HVX ccd size and pixel count, what is it?
- Thread starter mthornton
- Start date
Jan_Crittenden
Veteran
Hi,
All pixel shift does is cover the non-photosensitive areas, the registers, and give more resolution. Since the camera is working in the analog domain at that point it would not have any ill effect, as once it is gated or delayed back into the corect time, then it is turned into the digital signal. Be that 4:4:4 or 4:2:2. At least that is the way I understand it.
Best,
Jan
All pixel shift does is cover the non-photosensitive areas, the registers, and give more resolution. Since the camera is working in the analog domain at that point it would not have any ill effect, as once it is gated or delayed back into the corect time, then it is turned into the digital signal. Be that 4:4:4 or 4:2:2. At least that is the way I understand it.
Best,
Jan
Hm. Bizarre all the pixel-shift pushing in this thread. There's a lot of information here that doesn't strike me as correct. I for one would want to know the resolution of the CCDs.
Pixel-shift allows for bigger sensors - therefore better low light sensitivity and lattitude. This only makes sense if there are areas on CCDs where there is no light sensitivity due to other structures required for the CCD to operate, and the size of these "operational" areas doesn't scale with pixel density. The above sentence DOES make sense, so that argument is all good.
The "resolution gains" from pixel shift are conditional - not absolute. The apparant resolution from a pixel shifted CCD varies with what you shoot - and some objects will image better with this technique than others.
I don't understand what Jan means in the post above about an "analog domain". My understanding is that a CCD is essentially a rectangular grid of picture elements. Each pixel is independent of the others, and in this way, are discrete. Reading from any one NxM CCD ought to produce an NxM matrix. The "digital domain" enters when the accumulated charge on each pixel element is tossed through an Analog to Digital converter at some bit-depth (incidentally, accurate and high-bit depth ADCs are very expensive). With a 3CCD system, you get 3 such NxM matrices to combine - it isn't some "all analog" pattern.
Pixel shifting would then occur as you combine the 3 matrices (digitally most likely). You could pretend at this point that the full-colour matrix is 4:4:4, but if 1 or 2 dimensional pixel shifting is used, the actual colour depth will be 4:2:2 or 4:1:1 respectively.
I think pixel-shifting is a valid engineering tool, and a good use of CCD's... but if it could be avoided altogether, the resulting images would be superior.
-Spiff
Pixel-shift allows for bigger sensors - therefore better low light sensitivity and lattitude. This only makes sense if there are areas on CCDs where there is no light sensitivity due to other structures required for the CCD to operate, and the size of these "operational" areas doesn't scale with pixel density. The above sentence DOES make sense, so that argument is all good.
The "resolution gains" from pixel shift are conditional - not absolute. The apparant resolution from a pixel shifted CCD varies with what you shoot - and some objects will image better with this technique than others.
I don't understand what Jan means in the post above about an "analog domain". My understanding is that a CCD is essentially a rectangular grid of picture elements. Each pixel is independent of the others, and in this way, are discrete. Reading from any one NxM CCD ought to produce an NxM matrix. The "digital domain" enters when the accumulated charge on each pixel element is tossed through an Analog to Digital converter at some bit-depth (incidentally, accurate and high-bit depth ADCs are very expensive). With a 3CCD system, you get 3 such NxM matrices to combine - it isn't some "all analog" pattern.
Pixel shifting would then occur as you combine the 3 matrices (digitally most likely). You could pretend at this point that the full-colour matrix is 4:4:4, but if 1 or 2 dimensional pixel shifting is used, the actual colour depth will be 4:2:2 or 4:1:1 respectively.
I think pixel-shifting is a valid engineering tool, and a good use of CCD's... but if it could be avoided altogether, the resulting images would be superior.
-Spiff
Jan_Crittenden
Veteran
Spiff said:
Well I am not sure about which information you are referring to as incorrect, but the fact is that the pixel shift is using the entire ouput from the three chips. The red and blue are offset by half a pixel. And then time delayed to bring them back into time. This covers the non-photosensitive parts of the cuips, the spaces between the pixels. In our DVC30, and we call it a spatial offset, we do a horizontal and a vertical offset. The benefit is that the 1/4" chip set of the DVC30 has resolution like its big brother the 1/3" DVX100A.
