I just wouldn't be surprised if Sony soon introduces a format that uses high-bitrate, long-GOP H264 encoding, to replace XDCAM for the future - would really make a lot of sense.

That's the point though, thanks to Convergent Designs it's not a question of agreeing or disagreeing, it's been proved that it can be done - and it doesn't need to cost very much, and can even go upto 220 mb/sec!
Steve

1. The Nanoflash records off the SDI port, not the internal bus
2. The Nanoflash is not using SDHC... it's using Compact Flash
3. Sony have several high bitrate codecs already... just not in the EX1(r)

So of course it can be done, but there are physical limitations in the EX1 that prevent it from happening on certain media.

But to get to the SDI it has to go through the internal bus.
SxS can go to very high bitrates, so you could use that for quality work (as you likely would anyway) and SDHC for low bitrate stuff.
It's just impossible to argue against it.
Steve

How much sense does it really make for Sony to stick with MPEG-2 much longer though? After a few more new CPU generations (think CPUs with a dozen or more cores that are at least twice as efficient as i7 per core), MPEG-2 just won't offer much comparative advantage for editing purposes, but does suck up significantly more bandwidth (and space). In a few short years things might flip things around a bit, since HDD speeds aren't getting faster nearly as quickly as CPU power - with the lower bandwidth of AVC (faster HDD reads and writes) providing for an overall performance advantage once CPUs can handle AVC as easy as melting butter on a hot griddle.

Agreed. But we aren't there yet. And for right now, the XDCam codec is able to be edited natively for those people who need to do that for speed purposes. When we get to the i9 or i11 or whatever, and AVCHD presents no significant burden for editing then it might be prudent to move along.

But at this point, there's no compelling reason to leave the current codec. It meets broadcast spec, it's easy to edit, and it looks pretty darn good at higher bitrates. I sincerely hope the next jump is to wavelet and not AVC based.

EX codec doesn't meet EBU broadcast specs, needs to 50 mb/s.
Steve

I don't know why wavelet compression has been so largely ignored. It must have to do with money interests, since the technology pretty much makes more sense fundamentally (and has for awhile). I sometimes wonder why Panasonic didn't just buy out Cineform, and use that codec instead of introducing AVC-I. They could have got much faster adoption, simply because it isn't as taxing on current day CPUs (if for no other reason).

Could be down to processing power (and hence power draw) of doing it realtime. Remember the Grass Valley Infinity? Recorded JPEG2000 and drew an absolutely huge amount of power.

Well, the EBU recommendations. Remember they aren't hard and fast specifications. Just recommendations.

A modern quad core CPU can do encoding with Cineform's codec in real time a whale of a lot easier than H264. If a general purpose CPU can do it pretty easily, should be no problem whatsoever to create a dedicated encoder chip specifically designed for it.

But they do tend to be valued by broadcasters.
Steve

The HPX170 chipset is well past it's sell by date, and now 1920x1080 displays and codecs are common that's becoming increasingly obvious.

I feel 1 megapixel may be the best compromise for 1/3" chips - twice the number of photosites as the HPX170, but without compromising individual photosite size too severely. But best of all are obviously full 2 megapixel (1920x1080) on a 1/2" chip. Regrettably only Sony seems to offer that in a prosumer grade camera.
I disagree - it depends what criteria you use. Is your test one with a lot of motion or a lot of fine detail and sharp edges? Current implementations of AVC-HD by Panasonic seem to cope well with motion, but much less so with mosquito noise, especially in 1080i mode. (There are many reports arguing that for the HMC150 it's best to use 720p rather than 1080 - 720 mode captures all the detail the chips are capable of producing, whilst stressing the codec less.)

And the mosquito noise can be argued to be more objectionable than codec failings due to motion - it's there all the time, often "twinkling" around the edges of static detail.

The ratio of I frames to difference frames in a codec is not fixed, it's easy to see how static performance can be traded off for better motion performance. With AVCCAM they seem to weight in favour of motion.

What is really needed for the prosumer is a camera styled like JVCs HM700, but 1/2", full 1920x1080 chips, and a codec to the standard of XDCAM 50Mbs or AVC-Intra 100. And ideally recording to both "full spec" media (P2 or SxS) AND consumer memory like SDHC or CF. There is no technical reason why that is not possible.

Nothing currently meets all the criteria. Of what's available, I'd argue the EX is the obvious first choice at the moment, since at least you can get it to meet those criteria with an external recorder.

