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

[Piotr Wozniacki]

Dear Dan and Luben,

I used to have the trial version of PluralEyes that expired now, but can tell you I am usually able to synchronize manually (using waveforms) as precisely as PluralEyes can - yet in the case of EX1 35 Mbps (native) vs. nanoFlash 35 Mbps, this i s still impossible...It's close, but still the frame match is not quite spot on.

Dan, I wonder: how many frames long is the nanoFlash GoP in case of PAL (or 50 Hz Area) files, like my 1080/25p? It should be 12 frames (vs. 15 frames for NTSC) - but perhaps it isn't? If CD doesn't differentiate between PAL and NTSC GoP length, this could be one reason for the lack of frame synchronization with the EX1 native clips.

Also interesting would be to know what the number of B frames is, how the mpg2 non-linear quantization is performed, etc, etc...

BTW, it's a pity that a common utilities like MediaInfo cannot display all the CD codec info (it only displays bitrate of the video stream). No other data of the video stream can be read; MediaInfo doesn't recognize the audio streams, either.

[Bob Grant]

Now I'm getting worried, not about the noise but why you cannot align frames.
Surely you can just line two clips up using the source timecode?

[Piotr Wozniacki]

Originally Posted by Bob Grant

Surely you can just line two clips up using the source timecode?

Bob, you're a genius, and I'm an idiot.

I just tried the Multicamera tools in Vegas and managed to perfectly align a native EX1 clip with the simultaneously recorded, 35 Mbps nano clip, using "Lay out tracks using media timecode".

So, provided nanoFlash is using embedded timecode from the camera's SDI, it IS possible - at least with both clips being the same bit rate.

PS. It's also possible with different bitrates - so, looks like we have this problem (which branched off a little OT in this "noise thread") solved.

I was so focused on the noise issue that I never thought of this obvious method of alignment. Probably also due to the fact that in my multi-EX1 projects so far, I never relied on this tool in Vegas as the timecode tended to be a little off between the individual cameras, making it a must to manually synchronize using sound waveforms...


Thanks, Bob!

[Piotr Wozniacki]

OK, I have revised my test material, and think I owe you all some very important update/conclusion.

Thanks to:

a) Rafael and Dan, who pointed out the importance of comparing I-frames for any conclusion about the noise to be valid,


b) Bob, who reminded me how to properly align two clips with frame accuracy, using internal timecode (shame on me !!!),

- I can more precisely conclude now that:

1. With the "shimmering" nature of the chroma noise (whose intensity and form changes from frame to frame in any Long-GoP file), some of my previous comparisons were invalid in that they didn't compare the same, precisely aligned frames of the EX1-native vs. nanoFlash clips. After having them aligned properly using internal TC (thanks Bob again), the highest bitrate nanoclips (like those at 180 Mbps) do show considerable chroma noise, but I'm now more prone to say - with a high level of certainty - that it's not so much the nanoFlash augmenting chroma noise, but the EX1's native codec hiding it (along with detail), due to its much coarser compression which smooths out the picture

2. Properly aligned comparison of the nanoFlash high bitrate L-GoP frames vs. EX1-native frames, does NOT show more chroma noise increase than I-Frame Only clips comparison with EX1-native frames - unlike I wrongly concluded earlier.

3. While apologizing for even more ugly pictures, I'm attaching one more pair of blown-up frames - this time, the EX1 native (left) vs. nanoFlash I-Fo at 220 Mbps. While the higher level of chroma noise in the latter is evident, so is the much lesser macroblocking - especially around the high-contrast shadow patches on the roof, or around the metal nail heads in the level board.

[Dan Keaton]

Dear Piotr,

I will look into the 12 versus 15 GOP question you raise.

In reference to "MediaInfo":

We have to label certain of our higher bit-rates as 50 Mbps, otherwise many NLE's would not recognise our files.

Of course, you get the actual benefit of the bit-rate you choose.

I do not know why the audio info does not show up in MediaInfo".

[Piotr Wozniacki]

Dear Dan,

Actually, the newest MediaInfo version DOES recognize and display the bitrate correctly. It doesn't report other details (like the GoP structure and length), or any audio info at all - please see attached screen for a nanoFlash 180 Mbps L-GoP clip.

[Rafael Amador]

Originally Posted by Piotr Wozniacki

We have both agreed it's very difficult to synchronize the picture from EX1 and the nano perfectly when on separate tracks, one above the other, in an NLE. I do a lot of multicamera projects, and when only using material from several EX cameras, I'm able to synchronize them with frame accuracy using spikes in the audio waveforms.

piotr,
The problem is not to match the SxS picture with the Nano picture.
The problem is that the same frame in both devices are not necessarily the same kind of frame (I, B,P).
The camera processor and the Nano start to record in different moments.
The GOPs are not the same in both devices.
What is an I frame in the SxS file, may be a P or B frame in the Nano file.
rafael

[Peter Moretti]

Okay, I just want to say that this thread really shows how wonderful the nano is, and how excellent the guys at CD are. They patiently answered the ?'s posed here and came up w/ meaningful suggestions and good explanations.

And Piotr, thanks for all the images. This may have been a little bit of a chase down a rabbit hole, but I know I learned a lot reading through this.

Thanks again everyone and for letting me stick my nose into this. As much as the middle of this thread got a little muddy, the last two frames posted by Piotr are ringing endorsement for the nano.

[Garrett Low]

Wow,

Those last two posted pictures form Piotr really are a huge endorsement for the nanoFlash. Even though I think the picture from the EX is one of the best for any sub $10K camera out there. But add in the nanoFlash and you end up with one incredible system. Very much worth the extra money and time of setting up equipment.