This is where you may not understand how this system works as it always works to the benefit of the camera. Without pixel shift/spatial offset, you would have vertical stripes in the pictures, which once the light became low you would see. This was very evident in the early days of chip cameras, but today since it is a methodology that all manufacturers use. Canon was the one that brought the discussion into the open as they used our 270,000 pixel imager against the PD-100 which had 380,000. The Canon actually was sharper. Why? Pixel offset, but since many customers just can get the idea that the cip number count is not the deciding factor, Canon made an marketing issue out of it.
The reality is from the WV-F200/300 from Panasonic, which circa 1987 we have been doing this and from around there on, everyone else did as well, we just didn't talk about it.
Here is where you really misunderstand camera technology. There is not a pixel for pixel match on the chip to the recording. The Chip is an analog device. The signal generated is then sampled for the digital domain. The Sampling Frequency, will have a lot to do with resolution. Imagine if I sampled a signal only 12 times a second versus 7200 times. Which is going to give me more information? Of course, the 7200. From there it is quantized, what is the bit depth, 10 bit, 12 bit or 14 bit? Which is better, 14 bit. The chips are analog devices.
It is analog up and to the Black shading and White chading domain. Then it makes the transition.
I
Without it we would have stipes in the video.
Best,
Jan
Hi, I forgot my password etc., and had to create a new account.
First off, here's a decent breakdown of how CCD's work. It may not be entirely up to date, but I assume it's about right.
http://www.reed-electronics.com/tmworld/article/CA187479.html
No disagreement that pixel-shift uses the output from the three chips, and that the blue and red pixels are offset. The time delay you refer to is just the read-out and shift-register business - which, while required to make a chip work - doesn't affect my argument.
I don't think you're talking about "1/2-pixel shifting" the resolution enhancing tool here, but rather simply placing the 3 sensors so that there is no part of the imaging plane that is ignored. Since there are only finite active areas on the surface of the CCD, very minor repositioning of each CCD such that there aren't gaps. I imagine to do this on a one chip design, some low-pass filtering would be used to smear across non-active areas?
I would have thought that sampling frequency is simply the frequency required to sample each pixel independently in the serial stream of data coming off the CCD. Is this not the case? For example, let's say I have a 960x1080 CCD, that I am reading 24 times a second. That's 1036800x24 pixels per second, or about 24 MHz. The charge is then read by say a 14-bit counter and is then given a value between 1 and 16384.
The only a way a CCD would be an analog device as you suggest is if you can't maintain the integrity of the charge from an individual pixel. In this case, rather than get a discrete set of charges to sample, you get a continuous series of charges to sample... and in sampling you simply choose a time to sample for each "image pixel". This would effectively smear the image on the CCD... making it moreso an analog device than it actually is. The benefits to doing it this way are not at all obvious to me.
Now... when you decide to place the ADC is up for grabs. High and low pass filtering, amplification and a variety of other operations are easily done in either the analog or the digital domain. It may be beneficial to do some of them (most certainly amplification) before sampling. My first choice would be to do most of the filtering digitally, following amplification and high-bit-depth ADCs. After converting to digital, do whatever colour correction and filtering you want at 14-bit, and then downsample the resulting image to the 4:2:2 8-bit codec.
I have very little doubt in my mind that you're not talking about 1/2-pixel shifting here, but much smaller shifts. I can certainly see how this technology would be beneficial, but it wouldn't change the image resolution. I'm not saying "don't use small pixel off-sets to get rid of discrete pixel artifacts" - I'm saying: 1/2-pixel shifting produces conditional gains in resolution and colour space, while using native resolution CCDs does not. 1/2-pixel shifting is better than "no" pixel-shifting, when the resolution of the CCDs are significantly less than the output format of the camera.
-Spiff
First off, here's a decent breakdown of how CCD's work. It may not be entirely up to date, but I assume it's about right.
http://www.reed-electronics.com/tmworld/article/CA187479.html
No disagreement that pixel-shift uses the output from the three chips, and that the blue and red pixels are offset. The time delay you refer to is just the read-out and shift-register business - which, while required to make a chip work - doesn't affect my argument.