Perhaps you are looking at Panasonic's consumer AVCHD? Apparently the pro cameras get a much better AVCHD encoder (branded as AVCCAM) than the consumer camcorders (or the GH1, for that matter). Barry Green did a very direct comparison of AVCCAM AVCHD (the branded AVCHD for the pro cams), by attaching an AG-HMR10 to an EX1, recording XDCAM EX in-camera and Panasonic's AVCCAM AVCHD with the AG-HMR10 (recording from exactly the same source coming off the imaging block - true apples to apples comparison). In the article he wrote, he showed comparison frame grabs, where the codecs were stressed, and the AVC looked a tad better on the whole. He didn't post frame grabs of the more typical ("unstressed") footage, but did state that both codecs produced very good images that were quite comparable in most of the footage. When I look at blown up images shot with my HMC40 (full raster chips), it looks awfully good (just a whale of a lot better than HDV) and does seem (to me) somewhat difficult to imagine XDCAM EX encoding being a whole bunch better at maintaining image fidelity from the imaging block. I don't shoot 1080i though.

I'd love to see a similar (apples to apples) test (XDCAM EX vs AVCCAM with an EX1 or EX3) panning a res chart slowly.

No - the mosquito noise was on an HMC150.
If that's the test I think it is, the only comparisons I remember seeing were regarding how they compared in areas of high motion. (When yes, AVC-HD stood up well.) The comparisons of static detail I believe I saw in a separate test, and in this respect AVC-HD was inferior.

I've also seen that effect personally on an HMC150 in 1080i mode.

There are many ways of "stressing a codec" other than with motion, and the two most obvious are fine detail/sharp edges, and slowly changing gradients (especially coloured). Just because a codec performs well in one respect doesn't mean it will in the others.

(And different coders, different hardware, can give widely differing results, even if they are the same codec and same bitrate.)
Yes! And stationary as well. A sheet of newsprint might be an even better test as much of a res chart consists of blocks of uniform grey. The mosquito noise I observed was most obvious around high contrast sharp edges.

An HMC150 is going to show aliasing in 1080 line recordings of detailed images, even if recording uncompressed. The imaging block just can't resolve a high level of detail (low res chips).

This wasn't aliasing, it was a compression issue. If it had been aliasing it would have been just as visible in 720 mode as 1080 - it wasn't. I was able to step frame by frame through the material and see a constant mosquito noise pattern for a number of frames before the pattern jumped, then be constant for a number more frames, then another jump.

I'm pretty sure the jumps corresponded to GOP intervals. Hence pretty sure that the codec is fairly fragile in respect of static detail, whilst being quite robust with respect to inter-frame movement.

Of course, this only applies to the coder in the HMC150 which is several years old in design now. That in the forthcoming Canon camera is newer, hence likely to be better.

I believe aliasing should show up more on 1080 line recordings (which goes beyond the HMC150 imaging block's ability to resolve detail cleanly).

How much is the RED drawing? It's writing essentially the same thing. And how much is Cineform's new recorder drawing? it's writing wavelet also. How about the SI-2K which is writing wavelet (Cineform RAW).

You are confusing codec with bitrate. That same codec is in the NanoFlash. And I can absolutely guarantee the codec in that implementation exceeds every broadcast standard on the the planet.

Can you imagine how badly H264 encoded 4K footage from a RED camera would bog down even the fastest desktop computer?

Well, depends on the bitrate used. The 5D/7D do a fine job of hammering my 8-core machine! But I have little problem editing RED 4K footage. Wavelet is just SO much better in so many ways...

It's not bitrate that hammers you with H264. It's the number of pixels in the images being encoded or decoded. 4K, even at just 4:2:0, is 4 times as much work for a CPU.

There's h.264 and then there's h.264.

A full implementation will grind most anything to a stand still. It can include sub-pixel motion estimation and a number of other obscure tricks to cram everything into as few bits as possible.

The way codecs are spec'd is by defining the decoder. The decoder has to be able to decode every trick in the spec. No feature can be left out.

The encoder, on the other hand, has no spec at all. As long as the files are decodable by the standard decoder, it's legal.

A simple h.264 encoder isn't much different than a basic MPEG-2 encoder. That's what we get in the 7D/5D2/1D4. It's simple and fast and can run on battery power. And to get a good image, you need lots of bits. Those cameras encode 40-48mpbs. A two-pass, kitchen sink encoder can probably get the same quality in about 6mbps, but you can't spec that in a handheld camera.

So, keep in mind, when you say h.264, you're really talking about the decoder and compatible files. The encoder can vary widely.

I agree with Chris about the market. The biggest part of the market has a lot of specific needs. Two of the big ones are:

• Low light and deep DOF at the same time.
• Motorized lenses with huge zoom ratios, low breathing, etc.

