Your thoughts on the hvx200 sensor

[Barry Goyette]

While we all have nothing much to do but stare at the "bride" photo and wonder the wonders of DVCPRO-HD, I'm curious about what you all think about the sensor, and how it all relates. My guess is that we'll be seeing a 1/3 inch sensor, but I have a few questions that I think are worthy of a little speculation, or (from some incredibly attractive official source) some answers.

I think many were surprised that this camera would output in 1080p, in addition to 1080i and 720p in their various framerates. Specifically, do you think we'll be seeing a native 1080 resolution 1/3inch sensor, or will it be achieved through upsampling via pixel shift (or some other scheme) from a 720 pixel sensor? In other words, as someone said about the 1080p, "wow..thats the same as star wars!"...so I guess I'm wondering if it's likely that we will really see this kind of resolution from this diminutive camera?

In either case, what does this mean for sensitivity? We know that packing in smaller pixels usually creates a lower S/N ration and thus potentially more problems in moderate to low light. Will this be a "broad daylight or studio lighting only" camera, or could it be that panasonic has come up with some magic that will give the "birthday cake" crowd something to cheer about?

I suppose we could all just wait till the 18th and find out with everyone else, but what fun would that be.

Barry

[Luis Caffesse]

"or could it be that panasonic has come up with some magic that will give the "birthday cake" crowd something to cheer about?"

I seriously doubt many in the 'birthday cake' crowd plan on purchasing a camera anywhere near 10,000 dollars that shoots to solid state media.

Just a guess here.

As far as the CCD block goes, this month's DV magazine had a little bit on it, saying:

"The camera is built on 3 newly built 16:9 CCDs"

Not sure if that's actually accurate, but there it is.

Does any other cam in this range use 16:9 CCDs? Or are they all still using 4:3 (like the XL2 does).?

[Steven White]

The FX1/Z1U CCDs are 16x9, and chances are the new JVC will be too - I don't know about the original JVC though.

EDIT: As for "this range" - really... is there any other camera that's actually in "this range?" - we'll find out when it's priced I guess.

Personally, I'm hoping it's 3 1280x1080 CCDs, so that the 1080 lines at least are real. That would mean little to no interpolation to the 1080p DVCPRO-HD codec. I couldn't care less about the 720p functions of the cam, or any of the SD features. With decent 1080(i/p) content, the rest are acheivable in down conversion.

If it were really 3 1280x720 chips or less, the "1080p" features of this camera would be a bit of a joke, don't you think? The only point of recording 1080 24p would then end up being to use the 80 Mbps of that format - but crippling it with interpolation.

[Stefan Scherperel]

Not necesarily. IF the sensors are 1280x720 and it uses pixel shift to achive 1080 pixels, then you are not loseing resolution but losing color information. HOwever, the codec uses a 4:2:2 color space anyways, so this would in theory negate the color loss from the original 4:4:4 sampling. But I do agree with you, 1280 or 1440x1080 pixels would be great, but I don't see that happening with a camera that is in this price range.

[Anders Holck Petersen]

Even Panasonic's $42,500 HDX-400 use 2/3" 1280x720 CCD's for the 1080 signal, so don't expect this camcorder to use anything higher...

At 1/3" I believe a 1280x720 CCD's with pixelshift is a good choice because you get better sensitivity and lower overall noise.

[Barry Green]

Total speculation at this point, as we don't know anything about it (and we all know that).

But... while I'd love to have raw native resolution, it's hard to argue with pixel shift as employed on the DVC30 and Z1. Adam Wilt's review of the DVC30 said that the resolution was pretty much indistinguishable from the DVX. The Z1 has 960 pixels horizontally, but it sure puts out a lot more detail than 960 pixels worth! If the 960 were raw, the most you could see would be 480 distinguishable pairs of lines, but the Z1 easily handles 700, more like 800, and doesn't look too bad at all at 900.

So, I'd like to see at least 1280x1080, which is the luma grid resolution of DVCPRO-HD when recording 1080. If you had 1280x1080, you'd get native pixel resolution for 1080 mode, and a small downrez for 1280x720 for 720p mode.

But, another option might be to use 1280x720 CCDs. With 1280x720, you'd be looking at employing a 50% pixel shift on both the horizontal and the vertical. 720 x 1.5 = 1080, and 1280 x 1.5 = 1920. That's a lot of pixel shifting, but near as I can tell from the Z1, it does work. So even if it had 1280x720, I think they could still do a killer 1080 picture from it. You'd get the full horizontal resolution that DVCPRO-HD records, and only be shifting up for the vertical; and for uncompressed output, employing 50% pixel-shift on the horizontal.

