Alan Roberts BBC test report for the F3 is now available at BBC R&D White Paper WHP034 - Alan Roberts (http://thebrownings.name/WHP034/)

So by his measurements, the S/N ratio is -48ish?

And the AF100 is -49ish?

What?

I'd sure really like to understand his methods better, because his results don't match my eye. In addition by his own admission, he states that because pretty much all cameras have gamma correction before any sort of output that you can measure, direct measurements won't be accurate. I don't know if this is really true, or?

Anyway, sure glad I tested both AF100 and F3 before I bought, because if all I did was read this document to make a choice, I'd be missing out on a whole lot.

Yea and the EX1/3 are -47db at +3db gain on their report.. Kinda odd.

I don't mean to nit pick too much, but I see two errors right on the first page:

". . .but the supplied BP-U60 has a nominal capacity . . ."
Unless they are bundling a battery in the UK, the camera does not come with one.

" . . .the lens mount is standard PL, and has hot connections for the supplied Sony lenses . . ."
Does he not realize he's looking at an adapter? The native mount is an F3 mount.

I'm wondering now if he has a typo and ment to say the F3 is -58.5db.

I have recorded 56 hours of F3 material on both Nano and internal. Only once did I see aliasing, and it was from a tight patterned shirt a grip was wearing during my last trip to London. It showed up in a fold of the shirt, changing the presentation of the H/V pattern more diagonal. This seems to make sense from Adam Wilt's Mega Trumpet test. See ProVideo Coalition.com: Camera Log by Adam Wilt | Founder | Pro Cameras, HDV Camera, HD Camera, Sony, Panasonic, JVC, RED, Video Camera Reviews (http://provideocoalition.com/index.php/Awilt/story/ag-af100_and_pmw-f3_on_the_charts/P1/) .

Ill be returning to London and hiring the same grip with the same shirt. I will try to repeat the situation. I'll set Detail Level to 0, Detail Frequency to +99, and Aperture level to +20 as per pp10-11 Roberts Addendum 68 and see what we get. If it is still pronounced, I'll buy his shirt and burn it :) .

Mean while most of the material from the F3 has been posted for three network prime time programs (Sorry no frame grabs due to property rights), with very favorable reviews.

I don't mean to nit pick too much, but I see two errors right on the first page:

". . .but the supplied BP-U60 has a nominal capacity . . ."
Unless they are bundling a battery in the UK, the camera does not come with one.

" . . .the lens mount is standard PL, and has hot connections for the supplied Sony lenses . . ."
Does he not realize he's looking at an adapter? The native mount is an F3 mount.

There was some other factual errors like not being able to be remotely controlled (RMB-150 & 750 both work with F3). Maybe his criteria for "remotely controlled" is more stringent? Overall it seems like his time with the camera was really limited and he didn't do much research along the way.

Anyway, I don't want to pick his paper apart in fanboy fashion, because I really do believe by the tone of his writing that he doing his best to be scientific and impartial. There is however that one big head-scratcher with the noise level....

I still say it has to be a typo. That is a HUGE difference from the claimed S/N ratio. Greater than any other camera he tested I bet.

Roberts is also claiming that "the sensor has approximately 12.9 Megapixels, typical of a digital stills camera." Sony's F3 brochure states that the 'effective resolution' of the sensor is 3.3 Megapixels (2.4K)? What's going on with this report?

Here's another error:

"Sensitivity was not measured directly. The specification claims it to be T/11 at ISO800, and since ISO800
corresponds to 0dB gain, this means that the sensitivity is very . . ."

0db corresponds to ISO 400 not 800.
And even that is bogus. The ISO varies depending on the paint settings.

Roberts is also claiming that "the sensor has approximately 12.9 Megapixels, typical of a digital stills camera." Sony's F3 brochure states that the 'effective resolution' of the sensor is 3.3 Megapixels (2.4K)? What's going on with this report?

Apparently he missed it when Sony clearly said the pixels are 4 times bigger than a typical SLR's pixels.

