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

[Piotr Wozniacki]

Originally Posted by Piotr Wozniacki

OK, since the weather here has changed so that it would take some time to get identical lighting - I have browsed my older shots, and found a couple of scenes where EX1 can be compared directly to 100 Mbps Long-GoP and 220 Mbps I-frame only from nanoFlash. I will not post all of them, as they've been taken without the test in mind so they represent different scenery, lighting, and camera settings - but I can tell you that the conclusions can be summarized as follows:

1. The higher bit rate of the nanoFlash, the more detail and grain in its picture when compared to the EX1's own encoder.

2. Generally, the above tendency is more pronounced with long-GoP than with I-frame only compression

3. When comparing some perfectly lit and exposed scenes, containing negligible noise - the difference is still visible, but this time the advantages of nano's high data rate prevail: the picture contains more detail with less mosquito noise, and comparable grain level as the EX1.

To confirm my above conclusions, the point 2 in particular:

I have repeated extensive tests today, comparing SxS noise against that of the nanoFlash at 220 Mbps - and can assure you the nanoFlash is much cleaner. So, regarding the camera noise in low-lit or over-shadowed areas, here are my findings:

1. The nanoFlash I-frame only at 220 Mbps (my cards are not fast enough to test 280) is the cleanest by far !!!

2. The SxS (i..e. EX1's own 35 Mbps 4:2:0) is as clean as nanoFlash Long-GoP at up to 50 Mbps

3. With bitrates of 100 Mbps and above, the nanoFlash Long-GoP becomes more and more noisier than SxS.

My previous conclusions/recommendations, based around Long-GoP recordings mainly, should be adjusted accordingly.

Hope it helps,

Piotr

[David Issko]

Hi Piotr,

2 questions:

1. Can you get hold of another nanoFlash where you live for a comparison?
2. You recommend 50Mbps for run & gun and 100+ for controlled situations. What about 80Mbps? Does that not offer any advantage over 50, but still is just shy of the 100Mbps?

Cheers

[Piotr Wozniacki]

Hi David,

To answer your questions:

1. I know of only one other nanoFlash owner in Poland (Dan, is there more?), but haven't contacted him yet. I probably will, just to make sure my unit is up to the specs (but frankly, I'm 99.9% sure it is OK)

2. Frankly, I didn't test all bitrates as carefully as those extreme or typical ones (hence I did the 50 as a minimum that makes sense and is accepted for broadcast, the long-GoP 180 / I-Frame 220 as the maximums for my CF, needed for heavy grading in post, and the long-GoP 100 as the sweet spot, recommended by CD.

The noise increase (or lack of it) depends on so many other variables, that introducing additional nF speeds (like 80 or 140) was not of particular interest to me at this stage - but perhaps I'll include those speeds in the future tests.

[Julio Veas P.]

1. The nanoFlash I-frame only at 220 Mbps (my cards are not fast enough to test 280) is the cleanest by far !!!

sorry for the dumb question, but you mean by this: 220 Mbit is cleanest than the native ex-1 codec or the cleanest of the nanoflash flavors that you are able to test by the cards speed.

greetings

[Gints Klimanis]

Here are two examples from the same scene shot with 720p60 and Gain at + 6dB. The Nanoflash (140 MBps, LongGOP) is showing greater detail/less macro-blocking on the face and less mosquito-noise on the moving blade than the SxS (35 MBps) . Left side is SxS and right side in Nano. Overall, I'd say that the Nano is like adding a better lens.

http://www.gentlemensfightingclub.co...e_RVface4x.tif

http://gentlemensfightingclub.com/Im...re_Blade2x.png

Full frame grabs here :
Index of /Images/Video/NanoflashTest

[Piotr Wozniacki]

Originally Posted by Julio Veas P.

