Noise comparison: 35/4:2:0 vs. 180/4:2:2

[Piotr Wozniacki]

Originally Posted by Alister Chapman

If you consider that the grain in the NanoFlash sky is moving from frame to frame it is easy to see that this could react with the detail enhancement circuits of a consumer TV to create "shimmer" while the blocky artefacts from the 35Mb/s EX would not have the same effect.

Alister,

Thanks for chiming in.

Your explanation of what's going on with nanoFlash vs. EX1 encoding, and viewing results, is the most convincing so far. The more so that you also admit the EX1 is noisy in the first place...

The excerpt I quoted has drawn my attention because - even though I have all the user-accessible "enhancements" turned off - there still may be some doing their lousy job in my plasma.

[Piotr Wozniacki]

To all that might be concerned:

When I started these tests I initially focused on the augmented noise in low-light areas. I concluded that from this viewpoint, the L-GoP format is only good for up to 100 Mbps; otherwise the I-Frame only format should be used. This was also confirmed by Dan and Mike of Convergent Design.

Later, when I started using I-Frame at 140 Mbps, I noticed the shimmering grain all over the picture - not just in the shadow areas. Its nature has been explained and confirmed by Alister. In order to get it mitigated, very high datarate is required - in fact, only at 220 Mbps does the overall level of shimmering (low-light and otherwise) become comparable with L-GoP at 100 Mbps.

Therefore, I'd like to inform all those interested that when using with a noisy cameras like the EX series, the original recommendations of Convergent Design are correct, after all:

- to keep storage low, use L-GoP at 100 Mbps
- if you need I-Frame only, record at 220 Mbps or higher.

I'd like to apologize to Dan and Mike for introducing some confusion here. That said, I still am a little dismayed with the results one can obtain with the nanoFlash, if either the camera source or the lighting and scene composition are not of the highest quality imaginable.

Thanks all,

Piotr

[Olof Ekbergh]

Piotr, a couple thoughts.

In my suite with JVC DT series pro evaluation monitors the footage from the EXcams looks much better than my Sony 46" Plasma and my Panasonic consumer LCD's. I think Alister's point is very valid, others have noted this in this thread before.

If you view a large plasma at less than 3X the distance to screen size, you will see all kinds of junk.

Grading footage before delivering product is a big part of production. This is where the NanoFlash footage really shines. It will hold up much better than the SxS footage. There are also many ways of removing "grain" in post.

Also bear in mind how the product is delivered. SD DVD, BluRay, Video server, Internet streaming or broadcast. There are probably more options as well. But the finished product is really what counts.

I think the NanoFlash footage has an advantage in all these cases as the basic GIGO effect is always multiplied.

This is not to say that a better camera does not help. Of course it does. But it really is pretty remarkable that we can spend less than $10,000.00 on a camera package that produces the kind of images we now take for granted.

[Piotr Wozniacki]

Thanks Olof for your remarks - appreciated. I realize the true advantage of the nanoFlash encoding only comes into play with heavy grading, multiple renders, etc. But I must admit I still haven't established my best workflow using Vegas (where the output possibilities are limited, and even "raw" clips at anything above 100 Mbps can hardly play back at full speed/quality).

Olof and Alister - could I ask you for a favor of proposing some typical workflows for nanoFlash material, with different editing extent and delivery methods in mind? Of course, it could be a separate thread... I'm sure many nanoFlash users could benefit!

Back to this thread: I have done a very extensive and exhaustive series of tests by now, the conclusions being what I have written above.

While I realize that - in the first part of my thread (mainly on the low-light noise augmenting) - I have speculated I-Fo might actually be better than L-GoP for bitrates above 100, it still beats me why CD guys (whose participation and support I value very high indeed) jumped so quickly from their former recommendations of L-GoP over I-Fo, to stating explicitly that I-Fo from 140 Mbps up is superior (?!?).

This is absolutely not the case; in fact, nanoFlash I-Fo at 100 or even 140 Mbps has more artifacts (not just noise) than the native EX, 35 Mbps 4:2:0 codec! I said "not just noise", because the EX codec is known to be very good at suppresing it (at the expense of detail and macro-blocking). But even the mosquito noise and macro-blocking is worse from the nanoFlash at those speeds using I-Fo!

Only at some 180 Mbps, I-Fo is beginning to look as good as the XDCAM EX. Only at 220 Mbps, can it be compared to the L-GoP at 100 Mbps. Anything above these datarate (respectively for both formats), the quality gain becomes irrelevant against the increase in storage space required.

[Gints Klimanis]

Alistair Chapman wrote "If you consider that the grain in the NanoFlash sky is moving from frame to frame it is easy to see that this could react with the detail enhancement circuits of a consumer TV to create "shimmer" while the blocky artefacts from the 35Mb/s EX would not have the same effect."

My Sony Bravia 52" (2007, 1st gen) 1080p TV excessively amplifies the block edges even after disabling all possible image DSP controls from the user menu. A convincing test is to record LongGOP from 18 MBps, 35 MBps and 100 MBps. 18 MBps is horrendous on that TV, and there is no way that this is due to noise in my EX1.

Piotr is testing on a large plasma while I test with a large LCD. Many LCD reviews include statements about "sparkle" when viewing DVD. LCDs in the past did not have adequate reaction times, and I thought these sparkles mentioned were byproducts of algorithms used to increase the perceived reaction time of the LCDs. Sometimes the discussion about sparkle is really about macroblock edge, deinterlacing and sharpening artifacts. Plasma TVs are known for faster response times, but as Alistair writes, who knows that the image DSP is doing.

