XQD vs SDHC comparison

[Steve Siegel]

I've been putting my new PMW 300 through its paces today by going to the local zoo and filming birds in the aviary. It's really a wonderful place, if you are ever in Miami. Asian birds are featured, all free flying and gorgeous.

Anyway, this test was precipitated by a Sony telephone rep who told me that I could only use SXS cards in the PMW 300, or the still very expensive XQD S series cards if I intended to take advantage of 4:2:2 50 mbps filming in the UDF mode (as opposed to the FAT mode with the old, and cheap, SDHC cards at 4:2:0 35 mbps). I didn't want to believe that, because there are XQD N series cards that are half the price of the XQD S cards. The rep wouldn't answer me regarding XQD N, so I bought one ($99 US for 32 GB). The good news is that it works perfectly. A 32 GB card gives an hour of filming. You need a $30 adapter, but you need it for the XQD S card too.

I proceeded to shoot the same birds in similar or the same light using XQD N in UDF (422) and SDHC in FAT
(420). I used both my Nikon 80-400 zoom, and the Fujinon stock lens. I shot at 1080 29p and 720 60p for slow motion. Hypergamma 2 for all. Back home I applied several effects in Premiere Pro CS5, especially trying to reduce burnout of the whites in the 3 way color corrector, looking for inherent differences in color saturation, and by blowing up the view in the monitor window, looking for when pixellation became evident.

I could see little or no difference between 420 and 422 with any of these maneuvers. My question is: What is the big deal with 422? Am I missing something? Do you need fancier equipment than my computer monitor set at 1920x1080 resolution to appreciate the difference? Spending all this money on the latest and greatest would hurt a lot less if the advantages were at least obvious to the eye.

[Brian Murphy]

There will certainly be a difference if you were to grade 4.2.0 compared to 4.2.2. The technical advantages of the latter are well documented. I will also be running XQD cards soon, but only for certain clients. My broadcast clients are still using 4.2.0 because it takes less bandwidth and that means dollars to them. We will also need to arm our PMW 300s with XQD when Sony finally releases the firmware update as I am pretty sure it will like the other Sony XAVC cameras need XQD to record that codec.
Love the camera! Just wish someone would make a less expensive baseplate than Vocas.

[Steve Siegel]

Brian,

I think manipulating the footage in PPro constitutes grading, doesn't it? Adjusting white levels, color levels, etc. I did that. Can you suggest a grading manipulation where the difference would be evident?

[Matt Davis]

Steve - aside from certain UK/EU broadcasters' technical requirements, 4:2:2 really comes into play where strong colours are found: theatrical lighting, concerts, and red fabric especially. Chromakey also benefits from 4:2:2 origination, even though the 'hybrid' keyers such as FCPX/Motion and Keylight do a great job with simple 'taling head', anything involving translucency, reflections and full length pale coloured clothing has to be done in 4:2:2.

The other side of the equation is the 50 Mbit thing. XDCAM is extremely robust at 35 Mbits, but with highly detailed images and fast moving subjects, or anything with a lot of gain pumped in, benefits from the extra bitrate so that every frame contains clean images free from mosquito noise, blocking and so on.

FWIW, XAVC is lovely, but it's a bitrate hog for some jobs. XAVC-S may also be worth keeping an eye on for long form recording. I did some torture testing between XDCAM, XAVC-S and AVCHD and the winner was clear.

Very interested in how the cheaper XQD media fares in SxS adaptors for 50 Mbits. Should be well within the comfort zone even for S&Q.

[Brian Murphy]

Steve,
I think Matt has outlined what I was eluding to. Though my experience with this codec is limited, my previous expereiece with 4.2.0 had me dealing with issues when working with blacks and shadows that I had to solve by using a Ninja.
All the best!

[Steve Siegel]

I'm way out of my league here! My cameras have never been in a concert hall, or a theater. I have sort of used chroma keying once. I don't even know what a Ninja is (other than a turtle or bad guy). Guess it's just stick with 420 FAT and let the birds do what they will.

Thanks for the inputs, however.

