with the next generational leap).

Andrew

[Dave Sperling]

Remember, as new sensors are developed with ever increasing dynamic range, log-based recording formats become more necessary when you want to capture all the dynamic range from the sensor and squeeze 0-1200+ into a bucket that's only 0-108 in size -- and while 8 bit can be very useful and efficient for images with the desired look 'baked-in', log compression/ decompression looks significantly better when using 10 (or more) bits of data.
Since the basic XDcam formats are 8-bit, it made sense for the next generation codecs to be at least 10 bit.

[Christopher Young]

Whilst it is great to have the 10-bit range in post to work with the real world scenario in TV, not film, of being locked into delivering an image that is locked down to a Rec 709 six stop dynamic range signal that in most countries is transmitted as an 8-bit, 4:2:0, 1440 x 1080 signal, yes not 1920, really can cause problems.

I’ve seen 10-bit material in post that looked great in the grade. Nice skies, nice graphic grads, nice smooth black areas and then seen them all go to the dogs when transmitted. By the time the down sampling and many steps of cross coding take place in the transmission path that lovely 10-bit image can be worse off than an original 8-bit image that suffered far less in that 8-bit transmission chain. If in post you can get an 8-bit 4:2:0 image to look decent on its graduations there is a good chance it will hold up better in the widely prevalent 8-bit 4:2:0 transmission path.

A really good explanation of these problems and solutions being developed to overcome them is explained quite clearly here:

http://www.fujitsu.com/downloads/COM...enation_wp.pdf

It is in part because of these sorts of problems, plus data size in HD, that Sony went the 8-bit MPEG-2 path with its XDCam HD kit. This was a pronounced step back from its previous 10-bit Digi-Beta 4:2:2 design philosophy in SD where the transmission payloads and compression were far less.

I recently saw the problems that can be caused in the transmission chain with a ten camera show we posted where two of the cameras were 10-bit 4:2:2 DVC Pro 100-mbit AVC intra cameras, beautiful pictures but... The rest of the cameras used were XDCam 50-mbit 8-bit 4:2:2 and Sony EX 8-bit 4:2:0 cameras shooting at 35-mbit. The camera images that suffered the worst and that was from smearing, were the AVC Intra cameras once they were brought down to that common denominator of an 8-bit, 4:2:0, 1440 x 1080 transmission chain.

Most of our lovely 1920 x 1080 flat panel TVs are the last step in affecting the quality of the viewed signal as we are dependent on just how good that 1440 upscale to 1920 is.

The worst case scenario we see is when using the fantastic looking 10-bit 4:4:4 graphics out of the Quantel and Pixel Power TV CG / graphics paint boxes. In short they are almost unusable on air without Gaussian dithers applied to avoid banding on highly compressed 8-bit productions.

Love the 10-bit to work with but in many cases in the TV world its extra data payload and concatenation problems in the transmission chain make it easier and cleaner to work in 8-bit front to back.

Chris Young
CYV Productions
Sydney

[David Heath]

Originally Posted by Christopher Young

I’ve seen 10-bit material in post that looked great in the grade. Nice skies, nice graphic grads, nice smooth black areas and then seen them all go to the dogs when transmitted. By the time the down sampling and many steps of cross coding take place in the transmission path that lovely 10-bit image can be worse off than an original 8-bit image that suffered far less in that 8-bit transmission chain. If in post you can get an 8-bit 4:2:0 image to look decent on its graduations there is a good chance it will hold up better in the widely prevalent 8-bit 4:2:0 transmission path.

A key point is that "blocking" can be caused via two means. Yes, insufficient bit depth, especially with post manipulation, is one, and maybe the one that most people are aware of, and the mechanism is relatively obvious.

But it can also be caused by compression itself, especially chroma compression and compression of the chroma signal may be more severe than applied to the luminance channel.

If that doesn't seem obvious (well, it isn't!), try an experiment in Photoshop. Form a gradient, then in "levels" bring the white level right down (to reduce the bit depth) then expand back up again. You can expect fairly severe banding.

Now start again, but this time just save the gradient as a highest compression JPEG and see what happens. Yes, banding again, but would that be what you'd have expected? It's a good illustration of why banding is not necessarily a function of low bitdepth.

And that's why your comments about AVC-Intra 100 and XDCAM 422 shouldn't necessarily be a surprise.
It's common to think 10 bit can only be good - but it means 25% extra data to compress, so the compression rate of AVC-Intra 100 is therefore equivalent to the same system using 80 Mbs for 8 bit coding. It has the advantages of H264 compression tricks - but they don't offset the long-GOP efficiencies that XDCAM422 has. Result - higher compression, and that can give rise to gradient blocking as in the second Photoshop example above. 10 bit acquisition may well be making the situation worse than 8 bit, assuming the same datarate - it may be more efficient to use the datarate to lower overall compression - not increase the bitdepth.

So I agree with your conclusions. 10 bit is a great idea in theory - but it needs sufficient bandwidth. If that's limited, it may well be better to use 8 bit and reduced compression.

[Andrew Smith]

Awesome post. Much appreciated.

Andrew