EX3 or HPX500?

[Tim Polster]

I would think it would be 1080p and 720p where the differences would show up.

[Steve Phillipps]

Yes Tim, I tried 1080P vs 720P, don't think anyone shoots in 1080i anymore!
John, don't see myself why you'd prefer 2/3" chips over 1/2" really, unless it affected the image quality. Yes you get a bit more light gathering (though I'd guess the EX1 and HPX500 are not too disimilar), theoretically you get less noise (but again I don't think that's the case EX1 vs HPX500). You can get wide angle lenses for the EX1 so no problem there. If you want shallow dof then neither are really going to give you that so on either of them you'd want to use a Redrock/Letus etc. On the other hand if you want telephoto the 1/2" chip's a winner as you get about 50% more magnification. Plus smaller chip smaller lighter lenses, smaller lighter camera, less power draw so smaller lighter batteries.
The only concern over the EX series I have is the rolling shutter. I didn't like what I saw when I tested it, but so many people seem to think it shouldn't be an issue that I'm going to look at it again.
Steve

[Noah Yuan-Vogel]

Originally Posted by Chris Hurd

Pixel Shift is a benefit, not a drawback.

Not sure what you mean by "true" HD, but I don't think it's been conclusively stated that Sony isn't using some form of pixel offset in the EX series.

Could you explain further what you mean? I'm not sure I see how pixel shift is preferable over full resolution sensors. By full resolution I mean RGB sensors that are each natively the same or greater resolution compared to the luma res of the output video format. Usually pixel shift is used to make up for sensors being low resolution. Do you mean you prefer to have lower resolution sensors because bigger pixels potentially allow better light gathering and greater dynamic range? This doesnt seem to be the case with the HVX as from what I can tell it is bit noisier (less sensitive, less latitude) than most of its competitors (canon, jvc, sony).

i would imagine the less interpolation needed, the better. seems like the ideal would be a camera that has as close to the ~6million photosites used by 1080p yuv444 as possible. a single sensor full HD camera would have 2million photosites. HVX and HPX have 960x540x3= ~1.5million photosites doesnt really seem like enough to get the job done even with interpolation and shifting.

I'm guessing by "true" HD what is meant is fullHD (1920x1080) or at least natively some standard HD delivery resolution so as not to require scaling (1920x1080 or 1280x720)

[David Heath]

Originally Posted by Noah Yuan-Vogel

Could you explain further what you mean? I'm not sure I see how pixel shift is preferable over full resolution sensors.

Resolution measurement is far more complex than simply stating a simple number, it will tend to "tail away" rather than suddenly go from perfect to nothing at a certain point.

And THEORETICALLY there is the benefit to using pixel shift with full raster 1080 sensors such as the EX uses. It should mean that the "tail off" is less than if not used, even if the maximum resolution is still limited by the system. To put it another way, to mean the contrast of fine detail is improved, even if no finer detail can be seen. I believe what Chris is trying to say is not that low-res chips with pixel shift are better than full raster res chips, but best of all is full-res chips TOGETHER WITH pixel shift - at least in theory - and I'd agree.

PRACTICALLY, I doubt it would be used (at least for a camera of this price range), since the signal processing demand then is just likely to be too high to be realistic - likely to need one working at 3840x1080 for H shift alone. Use lower than full raster chips, but full raster signal processing, and pixel shift becomes a viable proposition.

All else equal, will 960x540 chips with pixel shift be better than 1920x1080 chips without? A resounding no. Will they be better than the same chips, but pixel shift not used? A resounding yes.

[Chris Hurd]

Originally Posted by David Heath

Resolution measurement is far more complex than simply stating a simple number, it will tend to "tail away" rather than suddenly go from perfect to nothing at a certain point.

Indeed, and thanks for stating it so well. It seems that some people tend to obsess over a single particular spec, such as the number of photosites on a chip, instead of considering the entire system. It's much easier to just focus on pixel count, but that's not at all a very accurate way to assess image "quality" or capability to any real degree.

For example, I own two vehicles that have 23hp engines. Since they're both 23 horsepower, does that mean they both have the same capability? No. Because one vehicle is a little riding mower with a 46" cutting deck and the other is a much larger farm tractor with a five-foot shredder. I can use one 23hp vehicle for working only around the yard that surrounds my house. It can't work where the other 23hp vehicle can, which is out on the rest of the property where the grass is tall and thick with huisache. Also, one of my 23hp vehicles -- the farm tractor -- can pull the the other 23hp vehicle with ease, but the riding mower would have a hard time pulling the tractor. Their 23hp engines are alike only in one specification on paper, and that spec doesn't really mean very much in a real-world comparison.

If you consider only a single spec -- such as the number of photosites on a chip -- you're not getting an accurate assessment of what the camera can do. In fact, it can be very misleading. As David smartly points out, it's much more complex than that.

Consider the idea of "full HD" or full-raster chips. The photosite count might be there, but that certainly is not the resolution you're actually getting from those chips. Remember you have to consider the *entire* chain, not just one spec. There's an optical low-pass filter on the face of the prism block (or the face of each chip) which cuts resolution -- on purpose -- in order to avoid Moire and other aliasing issues. The OLPF is easy to forget about when discussing sensor resolution, but it guarantees that a full-raster chip isn't really getting a full-resolution image. And it's made that way by intentional design.