1. The nanoFlash I-frame only at 220 Mbps (my cards are not fast enough to test 280) is the cleanest by far !!!

sorry for the dumb question, but you mean by this: 220 Mbit is cleanest than the native ex-1 codec or the cleanest of the nanoflash flavors that you are able to test by the cards speed.

greetings

Julio,

The I-frame 220 Mbps is cleanest of all nf flavors I can test with my CF cards, AND cleaner than the native EX-1 codec. Of course, we're still talking noise here.

[Piotr Wozniacki]

Originally Posted by Gints Klimanis

Here are two examples from the same scene shot with 720p60 and Gain at + 6dB. The Nanoflash (140 MBps, LongGOP) is showing greater detail/less macro-blocking on the face and less mosquito-noise on the moving blade than the SxS (35 MBps) . Left side is SxS and right side in Nano. Overall, I'd say that the Nano is like adding a better lens.

http://www.gentlemensfightingclub.co...e_RVface4x.tif

http://gentlemensfightingclub.com/Im...re_Blade2x.png

Full frame grabs here :
Index of /Images/Video/NanoflashTest

Gints,

Thanks for this. I never doubted in nanoFlash beating the EX-1 native codec as far as most compression features are considered (mosquito noise, macro-blocking, edge and color definition), in all its modes and data rates. It's only the noise on low-signal picture areas that can be worse that on SxS, and only with 100+ Mbps Lon-GoP.

[Bob Grant]

Attached is the results of a quick test I did yesterday.
EX1, iris and lens hood closed, 18dB gain, ex factory camera settings. Waveform monitor built into Panasonic BT-LH1760 17" monitor.
L.H. screenshot is the HD-SDI direct from the camera. R.H side is the same recorded and played back over HD-SDI into the monitor. I also tried a normal image at 18dB gain and noted the same result.
Within the resolution of the waveform monitor no difference in the noise levels between the signal fed from the camera over HD-SDI and that recorded by the camera.

I would suggest that if evaluating artifacts such as mosquito noise then a HDTV should not be relied on. Real instruments are what is needed but unfortunately I do not have ready access to a hardware HD-SDI scope good enough to perform those kind of tests. Given the complexities of mpeg-2 compression even if I did I have no clue as to how to create a proper test scenario either.

[Piotr Wozniacki]

Comments please, Bob :)

Because as Alister puts it:

"The very act of compressing a signal will reduce the appearance of noise as part of the compression process is to discard some picture information, so noise often gets "smoothed out".".

And I'm prone to agree with his statement, so more compression by EX1's own encoder -> less noise from SxS than from the nanoFlash at high bit rates (less compression).

[Rafael Amador]

Therefor I don't see the point in making tests while tweaking the camera settings.
Changing the PPs we are changing the system noise.
I would make a test with the Factory Default setting. I would compare the picture on the SxS with the one on the Nano and an SDI capture through a video card to Prores or so.
IMO if its happens that the SxS shows less noise I can only think that the in camera MPEG-2 processor introduces some kind of de-noising prior to compression.
I'm not concern about noise (I de-noise everything) but in this thread there are many interesting observations statements that I would like to check.
rafael

[Piotr Wozniacki]

Originally Posted by Piotr Wozniacki

Therefore, I'd be very grateful indeed to Dan (or anybody else from Convergent Design), if the following was addressed firmly, and after giving it some thought is deserves in my opinion:

1. Can you confirm my conclusions in this thread - i.e., is what I've described confirmed with your own testings? Dan - I'd really like to exclude hardware malfunction from the equation, and I still seem to be the only one bothered with the problem :) Alister Chapman, for instance, said that "he rarely sees a huge difference between noise in his 35Mb footage and NanoFlash footage" in this thread: http://www.dvinfo.net/forum/sony-xdc...ml#post1534295

Dear Dan,

Could Convergent Design please answer the above question of mine authoritatively? I'll reword it to make it simple:

- is it normal in your testing, that at high bitrates like 180 Mbps, Long-GoP nanoFlash files tend to contain noise higher than the camera's own encoder (like the EX1/3 35 Mbps, 4:2:0)?


