Ian’s primer could help when buying a digital camera – Ian Anderson
I originally wrote this for astro-photographers but I felt it may also be of interest to Overclockers.com readers. This is something to think about next time you are in the market for a digital camera.
The basic functioning of a digital camera is relatively simple; the information is easy to find on the internet but it is so heavily loaded with technical jargon you would need to know how they work before you could understand the description. Without further ado I present my (patronizing) oversimplification.
Imagine a solar panel attached to a battery. If the panel is exposed to light for a period of time, it will charge the battery. By measuring the charge on the battery, I can tell how much light fell on the panel. If I know how efficient the panel is, then I can tell you exactly how many photons fell on it. If I arrange a few thousand of these in a grid, I can then tell how much light fell on each part of the grid.
Given that I don’t have that much time, I can just as easily have a computer do it for me. For the sake of convenience, let’s say I shrink down the grid into something I can fit in the palm of my hand. Now I can use a lens to project an image onto the grid. The amount of energy each panel produces depends on the brightness of the image at that given point.
Getting this information out of the array in a timely fashion can be somewhat challenging. There is simply no logistical way to have one lead (connection or wire) for each pixel. Instead, each pixel in a given array is read out one at a time through the serial output pin.
Think of it like elementary students filing out of an auditorium exiting row by row. The first row stands up and the first student leaves the auditorium then each student advances one seat. The next student then exits and the process repeats until the row is empty. Then the next row stands and exits in the same manner. The charge gathered by each pixel must be stored somewhere before it is read; hence, they all must advance by one pixel after one is read.
Knowing the amount of energy in each pixel is useless unless you can convert this into luminance levels. To do this you need to know the quantum efficiency of each pixel.
Quantum efficiency (QE) is the efficiency of not just each individual pixel (or panel in the solar panel analogy) but the entire grid including the space in between and the wires that connect them. If each solar panel is 12″ square with a 4″ space and 1″ of wiring, then the grid can have at most 64% efficiency; but if each panel is 6″ square, then the grid can have at most 44% efficiency.
If the individual panels are 80% efficient, then the final efficiency is the maximum efficiency times 0.80. For maximum QE (sometimes as high as 99%), some sensors are thinned and back illuminated. What this means is the top of the sensor is glued to a substrate, then the silicon that once was the base is etched off. This places the connections between the pixels on the underside of the sensor, leaving nothing but active pixel surface on the back of the chip. This is rarely done because it is very expensive. Expect a blank look from the clerk if you ask for this in a camera store.
The maximum brightness for a given pixel is determined by the size of its electron well.
In the solar panel analogy, the battery is the electron well. The more electrons a well can hold, the greater difference between the lightest part and the dimmest part in a given photo can be. The size of the electron well is determined by the size of the pixel. A 9 micron pixel might have around 60,000 electron capacity, whereas a 24 micron pixel will have a capacity closer to 500,000.
Dark current is the amount of electrical noise each pixel naturally gains during an exposure, even if it receives no light. This number is based on temperature and length of exposure. It has nothing to do with the size of the electron well or QE of the pixel.
What this means is dark current will have much more effect on a small pixel than a large one. To increase the size of the electron well, some cameras have a function known as binning. What this does remove the electrostatic wall between two or more pixels to effectively combine the two electron wells into one. The down side of this is the two pixels become one, reducing the effective number of pixels and resolution.
When the electron well fills, it will begin to spill over into neighboring pixels. This effect is called blooming. An anti-blooming gate is a circuit that will drain the electron well before it begins to overflow. There are two drawbacks to this:
- First, it reduces the capacity of the electron well. Think of it like poking a drain in the side of a bucket to prevent water from spilling out the top.
- The second drawback is the area they take up on the surface of the sensor further increasing the space between each pixel reducing QE.
The bottom line for cameras, though, is image quality. For digital imagers it comes down to contrast and dynamic range.
These are two very much different, yet equally important, factors. The ADC (analogue to digital converter) determines the maximum level of contrast. In the solar panel analogy, this would be the processor that reads the amount of electricity in the battery and then converts it into a digital signal. With a 16-bit ADC, the charge in the electron well will be converted into a value between 1 and 65535. Dynamic range on the other hand is limited by the size of the electron well. The larger the well, then the greater the difference between the brightest object and the dimmest object in a given scene can be.