[Alister Chapman]

Contrast, brightness, black levels, saturation etc are no different with 422 or 420.

The difference is in the chroma resolution or color detail as well as some difference in color noise. If you shoot interlace the difference between 420 and 422 is more evident than in interlace.

Both 420 and 422 sample Y (brightness/luma/contrast) at every point in the image, so these are the same.
422 samples color (CbCr - color difference from Y of blue, color difference from Y of red) every other pixel on every line.
420 samples color every other pixel on every other line, it samples Cb on the first and the Cr on the following line. This results in noticeable banding in highly saturated colors. In interlace, in each field the color is sampled every other pixel, every other field line so there is a 2 line gap between samples in any given field.

See the attached example, this is actually two progressive frames, in interlace the 420 would be even worse. You have to "pixel peep" to see this, or be viewing on a large screen, but there is a difference. Whether it bothers you or not is another matter. Most web delivery and digital broadcasting uses 420, so often if you do shoot 422 the benefit may be lost further down the line.

[Matt Davis]

Originally Posted by Alister Chapman

You have to "pixel peep" to see this, or be viewing on a large screen, but there is a difference. Whether it bothers you or not is another matter. Most web delivery and digital broadcasting uses 420, so often if you do shoot 422 the benefit may be lost further down the line.

Absolutely - but that's what broadcasters are looking for - resilience.

Shoot and deliver 4:2:2, and when the inevitable down conversion happens, you don't suffer the concatenation effects of repeatedly re-encoding like you have in the broadcast chain. There's not the sudden drop in bit rate, but the re-encoding process hurts the original.

Delivering to the web is also an interesting issue - we've got to present the best quality to the video host's compressor - who knows what it will do to our footage, so we present as pristine an image as we can. The trouble is, that's still a 4:2:0 codec (H.264 usually at 5-20 Mbits using high profile).

I would prefer to make that master 4:2:0 file from a 4:2:2 source that is mercifully free from noise, motion artifacts and oversharpened images, so we still come back to 4:2:2.

Agreed that the average talking head or still pack shot is not going to immediately exhibit telltale 4:2:0 concatenation issues, and quite frankly will be 'good enough', but it's the ease at which one can shoot 10 bit 4:2:2 internally where an external recorder would be a real drag - OB B-Roll Run & Gun onto XQD at XAVC vs lug a NInja or Samurai? :-)

[David Heath]

Originally Posted by Steve Siegel

I could see little or no difference between 420 and 422 with any of these maneuvers. My question is: What is the big deal with 422? Am I missing something?

In HD, a lot of the desire for 422 applies specifically to interlace systems - use progressive and it's less of an issue. A lot of the rationale goes back to SD days, and then was much more of an issue in the US than the UK.

The reason is that for NTSC countries, when digital recording first came in at the lower end (DV, DVD etc) then most systems tended to be either 420 OR 411. Typically 411 for recording (DV, DVCAM, DVCPRO), 420 for distribution (DVD, broadcast etc)

The problem there was that one quarter the potential chroma information was discarded in the 411 coding, then half of that was further discarded in the move from 411 to 420. The result was effectively equivalent to "410" - or only one pair of chroma samples for every EIGHT luminance! So in NTSC countries, 422 was much preferred largely because it wasn't 411 rather than it was 422, if that makes sense!

In PAL countries, 411 coding was hardly used - DV, DVCAM were both 420, unlike in NTSC. Hence, for straightforward work, with a 420 end product, there was no difference between starting with 420 or 422. Such as a DV to DVD recode didn't mean this further halving of chroma information.

That said (and even in PAL countries) then if much post work is to be done, 422 has much greater relevance in an interlace system than a progressive one - and that's true for HD and 1080i systems as well as SD. And that's why 422 is largely insisted on by many broadcasters, who are most often likely to transmit 1080i/25. (Though note 420 (via XDCAM 35Mbs) is considered broadcast standard for "journalism", and heavily used for news.)

Move to HD progressive, and a 422 system will still have some advantage over a 420 one - but it's not anything like the level of importance that it is with interlace, and especially when that was interlace and NTSC.