In fact the whole report is riddled with typos and mistakes including this one: " . . . and audio via SLR connectors."
Sloppy work.

Seems odd he says "The lens mount is standard PL, and has hot connections for the supplied Sony lenses and for the Cooke /I range of lenses." but hen he makes no mention of the Arri LDS interface.

he also says " Noise performance is good, 18dB gain may be reasonable for best quality programme-making."

So forget about dbs and numbers

I don't mean to nit pick too much, but I see two errors right on the first page:

". . .but the supplied BP-U60 has a nominal capacity . . ."
Unless they are bundling a battery in the UK, the camera does not come with one.

" . . .the lens mount is standard PL, and has hot connections for the supplied Sony lenses . . ."
Does he not realize he's looking at an adapter? The native mount is an F3 mount.

I think he's referring to the camera as supplied to him, rather than how you'd purchase the camera. I guess he's interested in the high end sector and how the camera is set up for his test.

The noise levels are interesting, his graphs indicate the same low figure. It could be his test equipment's calibration is off? In practise, no one has commented on the F3's noise levels, whereas the AF 100's have been commented on. The reference to "18dB gain may be reasonable for best quality programme-making" seems contradictory.

I don't get the noise figures either. I don't quite understand why he off hand ignores the +2db bump in his noise plot at 0db say it's "not significant" when if you follow the otherwise reasonable looking noise/gain plot you would see an improvement of around 4db in the noise figure. The plot does not make sense, noise and gain almost always go hand in hand and most of the plot makes complete sense, the big bump at 0db is very odd. I would have a much easier time believing -52db (strangely close to that of the FS100) than -48db. Did he have +9db programmed and not zero db??


Alan is also implying equal R G and B pixels which is not typical of a bayer sensor which normally has 2 green for each red and each blue. Now Sony have not said that the F3 is bayer, so perhaps this is something different. Zone plates don't normally lie.

I also don't follow the pixel count logic:
quote
"implying that the sensor has dimensions which are related to those numbers, probably 2200x1238"
then
"and that the sensor resolution is probably much higher than 2200x1240"
and
"the native resolution of the camera does not reach the limits of the 1920x1080 format"

Now I know that a bayer sensor will produce a final resolution lower than the pixel count, but it all seems confused and the final conclusion that it must be a 12 mega pixel sensor because it has similar sensitivity to a 3 chip camera, well that's conjecture beyond belief. Also I don't see how 4,800 x 2,700 is a "reasonable fit" with 2,200 x 1,238, there is a pretty big discrepancy even if you do multiply 2200 x2. Anyway we know the sensor to be 3.5MP gross with 3.3MP active.

What we might be seeing is conventional conclusions about an un-conventional sensor?

looking at the noise plot again, Alan is suggesting that the F3 is noisier at 0db than at +6db. That really does not tie in with my visual observations. It implies we should be using either +6db or -3db but not 0db.

I was a little surprised to see that Alan forgot to add the customary conclusion to his report. Reading in between the lines, is it fair to assume that along with an external recorder the F3 is 100% compliant with the BBC's (and Discovery/Nat Geo) HD requirements?

I always wonder how the heck are they going to know what whether you used a nanoflash or SxS cards?

I think there is a machine that spots the characteristics of different cameras and recording systems. All part of checking SD or up-rezzed material within an HD programme.

Can they really tell SxS 1080 4:2:0 from 4:2:2 ? I'm dubious but very interested.

You can if you know what to look for very easily spot the difference between 4:2:0 interlaced and 4:2:2 interlace. With progressive it's a lot harder, but the difference can be seen as a difference in chroma resolution from horizontal to vertical.

Have you guys heard of shows getting busted for trying to pass?

I don't know about busted, but I've heard of BBC shows needing to redo material and they weren't even HD.

Yes, a major UK based production company is having to re-shoot an entire series after trying to pass off XDCAM EX as XDCAM HD. They were contracted to shoot on XDCAM HD but used EX instead. It's possible that they may also get sued.