[Gints Klimanis]

Originally Posted by Piotr Wozniacki

This is absolutely not the case; in fact, nanoFlash I-Fo at 100 or even 140 Mbps has more artifacts (not just noise) than the native EX, 35 Mbps 4:2:0 codec! I said "not just noise", because the EX codec is known to be very good at suppresing it (at the expense of detail and macro-blocking). But even the mosquito noise and macro-blocking is worse from the nanoFlash at those speeds using I-Fo!

Only at some 180 Mbps, I-Fo is beginning to look as good as the XDCAM EX. Only at 220 Mbps, can it be compared to the L-GoP at 100 Mbps. Anything above these datarate (respectively for both formats), the quality gain becomes irrelevant against the increase in storage space required.

These are my conclusions as well, but for the scenes I've recorded, I would still say that I-Frame100 offers more detail, less crunchiness and less mosquito noise in motion areas than 35 Mbps. If you're picking on noise on a solid color, then I would agree with you. But when I-Frame220 is used, most of the noise and MPEG LongGOP-induced shimmering artifacts Piotr has pointed out are reduced.

I think what we really want is constant quality encoding rather than constant bitrate. We are noticing the interframe quality fluctuation starting at 100 Mbps LongGOP and are dismayed that the solution is to run I-Frame MPEG2 at 220+ Mbps, which is double the bitrate. Flash cards have increased in speed immensely in the last year and can handle variations in data rates at the higher Nanoflash encoder bitrates. Is constant quality an option on the Sony XDCAM chipset ?

[Mark Job]

Hi Gints & Piotr:
I have been the lone voice in the wilderness maintaining from earliest times that I frame 220 and in particular 280 is far superior to any Long GOP bitrate you care to use. In fact, if we could get CD to give us a 300 or 320 Mbps I-Frame setting, then you would have a perfect MPEG 2 8 bit encode ! The CF cards are bigger, faster, and now cheap enough to justify this rate. Long GOP 50 is the only setting in Long GOP which matters - PERIOD !

[Gints Klimanis]

Yes, Mark. I remember your first posts about 220 MBps. Keep in mind that the rest of us were just getting used to 3-4x storage requirements when moving from HDV and Sony SxS at a merely LongGOP 100 Mbps.

Would 300-320 Mbps be enough? You probably want 400-500 MBps (only 50-60 MBytes/second) to see a difference, right? Uncompressed 1920x1080p30 4:4:4 is about 1.5 BGps and 4:2:2 is about 1.0 GBps.

[Mark Job]

Hi Gints:
I think I-Frame Intra encoding scheme is the only way to reduce the artifacts which translate as "translucent" on some of the larger screen TV's (?) Perhaps VBR encoding method could also be used to reduce overall file size somewhat for the higher data rate files encoded @ 220 & 280 Mbps I-Frame. Then again, I do not know if the Sony XDCAM HD encoding engine contained in the Nano & XDR allow for VBR encoding method ? I haven't gone back to the Convergent Design web site to see if VBR was part of their original spec. Everything I'm getting @ Long GOP 50 and I-Frame 220 and 280 looks pretty spectacular to my eye on my Sony HD SDI broadcast monitor, although after reading Piotr's posts I'm very curious to go out and watch my stuff on some really big TV's.

[Mark Job]

Originally Posted by Gints Klimanis

Yes, Mark. I remember your first posts about 220 MBps. Keep in mind that the rest of us were just getting used to 3-4x storage requirements when moving from HDV and Sony SxS at a merely LongGOP 100 Mbps.

Would 300-320 Mbps be enough? You probably want 400-500 MBps (only 50-60 MBytes/second) to see a difference, right? Uncompressed 1920x1080p30 4:4:4 is about 1.5 BGps and 4:2:2 is about 1.0 GBps.

...Yeah, I asked for 300 & 320 I-Frame data rates a few months back and Dan was not impressed.

EDIT: Then again, I asked to put a switch on the menu to bypass encoding as an option altogether (At least possible on the old XDR), and that was even a more popular request at CD then my high data rate request ;-)

[Gints Klimanis]

Mark, given the CF interface, what would you record uncompressed HD stream to? We would have to wait for a CF device capable of sustained 120-150 MBytes/second, right?

[Gints Klimanis]

Originally Posted by Mark Job

after reading Piotr's posts I'm very curious to go out and watch my stuff on some really big TV's.

Mark, try recording some small water geyers at a nearby park for 10 seconds each at bitrates from 18-280 MBps, plugin them into to a TV via the HDMI port and sit back with one of those potent 9-10% ales, such as Sierra Nevada Bigfoot or Dogfish 90 Minute Pale Ale.

[Mark Job]

Originally Posted by Gints Klimanis

Mark, given the CF interface, what would you record uncompressed HD stream to? We would have to wait for a CF device capable of sustained 120-150 MBytes/second, right?

...Hey Gints:
Not necessary. Try Raid 0. Problem solved.

[Mark Job]

Originally Posted by Gints Klimanis

Mark, try recording some small water geyers at a nearby park for 10 seconds each at bitrates from 18-280 MBps, plugin them into to a TV via the HDMI port and sit back with one of those potent 9-10% ales, such as Sierra Nevada Bigfoot or Dogfish 90 Minute Pale Ale.

....Hey Gints:
I'll better that ! ;-) Try drinking 16 cups of fresh bean Moca Java slow percolated and sit down and watch the same playback. By then end of the playback you will understand the full meaning of 2001:A Space Odessy ;-) !

[Gints Klimanis]

Mark, the most I've ever done is an Octo Espresso, which is two Quattros. How would you even order sixteen? "I'd like a HexaDeco Espresso, please?"