I don't think you're talking about "1/2-pixel shifting" the resolution enhancing tool here, but rather simply placing the 3 sensors so that there is no part of the imaging plane that is ignored. Since there are only finite active areas on the surface of the CCD, very minor repositioning of each CCD such that there aren't gaps. I imagine to do this on a one chip design, some low-pass filtering would be used to smear across non-active areas?
I would have thought that sampling frequency is simply the frequency required to sample each pixel independently in the serial stream of data coming off the CCD. Is this not the case? For example, let's say I have a 960x1080 CCD, that I am reading 24 times a second. That's 1036800x24 pixels per second, or about 24 MHz. The charge is then read by say a 14-bit counter and is then given a value between 1 and 16384.
The only a way a CCD would be an analog device as you suggest is if you can't maintain the integrity of the charge from an individual pixel. In this case, rather than get a discrete set of charges to sample, you get a continuous series of charges to sample... and in sampling you simply choose a time to sample for each "image pixel". This would effectively smear the image on the CCD... making it moreso an analog device than it actually is. The benefits to doing it this way are not at all obvious to me.
Now... when you decide to place the ADC is up for grabs. High and low pass filtering, amplification and a variety of other operations are easily done in either the analog or the digital domain. It may be beneficial to do some of them (most certainly amplification) before sampling. My first choice would be to do most of the filtering digitally, following amplification and high-bit-depth ADCs. After converting to digital, do whatever colour correction and filtering you want at 14-bit, and then downsample the resulting image to the 4:2:2 8-bit codec.
I have very little doubt in my mind that you're not talking about 1/2-pixel shifting here, but much smaller shifts. I can certainly see how this technology would be beneficial, but it wouldn't change the image resolution. I'm not saying "don't use small pixel off-sets to get rid of discrete pixel artifacts" - I'm saying: 1/2-pixel shifting produces conditional gains in resolution and colour space, while using native resolution CCDs does not. 1/2-pixel shifting is better than "no" pixel-shifting, when the resolution of the CCDs are significantly less than the output format of the camera.
-Spiff
Clint Johnson
Active member
Panasonic says 1/3" but hasn't said anything yet on the resolution but my money is on 1280x1080... or at least that is what I am hoping for.
fiercecurry
Well-known member
Jan......any info about these specs regarding the resolution? Or are we just going to have to wait?
Loki
Veteran
I am interested in why it is so challenging to create a lens that "resolves" 1920x1080 lines of resolution..
is it because of the chip size? how does this actually work? I just thought a lens focused the light onto the film plane (or onto the CCD or into the prism in our case)...
I wasn't aware lens could "resolve" anything... so would a 35mm prime lens resolve something equivalent to a 4k resolution?
Obviously there are differences between film and digital.. but I assumed the optics were rather similar... just a finer piece of glass for digital since we are working with much smaller sensors..
I also would like to know the exact pixel count of the CCD for the HVX... so I can determine whether it is a good chip for low light use..
The thing i want the most, is more latitude from the camera... Since it is working in HD, I am hoping that the DVCPRO-HD codec doesn't clip the **** out of the image like the DV Codec does...
I don't care that much about actual resolution... latitude is far more important to me..
well enough of my ranting..
oh ya... and pixel shift seems like quite an interesting technology.. I am wondering if Canon could upgrade the firmware of the XL2 to give a HD Image..? would be interesting no...
is it because of the chip size? how does this actually work? I just thought a lens focused the light onto the film plane (or onto the CCD or into the prism in our case)...
I wasn't aware lens could "resolve" anything... so would a 35mm prime lens resolve something equivalent to a 4k resolution?
Obviously there are differences between film and digital.. but I assumed the optics were rather similar... just a finer piece of glass for digital since we are working with much smaller sensors..
I also would like to know the exact pixel count of the CCD for the HVX... so I can determine whether it is a good chip for low light use..
The thing i want the most, is more latitude from the camera... Since it is working in HD, I am hoping that the DVCPRO-HD codec doesn't clip the **** out of the image like the DV Codec does...
I don't care that much about actual resolution... latitude is far more important to me..
well enough of my ranting..
oh ya... and pixel shift seems like quite an interesting technology.. I am wondering if Canon could upgrade the firmware of the XL2 to give a HD Image..? would be interesting no...