Alister,
Do you know how they were caught? Was it the footage itself, or did they get busted in some other way - i.e word of mouth, set photos, etc.

No, I don't know.

As a documentary editor when you look at material you get a feel for picture characteristics. When cutting a mixture of formats together although there is nothing wrong it can be easy in many cases to spot EX1 material for example in amongst HDCAM pictures as the lens quality can jump out at you if nothing else.

The engineers tend to use higher quality monitors, which reveal flaws giving clinical, rather than the nicest looking pictures.

Lens quality is pretty easu to spot, but for example, would you be able to easily spot EX or F3 footage shot on a Nanoflash at 8 bit 4:2:2 from an SxS card at 8 bit 4:2:0. Lots of people have said you generally only see it in the grading.
Who's to say how you shot the camera unless there is a real gotcha. Do they do an investigation with interviews and make you swear on the bible?

I've been asking this of a few people recently and no one has given me a clear answer except they have good monitors and are smart engineers. No doubt but what is it they will look for if the images look good as we know they can from either an EX or F3 if shot right in the first place. Certainly as good as lots of stuff shot on 720 HDCam did.

There must be people BS'ing this and getting away with it but I guess they're not about to come forward are they. Kind of like cheating on your taxes.

I was recently involved with a documentary for a major UK broadcaster who insisted on XDCAM HD 50Mbps. Fair enough as they are the commissioner.

I explained the additional cost to the producer who had planned to shoot XDCAM EX (PMW350/EX1/3). The producer explained to the commissioner that XDCAM HD would cost x pounds more. They then decided that XDCAM EX would be fine. The additional cost was small, I think we added about 50% cost of the Nanoflash and CF cards but it was enough to make XDCAM EX "good enough."

I would never lie to a client but if if they want higher end kit, the budget should reflect those demands.

Regarding lens quality....

On F3 footage, regardless of using a nano or not, a poor lens will taint the image more than a sub-par recording media. I suggest folks budget for renting good lenses, prior to shooting important projects.

Timur,

Exactly my question - How the heck do they know what you shot on? Can they spot a poorly shot film with a weaker camera and crappier lenses with the correct codec, as opposed to a well shot film with sharp lenses but the EX SxS 4:2:0 codec - and then reject the latter?

I'm not trying to be difficult or obnoxious here i just don't get it.

Lenny

I don't know if this would be how they'd do it but looking here:

Chroma subsampling - Wikipedia, the free encyclopedia (http://en.wikipedia.org/wiki/Chroma_subsampling)

There are artifacts in 4:2:0 that a program could possibly be used to detect.

I'm not saying this is what they'd use, but it's an example of the test equipment available to engineers.

http://www.cnrood.com/PHP/files/video_pdf/Tektronix-PQA500.pdf

I shot a lot of stock footage for Corbis Motion; now Thought Equity, and their technical engineering department picks apart the footage and sends me a detailed report of all that did wrong ie.... illegal over saturated colors and whites too high above 100IRE, usually accompanied with a hand written note that says "Looks amazing!". Their standards are 1920x1080 minimum, 100mbps, 422 color space. Technically speaking the HVX200 was the only non 2/3" camera they accepted a few years ago when i was shooting for them. They did not accept 720p in any shape or form. I did a bunch of 60p slow motion work with a SGproREV1 and too kthe 720p, brought it in to Final cut, and output it as 1080 24p, and submitted it with fingers crossed. Thy bought the entire lot of footage, with the usual "illegal whites" and a note that said "The 720p holds up well....." So go figure. I think its more about content.

will sony bring out a shoulder mount version of this camera,surely it'll become the indi industry standard if they did, using the pmwf3 as the B camera???

Yes. It's called the SRW9000 PL.

thats an $80k camera, not $20 odd k like the f3

You'd need to use HDCAM SR or similar for a shoulder F3 to become the industry standard. With the current codec it's fine for non HD broadcast work, or the B camera.

I think you'll just have to pimp your F3 in the short term, unless Sony decide to repackage the camera into a shoulder body as they've done in the past with some of their handicam style offerings.

