 |
||||||||||
|
 |
|||
|
|
|
|
|
Written as an article for ZERB, the magazine of the Guild of Television Cameramen, the following explanation of various HD related jargon may be of use: |
|||
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||
|
Meanwhile…in a production meeting somewhere in the UK…
“You’ll need to shoot this one 4:4:2 , mostly at 24.98spF if you’re going to avoid burning on your highlights, and a 24 bit DA gives you way more latitude if you’re taking it to a film-out for US deliverables, and of course Discovery HD will only accept MPEG long LOG de-compression via a Cineon curve LUT that’s been tweaked for D-log E compatibility with a CCMCR Rec 707 matrix. I use a curve that Lucas’s DIT created for Return of the Noddii, gives me 18 stops of attitude if I use it on my HDVFX-9790, which I’ve fitted with a single Foveon TTL sensor which accepts the vintage Schneider vari-focal prime zoom lenses…they’re the only ones with the low enough mod depth to bring out the depth of field we needed for deepening the depth we needed in the field. Of course I’ll be shooting my next feature 8K scope uncompressed direct to floppy disc, using a new compression technique, invented by an eccentric billionaire from Barnsley who made a fortune in designer tripe fettling machinery. It converts images and dialogue to a text file that describes the scene, which is later ‘printed’ to a cellulose based wafer at 1 bit (black or white) and read directly into the brain. Haven’t come up with a name for this ‘Bit One Only Kinematograph’ yet, but if I come up with a good acronym I’ll blog it with my elderberry blah blah blah”
Sounds familiar? Makes sense? Well if it makes sense, then you’re in trouble. You may need the following short and hugely over-simplified guide to decoding the latest industry jargon:
PsF You’ll often find these letters used as part of the description of a frame rate; 25PsF for instance. The P is for progressive scan, and is probably familiar terminology. Sometimes though, you’ll find the sF suffix tacked on the end for no apparent reason. Segmented Frame is a way of breaking a progressively scanned image into two pieces in order to make it easier to move the information around within an engineering infrastructure that is mostly based around 50 fields per second of interlaced video. Oddly enough the two frame segments that your image is split into look remarkably like the fields of interlaced video. Don’t worry though…it doesn’t mean ‘it’s not proper progressive scan’. When your image pops out the other end of the system it’s stuck back together again exactly the way it was created, as a complete frame acquired at a single instant in time. Segmented frame is simply a way of splitting your frame into two pieces whilst it’s transported round the broadcast infrastructure. It’s glued back together just the way it was when it reaches the display.
4:4:4
The human eye does not perceive colour information at as high resolution as luminance information. The retina has more sensory perception of luminance information than chrominance. Many recording systems have taken advantage of this fact, by allocating less data to describing the colour information in a scene. The colour is slapped with a slightly broader brush, because it saves data, and we never notice. This process is called colour sub-sampling. All those 4:4:4, 4:2:2, 3:1:1, 4:2:0 numbers that you may have come across are describing the various ratios at which data is generated to describe the colour and luminance information in an image. 4:2:2 colour sub sampling has always been used in for instance Betacam recording systems. The numbers here are describing the weighting of the sampling in the Luminance, R-Y, and B-Y channels. By recoding the RGB from the camera into luminance and chrominance components, (Y, Pb, Pr) you can then allocate less data to describing the colour content of your images, and save 1/3 of the data. For every 12 bits in a 4:4:4 system, you record 8 bits in a 4:2:2 system. Sometimes the image is not being viewed by the human eye….a chroma-keyer for instance in a blue screen set-up. Here you could really do with the best resolution possible in the blue channel, in order to get a nice clean key. This is where 4:4:4 colour sampling could come in useful. In this case the numbers tell you that the red green and blue colour channels are all sampled equally. 4 samples of red, for 4 samples of green, for 4 sample of blue. You will need a camera that can generate a 4:4:4 output, and a recorder that can cope with the extra data, but if you’re looking for a really clean key on a commercial, or doing multiple layers of effects in a movie, it may be worth the extra cost. For a narrative drama it’s harder to see the benefit, though some grading systems will give a better result in 4;4:4. Why ‘4’ though? The 4 is a multiple of a sampling frequency lost in the mists of time that fits in with the PAL and NTSC line and colour sub-carrier frequencies. Just think of it purely as a ratio. The one that usually confuses people is 4:2:0 sampling, as used in DVCAM. It doesn’t mean there is no information recorded in one of the colour components, but alternate lines record R-Y or B-Y. (It actually goes 4:2:0, 4:0:2, 4:2:0 on alternating lines.) Vertical colour resolution is lost, rather than taking more resolution out horizontally. This is a sensible compromise for PAL TV, which has good vertical resolution. The NTSC flavour of DVCAM records 4:1:1, as they don’t have as much vertical resolution, being a 525 line system.