Garrett

[Bruce Rawlings]

I'm pleased everything in the garden is now rosy. I spent the weekend testing out my just arrived nanoflash and am very satisfied with the results to say the least.

[Piotr Wozniacki]

Originally Posted by Rafael Amador

piotr,
The problem is not to match the SxS picture with the Nano picture.
The problem is that the same frame in both devices are not necessarily the same kind of frame (I, B,P).
The camera processor and the Nano start to record in different moments.
The GOPs are not the same in both devices.
What is an I frame in the SxS file, may be a P or B frame in the Nano file.
rafael

Dear Rafael,

I fully understand your point - there are ways in Vegas to place the cursor on the I-frame exactly. The problem had been that I couldn't find a cursor position that would be at the corresponding I-frames for BOTH clips; once synchronized though, the frame comparison is now fair.

And the results are as I described above - predictable. In both nanoFlash formats, there is more detail and more noise than in the native EX1 clip; the higher the bitrate used, the more visible the noise.

Being aware of this, one must just decide on his optimum settings (of the scene, camera settings, and nanoFlash bit rate) which may vary depending on the intended destination of the video, CF/HDD storage space available, delivery format, extent of post-processing, using or not color correction in post, using or not noise reduction in post, etc.

There is just one issue left that still requires shedding some more light at, and I'm counting on some technical suggestions from Dan: the whip pan example, where the movement blur was evidently accompanied with not just excessive noise, but also heavy pixelation of the nanoFlash frame.

I'd like to thank everyone participating in this little "investigation" of mine for their friendly advice, and their patience with me :)

Piotr

[Dan Keaton]

Dear Piotr,

I do not like to speak beyond my level of expertise and I like to defer to others when evaluating images.
(I actually learn quite a lot by reading the posts of true experts.)

This, of course, does not mean that I do not have an opinion.

I will offer my analysis of the two images in Post 103, but only since you have specifically requested it.

You were doing a whip pan, according to another post in this thread.

When you record at 35 Mbps 4:2:0 in the nanoFlash or Sony EX camera you will see less detail, the image will have a smoothing effect.

With a fast whip pan, you can present more detail than the 35 Mbps can handle, thus it can appear smooth.

With a 100 Mbps codec, the signal presented by the camera will be recorded more accurately.
(This is not just marketing speak, this is based on a very thorough analysis by Mr. Richard Wolnowski who tested our 100 Mbps codec for 6-weeks, throwing excessive detail and excessive motion to the Flash XDR. Richard's testing was with the Thomson Viper and not the Sony EX, but the principle stays the same.)

In Post 103, I do not see pixelation, but this is one of the reasons that I do not like to analyze images. Others may call it pixelation, and they may be 100% correct.

Also, I do not know how this image plays back at normal speeds.

I assume that this was shot at very low light levels; you have provided much of the camera detail about this shot in a previous post.

May I suggest that you try some whip pans, if you are going to include them in your work, under more normal lighting conditions.

Also, since I do not present myself as an expert at image analysis, I welcome others to join in, in analyzing the images in Post 103.

I hope this helps.

[Piotr Wozniacki]

Dear Dan,

Thanks for the clarification of your standpoint on this. Like you, I'm inviting everyone to discuss this issue. Myself, I'll repeat the "whip pan" experiment in a more controllable fashion.

In the meantime, however, I'd like to draw you attention to, and ask your opinion on, the following observation that is a little off-topic in this thread, but has surfaced as a side-result of my investigation described herein. As I mentioned in previous posts, I did have some difficulties in perfectly aligning simultaneously recorded clips from the EX1 and the nanoFlash. The method I've always used before with EX1 clips was based on the audio waveforms, and it doesn't work when one of the clip is a native EX1 clip, and the other comes from the nanoFlash: after aligning the waveforms precisely, the picture frames are slightly off.

At some point, Bob grant has reminded me about the obvious method based on the clips' internal timecode; using it it's now possible to align the video frames perfectly, but I'm seeing the following now:

1. The nanoFlash clip offset is some 9 frames, i.e. after precise video alignment, the nano-recorded clip starts 9 frames later in time than the EX1 "original". Some SDI latency is normal, I suppose.

2. What surprises me though is that with such a perfect picture frame alignment, the nanoFlash-recorded audio is further offset by an additional 1 frame (so the total audio offset between the nano clip and the EX1 original is some 10 frames). It would also explain why I couldn't get frame-accurate video alignment when using the audio waveforms!

The above numbers I can see with a 100 Mbps, Long-GoP clip. I haven't checked whether the offsets are the same with other formats/bitrates.

Could you please shed some light on it?

Piotr

[Dan Keaton]

Dear Piotr,

I am assuming that you are triggering the nanoFlash on Incrementing Timecode.

Yes, there is a slight delay in recording, a few frames. Of course, if these few frames are important, then you could use our pre-record buffer.

Generally, I recommend against using the pre-record buffer unless there is an actual need for it. It is easier to edit, in my opinion, if the extra seconds at the beginning of each take (the pre-recod buffer footage) is not there.

Is you nanoFlash in an idle mode (not playing back) when you trigger the camera to record?

If you nanoFlash is playing back, it takes longer to switch the nanoFlash (actually the Sony Codec), into record mode. We have to switch the Sony Codec from Playback Mode to Record Mode.

What firmware version are you using? This has significance in your audio/video sync issue.

[Piotr Wozniacki]

Dear Dan,

Of your questions, the only one relevant to the audio/video delay difference is the firmware version I'm using - and this is the current 1.5.126.

I don't care about the overall delay - as I said, this seems normal to me.

Thanks,

Piotr