Barry_Green
Moderator
It's not necessarily difficult to design that lens. It just depends on how big the target sensor is. The larger the sensor, the easier to design a lens that can handle it.Loki said:
Yes. With standard-def video, 480 lines tall on a 2/3" chip, you're talking about a total resolution of 36 line-pairs per millimeter. The chip is 6.6 millimeters tall, and so you get about 72 lines per millimeter of resolution, or 36 pairs of lines (as in, one black line and one white line = one pair of lines, or one line-pair), which is expressed as 36 line-pairs per millimeter.
So a standard-def lens usually resolves somewhere around 35 to 40 line-pairs per millimeter. And a standard-def lens, for a 2/3" camcorder, might cost around $8,000.
A high-def chipset packs well over twice as many lines per millimeter -- 1080 lines of resolution. That, plus they're using 16:9 chipsets which mean even tighter packing on the vertical, means that you need to resolve around 100 line-pairs per millimeter! The chip is 5.4 millimeters high, and you need to cram in over twice as many lines (1080 of them), so the chip is very, very tightly packed. 100 line-pairs per millimeter is three times as much precision as that which is required for standard-def! That's why you'll find that high-def lenses start at around $25,000, and may cost around $60,000.
Okay, now make it much harder: use a tinier chip. Just as many pixels, but half as many millimeters of physical size, right? 1080 lines in about 2.95mm of space... yes, we're talking about needing the ability to resolve nearly twice as much resolution as a regular high-def lens! We need the ability to see about 183 line-pairs per millimeter, for the small 1/3" chips.
And, mind you, that's in each direction. That means the lens needs to be able to resolve four times as much detail, just because the chip is smaller. Pack the same amount of data into 1/4 the space, and you'll need 4x the resolving power.
Very tough task. Now, on top of that, we have to do so with an extremely limited budget. Can't charge $60,000 for this lens when the whole camera+lens costs maybe 1/10 that! So JVC is valuing their lens at about $800. Can someone actually make a lens that resolves 4x as much, for 1/30th the price?
It will be interesting to see.
Depends on how big the sensor is. If you're talking about a massive 36mm x 24mm sensor, then you'd only need to be able to resolve 65 line-pairs per millimeter in order to get 4000x3000 resolution.
But if you put that same lens in front of a 1/3" chip, the most it could deliver would be about 600 x 400. 65 line-pairs per millimeter don't go very far when you only have about 4mm x 3mm to work with!
Yes, exactly. But when you say "finer", you're just beginning to scratch the surface. It has to be exceptionally unbelievably way way way sharper, in order to extract the detail being asked of a 1/3" high-def chip.
But... wouldn't a more valid test just be to use it in low light, and see what the footage looks like? That'll answer the question far more accurately than knowing what the pixel count is.
To renew my analogy of the cars -- knowing how many horsepower a vehicle has doesn't tell you how fast it is. A motorcycle might have 80 horsepower, and a Hummer might have 325. But when it comes time to race... well... it's no contest, not by a hundred bike-lengths.
Jan has said that they are working to maximize the latitude, sensitivity, and resolution. They'll come up with the best combination that delivers the most you can get.
Then you would be best served by looking at bigger-sensor cameras. The bigger the chip, and the lower the resolution on that chip, the better the latitude will be. I haven't seen footage from an HVX yet, but I can almost guarantee that a SDX900 (standard-def, 2/3" chip) will have much better latitude than any high-def 1/3" chip camera. It's just laws of physics, at that point.
I don't think that'd be possible, they'd have to design the chip in the first place to take advantage of the potential possible by pixel-shift. I don't think it's something that could be done with firmware.
redindian
Well-known member
Barry_Green said:
Isnt it the other way round ?
The smaller the sensor, the easier and cheaper to design a lens around that.
For instance, Nikon introduced the DX line of lenses designed exclusively for smaller sensor size so that they can offer complex lens designs(like extreme WA) at a cheap price.... like 12-24 DX, 10.5 DX, 18-70DX
Sigma has DC, Tamron has DT...etc
ram
thisiswells
Retired
No, Barry's right. Nikon will be in serious trouble above 14MegaPixels.
Those cameras and lenses have to be a short term solution.
http://www.bobatkins.com/photography/digital/size_matters.html
Those cameras and lenses have to be a short term solution.
http://www.bobatkins.com/photography/digital/size_matters.html
redindian
Well-known member
obviously yes - bigger sensor is always better...
don't steal my thunder - not everyone gets to correct Barry!
i said designing lens for smaller sensor is easier/cheaper... not harder..