Ironically, low light is precisely the area where large sensors face the biggest problem. Most people think of the 5D2 as pretty good in low light, but that's only because they use very thin DOF. As soon as you start using it at the same DOF as the 3-chip cameras, the low light performance goes out the window. Even my XH-A1 blows the 5D2 away in low light at deep DOF. The large sensor advantage hinges entirely on thinner DOF, but if you don't want thinner DOF, then that is not an advantage. Hoisting a thin DOF camera on videographers that aren't used to it would result in a huge backlash from shots with missed focus, IMHO.

But I think the problem can be solved. Think of it this way. Right now you have the XH-A1 and it gives you deep DOF in low light with a certain level of noise (Say, f/1.6 at +12dB gain.) If you want even less noise, there's no option; nothing you can do. Even if you were willing to accept thinner DOF, there is no choice available to do it.

We can improve on this with a large senosr. Have the software default to f/7, which gives APS-C users (e.g. 7D) the same DOF as f/1.6 on the XH-A1. They'll still have very poor low light performance (likely even worse than the XH-A1), but at least they'll retain the same DOF they had before. But this time, they can take manual control over the camera (and with the right lens) open up to a faster f-number, reducing the noise level.

There are other applications where the advantages would be huge, such as in ample light. Higher contrast, more dynamic range, increased color depth, and more. In ample light we can get the deep DOF needed without the noise penalty. In fact, it means we'll be able to stop using so much ND.

I'm no expert, but I think it's possible to do the same for FF35 and APS-C. From what I've read, it's actually easier to build lenses of similar capability for larger formats. For example, in order to match the capability of the XH-A1 lens on the 7D, we'd need a 20-400mm f/7-15. 20mm f/7 on the 7D has the same angle of view, depth of field, and total amount of light as 4.5mm f/1.6 on the XH-A1. Same with 400mm f/15 on 7D vs 90mm f/3.4 on XH-A1. If you look at lenses on the market, we already have ones like the Tamron 18-270 f/3.5-6.3 for $550. Now of course it is not motorized, has breathing issues, etc., but at the same time it's more than two stops faster than it needs to be to match the capabilities of XH-A1 lens. But it's still only 3.9 inches and 1.2 pounds. Now, they could not make an f/15 still lens because it would not work with phase detect autofocus. But video doesn't use PDAF, so that becomes an option, and that could really reduce cost and weight. I discussed this in more length in a different thread:

http://www.dvinfo.net/forum/canon-eo...eeper-dof.html

Think about back focus. Single chip cameras have a huge advantage in lens design: their back focus can just few millimeters (say, 4mm), even for sensors that are 36mm wide. That's the primary advantage that rangefinders have over SLRs, and the reason why their wide angle lenses have so much better performance. 3-chip systems, on the other hand, require a huge amount of back focus relative to the size of the sensor -- way more than even SLRs. This forces huge compromises on the lens design that greatly increase cost and aberration. On top of that, specific changes have to be made for the prism itself that are not necessary in normal lens design. Both of these work in the favor of single-chip lenses.

Consider parts tolerances. The manufacturing and build quality of 1/3" lenses have to be almost flawless, because even the tiniest movement out of place (0.1mm?) has a large effect on the image. That includes movement of the focus (with lens groups moving only in the most amazingly tiny increments). But in a FF35 system, a component that is off by the same exact amount will be invisble. I think that will greatly reduce the cost of motorizing the lens to the same capability. Then there's MTF (contrast). I think Adam Galt said that lens designers found it more difficult to achieve the same high MTF with 1/3" lenses as they could with larger formats, because it had to occur at correspondingly higher spatial frequencies.

All that is to say that I think it's possible to build a FF or APS-C lens that will match the capabilities of the XH-A1 for the same price class (under $4k -- whatever part of that is the lens cost). It would be a fully motorized 20-400mm f/7-15 on APS-C, and 32.5-650mm f/11-24 on FF with controlled breathing, parfocality, etc. Same weight too. But I'm no expert, and I'm sure there's factors that I don't know about. As for the rest of the camera (body, image processor, etc.), I'd like to think they could do that for about the same price with a line-skipping sensor, the 500D is only $650 after all.

But I don't think they will do it at all until they can do it right -- and that means no line skipping. So I think it will be quite a few years yet.

No line skipping is easy:

1) use bigger pixels,
2) use a lower frequency anti-aliasing filter.

The problem is that it won't have the volume of a DSLR. Dang.

Also, the low light capability of the 5D2 could be much, much bigger. There are artifacts in both the vertical and horizontal dimensions. If it's skipping lines and columns, then it's missing 8 of 9 pixels, the sensitivity could be nine times higher(!) If it's just line skipping, make it three times higher. Start with the 1D4 sensor and you get good performance at 19,200 ISO - or 57,600 ISO(!)