Perhaps Sony may bring out a tapeless HDCAM SR camera to compete against the Epic for those producers who don't want a RAW workflow.

I would say Sony will release a shoulder mount using the F3 sensor at some stage, but to make it worth the size increase and not to under cut their higher end it would probably included a fixed SR-R1 type deck, S-Log and a commensurate cost increase.

Perhaps Sony may bring out a tapeless HDCAM SR camera to compete against the Epic for those producers who don't want a RAW workflow.

Excuse me if I'm misundertanding what you are speculating about, but Sony has already announced a new version of the 9000PL that will use 1TB SR memory cards instead of tape. That's the same cards that the upcoming SR-R1 will use. THAT will be the camera to have.

That could be the camera I had in mind, I haven't been following the Sony 9000 developments.

I also don't follow the pixel count logic. What we might be seeing is conventional conclusions about an un-conventional sensor?

The problem is that while Sony refuses to provide specifications (see Juan's video where actually say he can't provide the numbers) they have said that 3.37Mp are "used." That forces assumptions to be made.

But Allan's paper -- if you read it carefully -- works through the process very clearly. Here is my parsing of it:

Figure 1 shows the luma resolution when the camera was in factory default settings for detail enhancement.

The result is not free from spatial aliasing, there [ARE] both horizontal and vertical visible aliasing. The aliasing, is, [UNUSUAL], both first and second order, i.e. frequencies are reversed and then reversed again.

The first alias [IS] centred on 1920 and 1080,

WHICH IMPLIES IT IS FROM THE RECORDED FRAME-SIZE.

while the second aliases are centred on about 1100 pixels and 619 lines, implying that the sensor has dimensions which are related to those numbers, probably 2200x1238.

MY ESTIMATES ARE 2456 AND 1380. HOWEVER, THE VERTICAL ESTIMATE IS TOO HIGH.

THIS IS A 3,389,280 PIXEL FRAME. SONY SPECS 3,370,000 PIXELS.

THIS IS BOTH: THE FRAME THAT RESULTS FROM THE "DOWNSCALE" OF THE PHOTOSITES FRAME AND THE PRE-DEBAYERING FRAME. THIS FRAME , THROUGH DEBAYERING, BECOMES 1920x1080.

The F3 specification [IS] T/11 at ISO800, and since ISO800 corresponds to 0dB gain, this means that the sensitivity is very similar to that of a 3-sensor 2/3-inch sensor,

IN HIS EXPERIENCE, THIS VALUE, IS WHAT HE TYPICALLY SEES FROM 2/3-INCH CHIPS

which in turn implies that the pixels are about 5μm square.

THESE CHIPS, HE STATES, HAVE A TYPICAL SIZE OF 5μm

Given that the sensor is ‘super 35mm’ size, it must be 24x13.5mm. [Not the 23.5x15.6 of an APS-C]

SIMPLE MATH, GIVEN THE DIMENSIONS

For the pixels to be 5μm spaced, the sensor width must be about 4800 pixels, making the sensor approximately 4,800x2,700.

MY ESTIMATES ARE 4912x2760 WHICH MEANS THEY WOULD BE SLIGHTLY SMALLER AT ABOUT 4.784 WITH AN AREA OF 9.568

This fits reasonably well with the estimations in section 1.2.1, and means that the sensor has approximately 12.9 Megapixels, typical of a digital stills camera.

MY ESTIMATE IS 13.56MP.

It also explains why there is little or no coloured aliasing,

and why the red, green and blue signals all have the same resolution and aliasing.

THE LATTER "MAY" BE A GOOD CLUE BECAUSE THERE ARE TWO WAYS A CHIP COULD WORK.

PERHAPS YOU ARE CORRECT, ALTHOUGH THE CHIP HAS A BAYER FILTER, IT MAY NOT BE BEING READ-OUT AS SUCH.

Figure 3 shows the result for setting progressive, again with factory detail settings. There appears to be no difference in resolution between progressive and interlace, a sure sign that the native resolution of the camera does not reach the limits of the 1920x1080 format.