Also, why do you think it only happens with Long-GoP format (I-Frame only files at up to 220 Mbps do not show noise increase)?

Thanks,

Piotr

[Dan Keaton]

Dear Piotr,

In my opinion, the proper test is to connect a professional HD-SDI monitor to your camera's live output and check for noise.

While you are checking for noise, then you can record using the nanoFlash at whatever bit rate you wish.

Then, playback, from the nanoFlash the same footage.

I have done this test hundreds of times, using a $5,000 Sony LMD-2460wHD monitor, at trade shows. I have tried all types of scenes, and a wide variety of bit-rates/options, using the nanoFlash.

All of this testing was done for hundreds of industry experts. Most said, "Wow, that is really clean".

Others had the same opinion but asked for further tests. One wanted to see deep shadows, so I focused on a very dark area of the ceiling at NAB in 2009. The result was the same: Very Clean!

One expert spent 30 minutes checking the images. He asked to see just the blue channel; it was clean also.

Priotr, you can run another test:

Set up the nanoFlash for 35 Mbps (1920 x 1080) and compare the footage produced via the nanoFlash and your camera.


Please note that high compression / low bit-rate codecs, such as the 35 Mbps codec will hide/remove some detail (and noise). Higher bit-rate codecs will preserve more of this detail and noise. This is perfectly normal.

The proper test, in my opinion, is the examine the live versus recorded images on a very high quality monitor.

It is not suprising to me at all that most any camera can be adjusted/misadjusted so that noise may be visible in the extreme shadows at 600% magnification.

[Piotr Wozniacki]

Dear Dan,

Thanks for your answer. What you are saying is describing what things are supposed to be with the nanoFlash, unfortunately it's not how they actually are with my unit.

I know I should be using a proper, calibrated, professional HD-SDI monitor, which I haven't got unfortunately. Also, perhaps my previous examples of enlarged crops exaggerated things. Therefore, I'm posting two screen-grabs now that have NOT been enlarged. Please look at the back-lit surface of the house wall, and tell me where the "WOW" factor is...

The left hand one is from the EX1 (35Mbps, 4:2:0). The right hand one is from the nanoFlash at the mere 100 Mbps, 4:2:2, Long-GoP (previous examples of enlarged crops used the rather extreme 180 Mbps). I don't think I need a professional display to see the difference, Dan. While enlarged crops clearly show less macroblocking and better color resolution of the nanoFlash files, one doesn't need to enlarge the nanoFlash Long-Gop clips to see they are more noisy!

So, again - is what I have posted now normal? Because frankly, you're contradicting yourself in your answer, speaking first about how "much cleaner" the nanoFlash is, only to later admit that more noise is "normal"...

Thanks,

Piotr

PS. Disclaimer:

Please note I'm by no means bashing, ranting, or splitting hairs. I just want to understand things better, make sure my hardware is not malfunctioning, and find the best settings to use.

[Dan Keaton]

Dear Priotr,

I will let others critique those two images.

To be clear:

If you create a noisy/detailed image with your camera, the Sony EX1/EX3 will mask some of that noise/detail when you record it in camera at 35 Mbps.

If you record it in the nanoFlash, more of that noise/detail, will be preserved when recorded at higher bit-rates.

I currently see no reason to suspect that your nanoFlash is malfunctioning.

[Piotr Wozniacki]

Originally Posted by Dan Keaton

I currently see no reason to suspect that your nanoFlash is malfunctioning.

Dear Dan,

I'd like to make it clear that it's not necessarily the nanoFlash, but my camera HD-SDI port, or cables, that might be malfunctioning. My problem is I have no means of excluding any from the equations, hence my request to have somebody say "yes, this increase in noise is normal" or "no, there should be no such an increase in noise"...

Thanks, anyway.