Dual Link
Stuff like 4:4:4 colour sampling (above) is all very well, but it comes at a price. That price is usually the inconvenience of dealing with the extra data generated. In terms of cables and connections, the HDSDI output of your camera goes up from about 1.5Gbps for the normally used 4:2:2 system to about 2.2Gbps. It’s already difficult getting long cable runs of normal HDSDI, but at this increased data rate you really have to use two wires. Still normal BNC cables (use good quality cable and proper 75 ohm terminations) but you need two of them. Future standards are likely to enable data to travel over a single cable at up to 3Gbps (SMPTE 372M). A typical application is using the 4:4:4 output from a Sony F23, Arri D20 , Panavision Genesis, or GVG Viper, and connecting via 2xBNC to the SRW-1 HDCAM SR recorder.
LUT A look up table is a means of modifying a signal with for instance a gamma curve. At its simplest it’s a list of multipliers for a range of input values. The process has been used for many years in digital cameras, and has been the reason why it’s a relatively simple process to match gamma curves on different cameras. You put the same numbers into the look up table on two cameras, and you get the same result.
Performed in the digital domain, a look up table gives completely consistent results, and can for instance allow you to look at images that have been encoded with a ‘log’ type gamma curve, on a monitor with a traditional ‘power law ‘ type gamma characteristic. Without a LUT, your image will typically have very low contrast, giving you plenty of scope for grading, but a very flat look to location monitors. Using an appropriate LUT can give you a better idea of the final look of the image. Look up tables are often provided in editing and grading software, and are starting to be incorporated in hardware.
DI Digital Intermediate. Image data is often stored, manipulated and graded in a digital format that contains the maximum possible amount of information captured from the camera. This format is unlikely to be suitable for transmission or programme interchange, because of the volume of data involved. Examples would be 10-bit log files from a 4K telecine, and a typical file format might be a DPX file. After the post production effects and grading has been completed, the DI would typically be mastered to a format such as HDCAM SR for delivery. Any format of video or film can be converted to a DI for further processing, but of course the final quality will always be limited by the originating format.
S-Log gamma curve
Strictly speaking, a gamma curve is built in to a camera to compensate for the non-linear response of CRT monitors. ‘Tweaking’ the curve, with knee circuits, alternative gamma curves, black stretch etc. has always been part of the toolkit for getting the most from a camera. Curves such as the S-log gamma available in the Sony F23 do not obey a traditional gamma law, but are based on a log law, and are becoming more commonly used. The idea of these log curves is to capture maximum dynamic range, but they rely on a post production grading process to extract the best possible look from the rather flat, low contrast images captured by the camera. The ‘S’ bit relates to the shape of the log curve used to specify film stock responses. You’ll typically need a viewfinder or monitor with a LUT capable of applying an anti-log curve, unless you’re happy to look at the low contrast un-corrected image from the camera, knowing that it will look different after post-production. The new BVM series LCD monitors from Sony will have an S-log gamma installed, as does the viewfinder feed on the F23 camera.
Pre –knee Before the output of a CCD is converted to digits, it is often put through a pre-knee process that ensures that the sampling levels available in the analogue to digital converters are spread most effectively over the useful dynamic range of the CCD. With the increased availability of 14 bit A/D converters pre-knee may not be required.
Knee Knee compression is applied within the digital signal processing of a camera, and is used to compress the dynamic range captured by the A/D converters into the standard signal amplitude required for recording, display, and interchange. It’s often regarded as a user control, and you can set the onset point of knee compression, and the slope or attack of the knee in the paint menus. Auto knee (DCC) is also usually available, which moves the knee point automatically according to picture content. (Be careful if you’ve got a flashing lamp in shot…your knee point will bounce up and down as the light goes on and off if using auto knee.) Auto knee was often regarded as ineffective in early cameras, but has come a long way since then, and will usually help with handling burnt out highlights.