WHICH IS TRUE, GIVEN THE REZ HAS BEEN MEASURED MUCH LESS THAN 1000TVL/ph.


IF you assume the chip is a Sony 16MP chip that is being read-out slightly differently, it is possible for everything Sony has said is true while Allan's measures are valid and his speculations are very good.

I've got to do more digging on the how the increased sensitivity is achieved. Juan seems to imply it is because the photosites are 4X bigger. Of course, the critical question is bigger than WHAT? Did he say?

Sony IMX046 chip is a 1/3.2 type 8MP sensor that has a very tiny 1.4 µm photosite x 2 is 2.8sq x 4 >> 11.2sq >> 5.6 µm

But, HOW the data are read-out could be contributing as well.

Sorry for the paraphrase Steve but Juan doesn't "imply" that pixels are four times the size, he flat out says they are. Plus he goes to a lot of trouble to point out the sensor is not from a stills camera, it is a moderate resolution chip purpose built for video.

He says that, and has pictures to make his point even more. He would have to be grossly misinformed (or lying) to be wrong.

I am no expert, I am just repeating what was said.

As I mentioned on the FS 100 thread, Alan Roberts' assumption seems to be based on an ISO of 800, which is a rating similar to that of the Alexa and Epic, which do have rather different pixel counts.

I said;

"I've got to do more digging on the how the increased sensitivity is achieved. Juan seems to imply it is because the photosites are 4X bigger."

It = the greater light sensitivity.

And, where do I say it IS from a still camera?

Where does Allan say it IS from a still camera?

Yes, Alan doesn't say it's a still camera sensor, just his guessimate at the pixels based on his line of thought and a passing comment on that number: "the sensor has approximately 12.9 Megapixels, typical of a digital stills camera".

That figure is roughly the same as a Sony F35 or Panavision Genesis sensor.

If his guessimate is accurate is another matter, but he gave his reasoning behind the figure, which may or may not not apply in practise. However, that's in the nature of these things.

And, where do I say it IS from a still camera?

You basically say that in the post below, and perhaps other places:
http://www.dvinfo.net/forum/sony-avchd-nex-fs100/493852-sony-nex-fs100-camera-test-5.html#post1639132

I'm not trying to start a fight so I'll stay out of this...

Someone did find this about the sensor. It mentions gross pixel count vs effective. It is talking about the FS100 but still they both use the same sensor.

page 4.
http://www.sony.co.uk/res/attachment/file/59/1237480643259.pdf

There is just too much confusion surrounding this sensor to know for sure.

What are "effective pixels?" Why is the pixel count "approx.?" Are "pixels" the same as photosites?

This sensor could be anything, and convincing arguments have been made which run the gamut.

page 4.
http://www.sony.co.uk/res/attachment/file/59/1237480643259.pdf

Gross pixels Approx. 3,530,000 pixels
Effective pixels of moving pictures in 16:9 Approx. 3,370,000 pixels
Effective pixels of moving pictures in 4:3 Approx. 2,530,000 pixels

While many have assumed pixels are photosites, there are reasons to think this is not true:

Juan say's the image is "grossly" over-sampled. There's nothing grossly over sampeled with 3.37MPixels if the photosite count is only 3.53 million.

PS: from the same PDF:

The sensor also features an optimum number of effective pixels for shooting HD
moving images. At a high 50fps in >>> all-pixel scan mode <<<, this allows images to be
captured with less color aliasing, jaggedness and rolling shutter than DSLR
cameras, while providing the sensitivity to shoot at a minimum illumination of
0.28lux.

With regard to sensitivity, I don't think we can draw any meaning full conclusions over pixel size from sensitivity alone. There are far too many other factors that can skew any conclusions, Q-factor, gain settings and noise reduction, microlense size and usage, binning etc etc.

The F3's effective resolution does seem to tie in quite well with what I would expect from a 3.3MP Bayer sensor. If it really was using 4800 x 2700 I would have expected higher measured resolution.

Bottom line: It does produce a good image in the real world.