With that kind of sensitivity, one can stop down the lens and get back a lot of that DOF.

Regarding the power requirements if the RED One, it's batteries seem to last a similar time to a HDCAM camera - it's pretty power hungry. That's a lot more than any prosumer camera.

Given that people seem to want the 35mm sensor cameras for a shallow DOF, why would they want a large zoom range lens with a limited max aperture? Besides an aperture ramp of more than a stop is not a good idea on a video camera - personally I find the ramp on the Z1 lens too much

I think Brian touched upon the main reason this cameras has not surfaced or will ever surface. There are really two markets here - Video Video cameras and Film Video cameras.

They each have different styles of shooting. So one design will not help both out.

The film video shooter could get away with a fixed focal length design, the video video shooter could not.

The film video shooter really wants very shallow DOF, video folks want that from time to time, but for a majority of work this would be too thin for reliable focus.

The bigger problem is the film video camera will require all of the R&D and it is the smallest market.

So the manufacturer has to swim upstream on a concept which is a lot to ask.

If technically possible, a dual sensor camera would be great. If one could flip between a smaller and larger sensor depending upon the situation this would make a lot of people happy.

Yes Tim, what you say here looks to be the reality of it all.

Agreed.

Because that's not the only reason to want a 35mm camera. Is that the only reason why people want 1/2" or 2/3"? There are other reasons too, including improved dynamic range and (sometimes) better contrast.

[EDIT: I should point out that the XH-A1 lens for a FF35 camera could just be one option among many. People who want the regular camcorder experience could buy it, but those who want more control over DOF and/or low light could buy a lens with a smaller zoom ratio, faster f-numbers, like a motorized 24-105mm f/4.]

Well, if it's only the ramping that you dislike, that can be easily fixed by camming the lens to stay at the slowest f-stop. The 20X XH-A1 would be a constant f/3.4 that way instead of ramping from f/1.6 to f/3.4. But I suspect what you mean is that you want the iris to be constant *and* fast. That is possible, but something else has to give: zoom range, price, or aberrations. Whatever choice you feel is appropriate, my point is that the same thing can be done for larger sensors.

Unless there is a single design that can satisfy both requirements, and there is. In fact, many of the side benefits of the cine-style cameras would be a big benefit to vidoegraphers, such as increased dynamic range. (They could have 5 stops of highlight headroom instead of just 2 like they have now -- which means far fewer clipped highlights.)

They already invented that -- just use the f-number. On FF35, you get the exact same effect by switching between f/2.8 and f/24. f/24 (and high ISO) gives you the same (poor) low light performance and deep DOF as 1/3", while f/2.8 (and low ISO) gives you the super thin DOF and amazingly good low light performance of super large sensors.

Not so - aliases should be equally visible on 1080 and 720 recordings. (At least in this case.)

Aliasing is a result of not sampling at a high enough rate, either in time or in space. In the case of the HMC150 the sampling is at the horizontal rate of 960 across the frame width, so it could theoretically resolve up to 480 line pairs before aliasing. Any finer than that, and the result will not be true but an alias, and will manifest itself as a *lower* frequency than 480lp, folding back round the Nyquist point. In general terms, no matter how high a frequency you put in, you'll never get a higher output frequency than the sampling rate. Hence, recording 960x540 will capture the lot - "real" detail and aliases.

That's true on a chip by chip basis. Taken as a whole, the H & V pixel shift complicates matters. Without going into vast detail, what it means is that a fine line pattern can be present on the output as both real detail and an alias, the latter most likely being coloured. Either way, recording a 1280x720 raster will capture the lot, going to 1920x1080 won't resolve any more detail (or aliasing).

What I saw was a compression issue, nothing to do with aliasing.

Not really, if you stop down too much you'll start getting massive problems with diffraction. Even on a 35mm sensor diffraction starts coming in at around f8 I seem to recall.
Steve

Apart from the dual sensor quote being incorrectly applied to me, the main reason that many people seem to be get excited about the 35mm sensors is the shallow depth of field. They don't get seem to excited about other qualities like the dynamic range. However, the video from the DSLR is currently not that impressive in that particular area, apart from the DOF their other strong point is the sensitivity.

The argument can be made for 35mm sensors, the question is more if it's going to be a motion centred design or stills camera with a video capability, but with compromises to keep the price right for the large stills market? For many uses the DSLR video route is fine, but for others the compromises make them unsuitable.

Often the reason for the aperture ramping is to keep the lens size down - a constant aperture zoom can get pretty large even on a 2/3" camera, which is why almost all of them have some ramping. I suspect the 20x on the 1/3" Canon would be somewhat larger than that present.