Latitude Latitude is a term used mainly within the film world to describe the range of exposure over which a film stock will give a perceptible result. The minimum satisfactoryexposure is one in which good tone separation is justattained in the deepest shadow areas. The maximumsatisfactory exposure is one in which detail is justretained in the brightest highlight. The difference between the two is the latitude. It’s a bit subjective, as it depends on ‘just noticeable difference’, and allows a non linear relationship between input and output….you may need to put a lot of light in to get that last noticeable difference out. Because of this, you may get a different Latitude measurement compared to Dynamic Range.
Dynamic range Dynamic range is a similar measurement to latitude. In a video camera, it is typically determined by the difference between the noise floor of the camera, and the maximum possible input for which the output is directly proportional to the input. It’s a little different to latitude, in that it’s only valid over the linear part of the curve (if you can have a linear curve!). Usually measured in ‘stops’, where a stop is equivalent to 6dB, or a doubling of amplitude. Somewhere between 10 and 13 stops is typical for an electronic camera.
2K
2K resolution can be thought of as HD, though the pixel count for the 2K digital cinema standard is 2048 x 1080.
4K
Still acquiring images at 2K? How quaint! 4K is the new 2K, with many film scanners being able to capture at 4K resolution, and 4K being a fairly common process for digital intermediate work. Sony are now shipping the SXRD 4K projectors, so in theory all the pieces are in place. 4K cameras are still a bit of a sticky area though. Most 4K cameras actually have a sensor with 4K horizontal pixels…but there’s only one sensor, so it’s a Bayer pattern output, and will not of course have the same resolution as a traditional 3 sensor camera.
Meanwhile…in NHKs back room in Japan…ultra high def is taking shape. It has already been demonstrated in Japan, though it’s relevance to us mere mortals in the broadcast industry is hard to imagine. The sound system that goes with it is no longer anything as primitive as 7.1. Anyone for 22.2 location recording?
Bayer Pattern Filtering Single chip cameras have long been regarded as the poor relations of the family, with 3 CCD cameras used for most broadcast applications. If, however, you can create a single chip sensor with enough pixels, you can reasonably expect to be able to use on chip colour filtering rather than a prism block and dichroic filters to create a good quality image. The simplest way of creating a colour image from a single sensor is to use an RGB vertical stripe filter. Every third pixel responds to either R,G or B. There are more efficient ways to derive an image: the Bayer filter in the diagram below relies on the fact that most of the luminance information is held in the green channel, and so can give you more resolution for the same pixel count once decoded. It’s a kind of colour sub-sampling process in a way. None of this is new. Sony has used an even more complex filter based on the complementary colours and luminance sampling in single CCD cameras for many years. Single chip security cameras generally use the CMYW (cyan, magenta, yellow, white) filter, as it’s sensitivity is particularly good. What is new is being able to manufacture sensors with enough sensing area, and enough pixels, sufficiently close together, to give performance suitable for high end cameras. Don’t forget that a single chip camera will need a lot more pixels on its sensor to match a 3 chip camera…perhaps not 3x as many, because of clever tricks like the Bayer pattern, but not far off.
SMPTE Fibre
HD cameras generate a great deal of data. As mentioned previously, the HDSDI output of a camera will be 1.485Gbps, and will not run far on copper cables, whether co-ax or tri-ax. This is of course a serious problem for outside broadcasts, where cable runs round a venue may be several kilometres. You have two choices…use bandwidth limited analogue transmission down a traditional triax cable, or send the HDSDI signal down a fibre optic cable. Fibre optic transmission has long been used in the IT and military worlds, but a different type of cable was required for broadcast HD. SMPTE 311M is the specification for a composite cable containing two single mode fibre optic cores for HD video, copper cores for power transmission and signalling, and a stainless steel strength core. Cable runs are more likely to be limited by power requirements than data bandwidth. You do need to keep your connectors clean though, and you can’t fix a fibre cable with a soldering iron.
I hope this helps decode the jargon flying around in your next production meeting. If not, just nod thoughtfully, and explain that you just need the technology to push back the barriers to your creativity, and isn’t it time for lunch…
|
A commonly used abbreviation for a resolution of about 2000 horizontal pixels. I say ‘about’ because it was originally used to describe the scanning of film through a telecine machine, which depends on the gate used, and the type of Telecine, and probably ends up as 1920 at best, which is of course the horizontal pixel count of HD.
|
|
|
[Home] [Contacts] [Training] [Background] [Links] [HDC Systems Cameras] [Time for some filing] [HD Production Guide] [Decoding the Jargon] |