The other factor easily forgotten is that a CCD is an *analog* device. It's outputting voltage which is then converted into a digital signal further down the chain. Because it's analog, the number of photosites doesn't have to equal the number of pixels in the recording or display formats. The bit depth and efficiency of the A/D converter, not the sensor itself, is what makes the pixels that are to be recorded. You could have three cameras each recording to the same format but with a different number of photosites on their sensors: one with less photosites than the resolution of the recording format, one with the same number as the recording format, and one with more. The actual quality of the recorded image will be determined by a number of factors, from the MTF of the lens to the efficiency of the A/D converter, and it's easy to have a situation where the camera with fewer photosites delivers a better image, because it's a complex process with many variables to consider other than pixel count.

[ Re: The concern over sensor resolution is misplaced. Even the full 1920x1080 chips aren't receiving that much res from the glass. Remember there's an OLPF (optical low pass filter) on the face of the sensor, so even if the glass in the lens really does resolve that high, that's not what's reaching the chips -- and that's by design. What happens at the sensor does not ultimately define the image. Far too many other elements affect the process. ]

I believe what Chris is trying to say is not that low-res chips with pixel shift are better than full raster res chips, but best of all is full-res chips TOGETHER WITH pixel shift - at least in theory - and I'd agree.

What I'm trying to say is that you could also have a situation where a lower-res sensor block with pixel offset might actually be better than a higher-res sensor block without. The bottom line really is that there are too many variables along a very long chain (from the differences in the glass to the differences between recording formats) for the photosite count, or the presence or lack of pixel offset, to be that much of a real-world issue.

In other words, the critical pre-purchase question should not be "does this camera use pixel shift?" but rather "how does this thing feel on my shoulder?" and "do I like the image it makes on an HDTV display?"

But yes, a very good situation would be where the sensor res plus pixel offset is greater than the native recording res. This is how the sensor block works in the Canon XH and XL line of HDV camcorders: their chips are native 1440x1080 (which is the resolution of HDV2 or, for that matter HDCAM) plus they also employ H-axis Pixel Shift for a resolution boost equaling about 1920x1080. The problems encountered when going much beyond this are primarily the limitations of processing power and cost. That's why I refer to sub-$15K or sub-$10K camcorders, because when there's such an obvious cost limitation, you run into a wall in terms of processing capability and efficiency. Pixel offset, in one form or another (whether it's H-axis, dual-axis or ClearVid), is understandably a big advantage where cost is a factor. In fact it's an advantage for all three-chip systems, but the reason why some native-res chipsets aren't using it is due either to a processing limitation (such as JVC ProHD) or a cost limitation (most likely with Sony EX), or both. Marketing spins this as "we don't need pixel shift." Well, relative to lower-res three-chip systems that do need it, that might be true. But the reality is that lower-costing native-res three-chip systems would also benefit from it as well, but they can't afford to include it.

[David Heath]

Originally Posted by Chris Hurd

There's an optical low-pass filter on the face of the prism block (or the face of each chip) which cuts resolution -- on purpose -- in order to avoid Moire and other aliasing issues. The OLPF is easy to forget about when discussing sensor resolution, but it guarantees that a full-raster chip isn't really getting a full-resolution image. And it's made that way by intentional design.

Chris - what you say above is true, but an optical low pass filter SHOULD be designed such that a chip doesn't receive any more resolution than it's pixel dimensions. So for 1920 H pixels, no more than 1920 lines of resolution should pass through, but for a chip with 960 H pixels, it shouldn't pass more than 960 lines. Use the 1920 LPF with a 960 chip and it will do nothing at all to limit aliasing.

Which is a problem - because if you want to make a design with (say) 960x540 chips, and you want to stop all aliasing with optical filtering, you will inevitably filter off all the detail that pixel shift techniques depend on!

So what gives? My understanding is that in practice optical lpf's only exist on the top end of the market. For sub $10,000 cameras, their use is just not viably justified for the improvement it gives. The hope is that the lens performance tails off to suit anyway, but the unfortunate truth is that the better pixel shift works within a design, the worse an aliasing problem it is likely to have.

And the most unfortunate aspects of aliasing only really come to light with motion, as aliasing artifacts move in the opposite direction to the moving object! Bad enough in itself - but made even worse if you then put it into a motion sensitive coder. And even more unpredictable if it gets recorded and edited on I-frame only codecs, then coded to a lowish bitrate long-GOP codec for transmission. The artifacts remain relatively "hidden" through the production process, but come out to bite at the last moment. Nasty.

[Jad Meouchy]

Originally Posted by Chris Hurd

The other factor easily forgotten is that a CCD is an *analog* device. It's outputting voltage which is then converted into a digital signal further down the chain. Because it's analog, the number of photosites doesn't have to equal the number of pixels in the recording or display formats. The bit depth and efficiency of the A/D converter, not the sensor itself, is what makes the pixels that are to be recorded.