To increase the number of pixels in a camera, it is necessary to either increase the physical area of the sensor or to shrink the pixels. The cost of fabricating a sensor is based on the amount of silicon used – more pixels, more money. Reducing the size of the pixels as is most often the case for consumer cameras has its own costs.
The dynamic range will be drastically reduced because the size of the electron well is limited by the area of the pixel. This will in turn make the dark current more pronounced, resulting in a grainy image. In most cases the image will begin to suffer below 6 microns.
If you want professional images, you should look for a sensor with at least 9 micron pixels.
A number of individuals have expressed concerns with the resolution of cameras with 9 micron pixels and how they stack up against film. The short answer is:
You will never match the resolution of Ansel Adams or Alfred Stiglitz using a digital camera.
A good professional quality film will have a grain size comparable to a wavelength of light. Even the lowly daguerreotype of 1839 had a sub-micron resolution. This means film is not limited by its intrinsic resolution but by the lens itself.
The smallest possible point a lens can produce is given by an extremely simple formula:
1.35 x (your lens’ F-ratio) gives the spot size in micrometers [wikipedia.org].
If you have read some of my other articles (particularly An Explanation of the Impact of Optical Restrictions on Semiconductor Yield Rates), you know this is grossly simplified but it is a good rule of thumb for photography in visible light at reasonable F-ratios.
In short the larger the aperture the smaller the spot size and the longer the focal length then the more this spot is magnified on the image surface. Because these two parameters are linear, the spot size is given not by absolute value but by their ratio.
Another rule of thumb you should remember from the same article is if you halve the F ratio of a given lens design, you will square the price of the lens to maintain quality.
Whether you are using film or a digital camera, if you are using an F8 lens you are limited to a resolution of 10.8 microns. I chose F8 because it is difficult to find a lens with a shorter F ratio that can achieve its theoretical maximum. Most lenses on consumer cameras (often up through prosumer SLR) are limited to the resolution they see at F8 due to errors in the lens design and polish. In fact, most of these cameras do better if they are stopped down a little.
Knowing the resolution of an F8 lens (we will round it down to 10 microns to make the math easier), we can see the equivalent resolution of film:
- 135 format (35mm) is 24mm x 35mm, giving 2400 x 3500 pixels (8.4 MP)
- 120/220/ is what most professionals use with a 56mm x 90mm frame, or 5600 x 9000 pixels (50MP)
- 4″ by 5″ sheet film mostly used for landscapes and some fashion photography gives roughly 10,000 x 12,500 pixels (125MP)
- 8″ x 10″ (which is what Ansel Adams used) would be 20,000 x 25,000 pixels (500MP)
And just for fun, the largest sheet film currently available is the Polaroid 20″ X 24″ used for recording large works of art (murals, tapestries etc.), which gives about 500,000 X 600,000 (3GP). If you want larger negatives than this, you will need to coat a glass plate yourself.
Bottom line: If your only concern is resolution, you should not be looking at a digital camera.
The only cameras I know of that come close to medium format resolution with a good pixel size are made by Phase One and Hassleblad. From what I understand, Fujifilm also makes one in this size but I have not seen one in person. You should be more interested in a 120/220 format camera or better yet, an 8″ by 10″ sheet camera.
There are three reasons to choose a digital camera over a film camera:
- First, and the reason most of us choose to make the transition, is to avoid the per exposure cost associated with film and the limited number of exposures on a role;
- Second is to avoid the dark room altogether in order to shorten the time between exposure and publication, and
- Third, because the length of an exposure on film would be prohibitively long to capture the event you are interested in, such as in sports or scientific research.
Ed Note: Having spent some time in the “film” world, I have been mightily impressed with digital cameras for many of the reasons Ian listed; however, I have enlarged prints to 16″ x 20″ from ¼ of a 35mm negative (B&W, fine grain) to presentation quality, which is impossible with digital “prosumer” cameras.