People will buy the upcoming Scarlets (2/3 & 35mm), but they're a higher price than the video capable DSLRs.

Actually, that's a popular misconception. Diffraction is the exact same accross all format sizes because it scales with the optical DOF . f/32 on a 36x24mm sensor has the exact same diffraction and DOF as f/3.2 on 3.4x2.4mm sensors. The only way to get less diffraction is to use less depth of field (either a larger format with the same f-number, or a faster f-number with the same format).

For example, here is 200mm f/32 on a 36x24mm sensor with 6.4 micron pixels:
http://thebrownings.name/images/2009...-100crop-1.png

And here is 70mm f/11.3 on a 12.5x8.5mm sensor with 6.4 micron pixels:
http://thebrownings.name/images/2009...-100crop-1.png

See how similar they are in contrast and sharpness? If anything, the f/32 is a bit sharper thanks to a higher pixel count (even though the pixel size is the same).

If you check a diffraction calculator, you'll see that the math agrees with these examples.

Fixed.

Perhaps I am not in tune with what most people are excited about. In any case, I think everyone would like to have more dynamic range, improved color depth (i.e. SNR over the used dynamic range), and the choice of improving low light performance and/or using a variety of lenses. Even if they are happy with their clipped highlights, poor color depth, and fixed lens, I think they would learn to greatly appreciate the improvement in a true APS-C camcorder once they had it in their hands.

I highly disagree. I've compared my 5D2 and XH-A1 and I can get 5 stops of highlight headroom out of my 5D2, with almost as many below middle gray, where as I struggle to get even 2.5 stops of highlight headroom from the XH-A1, and at just 5 stops below clipping at 0db the noise in the XH-A1 becomes very strong. Perhaps the video you've seen from DSLRs was not processed for high dynamic range?

If they can sell the 500D for $650, I'd like to think they can make an XL-series with an APS-C sensor and lens of the XL-capability for only ten times as much ($6,500) -- hopefully even low enough to hit the XH-A1 price point. But I don't know.

I'm not sure what you mean. The f-number ramps from f/1.6 to f/3.4 as I mentioned, which is equivalent to f/7-15 on APS-C.

Daniel, I think you'll find it's a fact that the diffraction limit gets lower the smaller the film/chip size. So with an 8x10 camera you're happy well down to f45, a Hasselblad is OK to f11, but 35mm starts to lose resolution at f5.6-8. For 1/3" chips it's down around f4! You obviously have access to science that I don't.
Steve

You have the correct principle (the f-number scales with chip size), but your math is still way off. For any given display size, the following have the exact same diffraction, depth of field, (and angle of view and focus distance of course):

f/4 on 1/3" (5.1mm width)
f/7.5 on 2/3" (9.6mm)
f/17.5 on APS-C (22.5mm)
f/28 on FF (36mm)
f/44 on Hasselblad 645 (56mm)
f/198 on 8x10 (254mm)

As I said, diffraction scales with DOF. When you said that diffraction prevents the possibility of stopping down FF35 to match the DOF of 1/3", you were wrong.

You obviously resort to insults when proven wrong.

Didn't mean it to be an insult, sorry you took it that way.
I still don't think I am wrong, in your example you say stopping down to f28 on 35mm, are you really telling me this won't cause the image to be affected by diffraction? I simply don't believe it.
Steve

Well I don't want this thread to get too out of hand here...

My experience with shooting with the XL H1S, 5D Mark 2, and 7D is that the XL/XH still seemingly have greater overall dynamic range. The SLRs seem to have less, despite the sensor size. I assume this has to do with the heavy pixel count on the sensors.

Wasn't it also Barry Green or Adam Wilt that measured the range as around 8 stops? I can't remember...

I was wrong to have taken it the wrong way. I apologize. Thank you for the clarification.

No, I am not saying it will not be affected by diffraction. I am saying that f/28 on 35mm will be affected by diffraction to the exact same degree as f/4 on 1/3". Look it up in a diffraction calculator, there are several of them around the web. Just be sure that you set the calculator to use the same CoC (i.e. same display size and resolution). For example, set both to a 1080p monitor or a 4x6 print.

If you prefer to look at test images instead of calculators, you can download the raw files for the images I posted earlier:

Comparison images

• Picture b1: 70mm, f/11.3, ISO 200, 12.5x8.5mm sensor
• Picture b2: 200mm, f/32, ISO 200, 36x24mm sensor

One of the images is f/32, which is even narrower than f/28. It shows the same level of diffraction as a smaller sensor at f/11.3.