From the link you sent:
Yes DX lens are stop gap arrangement till Nikon goes full frame
don't steal my thunder - not everyone gets to correct Barry!
i said designing lens for smaller sensor is easier/cheaper... not harder..
From the link you sent:
Yes DX lens are stop gap arrangement till Nikon goes full frame
Barry_Green
Moderator
Well, it goes both ways, really. Smaller sensors need a smaller circle of illumination, which means they can be made with smaller glass, which means that it can be cheaper. But the resolution demands are much higher -- can you get the resolution? That's the hard part.
Larger sensors need larger circles of illumination to cover them, which means bigger glass and bigger elements, etc. But, they don't have to be nearly so sharp in order to still provide 100% resolution on their targets.
Bigger sensors don't require the same level of super-fine precision resolution that smaller sensors do.
So I shouldn't have said that it's easier to design a lens for a larger sensor; that's probably not accurate. What I meant to say was, it seems like optical engineers would have an easier time making a 60lp/mm lens than they will designing a 200lp/mm lens. I don't know what the limit of lens/resolution are, but even Zeiss DigiPrimes, some of the best glass on the planet, only resolve around 100lp/mm, don't they?
Larger sensors need larger circles of illumination to cover them, which means bigger glass and bigger elements, etc. But, they don't have to be nearly so sharp in order to still provide 100% resolution on their targets.
Bigger sensors don't require the same level of super-fine precision resolution that smaller sensors do.
So I shouldn't have said that it's easier to design a lens for a larger sensor; that's probably not accurate. What I meant to say was, it seems like optical engineers would have an easier time making a 60lp/mm lens than they will designing a 200lp/mm lens. I don't know what the limit of lens/resolution are, but even Zeiss DigiPrimes, some of the best glass on the planet, only resolve around 100lp/mm, don't they?
The theoretical limit of lens resolution for conventional optics is governed by the wavelength of the light being imaged. Taking 500 nm as the "center" of the visible spectrum, you're looking at 0.5x0.5 microns. In a simplistic "optimal" case, you need 5 pieces of information to resolve a line pair, so the best you could ever do with light would be about 400 lp/mm. Non-linearities, manufacturing and aberration get in the way of this... as do my simplistic resolution requirement. I wouldn't be at all surprised if resolving much more than 100 lp/mm is completely implausible. While it's possible to build higher-resolution sensors - light won't co-operate.
thisiswells
Retired
Spiff--
For the discussion of lp/mm required per electronic sensors, how would you figure film emulsions into the equation? Is film "lens limited" as well? As a practical matter, what is the highest resolution scan possible in optimal conditions from 35mm motion picture film?
For the discussion of lp/mm required per electronic sensors, how would you figure film emulsions into the equation? Is film "lens limited" as well? As a practical matter, what is the highest resolution scan possible in optimal conditions from 35mm motion picture film?
Zig_Zigman
Well-known member
Slighty off topic, is the Varicam 2/3'' chip size? It's actually getting a bit long in the tooth, I imagine Panny is working on a Varicam successor, something with a full 35mm sensor.
thisiswells
Retired
That would be highly unlikely because no existing video lenses would function with a chip that big.
You would have to use cinema lenses with a chip that size; and it would need to be a single chip
camera to work with cine lenses... I don't think they're going that direction, perssonally.
You would have to use cinema lenses with a chip that size; and it would need to be a single chip
camera to work with cine lenses... I don't think they're going that direction, perssonally.
thisiswells said:
B&W or colour film?
Still camera lenses for "chemical film" are designed to offer a resolution of
100 lp/mm (average) up to 120 lp/mm (top quality).
Therefore, yes, film is "lens limited" as well, apart from all the other factors
(film movement during exposure, to name but one).
Graeme_Nattress
Red Team
It doesn't matter what the sensor is - bit it 1 chip, 3 chips, pixel shfited chips, CMOS or film - they can all be limited by the resolving power of the lens. The factor is the density of information. That's why you can get a higher resolution off a medium format camera, and that's why a lens for a HD 2/3" camera needs to be sharper than that of a 35mm camera, and why a HD 1/3" camera needs a sharper lens still!
Graeme
Graeme