If I had my druthers, I'd like to see 1920x1080 progressive-scan variable-scan-rate CCDs, using CMOS technology (or some other next-generation super-hyper-HAD technology) to get ultimate resolution and ultimate low-light/smear/noise performance from the chips.

[Aaron Shaw]

Even Panasonic's $42,500 HDX-400 use 2/3" 1280x720 CCD's for the 1080 signal, so don't expect this camcorder to use anything higher...

True, but then the Varicam doesn't shoot 1080p and it costs a lot more than the HDX400 so who knows?

[Jeff Kilgroe]

I'm probably wrong, but I'm actually expecting this to be a single chip cam using a 1/2" 1440x1080 CMOS sensor. Some of the newer CMOS censor designs out there can scan at some pretty fast rates and can cope with both interlaced and progressive scanning. But there would be no pixel shift.

Most likely I'm wrong anyway. If it does happen to be a 3-CCD camera, I hope they do use censors with a higher native resolution than 960x720.

I guess we'll know for sure in about 10 days...

[Aaron Shaw]

Could be. There have been several rumors around that have said 3 1/3" though. Still, if the rumor that the lens is 82mm in diamter (not 72) then this might very well suggest such a sensor.

[Graeme Nattress]

Barry: "The Z1 has 960 pixels horizontally, but it sure puts out a lot more detail than 960 pixels worth! If the 960 were raw, the most you could see would be 480 distinguishable pairs of lines, but the Z1 easily handles 700, more like 800, and doesn't look too bad at all at 900."

480 line pairs is 960 lines of resolution. A DV camera has a chip with 720 pixels but >500 lines horizontal per picture height.

That the Z1 measures 800 lines, not line pairs horizontal, even with the pixels shift is very telling. I'd hate to see what it is without the pixelshift!

Now, further to a post somewhere else on this forum, I managed to do a bit more learning about pixelshift via a "version" used in research to get more rez out of a black and white 1 chip camera taking an static image, but using video and, I think the natural vibration of the camera over time to effectively be multiple pixelshifts. They found out that when an image is undersampled by n times (in horizontal and vertical equally) you need n^2 pixelshifts to generate your high rez image. But the image onto their CCD was heavily aliassed as there was no filtering. Obviously, if the lens does not pass high enough detail, no amount of pixelshifting is going to boost your rez beyond what the chip or is capable of. You must, at least from what I can see be able to pass a high enough resolution through to the chip so that it aliasses, and the multiple pixelshifts can be calculated to produce a higher rez anti-aliassed image.

Now, in video we're using a static physical shift of 1 chip. This can be horizontal, vertical or both. We're dealing with colours, so it's not a perfect situation for information extraction like the black and white example and we only have 2 samples, which would give a root 2, or 1.414 undersampling factor, which is close to the 1.5 people often quote. Now this is for a practically perfect case which I don't see existing in a video camera due to the colour filters on the CCDs. If the lens is not capable enough, no amount of magic is going to pull out a higher rez, but you might get a less gradual contrast roll of as detail increases. This is what I think is happening with the FX1 and indeed, as the measured resolution doesn't pass the 960 pixels in the chip, from my above, admitedly limited understanding, is what I think is happening. As a test, I took some full rez images from the FX1, dropped them to 50% horizontal rez in photoshop and bicubically interpolated them back up to full rez. It is very, very difficult to tell the before and after apart, which would make sense if the horizontal rez < 960, and fits in very well with the above hypothesis. If any one can shed some more light on this, I'd be very, very keen to learn more.

So, back to the HDX200. The HDX400 uses 1280x720 chips. I think that the HDX200 will also use 1280x720 chips, with pixel shift not so much to go much beyond that in rez, but to boost the quality up to that rez, maybe a bit beyond if the lens is up to it, which should do really nice 720p and 1080i, and compromise 1080p a little but not too much.

I'd originally though the chips would be 1280x1080, and ofcourse, that would be higher rez still, but lower light sensitivity, so I think we'll see a noise / resolution trade-off here.

Graeme

[Michael Struthers]

The JVC is using 1280 by 720, I expect the panasonic to use them also.