Technically speaking, a light switch is analog. Obviously, it's not meant to be used in that manner either. The analog argument works very easily for recording media such as Hi8 vs. Digital8; it is more difficult to make for an imager with such a discrete signal. But, going back to our light switch, is the signal really so discrete? Generally, for this application, it is (not so for the light switch as the bulb and high voltage does some of the debouncing for us). The analog signal path on a modern camcorder is relatively simple, with heavy emphasis on "relatively." After a certain milestone of complexity, digital signal processing is cheaper than its analog counterpart.

The majority of the work is done to the digital signal, hence the bragging rights for 12/14/16bit DSP processing. The reason we do care about in-cam settings is because they occur prior to encoding/compressing the signal, the point of no return. Compare this to a raw output, where all you really care about is exposure, focus, and shutter.

The point: Chris is right that there is quite a bit going on between the imager and the output. Relying only on imager specs is not going to give a good comparison. How many people would argue that glass is just as important?

The EX3 will win the resolution test. But the HPX might win the subjective audience 'image quality' test because DVCPROHD is easier to grade in post and 4hr of color correction may render a 'nicer' image. With no time limit, the EX3 might win in post, I'm not sure.

These cameras should be compared in terms of their format and workflow. The rolling shutter is a huge factor as well.

[Ali Husain]

there's an OLPF in front of almost every digital sensor in every camera. it's just a blur filter. sampling won't quite work right without it.

pixel shifting does the same thing that a bayer array does. in both single and 3-chip designs you have to interpolate colors from the 3 sensors. except i'd argue that you get a better fill-factor with pixel shifting.

graeme nattress makes some good comments about these things:
http://luminous-landscape.com/forum/...howtopic=20367

Originally Posted by David Heath

So what gives? My understanding is that in practice optical lpf's only exist on the top end of the market. For sub $10,000 cameras, their use is just not viably justified for the improvement it gives. The hope is that the lens performance tails off to suit anyway, but the unfortunate truth is that the better pixel shift works within a design, the worse an aliasing problem it is likely to have.

And the most unfortunate aspects of aliasing only really come to light with motion, as aliasing artifacts move in the opposite direction to the moving object! Bad enough in itself - but made even worse if you then put it into a motion sensitive coder. And even more unpredictable if it gets recorded and edited on I-frame only codecs, then coded to a lowish bitrate long-GOP codec for transmission. The artifacts remain relatively "hidden" through the production process, but come out to bite at the last moment. Nasty.

[Greg Boston]

Originally Posted by Jad Meouchy

Technically speaking, a light switch is analog. Obviously, it's not meant to be used in that manner either. The analog argument works very easily for recording media such as Hi8 vs. Digital8; it is more difficult to make for an imager with such a discrete signal. But, going back to our light switch, is the signal really so discrete? Generally, for this application, it is (not so for the light switch as the bulb and high voltage does some of the debouncing for us).

Sorry Jad, but I have to disagree there. A light switch is a digital device. It has two states. It passes all the current, or none at all. Two states... 1 or 0, on or off, true or false, yes or no, etc.

A rheostat (dimmer) is an analog device which passes a variable amount of current. A stylus on a turntable is an analog device. They have continuously variable output.

CCD sensors are analog devices. A photo site on the ccd is really a photo transistor. Within limits, the output current from each transistor is proportional to the light striking it. IOW a continuously variable output.

True, the signal does not stay in analog form for very long as it hits the A/D converters, but it does start out that way. And as you rightfully pointed out, the number of bits used to convert that analog signal have an impact on image fidelity as in.... how many discrete numbers are utilized to represent a given voltage level from each photosite.

But as Chris points out, these are but one part of the chain and what comes out the tail pipe is what counts.

-gb-

[David Heath]

Originally Posted by Greg Boston

CCD sensors are analog devices.

Is a CCD analogue or digital? I tend to think of it in four dimensions - width, height, time and output level. Of those I'd say three are digital, and one analogue. Width and height are digital because the photosites are discrete (unlike a tube), time is digital because the chip gets sampled every frame or field, but the output is inherently analogue, only subsequently becoming digital via an A-D convertor.

Compared to a tube camera, it's the digital nature of width and height (discrete pixels) that makes it prone to aliasing.

[Chris Hurd]

Sorry David, CCD and CMOS image sensors are strictly analog.

In fact most of the signal processing chain is analog. It doesn't become digital until the A/D output.

Edit: Upon re-reading your post I can sort of see what you're getting at, but we really don't want to confuse people here. I've run into a lot of folks who have a hard time believing that these image sensors are not only analog, but monochrome as well.

[Matt Gottshalk]

Well, I couldn't put off the $2k rebate Panasonic was offering. The 500 ended up being $7999 after rebate...thats incredible. Also found a sweet HD Angenieux wide angle HD lens. The combo is amazing.

[Andrew McMillan]

I love you :')

[Andrew McMillan]

by the way was that lens over 10k?

[Matt Gottshalk]

Originally Posted by Andrew McMillan

by the way was that lens over 10k?

If I bought it new, it was, but I didn't buy it new. And quite honestly, I think it blows away any of the "kit" lenses.